[{"id":"doi:10.3390/mi16121422","type":"article-journal","title":"An Artificial Synaptic Device Based on InSe/Charge Trapping Layer/h-BN Heterojunction with Controllable Charge Trapping via Oxygen Plasma Treatment.","abstract":"Neuromorphic computing, an emerging computational paradigm, aims to overcome the bottlenecks of the traditional von Neumann architecture. Two-dimensional materials serve as ideal platforms for constructing artificial synaptic devices, yet existing devices based on these materials face challenges such as insufficient stability. Indium selenide (InSe), a two-dimensional semiconductor with unique properties, demonstrates significant potential in the field of neuromorphic devices, though its application research remains in the initial stage. This study presents an artificial synaptic device based on the InSe/Charge Trapping Layer (CTL)/h-BN heterojunction. By applying oxygen plasma treatment to h-BN to form a controllable charge-trapping layer, efficient regulation of carriers in the InSe channel is achieved. The device successfully emulates fundamental synaptic behaviors including paired-pulse facilitation and long-term potentiation/inhibition, exhibiting excellent reproducibility and stability. Through investigating the influence of electrical pulse parameters on synaptic weights, a structure-activity relationship between device performance and structural parameters is established. Experimental results show that the device features outstanding linearity and symmetry, realizing the simulation of key synaptic behaviors such as dynamic conversion between short-term and long-term plasticity. It possesses a high dynamic range ratio of 7.12 and robust multi-level conductance tuning capability, with stability verified through 64 pulse cycle tests. This research provides experimental evidence for understanding interfacial charge storage mechanisms, paves the way for developing high-performance neuromorphic computing devices, and holds broad application prospects in brain-inspired computing and artificial intelligence hardware.","author":[{"family":"Wang","given":"Qinghui"},{"family":"Wang","given":"Jiayong"},{"family":"Lu","given":"Manjun"},{"family":"Ma","given":"Tieying"},{"family":"Li","given":"Jia"}],"issued":{"date-parts":[[2025]]},"DOI":"10.3390/mi16121422","URL":"https://doi.org/10.3390/mi16121422","source":"europepmc"},{"id":"doi:10.1126/sciadv.aea1114","type":"article-journal","title":"Parallel nonlinear neuromorphic computing with temporal encoding.","abstract":"The proliferation of deep learning applications has intensified the demand for electronic hardware with low energy consumption and fast computing speed. Neuromorphic photonics have emerged as a viable alternative to process high-throughput information at the physical space. However, the simultaneous attainment of high linear and nonlinear expressivity poses a considerable challenge due to the power efficiency and impaired manipulability in conventional nonlinear materials and optoelectronic conversion. Here, we introduce a parallel nonlinear neuromorphic processor that enables arbitrary superposition of information states in multidimensional channels, only by leveraging the temporal encoding of spatiotemporal metasurfaces. We experimentally demonstrated the concept based on distributed spatiotemporal metasurfaces, showcasing robust performance in multilabel recognition and multitask parallelism with asynchronous modulation. Our nonlinear processor demonstrates dynamic memory capability in real-time responsiveness to canonical maze-solving problem. Our work opens up a flexible avenue for a variety of temporally modulated neuromorphic processors tailored for complex scenarios.","author":[{"family":"You","given":"Guangfeng"},{"family":"Qian","given":"Chao"},{"family":"Wu","given":"Ouling"},{"family":"Chen","given":"Hongsheng"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1126/sciadv.aea1114","URL":"https://doi.org/10.1126/sciadv.aea1114","source":"europepmc"},{"id":"doi:10.48550/arxiv.2602.23274","type":"manuscript","title":"Exploiting network topology in brain-scale simulations of spiking neural networks","abstract":"Simulation code for conventional supercomputers serves as a reference for neuromorphic computing systems. The present bottleneck of distributed large-scale spiking neuronal network simulations is the communication between compute nodes. Communication speed seems limited by the interconnect between the nodes and the software library orchestrating the data transfer. Profiling reveals, however, that the variability of the time required by the compute nodes between communication calls is large. The bottleneck is in fact the waiting time for the slowest node. A statistical model explains total simulation time on the basis of the distribution of computation times between communication calls. A fundamental cure is to avoid communication calls because this requires fewer synchronizations and reduces the variability of computation times across compute nodes. The organization of the mammalian brain into areas lends itself to such an optimization strategy. Connections between neurons within an area have short delays, but the delays of the long-range connections across areas are an order of magnitude longer. This suggests a structure-aware mapping of areas to compute nodes allowing for a partition into more frequent communication between nodes simulating a particular area and less frequent global communication. We demonstrate a substantial performance gain on a real-world example. This work proposes a local-global hybrid communication architecture for large-scale neuronal network simulations as a first step in mapping the structure of the brain to the structure of a supercomputer. It challenges the long-standing belief that the bottleneck of simulation is synchronization inherent in the collective calls of standard communication libraries. We provide guidelines for the energy efficient simulation of neuronal networks on conventional computing systems and raise the bar for neuromorphic systems.","author":[{"family":"Lober","given":"Melissa"},{"family":"Diesmann","given":"Markus"},{"family":"Kunkel","given":"Susanne"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2602.23274","URL":"https://doi.org/10.48550/arxiv.2602.23274","source":"datacite"},{"id":"doi:10.5281/zenodo.19357843","type":"article-journal","title":"Ep. 128: AI's Dial-Up Era: Looking Back from 2036","abstract":"Episode summary: In this forward-thinking episode of My Weird Prompts, hosts Herman Poppleberry and Corn kick off the year 2026 by traveling a decade into the future. They imagine a world in 2036 where the \"cutting-edge\" AI of today is viewed as an adorable, clunky relic of the past—much like we view the screeching sounds of dial-up internet today. From the death of prompt engineering to the rise of zero-latency, embodied intelligence, the duo breaks down why our current obsession with context windows and text boxes is just a passing phase. They dive deep into the transition from \"command-based\" to \"intent-based\" computing, where AI understands your needs without the need for complex instructions. Herman explains the shift from monolithic models to federated swarms of specialized agents, and how the \"hallucination\" bug of the 2020s will eventually be seen as a primitive technical limitation. Whether you're curious about the future of robotics or the evolution of persistent holographic memory, this episode provides a fascinating roadmap for the next decade of innovation. Tune in to find out why your current smartphone might soon feel like a rotary phone. Show Notes As the calendar turned to January 1, 2026, *My Weird Prompts* hosts Herman Poppleberry and Corn took a moment to look not just at the year ahead, but a full decade into the future. Prompted by a thought experiment from their housemate Daniel, the duo spent the episode \"time traveling\" to 2036 to look back at the current state of artificial intelligence. Their conclusion? The sophisticated tools we use today—the LLMs, the image generators, and the coding assistants—are destined to become the \"dial-up modems\" of the future. ### The Death of the Prompt One of the most striking insights from the discussion was the predicted obsolescence of \"prompt engineering.\" In 2026, users pride themselves on their ability to craft complex instructions, using delimiters and \"chain-of-thought\" techniques to coax the best results out of a model. Herman argues that by 2036, this will seem as primitive as using a rotary phone. We are currently in a \"lossy\" phase of technology, where we must translate human intent into rigid strings of text. Herman suggests that the future lies in \"intent-based computing.\" In this future, AI will possess such deep context regarding a user's life, professional history, and personal preferences that it will no longer require a three-paragraph explanation. A simple glance or a vague suggestion will suffice, as the machine will already understand the nuances of what \"professional\" or \"creative\" means to that specific individual. ### From Context Windows to Holographic Memory The hosts also tackled the technical limitations of modern AI memory. Today, developers and users celebrate when a model's \"context window\" expands to a million tokens. However, Herman describes the current state of AI as a \"brilliant assistant who gets hit with an amnesia ray every time you walk out of the room.\" By 2036, the concept of a \"window\" will likely be replaced by what Herman calls \"persistent, holographic memory.\" Instead of a blank slate at the start of every chat, a personal AI will have a continuous, decade-long relationship with its user. It will remember a casual comment about architectural styles from years prior and seamlessly apply that knowledge to a current project. The manual management of AI memory will become a relic of a more cumbersome era. ### Zero Latency and the End of the \"Thinking\" Pause One of the most relatable points of the episode was the \"dial-up screech\" of 2026: latency. Even the fastest models today have a slight delay as they generate tokens. Herman predicts that 2036 will be the era of \"zero-latency intelligence.\" Powered by specialized hardware—potentially optical or neuromorphic chips—AI responses will be instantaneous or even predictive. The duo joked about how future generations will find it hilarious that we used to sit and watch text scroll a","author":[{"family":"Rosehill","given":"Daniel"},{"family":"Tts","given":"Chatterbox"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.19357843","URL":"https://doi.org/10.5281/zenodo.19357843","source":"datacite"},{"id":"doi:10.48550/arxiv.2606.27365","type":"manuscript","title":"3D Imaging of Complex Skyrmion and Hopf Topologies in an Extended Sample","abstract":"Spin textures are key for emergent magnetic phenomena such as topological protection and underpin novel spintronic device paradigms based on racetrack memory, logic gates, and neuromorphic computing. Using a coherent diffractive imaging technique called vector ptycho-tomography, in combination with algorithms that are robust to noise, we image the 3D magnetic texture of skyrmion and Hopf topologies with no prior assumptions about the sample. This directly reveals experimentally for the first time an extended 3D skyrmion lattice, including the domain wall shape, topological charge, helicity, and Hopf index. Our findings demonstrate experimentally that dipole stabilized skyrmions in Fe/Gd multilayers exhibit barrel-shaped skyrmion tubes with a twisted helicity, transitioning from N$é$el-type winding at the surfaces to both clockwise and counterclockwise Bloch-type winding in the bulk, that can also be described as fractional hopfions. We image a lattice of 24 skyrmions with topological charge 1, average depth-dependent domain wall width of 23 to 40 nm, depth-dependent twisted helicity from $\\pm$155$°$ to $\\pm$30$°$, and fractional Hopf index of $\\pm$0.3. Over 10 TB of data were analyzed to yield a fully-resolved 3D reconstruction over a &gt;0.4 $μ$m$^3$ volume, with high fidelity down to the Nyquist limit of 8 nm. This method fills a key gap in the current landscape of magnetic imaging by enabling high-resolution, element-specific 3D reconstructions of full-field extended spin textures - offering a new route for exploring the topological complexity of magnetic materials in three dimensions.","author":[{"family":"Binnie","given":"I"},{"family":"Fang","given":"H"},{"family":"Shearer","given":"B"},{"family":"Grafov","given":"A"},{"family":"Jenkins","given":"N"},{"family":"Shao","given":"Y"},{"family":"O'leary","given":"C"},{"family":"Liao","given":"Y"},{"family":"Feggeler","given":"T"},{"family":"Oh","given":"A"},{"family":"Yazdi","given":"S"},{"family":"Zou","given":"J"},{"family":"Wang","given":"B"},{"family":"Cating","given":"EE"},{"family":"Montoya","given":"SA"},{"family":"Shapiro","given":"D"},{"family":"Miao","given":"J"},{"family":"Kapteyn","given":"HC"},{"family":"Murnane","given":"MM"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2606.27365","URL":"https://doi.org/10.48550/arxiv.2606.27365","source":"datacite"},{"id":"doi:10.5445/ir/1000194612","type":"article-journal","title":"Printed 1T1R Arrays for Next-Generation Electronics","abstract":"The increasing demand for energy-efficient and high-density hardware driven by artificial intelligence and the Internet of Things has exposed the limitations of conventional computing systems based on the von Neumann architecture. Consequently, neuromorphic computing has emerged as a promising paradigm, enabling in-memory data processing inspired by biological systems. The memristive array is one of the most promising hardware solutions for implementing various neuromorphic algorithms. Among the different unit cell configurations constituting a memristive array, the 1T1R architecture, combining a transistor and a memristor, has demonstrated significant potential due to its ability to provide controllable operation and reduced crosstalk. At the same time, printed electronics has gained attention as a low-cost and scalable fabrication approach, compatible with flexible substrates and low-temperature processing. However, the implementation of 1T1R structures using printing techniques introduces several challenges, including ink formulation, interfacial compatibility between printed layers, and multilayer alignment. This perspective provides an overview of 1T1R architectures with a particular focus on their fabrication through printing-based methods. The device structure, electrical performance, and integration strategies are discussed, along with the main challenges associated with printed techniques. The potential of 1T1R structures for applications in neuromorphic hardware, analog computing, and flexible IoT platforms is highlighted, outlining future directions toward scalable and low-power printed electronic systems.","author":[{"family":"Martins","given":"Raquel"},{"family":"Hu","given":"Hongrong"},{"family":"Pereira","given":"Maria"},{"family":"Cadilha Marques","given":"Gabriel"},{"family":"Kiazadeh","given":"Asal"},{"family":"Aghassi-Hagmann","given":"Jasmin"},{"family":"Carlos","given":"Emanuel"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5445/ir/1000194612","URL":"https://doi.org/10.5445/ir/1000194612","source":"datacite"},{"id":"doi:10.48550/arxiv.2511.08065","type":"manuscript","title":"I2E: Real-Time Image-to-Event Conversion for High-Performance Spiking Neural Networks","abstract":"Spiking neural networks (SNNs) promise highly energy-efficient computing, but their adoption is hindered by a critical scarcity of event-stream data. This work introduces I2E, an algorithmic framework that resolves this bottleneck by converting static images into high-fidelity event streams. By simulating microsaccadic eye movements with a highly parallelized convolution, I2E achieves a conversion speed over 300x faster than prior methods, uniquely enabling on-the-fly data augmentation for SNN training. The framework's effectiveness is demonstrated on large-scale benchmarks. An SNN trained on the generated I2E-ImageNet dataset achieves a state-of-the-art accuracy of 60.50%. Critically, this work establishes a powerful sim-to-real paradigm where pre-training on synthetic I2E data and fine-tuning on the real-world CIFAR10-DVS dataset yields an unprecedented accuracy of 92.5%. This result validates that synthetic event data can serve as a high-fidelity proxy for real sensor data, bridging a long-standing gap in neuromorphic engineering. By providing a scalable solution to the data problem, I2E offers a foundational toolkit for developing high-performance neuromorphic systems. The open-source algorithm and all generated datasets are provided to accelerate research in the field.","author":[{"family":"Ma","given":"Ruichen"},{"family":"Meng","given":"Liwei"},{"family":"Qiao","given":"Guanchao"},{"family":"Ning","given":"Ning"},{"family":"Liu","given":"Yang"},{"family":"Hu","given":"Shaogang"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2511.08065","URL":"https://doi.org/10.48550/arxiv.2511.08065","source":"datacite"},{"id":"doi:10.48550/arxiv.2606.24075","type":"manuscript","title":"End-to-End Radar and Communication Modulation Recognition with Neuromorphic Computing","abstract":"Although deep learning-based methods can achieve high accuracy in automatic modulation recognition (AMR) tasks, their high computational cost makes it difficult to strike a balance between accuracy and power consumption, thereby limiting their application on resource-constrained platforms. Neuromorphic architectures that perform spike-driven inference with modest energy budgets have recently been explored for vision and timeseries tasks. Motivated by these works, we propose EMRFormer, a novel end-to-end spiking nerural network (SNN) architecture that applies spike-driven transformer to the constraints of neuromorphic hardware for AMR. The model incorporates an adaptive spike encoder and Integer Leaky Integrate-and-Fire neurons to mitigate the degradation of effective information and enhance SNN representational capacity. By integrating spike-separable Convolution Neural Networks (SSCNN) into Spike-Driven Transformers (SpikeFormer), EMRFormer effectively extracts multi-scale temporal features from the raw IQ waveforms. We validate our approach across various mainstream datasets, the experimental results show that EMRFormer achieves state-of-the-art interms of accuracy, outperforming all the baselines. Furthermore, the model maintains strong performance in low signal-to-noise(SNR) environments and reduces theoretical energy consumption by over 90%. Finally, we evaluate our model on a KA200 neuromorphic chip. The results show that our model achieves up to 5 times reduction in power compared to running on a 3090 GPU or an Orin NX. This work demonstrates a promising pathway for AMR on resource-constrained devices.","author":[{"family":"Li","given":"Xiaohu"},{"family":"Qu","given":"Chongxiao"},{"family":"Lin","given":"Caiyong"},{"family":"Dou","given":"Chenxiao"},{"family":"Hua","given":"Wei"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2606.24075","URL":"https://doi.org/10.48550/arxiv.2606.24075","source":"datacite"},{"id":"doi:10.24406/publica-5477","type":"article-journal","title":"Hardware realization of neuromorphic computing with a 4-port photonic reservoir for modulation format identification","abstract":"The fields of machine learning and artificial intelligence drive researchers to explore energy-efficient, brain-inspired new hardware. Reservoir computing encompasses recurrent neural networks for sequential data processing and matches the performance of other recurrent networks with less training and lower costs. However, traditional software-based neural network (NN) require high energy due to computational demands and massive data transfer needs. A hybrid photonic-electronic reservoir computing (RC) system overcomes this challenge with neuromorphic photonic integrated circuits or NeuroPICs, combining passive photonic reservoirs with electronic readout layers. Here, we introduce a hybrid photonic-electronic RC solution for modulation format identification in C-band telecommunication network monitoring. The NeuroPIC is built on a silicon-on-insulator platform featuring a 4-port reservoir architecture. We discuss the NeuroPIC design, fabrication, experimental performance, and compare it with simulations. The system incorporates nonlinearity through a simple digital readout and achieves close to 100% accuracy in identifying quadrature amplitude modulation formats transmitted over 20 km of optical fiber at a 32 Gbaud symbol rate. The NeuroPIC performance is robust against fabrication imperfections like waveguide propagation loss, phase randomization, and delay line length variations. Experimental results surpassed simulations, attributed to enhanced ‘richness’ of signal interference in the physical system. By integrating an energy-efficient photonic analog processing with electronic readout, this hybrid system reduces the dependence on fully digital NNs, offering a pathway for high-speed temporal data processing in various applications. While system-level energy metrics require further exploration, our work highlights the potential of hybrid approaches to balance performance, fabrication tolerance, and computational efficiency in real-world applications.","author":[{"family":"Şeker","given":"Enes"},{"family":"Thomas","given":"Rijil"},{"family":"Hünefeld","given":"Guillermo"},{"family":"Suckow","given":"Stephan"},{"family":"Kaveh","given":"Mahdi"},{"family":"Ronniger","given":"Gregor"},{"family":"Safari","given":"Pooyan"},{"family":"Sackey","given":"Isaac"},{"family":"Stahl","given":"David"},{"family":"Schubert","given":"Colja"},{"family":"Fischer","given":"Johannes"},{"family":"Freund","given":"Ronald"},{"family":"Lemme","given":"Max"},{"family":"Unav"}],"issued":{"date-parts":[[2025]]},"DOI":"10.24406/publica-5477","URL":"https://doi.org/10.24406/publica-5477","source":"datacite"},{"id":"doi:10.48550/arxiv.2506.20015","type":"manuscript","title":"Neuromorphic Wireless Split Computing with Resonate-and-Fire Neurons","abstract":"Neuromorphic computing offers an energy-efficient alternative to conventional deep learning accelerators, particularly for real-time processing of time-series data. However, many edge applications, such as wireless sensing and audio recognition, generate streaming signals with rich spectral features that are not effectively captured by conventional leaky integrate-and-fire (LIF) spiking neurons. This paper investigates a wireless split computing architecture that employs resonate-and-fire (RF) neurons with oscillatory dynamics to process time-domain signals directly, eliminating the need for costly spectral pre-processing. By resonating at tunable frequencies, RF neurons extract time-localized spectral features while maintaining low spiking activity. This temporal sparsity translates into significant savings in both computation and transmission energy. Assuming an OFDM-based analog wireless interface for spike transmission, we present a complete system design and evaluate its performance on audio classification and modulation classification tasks. Experimental results show that the proposed RF-SNN architecture achieves comparable accuracy to conventional LIF-SNNs and ANNs, while substantially reducing spike rates and total energy consumption during inference and communication.","author":[{"family":"Wu","given":"Dengyu"},{"family":"Chen","given":"Jiechen"},{"family":"Poor","given":"HV"},{"family":"Rajendran","given":"Bipin"},{"family":"Simeone","given":"Osvaldo"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2506.20015","URL":"https://doi.org/10.48550/arxiv.2506.20015","source":"datacite"},{"id":"doi:10.48550/arxiv.2606.16896","type":"manuscript","title":"Neural dynamical systems on ferroelectric compute-in-memory for real-time forecasting","abstract":"Neural dynamical systems are expressive temporal predictors that capture continuous-time dynamics through fine-grained state updates. However, this sequential structure maps poorly onto digital hardware optimized for dense matrix operations, a mismatch that analog neuromorphic computing, with its native continuous-time dynamics, can resolve. We introduce FerroNDS, a neuromorphic system built from two analog primitives: an integrator for temporal accumulation and an oscillator for frequency-selective filtering. We map this system onto compute-in-memory hardware based on multi-bit ferrodiodes. A 128-neuron instance of FerroNDS computes short-time Fourier transform and forecasts a 500-ms horizon for periodic, quasi-periodic, and chaotic signals. The system achieves sub-watt real-time operation with per-neuron per-inference energy of 1.64 $μ$J (200 Hz) and 0.29 $μ$J (10 kHz), 25-40$\\times$ area reduction over SRAM-based digital systems, and per-layer latency of 3.18 ms (200 Hz) and 63.87 $μ$s (10 kHz). To our knowledge, this is the first end-to-end integration of a ferrodiode into a neuromorphic computational framework, establishing ferroelectric compute-in-memory as a practical substrate for analog neural dynamical systems.","author":[{"family":"Katti","given":"Keshava"},{"family":"Selvakumar","given":"Adithya"},{"family":"Chaudhari","given":"Pratik"},{"family":"Jariwala","given":"Deep"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2606.16896","URL":"https://doi.org/10.48550/arxiv.2606.16896","source":"datacite"},{"id":"doi:10.5281/zenodo.18305742","type":"article-journal","title":"SECURE AND INTELLIGENT IOT SYSTEMS: ARCHITECTURES, THREATS, AND DEFENSE","abstract":"This book brings together advanced research that explores the design, security, and sustainability of intelligent and connected computing systems. The chapters collectively reflect the rapid evolution of embedded intelligence, trusted digital environments, and Internet of Things (IoT) architectures in response to growing demands for efficiency, security, and scalability. The chapter Cognitive Microcontrollers: A Hybrid Neuromorphic– RISC Architecture for Ultra-Low-Power On-Device Intelligence introduces a novel hardware paradigm that enables intelligent processing at the edge with minimal energy consumption. This innovation is complemented by Engineering Trusted Computing Systems for Secure Digital Music Production Environments, which addresses the need for secure, reliable computing infrastructures in creative and digital content production workflows. Security and sustainability concerns are further examined in Malware Threats in Green IoT: Five Years of Attacks, Energy Impacts, and AI-Driven Defense (2020–2025). This chapter provides a comprehensive analysis of evolving cyber threats in energy-aware IoT systems and highlights the role of artificial intelligence in strengthening defensive mechanisms while preserving system efficiency. The final chapter, Architecture and Implementation of IoT Systems: Integration of ESP32, Communication Protocols, Platforms, Tools and Frameworks for the IoT, offers a practical perspective on building and deploying robust IoT solutions. Together, these chapters present a cohesive view of how intelligent hardware, secure computing, and scalable IoT architectures can be integrated to support next-generation digital ecosystems.","author":[{"family":"Mebarki","given":"Abdelkrim"},{"family":"Dabbabi","given":"Karim"},{"family":"Agoi","given":"Moses"},{"family":"Ogunsanwo","given":"Gbenga"},{"family":"Agoi","given":"Emmanuel"},{"family":"Bamidele","given":"Benjamin"},{"family":"Mishra","given":"Bimal"},{"family":"Alaeddine","given":"Hmidi"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18305742","URL":"https://doi.org/10.5281/zenodo.18305742","source":"datacite"},{"id":"doi:10.5281/zenodo.18305741","type":"article-journal","title":"SECURE AND INTELLIGENT IOT SYSTEMS: ARCHITECTURES, THREATS, AND DEFENSE","abstract":"This book brings together advanced research that explores the design, security, and sustainability of intelligent and connected computing systems. The chapters collectively reflect the rapid evolution of embedded intelligence, trusted digital environments, and Internet of Things (IoT) architectures in response to growing demands for efficiency, security, and scalability. The chapter Cognitive Microcontrollers: A Hybrid Neuromorphic– RISC Architecture for Ultra-Low-Power On-Device Intelligence introduces a novel hardware paradigm that enables intelligent processing at the edge with minimal energy consumption. This innovation is complemented by Engineering Trusted Computing Systems for Secure Digital Music Production Environments, which addresses the need for secure, reliable computing infrastructures in creative and digital content production workflows. Security and sustainability concerns are further examined in Malware Threats in Green IoT: Five Years of Attacks, Energy Impacts, and AI-Driven Defense (2020–2025). This chapter provides a comprehensive analysis of evolving cyber threats in energy-aware IoT systems and highlights the role of artificial intelligence in strengthening defensive mechanisms while preserving system efficiency. The final chapter, Architecture and Implementation of IoT Systems: Integration of ESP32, Communication Protocols, Platforms, Tools and Frameworks for the IoT, offers a practical perspective on building and deploying robust IoT solutions. Together, these chapters present a cohesive view of how intelligent hardware, secure computing, and scalable IoT architectures can be integrated to support next-generation digital ecosystems.","author":[{"family":"Mebarki","given":"Abdelkrim"},{"family":"Dabbabi","given":"Karim"},{"family":"Agoi","given":"Moses"},{"family":"Ogunsanwo","given":"Gbenga"},{"family":"Agoi","given":"Emmanuel"},{"family":"Bamidele","given":"Benjamin"},{"family":"Mishra","given":"Bimal"},{"family":"Alaeddine","given":"Hmidi"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18305741","URL":"https://doi.org/10.5281/zenodo.18305741","source":"datacite"},{"id":"doi:10.48550/arxiv.2602.18072","type":"manuscript","title":"HiAER-Spike Software-Hardware Reconfigurable Platform for Event-Driven Neuromorphic Computing at Scale","abstract":"In this work, we present HiAER-Spike, a modular, reconfigurable, event-driven neuromorphic computing platform designed to execute large spiking neural networks with up to 160 million neurons and 40 billion synapses - roughly twice the neurons of a mouse brain at faster than real time. This system, assembled at the UC San Diego Supercomputer Center, comprises a co-designed hard- and software stack that is optimized for run-time massively parallel processing and hierarchical address-event routing (HiAER) of spikes while promoting memory-efficient network storage and execution. The architecture efficiently handles both sparse connectivity and sparse activity for robust and low-latency event-driven inference for both edge and cloud computing. A Python programming interface to HiAER-Spike, agnostic to hardware-level detail, shields the user from complexity in the configuration and execution of general spiking neural networks with minimal constraints in topology. The system is made easily available over a web portal for use by the wider community. In the following, we provide an overview of the hard- and software stack, explain the underlying design principles, demonstrate some of the system's capabilities and solicit feedback from the broader neuromorphic community. Examples are shown demonstrating HiAER-Spike's capabilities for event-driven vision on benchmark CIFAR-10, DVS event-based gesture, MNIST, and Pong tasks.","author":[{"family":"Frank","given":"Gwenevere"},{"family":"Hota","given":"Gopabandhu"},{"family":"Wang","given":"Keli"},{"family":"Deng","given":"Christopher"},{"family":"Arora","given":"Krish"},{"family":"Vins","given":"Diana"},{"family":"Uppal","given":"Abhinav"},{"family":"Olajide","given":"Omowuyi"},{"family":"Yoshimoto","given":"Kenneth"},{"family":"Wang","given":"Qingbo"},{"family":"Yamaoka","given":"Mari"},{"family":"Leugering","given":"Johannes"},{"family":"Deiss","given":"Stephen"},{"family":"Gibb","given":"Leif"},{"family":"Cauwenberghs","given":"Gert"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2602.18072","URL":"https://doi.org/10.48550/arxiv.2602.18072","source":"datacite"},{"id":"doi:10.48550/arxiv.2606.12968","type":"manuscript","title":"Quantum-Driven Neuromorphic Computing for Million-Qubit-Scale Workloads","abstract":"We introduce Apollo, a 10000 node p-qubit neuromorphic processor fabricated in 16 nm mixed signal CMOS and operating fully at room temperature with a typical analog core power envelope of about 0.5 W. Its fundamental element, the p-qubit, is a bistable stochastic unit whose continuous time state fluctuations are driven by integrated quantum entropy units that inject true quantum derived randomness. This enables ultrafast stochastic transitions at low energy while preserving a classical state representation. Apollo combines these p-qubits with a high degree Hyperion 256 interconnect topology, allowing efficient embedding of dense Ising and QUBO problems with substantially reduced minor embedding overhead compared with sparse annealing platforms. We show that, through the Suzuki Trotter correspondence, the equilibrium statistics and annealing dynamics of the p-qubit network reproduce key properties of transverse field quantum annealing without cryogenic cooling, long lived coherence, or microwave control. Beyond device level validation, Apollo is evaluated on a three dimensional spin glass benchmark previously used to study quantum advantage in superconducting annealers. Across 300 disorder realizations, Apollo reaches substantially lower ground state energies than reported cryogenic quantum annealing hardware, while remaining distinct from classical simulated annealing and simulated quantum annealing. A 350 nm release candidate device experimentally validates the core p-qubit dynamics, thermodynamic sampling correctness, and continuous time annealing behavior. These results establish Apollo as a room temperature, industrially scalable platform for quantum driven energy based optimization, probabilistic inference, generative modeling, and hybrid classical quantum workflows.","author":[{"family":"Ivanov","given":"Adams"},{"family":"Rahmeh","given":"Samer"},{"family":"Nascimento","given":"Erick"},{"family":"Herrmann","given":"Daniela"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2606.12968","URL":"https://doi.org/10.48550/arxiv.2606.12968","source":"datacite"},{"id":"doi:10.48550/arxiv.2606.11703","type":"manuscript","title":"Integrated magnonic neural circuits based on nonlinear wave neurons","abstract":"Artificial intelligence is driving intense interest in alternative computing hardware capable of neural information processing beyond conventional charge-based electronics. Among emerging approaches, wave-based computing promises highly parallel and energy-efficient operation, but scalable physical neural hardware has remained elusive because wave systems generally lack cascadable nonlinear neurons with signal regeneration and phase-robust operation. Here we demonstrate integrated magnonic neural circuits based on nonlinear threshold neurons realized in nanoscale yttrium iron garnet waveguides. The neurons perform weighted summation of multiple spin-wave inputs, while a pump-controlled nonlinear activation defines continuously tunable firing thresholds. Owing to deeply nonlinear spin-wave dynamics, the activated neurons emit self-normalized outputs whose intensities are largely independent of the input amplitudes, while nonlinear phase self-adjustment suppresses sensitivity to the relative input phases, enabling deterministic neuron-to-neuron cascading without external signal restoration. We experimentally realize programmable threshold neurons, reconfigurable weighted classification and deterministic cascading between sequential neuronal stages, and further demonstrate reconfigurable physical pattern recognition in a seven-neuron integrated magnonic circuit through experimental classification of the binary letter patterns 'HUST'. These results establish nonlinear magnons as a scalable platform for integrated neural hardware and position nonlinear wave dynamics as a general paradigm for physical neuromorphic computing.","author":[{"family":"Guo","given":"Mengying"},{"family":"Jing","given":"Xudong"},{"family":"Davidkova","given":"Kristýna"},{"family":"Verba","given":"Roman"},{"family":"Zhou","given":"Zhenyu"},{"family":"Guo","given":"Xueyu"},{"family":"Dubs","given":"Carsten"},{"family":"Gao","given":"Chuan"},{"family":"Rao","given":"Yiheng"},{"family":"Cai","given":"Kaiming"},{"family":"Li","given":"Jing"},{"family":"Pirro","given":"Philipp"},{"family":"Chumak","given":"Andrii"},{"family":"Wang","given":"Qi"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2606.11703","URL":"https://doi.org/10.48550/arxiv.2606.11703","source":"datacite"},{"id":"doi:10.17169/refubium-52446","type":"article-journal","title":"A New Threshold Switching Device With Tunable Negative Differential Resistance Based on ErMnO3 Polymorphs","abstract":"Negative differential resistance (NDR) devices have emerged as promising building blocks for neuromorphic computing due to their inherent nonlinear and threshold-dependent electrical behavior. In particular, current-controlled NDR devices can exhibit abrupt switching, self-sustained oscillations, and volatility – key characteristics analogous to the spiking dynamics of biological neurons. The NDR is typically driven by thermal runaway effects, or by an insulator-to-metal transition, with materials such as VO2, NbOx, TaOx, and complex oxides like nikelates, cobaltites, or manganites. However, these systems often suffer from high forming and operating voltages, large Joule heating, limited device endurance, and no or little tunability. Here, we demonstrate a novel NDR device based on polycrystalline ErMnO3 comprising conducting orthorhombic and insulating hexagonal phases. Pt/ErMnO3/Pt memory devices exhibit unipolar, symmetric, forming-free threshold switching with a low threshold voltage of ~2.1 V, a memory window of 1.1 V, and endurance over 2 × 104 cycles. Joule-heating-enhanced Poole–Frenkel conduction occurs in the orthorhombic phase, while the insulating hexagonal phase prevents excessive heating and breakdown. Tunability is achieved by controlling the polymorph ratio and the orthorhombic phase conductivity. ErMnO3 polymorphs thus offer a versatile platform to engineer the electrical characteristics of NDR devices for low-power, reliable neuromorphic applications.","author":[{"family":"Wu","given":"Rong"},{"family":"Maudet","given":"Florian"},{"family":"Phan","given":"Thanh"},{"family":"Hamouda","given":"Wassim"},{"family":"Schroedter","given":"Richard"},{"family":"Demirkol","given":"Ahmet"},{"family":"Tetzlaff","given":"Ronald"},{"family":"Deshpande","given":"Veeresh"},{"family":"Dubourdieu","given":"Catherine"}],"issued":{"date-parts":[[2026]]},"DOI":"10.17169/refubium-52446","URL":"https://doi.org/10.17169/refubium-52446","source":"datacite"},{"id":"doi:10.5281/zenodo.20626393","type":"article-journal","title":"Threshold Dynamics of Silver Filament Formation and Rupture in Ag/SiO2/Au Memristors","abstract":"Memristive devices with tunable volatility represent a promising platform for neuromorphic computing and adaptive electronics, enabling temporal information processing and gradual memory consolidation. Understanding the threshold dynamics governing conductive filament formation and rupture is critical for precise control of resistive switching behavior. Here, we systematically investigate Ag/SiO2/Au memristor devices to elucidate the physics of silver filament evolution, revealing the asymmetric interplay between drift-dominated formation and diffusion-dominated rupture processes. Quantitative analysis of filament formation dynamics yields an activation energy of 0.77 ± 0.12 eV for ion migration with field enhancement factor β = 0.21, demonstrating field-assisted thermally activated silver filament formation. Our experiments confirm that formation and rupture share a similar intrinsic energy landscape but operate under different driving forces: field-accelerated drift versus thermally-driven diffusion. Interestingly, detailed filament dynamics analysis reveals the existence of a critical metastable regime during both formation and rupture processes, where drift and diffusion forces balance, and conduction occurs via trap-assisted tunneling through nanoscale silver clusters. This metastable state exists at specific voltages where drift and diffusion forces achieve dynamic equilibrium, leading to intermediate conductance states between the stable low and high resistance states. Beyond this regime, the system exhibits deterministic behavior: at higher voltages, filament growth proceeds through sequential conduction regimes where Joule heating drives morphological changes in silver nanoclusters that concentrate the electric field and accelerate complete filament formation. At lower voltages, diffusion dominates and drives complete rupture of filaments back to nanoclusters. This enables voltage-tunable operation from stochastic to deterministic modes. Combined with quantitative activation energies and the drift-diffusion framework, our work provides fundamental insight into threshold switching physics and establishes design principles for engineering devices with controlled probabilistic or deterministic behavior for neuromorphic and stochastic computing applications.","author":[{"family":"Ranade","given":"Varun"},{"family":"Veldhoen","given":"William"},{"family":"Gibbins","given":"Felix"},{"family":"Shradha","given":"Sajal"},{"family":"Murray","given":"Chris"},{"family":"Chunin","given":"Igor"},{"family":"Mccloskey","given":"David"},{"family":"Shvets","given":"Igor"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20626393","URL":"https://doi.org/10.5281/zenodo.20626393","source":"datacite"},{"id":"doi:10.24406/publica-5297","type":"article-journal","title":"Gate controlled EDL-based capacitor-diodes (G-CAPode) for switchable signal filters with logic gate functionality","abstract":"This work explores a novel application of the G-CAPode - a previously developed, 3D-printed gate-controlled electrochemical capacitor-diode - by demonstrating its ability to perform logic operations in ionic circuits. The G-CAPode integrates diode-like rectification, voltage-controlled switching via a gate electrode, and dynamically tunable capacitance within a single architecture. While prior studies demonstrated G-CAPodes to function as tunable capacitors - a varactor - in high-pass filters, this study extends its use toward logic operations. Specifically, frequency-dependent OR and AND logic behaviors are realized by connecting two G-CAPodes in parallel and series, respectively. In these configurations, logic output is determined by the impedance state of each device, which is dynamically controlled via gate bias. The OR-gate circuit yields a high output when at least one G-CAPode is in the ON (low-impedance) state, while the AND-gate setup requires both devices to be ON for signal transmission, consistent with Boolean logic. The switching behavior is controlled by applying specific gate biases, which modulate the impedance of each G-CAPode in real-time and therefore enables analog logic functionality using purely ionic components. The results underline the potential of the printed G-CAPode as a multifunctional platform for adaptive signal processing, offering a compact and silicon-free solution for emerging iontronic applications such as neuromorphic computing, adaptive filtering, and analog signal processing.","author":[{"family":"Gellrich","given":"Christin"},{"family":"Galek","given":"Przemysław"},{"family":"Shupletsov","given":"Leonid"},{"family":"Grothe","given":"Julia"},{"family":"Kaskel","given":"Stefan"},{"family":"Unav"}],"issued":{"date-parts":[[2025]]},"DOI":"10.24406/publica-5297","URL":"https://doi.org/10.24406/publica-5297","source":"datacite"},{"id":"doi:10.18721/jpm.191.126","type":"article-journal","title":"The role of charge carrier diffusion in halide perovskite luminophores with memory for optical computing","abstract":"The optical elements that can combine memory and signal modulation have become a topic of interest in the field of neuromorphic computing systems. In particular, the optical analogue of a memristor called memlumor exhibits promising features in which the change of the output luminescence depends not only on the excitation light signal but also on the state of the material. Metal halide perovskite luminophores are considered to be suitable for memlumor implementation as they exhibit modulation of photoluminescence due to the interaction of structure defects with the environment and previous interactions, thereby possessing memory. Luminescence in perovskite materials is described via Shockley–Reed–Hall model which takes into account charge carriers dynamics in the structure. Additionally, the diffusion of charge carriers also plays a key role and highly depends on memlumor’s size. This paper explores the perovskite memlumor’s functionality based on their characteristic size to identify optimal parameters and suitable designs for future optical neuromorphic computing architectures.","author":[{"family":"Ekgardt","given":"Alexey"},{"family":"Sapozhnikova","given":"Elizaveta"},{"family":"Verkhogliadov","given":"Grigorii"},{"family":"Pushkarev","given":"Anatoly"}],"issued":{"date-parts":[[2026]]},"DOI":"10.18721/jpm.191.126","URL":"https://doi.org/10.18721/jpm.191.126","source":"datacite"},{"id":"doi:10.48550/arxiv.2606.05768","type":"manuscript","title":"Electrolyte Bonding Engineering for Highly Uniform GeTe-based CBRAM and Parallel Hebbian Learning in Selector-free Hopfield Networks","abstract":"Hopfield networks offer a hardware-friendly framework for energy-efficient associative memory, yet their practical realization in memristor crossbar arrays is critically hindered by device-to-device (D2D) variability, which prevents reliable parallel programming. Here, we address this bottleneck through systematic composition engineering of the Ge-Te solid electrolyte in conductive bridge random access memory (CBRAM) devices. By varying the Ge:Te ratio, we identify Ge3.5Te1 as an optimal electrolyte composition that suppresses stochastic resistance variation by approximately three orders of magnitude compared to GeSe-based devices. Raman spectroscopy reveals that this dramatic improvement originates from a bonding network dominated by asymmetric-stretching GeTe4 tetrahedral units, which form interconnected free-volume channels that confine and stabilize Cu+ ion migration pathways. Leveraging this enhanced uniformity, we fabricate a selector-less 16x16 Cu/Ge3.5Te1 CBRAM crossbar array and demonstrate a 4x4 Hopfield associative network capable of learning and recalling binary pattern pairs via fully parallel programming using a half-selection scheme. Successful pattern recall is achieved for up to two stored associations despite the absence of selector elements, establishing a proof-of-concept for selector-free hardware implementations of associative memory. These results highlight the critical role of electrolyte bonding structure in determining memristor uniformity and provide a materials-driven pathway toward scalable, parallel neuromorphic computing systems.","author":[{"family":"Bang","given":"Jiin"},{"family":"Hwang","given":"Jingyeong"},{"family":"Kang","given":"Unhyeon"},{"family":"Oh","given":"Seungmin"},{"family":"Lee","given":"Kyungmin"},{"family":"Park","given":"Jaehyun"},{"family":"Lee","given":"Younghyun"},{"family":"Jang","given":"Hyun"},{"family":"Park","given":"Seongsik"},{"family":"Jeong","given":"Yeonjoo"},{"family":"Kim","given":"Inho"},{"family":"Park","given":"Jong"},{"family":"Lee","given":"Suyoun"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2606.05768","URL":"https://doi.org/10.48550/arxiv.2606.05768","source":"datacite"},{"id":"doi:10.48550/arxiv.2606.03935","type":"manuscript","title":"Quadratic integrate-and-fire neurons exhibit less fragmented loss landscapes and outperform leaky integrate-and-fire neurons in spike-based gradient descent","abstract":"The ability to train spiking neural networks is essential for modeling biological neural networks as well as for neuromorphic computing. However, for the extensively used leaky integrate-and-fire (LIF) neurons, arbitrarily small parameter changes can induce spike (dis)appearances that disrupt subsequent activity, leading to unstable neural representations and permanently silent neurons during exact spike-based gradient descent. Recent work shows that a class of neuron models, which includes the quadratic integrate-and-fire (QIF) neuron, avoids these discontinuities and enables continuous and even smooth spike-based gradient descent. However, it remains unclear whether these advantages translate into practice. Here, we demonstrate that they do so via a controlled comparison between networks of LIF and QIF neurons on the popular Spiking Heidelberg Digits dataset. Specifically, in a first step, we perform a thorough hyperparameter search to optimize both models, revealing a clear performance advantage of QIF neurons. In a second step, we visualize the loss and gradient landscapes. Consistent with their inferior performance, we find that the loss landscapes of LIF neurons, which are discontinuous, appear more fragmented and the related gradients more erratic. An analysis of the landscapes of single samples indicates that these features arise from changes in the temporal order of spikes, which often cause disruptive spike (dis)appearances. Overall, our results advocate replacing LIF neurons with neuron models exhibiting continuous spiking dynamics, such as QIF neurons, for gradient descent training.","author":[{"family":"Wenig","given":"Carlo"},{"family":"Memmesheimer","given":"Raoul"},{"family":"Klos","given":"Christian"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2606.03935","URL":"https://doi.org/10.48550/arxiv.2606.03935","source":"datacite"},{"id":"doi:10.48550/arxiv.2606.02931","type":"manuscript","title":"Second-Order Synaptic Memory using Inherent Plasticity of Moiré Superlattices","abstract":"Achieving synaptic functionality electronically in a single-element quantum material is a fundamental challenge, as conventional methods rely on the introduction of extrinsic charge-traps or polar components. Here, we demonstrate that twisted double bilayer graphene (tDBLG) moiré superlattices, composed purely of carbon, exhibit electronic hysteresis and plasticity in presence of twist-angle disorder. Inversion symmetry breaking at the moiré length scales also gives rise to second-order nonlinear electrical response via disorder-mediated extrinsic mechanisms. Such second-order nonlinearity is highly tunable in both sign and magnitude by varying carrier concentration and vertical displacement field. We harness the coexistence of electronic plasticity and second-order nonlinearity to realize a second-order synaptic memory device. Our findings establish strained moiré carbon systems as a powerful new platform for energy-efficient neuromorphic computing, demonstrating that complex electronic functionality can emerge purely from symmetry breaking physics in a single-element material.","author":[{"family":"Ahmed","given":"Tanweer"},{"family":"Watanabe","given":"Kenji"},{"family":"Taniguchi","given":"Takashi"},{"family":"Casanova","given":"Fèlix"},{"family":"Hueso","given":"Luis"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2606.02931","URL":"https://doi.org/10.48550/arxiv.2606.02931","source":"datacite"},{"id":"doi:10.48550/arxiv.2606.01135","type":"manuscript","title":"Spiking and Event-driven Neuromorphic Mamba Models for Efficient Speech Recognition","abstract":"Deep learning has greatly advanced automatic speech recognition (ASR), enabling widespread deployment on edge devices such as smartphones and smart home systems. However, the computational and energy demands of deep neural networks pose significant challenges for such resource-constrained deployments, introducing latency and limiting real-time interaction. Neuromorphic computing offers a promising solution by introducing activation sparsity through spiking neural networks (SNNs) and event-driven neural networks, converting dense operations into sparse computations. However, a study that evaluates the hardware benefits of different neuromorphic strategies remains lacking for ASR. This paper explores spiking and event-driven neuromorphic neural networks to improve activation sparsity in the state-of-the-art SpeechMamba model for ASR. We introduce an event-driven SpeechMamba with FATReLU activation, achieving over 60% activation sparsity with less than 1% accuracy degradation on LibriSpeech. We also propose a spiking SpeechMamba that attains over 70% sparsity while using 30% fewer parameters than comparable SNNs. Finally, we develop a cycle-accurate event-driven simulator enabling flexible algorithm-hardware co-exploration, which helps us identify computational bottlenecks and yields over 10% additional efficiency improvements.","author":[{"family":"Ahmed","given":"Tauseef"},{"family":"Sun","given":"Tao"},{"family":"Castrillon","given":"Jeronimo"},{"family":"Vadivel","given":"Kanishkan"},{"family":"Tang","given":"Guangzhi"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2606.01135","URL":"https://doi.org/10.48550/arxiv.2606.01135","source":"datacite"},{"id":"doi:10.48550/arxiv.2605.29942","type":"manuscript","title":"Reconfigurable Multistate MRAM Synapses with Vortex STNO based Neurons for Scalable In-Memory Convolutional Neural Networks","abstract":"Magnetic tunnel junction (MTJ)-based magnetic random-access memory (MRAM) is a promising platform for neuromorphic and in-memory computing owing to its non-volatility, high endurance, fast switching dynamics and CMOS compatibility. However, conventional spin-transfer torque and spin-orbit torque MRAM implementations for neural networks often suffer from high critical switching currents, large latency, thermal instability and significant read-write overheads. Here, we demonstrate a unified multistate MRAM-spin-torque nano-oscillator (STNO) architecture that integrates synapses and neurons on a single chip for convolutional neural network (CNN) applications. The system employs 1x8 multistate MRAM arrays as programmable synapses coupled with a vortex-based STNO neuron, enabling both individual and collective programming through fieldline-driven write channels. Multiple configurable resistance states are achieved by tuning internal and external magnetic fields together with bias currents, allowing quantized positive and negative synaptic weights for configurable kernel and pooling operations. The proposed architecture is evaluated through simulation on MNIST, SVHN, CIFAR-10, Google Speech Commands (GSC) and RadioML datasets, achieving accuracy of 99.76%, 87.93%, 78.14%, 87.96% and 56.46% respectively. Based on fabricated device dimensions, the complete architecture occupies ~6171.2 μm2 with an average energy consumption of 200.08 pJ per training and inference cycle for MNIST, highlighting its potential for scalable low-power neuromorphic computing","author":[{"family":"Raj","given":"Ravish"},{"family":"Richter","given":"Simon"},{"family":"Ivriq","given":"Saeed"},{"family":"Fridorf","given":"Oliver"},{"family":"Fernández-Khatiboun","given":"Darío"},{"family":"Rezaeiyan","given":"Yasser"},{"family":"Benetti","given":"Luana"},{"family":"Boehnert","given":"Tim"},{"family":"Ferreira","given":"Ricardo"},{"family":"Farkhani","given":"Hooman"},{"family":"Shreya","given":"Sonal"},{"family":"Moradi","given":"Farshad"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2605.29942","URL":"https://doi.org/10.48550/arxiv.2605.29942","source":"datacite"},{"id":"doi:10.48550/arxiv.2605.29127","type":"manuscript","title":"Field-Driven Hybrid Filament Formation Governs Switching in Ta-HfO$_2$-Pt Memristors","abstract":"Memristive devices have gained significant attention for their potential in next-generation non-volatile memory and neuromorphic computing architectures. Among emerging candidates, transition metal oxides have proven particularly promising. While the switching mechanism in Ta/HfO$_2$/Pt devices was long attributed solely to oxygen vacancy based filaments, recent experimental evidence suggests a more complex dual-regime: the diffusion of metal cations also contributes to the formation of a conductive bridge. However, the precise atomistic mechanisms governing this metal cation migration remain poorly understood. Additionally, the role of defects such as oxygen vacancies present in the transition metal oxide in determining the final filament size and shape is also not well understood. Here, we employ molecular dynamics (MD) simulations with dynamic charge transfer to provide a detailed analysis of the atomistic mechanisms governing the co-formation of Ta-cation and oxygen-deficient filaments. We clearly show how varying the initial oxygen vacancy concentrations and spatial configurations within the HfO$_2$ matrix influences the final morphology and dimensions of the conductive filament. The switching is governed by field-driven formation and rupture of a hybrid Ta-cation-rich, oxygen-deficient filament in HfO$_2$. Our simulations closely match experiment, validating the model as a robust framework for understanding switching in oxide memristors and guiding designs that reduce cycle-to-cycle and device-to-device variability -- key barriers to high-performance devices.","author":[{"family":"Amaram","given":"Ashutosh"},{"family":"Koneru","given":"Aditya"},{"family":"Sankaranarayanan","given":"Subramanian"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2605.29127","URL":"https://doi.org/10.48550/arxiv.2605.29127","source":"datacite"},{"id":"doi:10.15480/882.17213","type":"article-journal","title":"Beyond silicon: materials, mechanisms, and methods for physical neural computing","abstract":"Physical implementations of neural computation now extend far beyond silicon hardware, encompassing substrates such as memristive devices, photonic circuits, mechanical metamaterials, microfluidic networks, chemical reaction systems, and living neural tissue. By exploiting intrinsic physical processes, such as charge transport, wave interference, elastic deformation, mass transport, and biochemical regulation, these substrates can realize neural inference and adaptation directly in matter. As silicon GPU-centered AI faces growing energy and data-movement constraints, physical neural computation becomes increasingly relevant as a complementary path beyond conventional digital accelerators. This trend is driven in particular by pervasive intelligence, i.e., the deployment of on-device and edge AI across large numbers of resource-constrained systems. In such settings, co-locating computation with sensing and memory can reduce data shuttling and improve efficiency. Meanwhile, physical neural approaches have emerged across disparate disciplines, yet progress remains fragmented, with limited shared terminology and few principled ways to compare platforms. This survey unifies the field by mapping neural primitives to substrate-specific mechanisms, analyzing architectural and training paradigms, and identifying key engineering constraints including scalability, precision, programmability, and I/O interfacing overhead. To enable cross-domain comparison, we introduce a first-order benchmarking scheme based on standardized static and dynamic tasks and physically interpretable performance dimensions. We show that no single substrate dominates across the considered dimensions; instead, physical neural systems occupy complementary operating regimes, enabling applications ranging from ultrafast signal processing and in-memory inference to embodied control and in-sample biochemical decision making.","author":[{"family":"Fischer","given":"Stefan"},{"family":"Ay","given":"Nihat"},{"family":"Landsiedel","given":"Olaf"},{"family":"Mohammadi","given":"Esfandiar"},{"family":"Otte","given":"Sebastian"},{"family":"Renner","given":"Bernd"},{"family":"Russwinkel","given":"Nele"}],"issued":{"date-parts":[[2026]]},"DOI":"10.15480/882.17213","URL":"https://doi.org/10.15480/882.17213","source":"datacite"},{"id":"doi:10.57760/sciencedb.hep.00013","type":"article-journal","title":"Supplyment to the article\"Polarization-sensitive photo-synapses based on anisotropic β-Ga2O3 for dynamic visual perception\"","abstract":"Supplyment to the article\"Polarization, as a fundamental property of light, carries abundant environmental and target-specific information that is invisible to the human eye but crucial for advanced vision. In biological visual systems, such polarization information is effectively encoded, processed, and integrated to enhance scene perception and target recognition. Inspired by this biological paradigm, polarization sensitive optoelectronic synapses provide a promising pathway toward information perception and neuromorphic computing. Benefiting from its low-symmetry monoclinic lattice, β-Ga2O3 inherently exhibits strong optical absorption anisotropy, which can be used to enable photo-synapses. Here, the structural anisotropy of β-Ga2O3 single crystals with (100), (010), and (001) orientations was systematically investigated through atomic arrangement analysis, polarization-resolved Raman spectroscopy, polarization-dependent absorption, and XPS characterization. Compared with the (010) and (001) orientations, the (100) β-Ga2O3 crystal exhibited stronger optical absorption anisotropy and a higher density of oxygen vacancies, making it a favorable candidate for constructing polarization-sensitive neuromorphic synapses. A solar-blind polarization-sensitive optoelectronic synapse was developed in this work, demonstrating polarization-dependent excitatory postsynaptic currents (EPSC), paired-pulse facilitation (PPF), and learning-forgetting-relearning functionalities. Furthermore, the optoelectronic synapse enabled the construction of a neuromorphic visual system (NVS), achieving noise suppression, enhanced image quality, and handwritten digit recognition accuracy up to 98.74%. Beyond static recognition, integration of the arrays into a reservoir computing system allowed efficient motion direction perception with accuracies approaching 99.7%. These results highlight the potential of anisotropic β-Ga2O3 as a material foundation for next-generation solar-blind polarization-sensitive neuromorphic vision systems.\"","author":[{"family":"Shen","given":"Xianchun"},{"family":"Wu","given":"Chao"},{"family":"Liu","given":"Yanjie"},{"family":"Yu","given":"Zhihao"},{"family":"Wang","given":"Zhenyang"},{"family":"Guo","given":"Daoyou"}],"issued":{"date-parts":[[2026]]},"DOI":"10.57760/sciencedb.hep.00013","URL":"https://doi.org/10.57760/sciencedb.hep.00013","source":"datacite"},{"id":"doi:10.48550/arxiv.2605.20802","type":"manuscript","title":"ELSA: An ELastic SNN Inference Architecture for Efficient Neuromorphic Computing","abstract":"Spiking neural networks (SNNs) exploit event-driven and addition-only computation to substantially improve efficiency for intelligent computation. A key temporal property of SNNs, elastic inference, allows outputs to emerge progressively, enabling responses to salient inputs much earlier than full evaluation. However, existing SNN-specific accelerators cannot capitalize on this property. Layer-by-layer designs emit outputs only after all layers are complete, while time-step-by-time-step designs rely on coarse-grained, layer-wise pipelines that require synchronizing all spines/tokens within a layer. This barrier prevents results from being forwarded immediately, delaying the earliest possible response and forfeiting the benefits of elastic inference. To address these challenges, we propose ELSA, a near-SRAM dataflow architecture that realizes true elastic inference through a fine-grained spine/token-wise pipeline and hardware optimizations tailored to SNNs. ELSA forwards each spine/token immediately upon production, forming a continuous streaming pipeline that substantially reduces the latency to the first response. To enhance this lightweight execution, ELSA introduces a bundled address event representation protocol to lower communication traffic of network-on-chip (NoC), and leverages mini-batch spiking Gustavson-product to cut memory access and exploit inherent sparsity. Combined with mapping and scheduling optimizations, ELSA achieves efficient, event-driven computation without compromising accuracy. Experiments show that SNNs can outperform quantized artificial neural networks (QANNs) while maintaining on-par accuracy. For a 4-bit ResNet-50, ELSA achieves 3.4$\\times$ speedup and 13.6$\\times$ higher energy efficiency over the SOTA QANN accelerator (ANT), and 2.9$\\times$ speedup and 22.1$\\times$ energy efficiency gains over the SOTA SNN accelerator (PAICORE).","author":[{"family":"You","given":"Kang"},{"family":"Nie","given":"Chen"},{"family":"Yan","given":"Lee"},{"family":"Wei","given":"Ziling"},{"family":"Zou","given":"Cheng"},{"family":"Xu","given":"Zekai"},{"family":"Feng","given":"Yu"},{"family":"Jiang","given":"Honglan"},{"family":"He","given":"Zhezhi"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2605.20802","URL":"https://doi.org/10.48550/arxiv.2605.20802","source":"datacite"},{"id":"doi:10.17863/cam.119706","type":"article-journal","title":"Research data supporting \"Controlled Oxygen Vacancy Electrode Reservoir for Robust WO3-based Memory Devices\"","abstract":"This dataset contains the raw and processed data used to evaluate resistive switching, synaptic behaviour, electronic structure, optical response, and microstructure in the thin-film memory devices. The dataset is organized by figure number, containing the raw experimental measurement data used to generate the corresponding main-text or supplementary plots. The file names generally encode the sample stack, device identifier, measurement type, and, where relevant, voltage or pulse conditions. Details of folder contents are as follows: Fig1b: contains consecutive current–voltage (I–V) switching cycles. Fig1c: contains I–V data collected from multiple devices with different top-electrode diameters. Each subfolder corresponds to an individual device location and contains repeated I–V scans. Fig1d: contains processed data used for the forming-voltage summary as a function of ITO oxygen deposition condition. Fig1e: contains pulsed endurance measurement files. Fig1f: contains processed endurance statistics as a function of ITO oxygen deposition condition. Fig1g: contains the sources of the comparison used in the performance map. Fig2a: contains conductance data measured under repeated identical programming pulses. These files record conductance evolution during analog switching. Fig2b: contains multilevel retention data. Fig2c: contains conductance data obtained under pulse trains with gradually increasing amplitude. Fig2d: contains the sources of the comparison used in the performance map. Fig2e: contains paired-pulse facilitation and paired-pulse depression data. Fig2f: contains spike-timing-dependent plasticity data. Fig3b, Fig3c: contains Kelvin probe force microscopy data and related processed outputs used to evaluate surface potential changes. Fig3d: contains ultraviolet photoelectron spectroscopy data. Fig3e: contains four-point probe resistivity data. Fig3f: contains operando Raman spectroscopy data collected after increasing numbers of switching cycles. These spectra track changes in WO(_3) vibrational modes during repeated electrical stressing and provide evidence of defect-related structural evolution. Fig3g: contains operando photoluminescence data collected after increasing numbers of switching cycles. These files capture defect-related optical emission changes associated with oxygen-vacancy evolution during device operation. Fig4a: contains cross-sectional transmission electron microscopy images of the optimized device stack. Fig4b-c: contains transmission electron microscopy elemental maps and related compositional imaging files. Fig4d: contains Rutherford backscattering spectrometry data and processed composition profiles. Fig4e: contains X-ray photoelectron spectroscopy data. Fig4f: contains Raman spectroscopy data. Fig4g: contains photoluminescence data. SI: contains supplementary datasets corresponding to additional electrical, structural, spectroscopic, and surface-characterization results. These include extra retention and switching data, fitting results of spectroscopy data, extended transmission electron microscopy datasets, etc. The supplementary folder structure follows the same figure-based organization as the main dataset. File formats: .csv files: mainly contain raw or processed numerical data from electrical measurements, resistivity measurements, and spectroscopy measurement data. .xrdml files: raw X-ray diffraction data exported from the diffractometer. .tif files: raw electron microscopy images and elemental map. .spm files: raw atomic force microscopy-derived files. These files can be opened in Bruker NanoScope Analysis software. .txt files: readme or supporting text files describing figure content, fitting information, or data organization.","author":[{"family":"Yuan","given":"Ziyi"},{"family":"Bakhit","given":"Babak"},{"family":"Lu","given":"Jiahao"},{"family":"Liu","given":"Yixuan"},{"family":"Jan","given":"Atif"},{"family":"Li","given":"Xinjuan"},{"family":"Ducati","given":"Caterina"},{"family":"Di Martino","given":"Giuliana"},{"family":"Hellenbrand","given":"Markus"},{"family":"Driscoll","given":"Judith"}],"issued":{"date-parts":[[2026]]},"DOI":"10.17863/cam.119706","URL":"https://doi.org/10.17863/cam.119706","source":"datacite"},{"id":"doi:10.48550/arxiv.2605.19465","type":"manuscript","title":"Task-specific programming of chaos in neural circuits","abstract":"Chaotic dynamics have emerged as a versatile resource for neuromorphic and probabilistic computing, enabling high-dimensional nonlinear processing and classical analogues of quantum randomness. Exploiting chaos for computation requires task-dependent control over complexity, as demonstrated in reservoir computing, random-number generation, and probabilistic inference. Existing approaches have focused on tuning element-level parameters, leaving the collective, many-body origin of chaos largely unexplored as a design freedom. Here, we demonstrate programmable chaotic dynamics for task-specific reservoir computing. Using a continuous-time neural-circuit model, we show that tuning network topology drives an ordered-to-chaotic transition, accompanied by transitions in correlation timescales, stability characteristics, and signal propagation. By jointly controlling element-level properties and network topology, we establish a unified chaos-latency phase diagram, revealing that small-world connectivity enables low-latency on-off switching of chaos via edge rewiring. Supported by distinct reservoir-computing benchmarks across various topological regimes, our results demonstrate that network topology serves as a reconfigurable parameter for task-specific computation and tunable randomness.","author":[{"family":"Kim","given":"Jungyoon"},{"family":"Kim","given":"Kyuho"},{"family":"Park","given":"Kunwoo"},{"family":"Park","given":"Namkyoo"},{"family":"Yu","given":"Sunkyu"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2605.19465","URL":"https://doi.org/10.48550/arxiv.2605.19465","source":"datacite"},{"id":"doi:10.48550/arxiv.2605.17752","type":"manuscript","title":"Optical Neural Networks from Coherent Transient Dynamics in Waveguide QED","abstract":"Optical neural networks promise ultrafast, low-energy information processing by performing computation directly with photons. Current implementations, however, are largely restricted to steady-state operation and rely on high-latency electro-optical conversion for nonlinear activation. To address these limitations, we propose an all-optical fully connected neural network architecture in which the basic neuronal functions are realized by coherent transient quantum dynamics. Within this framework, phase-tunable nonlocal interference in a giant cavity implements programmable synaptic weights; an integrator operating in the bad cavity regime performs temporal summation by coherently combining sequential wavepackets; and transient Rabi dynamics of a driven two-level system provide nonlinear activation. Full-physics simulations demonstrate high classification accuracy on MNIST and colored-object recognition tasks. These results eliminate the optoelectronic activation bottleneck, reduce latency, and establish transient light-matter dynamics as a native physical resource for high-dimensional nonlinear information processing, paving the way toward fully optical neuromorphic computing.","author":[{"family":"Cao","given":"Jiande"},{"family":"Zeng","given":"Yexiong"},{"family":"Nori","given":"Franco"},{"family":"Xiang","given":"Ze"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2605.17752","URL":"https://doi.org/10.48550/arxiv.2605.17752","source":"datacite"},{"id":"doi:10.1002/advs.77116","type":"article-journal","title":"Neuromorphic Devices and Computing for Sensing, Memory, and Control.","abstract":"Neuromorphic devices are bioinspired electronic systems that mimic key structures and functions of the nervous system, enabling integration and communication between living tissues and machines. This review examines how neuromorphic devices and computing are designed to emulate the structure, organization, and function of the nervous system. For neuromorphic devices, we first describe strategies that mimic subcellular neural functions. We then highlight how device architecture recapitulates biological topology from subcellular components to brain networks. We next summarize how neuromorphic devices emulate sensory and sensorimotor (sensory-modulation) neural circuits. For neuromorphic computing, we review recent advances in artificial and biological neuromorphic computing, including spiking neural networks, bioinspired learning algorithms, and applications. We also discuss emerging biohybrid intelligence systems leveraging two- and three-dimensional biological networks as computational units. Finally, we outline key challenges, potential milestones, and future directions.","author":[{"family":"Zhu","given":"Zhengguang"},{"family":"Schaffer","given":"Nicholas"},{"family":"Yang","given":"Xiao"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1002/advs.77116","URL":"https://doi.org/10.1002/advs.77116","source":"europepmc"},{"id":"doi:10.48550/arxiv.2602.18198","type":"manuscript","title":"Adaptive transitions in FitzHugh-Nagumo networks with Hebb-Oja coupling rules","abstract":"Adaptive coupling in networks of interacting neurons has gained recent attention due to the many applications both in biological and in artificial neural networks, where adaptive coupling or synaptic plasticity is considered as a key factor in learning processes. In the present study, we apply adaptive connectivity rules in networks of interacting FitzHugh-Nagumo oscillators. Adaptive coupling, here, is realized via Hebbian learning adjusted by the Oja rule to prevent the network link weights from growing without bounds. Numerical investigations demonstrate that during the adaptation process the FitzHugh-Nagumo network undergoes adaptive transitions realizing traveling waves, synchronized states and chimera states transiting through various multiplicities. These transitions become more evident when the time scales governing the coupling dynamics are much slower than the ones governing the nodal dynamics (nodal potentials). Namely, when the coupling time scales are slow, the network has the time to realize and demonstrate different synchronization regimes before reaching the final steady state. The transitions can be observed not only in the spacetime plots but also in the abrupt changes of the average coupling weights as the network evolves in time. Regarding the asymptotic coupling distributions, we show that the limiting average coupling strength follows an inverse power law with respect to the Oja parameter (also called \"forgetting\" parameter) which balances the learning growth. We also report abrupt transitions in the asymptotic coupling strengths when the parameter related to adaptive coupling crosses from fast to slow time scales. These findings are in line with previous studies on spiking neural networks.","author":[{"family":"Provata","given":"Astero"},{"family":"Boulougouris","given":"George"},{"family":"Hizanidis","given":"Johanne"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2602.18198","URL":"https://doi.org/10.48550/arxiv.2602.18198","source":"datacite"},{"id":"doi:10.48693/883","type":"article-journal","title":"Continual familiarity decoding from recurrent connections in spiking networks","abstract":"Familiarity memory enables recognition of previously encountered inputs as familiar without recalling detailed stimuli information, which supports adaptive behavior across various timescales. We present a spiking neural network model with lateral connectivity shaped by unsupervised spike-timing-dependent plasticity (STDP) that encodes familiarity via local plasticity events. We show that familiarity can be decoded from network activity using both frequency (spike count) and temporal (spike synchrony) characteristics of spike trains. Temporal coding demonstrates enhanced performance under sparse input conditions, consistent with the principles of sparse coding observed in the brain. We also show how connectivity structure supports each decoding strategy, revealing different plasticity regimes. Our approach outperforms LSTM in temporal generalizability on the continual familiarity detection task, with input stimuli being naturally encoded in the recurrent connectivity without a separate training stage.","author":[{"family":"Zemliak","given":"Viktoria"},{"family":"Pipa","given":"Gordon"},{"family":"Nieters","given":"Pascal"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48693/883","URL":"https://doi.org/10.48693/883","source":"datacite"},{"id":"doi:10.48550/arxiv.2505.21185","type":"manuscript","title":"Constructive community race: full-density spiking neural network model drives neuromorphic computing","abstract":"The local circuitry of the mammalian brain is a focus of the search for generic computational principles because it is largely conserved across species and modalities. In 2014 a model was proposed representing all neurons and synapses of the stereotypical cortical microcircuit below $1\\,\\text{mm}^2$ of brain surface. The model reproduces fundamental features of brain activity but its impact remained limited because of its computational demands. For theory and simulation, however, the model was a breakthrough because it removes uncertainties of downscaling, and larger models are less densely connected. This sparked a race in the neuromorphic computing community and the model became a de facto standard benchmark. Within a few years real-time performance was reached and surpassed at significantly reduced energy consumption. We review how the computational challenge was tackled by different simulation technologies and derive guidelines for the next generation of benchmarks and other domains of science.","author":[{"family":"Senk","given":"Johanna"},{"family":"Kurth","given":"Anno"},{"family":"Furber","given":"Steve"},{"family":"Gemmeke","given":"Tobias"},{"family":"Golosio","given":"Bruno"},{"family":"Heittmann","given":"Arne"},{"family":"Knight","given":"James"},{"family":"Müller","given":"Eric"},{"family":"Noll","given":"Tobias"},{"family":"Nowotny","given":"Thomas"},{"family":"Coppola","given":"Gorka"},{"family":"Peres","given":"Luca"},{"family":"Rhodes","given":"Oliver"},{"family":"Rowley","given":"Andrew"},{"family":"Schemmel","given":"Johannes"},{"family":"Stadtmann","given":"Tim"},{"family":"Tetzlaff","given":"Tom"},{"family":"Tiddia","given":"Gianmarco"},{"family":"Van Albada","given":"Sacha"},{"family":"Villamar","given":"José"},{"family":"Diesmann","given":"Markus"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2505.21185","URL":"https://doi.org/10.48550/arxiv.2505.21185","source":"datacite"},{"id":"doi:10.48550/arxiv.2602.13261","type":"manuscript","title":"A feedback control optimizer for online and hardware-aware training of Spiking Neural Networks","abstract":"Unlike traditional artificial neural networks (ANNs), biological neuronal networks solve complex cognitive tasks with sparse neuronal activity, recurrent connections, and local learning rules. These mechanisms serve as design principles in Neuromorphic computing, which addresses the critical challenge of energy consumption in modern computing. However, most mixed-signal neuromorphic devices rely on semi- or unsupervised learning rules, which are ineffective for optimizing hardware in supervised learning tasks. This lack of scalable solutions for on-chip learning restricts the potential of mixed-signal devices to enable sustainable, intelligent edge systems. To address these challenges, we present a novel learning algorithm for Spiking Neural Networks (SNNs) on mixed-signal devices that integrates spike-based weight updates with feedback control signals. In our framework, a spiking controller generates feedback signals to guide SNN activity and drive weight updates, enabling scalable and local on-chip learning. We first evaluate the algorithm on various classification tasks, demonstrating that single-layer SNNs trained with feedback control achieve performance comparable to artificial neural networks (ANNs). We then assess its implementation on mixed-signal neuromorphic devices by testing network performance in continuous online learning scenarios and evaluating resilience to hyperparameter mismatches. Our results show that the feedback control optimizer is compatible with neuromorphic applications, advancing the potential for scalable, on-chip learning solutions in edge applications.","author":[{"family":"Saponati","given":"Matteo"},{"family":"De Luca","given":"Chiara"},{"family":"Indiveri","given":"Giacomo"},{"family":"Grewe","given":"Benjamin"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2602.13261","URL":"https://doi.org/10.48550/arxiv.2602.13261","source":"datacite"},{"id":"doi:10.48550/arxiv.2510.17392","type":"manuscript","title":"ReLANCE: A Resource-Efficient Low-Latency Cortical Neural Acceleration Engine","abstract":"We present a Cortical Neural Pool (CNP) architecture featuring a high-speed, resource-efficient CORDIC based Hodgkin-Huxley (RCHH) neuron model. Unlike shared CORDIC-based DNN approaches, the proposed neuron leverages modular and performance-optimised CORDIC stages with a latency-area trade-off. We introduce a novel Constraint-Aware Modular Parallelism (CAMP) with Precision &amp; Stability handling to leverage maximum speedup and utilisation of hardware through hardware software co-design. The FPGA implementation of the RCHH neuron shows 24.5% LUT reduction and 35.2% improved speed, compared to SoTA designs, with 70% better normalised root mean square error (NRMSE). Furthermore, the CNP exhibits 2.85x higher throughput (12.69 GOPS) than a functionally equivalent CORDIC-based DNN engine, with only a 0.35% accuracy drop relative to the DNN counterpart on the MNIST dataset. The overall results indicate that the design shows biologically accurate, low-resource spiking neural network implementations for resource-constrained edge AI applications. The reproducibility codes are publicly available at https://github.com/mukullokhande99/CNP RCHH, facilitating rapid integration and further development by researchers.","author":[{"family":"Kumar","given":"Sonu"},{"family":"Nair","given":"Arjun"},{"family":"Chaudhary","given":"Bhawna"},{"family":"Lokhande","given":"Mukul"},{"family":"Vishvakarma","given":"Santosh"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2510.17392","URL":"https://doi.org/10.48550/arxiv.2510.17392","source":"datacite"},{"id":"doi:10.17169/refubium-51237","type":"article-journal","title":"A brain-constrained neural model of cognition and language with NEST: transitioning from the Felix framework","abstract":"We introduce a brain-constrained neurocomputational model designed to simulate higher cognitive functions of the human brain, implemented using NEST, a widely used open-source simulator optimised for high-performance spiking neural network simulations. Previously implemented in the custom-built C-based Felix simulation library, transitioning the model to NEST enhances accessibility, reproducibility, and computational efficiency. At the cellular level, the model comprises spiking excitatory neurons and local inhibitory neurons, whereas at the network level, it replicates the structural and functional organisation of 12 cortical regions spanning frontal, temporal, and occipital cortices, along with their associated inter-area connectivity. Additionally, global inhibition mechanisms and neuronal noise are integrated. Learning in the model follows biologically plausible Hebbian plasticity principles, incorporating both long-term potentiation and long-term depression. To validate the NEST implementation, we replicated previous simulation findings obtained with the Felix-based model. The new implementation successfully reproduced the same topographical distribution of cell assemblies following associative learning of object and action words within action and perception systems, replicating a range of previous neuroimaging results. Although the NEST model produced larger cell assemblies than Felix, the overall topographical patterns remained similar, indicating preservation of fundamental network characteristics. Moreover, the transition to NEST significantly enhanced computational efficiency, reducing simulation runtime nearly sixfold compared to Felix. This improvement in computational speed is crucial for expanding the model to include additional cortical regions, such as extending to the right hemisphere, which necessitates increased computational resources.","author":[{"family":"Carriere","given":"Maxime"},{"family":"Dobler","given":"Fynn"},{"family":"Plesser","given":"Hans"},{"family":"Feledyn","given":"Agata"},{"family":"Tomasello","given":"Rosario"},{"family":"Wennekers","given":"Thomas"},{"family":"Pulvermüller","given":"Friedemann"}],"issued":{"date-parts":[[2026]]},"DOI":"10.17169/refubium-51237","URL":"https://doi.org/10.17169/refubium-51237","source":"datacite"},{"id":"doi:10.48550/arxiv.2602.07037","type":"manuscript","title":"Stochastic Spiking Neuron Based SNN Can be Inherently Bayesian","abstract":"Uncertainty in biological neural systems appears to be computationally beneficial rather than detrimental. However, in neuromorphic computing systems, device variability often limits performance, including accuracy and efficiency. In this work, we propose a spiking Bayesian neural network (SBNN) framework that unifies the dynamic models of intrinsic device stochasticity (based on Magnetic Tunnel Junctions) and stochastic threshold neurons to leverage noise as a functional Bayesian resource. Experiments demonstrate that SBNN achieves high accuracy (99.16% on MNIST, 94.84% on CIFAR10) with 8-bit precision. Meanwhile rate estimation method provides a ~20-fold training speedup. Furthermore, SBNN exhibits superior robustness, showing a 67% accuracy improvement under synaptic weight noise and 12% under input noise compared to standard spiking neural networks. Crucially, hardware validation confirms that physical device implementation causes invisible accuracy and calibration loss compared to the algorithmic model. Converting device stochasticity into neuronal uncertainty offers a route to compact, energy-efficient neuromorphic computing under uncertainty.","author":[{"family":"Zheng","given":"Huannan"},{"family":"Liu","given":"Jingli"},{"family":"Yang","given":"Kezhou"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2602.07037","URL":"https://doi.org/10.48550/arxiv.2602.07037","source":"datacite"},{"id":"doi:10.48550/arxiv.2602.07010","type":"manuscript","title":"Learning Alzheimer's Disease Signatures by bridging EEG with Spiking Neural Networks and Biophysical Simulations","abstract":"As the prevalence of Alzheimer's disease (AD) rises, improving mechanistic insight from non-invasive biomarkers is increasingly critical. Recent work suggests that circuit-level brain alterations manifest as changes in electroencephalography (EEG) spectral features detectable by machine learning. However, conventional deep learning approaches for EEG-based AD detection are computationally intensive and mechanistically opaque. Spiking neural networks (SNNs) offer a biologically plausible and energy-efficient alternative, yet their application to AD diagnosis remains largely unexplored. We propose a neuro-bridge framework that links data-driven learning with minimal, biophysically grounded simulations, enabling bidirectional interpretation between machine learning signatures and circuit-level mechanisms in AD. Using resting-state clinical EEG, we train an SNN classifier that achieves competitive performance (AUC = 0.839) and identifies the aperiodic 1/f slope as a key discriminative marker. The 1/f slope reflects excitation-inhibition balance. To interpret this mechanistically, we construct spiking network simulations in which inhibitory-to-excitatory synaptic ratios are systematically varied to emulate healthy, mild cognitive impairment, and AD-like states. Using both membrane potential-based and synaptic current-based EEG proxies, we reproduce empirical spectral slowing and altered alpha organization. Incorporating empirical functional connectivity priors into multi-subnetwork simulations further enhances spectral differentiation, demonstrating that large-scale network topology constrains EEG signatures more strongly than excitation-inhibition balance alone. Overall, this neuro-bridge approach connects SNN-based classification with interpretable circuit simulations, advancing mechanistic understanding of EEG biomarkers while enabling scalable, explainable AD detection.","author":[{"family":"Mamoń","given":"Szymon"},{"family":"Talanov","given":"Max"},{"family":"Crimi","given":"Alessandro"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2602.07010","URL":"https://doi.org/10.48550/arxiv.2602.07010","source":"datacite"},{"id":"doi:10.48550/arxiv.2511.00732","type":"manuscript","title":"FeNN-DMA: A RISC-V SoC for SNN acceleration","abstract":"Spiking Neural Networks (SNNs) are a promising, energy-efficient alternative to standard Artificial Neural Networks (ANNs) and are particularly well-suited to spatio-temporal tasks such as keyword spotting and video classification. However, SNNs have a much lower arithmetic intensity than ANNs and are therefore not well-matched to standard accelerators like GPUs and TPUs. Field Programmable Gate Arrays (FPGAs) are designed for such memory-bound workloads, and here we present a novel, fully-programmable RISC-V-based system-on-chip (FeNN-DMA), tailored to simulating SNNs on modern UltraScale+ FPGAs. We show that FeNN-DMA has comparable resource usage and energy requirements to state-of-the-art fixed-function SNN accelerators, yet it supports more complex neuron models and network topologies, and can simulate up to 16 thousand neurons and 256 million synapses per core. Using this functionality, we demonstrate state-of-the-art classification accuracy on the Spiking Heidelberg Digits, Neuromorphic MNIST and Braille tactile classification tasks.","author":[{"family":"Aizaz","given":"Zainab"},{"family":"Knight","given":"James"},{"family":"Nowotny","given":"Thomas"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2511.00732","URL":"https://doi.org/10.48550/arxiv.2511.00732","source":"datacite"},{"id":"doi:10.48550/arxiv.2602.06405","type":"manuscript","title":"A neuromorphic model of the insect visual system for natural image processing","abstract":"Insect vision supports complex behaviors including associative learning, navigation, and object detection, and has long motivated computational models for understanding biological visual processing. However, many contemporary models prioritize task performance while neglecting biologically grounded processing pathways. Here, we introduce a bio-inspired vision model that captures principles of the insect visual system to transform dense visual input into sparse, discriminative codes. The model is trained using a fully self-supervised contrastive objective, enabling representation learning without labeled data and supporting reuse across tasks without reliance on domain-specific classifiers. We evaluated the resulting representations on flower recognition tasks and natural image benchmarks. The model consistently produced reliable sparse codes that distinguish visually similar inputs. To support different modelling and deployment uses, we have implemented the model as both an artificial neural network and a spiking neural network. In a simulated localization setting, our approach outperformed a simple image downsampling comparison baseline, highlighting the functional benefit of incorporating neuromorphic visual processing pathways. Collectively, these results advance insect computational modelling by providing a generalizable bio-inspired vision model capable of sparse computation across diverse tasks.","author":[{"family":"Hines","given":"Adam"},{"family":"Nordström","given":"Karin"},{"family":"Barron","given":"Andrew"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2602.06405","URL":"https://doi.org/10.48550/arxiv.2602.06405","source":"datacite"},{"id":"doi:10.48550/arxiv.2509.05356","type":"manuscript","title":"Spiking Neural Networks for Continuous Control via End-to-End Model-Based Learning","abstract":"Despite recent progress in training spiking neural networks (SNNs) for classification, their application to continuous motor control remains limited. Here, we demonstrate that fully spiking architectures can be trained end-to-end to control robotic arms with multiple degrees of freedom in continuous environments. Our predictive-control framework combines Leaky Integrate-and-Fire dynamics with surrogate gradients, jointly optimizing a forward model for dynamics prediction and a policy network for goal-directed action. We evaluate this approach on both a planar 2D reaching task and a simulated 6-DOF Franka Emika Panda robot with torque control. In direct comparison to non-spiking recurrent baselines trained under the same predictive-control pipeline, the proposed SNN achieves comparable task performance while using substantially fewer parameters. An extensive ablation study highlights the role of initialization, learnable time constants, adaptive thresholds, and latent-space compression as key contributors to stable training and effective control. Together, these findings establish spiking neural networks as a viable and scalable substrate for high-dimensional continuous control, while emphasizing the importance of principled architectural and training design.","author":[{"family":"Huebotter","given":"Justus"},{"family":"Lanillos","given":"Pablo"},{"family":"Van Gerven","given":"Marcel"},{"family":"Thill","given":"Serge"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2509.05356","URL":"https://doi.org/10.48550/arxiv.2509.05356","source":"datacite"},{"id":"doi:10.48550/arxiv.2505.24161","type":"manuscript","title":"Proxy Target: Bridging the Gap Between Discrete Spiking Neural Networks and Continuous Control","abstract":"Spiking Neural Networks (SNNs) offer low-latency and energy-efficient decision making on neuromorphic hardware, making them attractive for Reinforcement Learning (RL) in resource-constrained edge devices. However, most RL algorithms for continuous control are designed for Artificial Neural Networks (ANNs), particularly the target network soft update mechanism, which conflicts with the discrete and non-differentiable dynamics of spiking neurons. We show that this mismatch destabilizes SNN training and degrades performance. To bridge the gap between discrete SNNs and continuous-control algorithms, we propose a novel proxy target framework. The proxy network introduces continuous and differentiable dynamics that enable smooth target updates, stabilizing the learning process. Since the proxy operates only during training, the deployed SNN remains fully energy-efficient with no additional inference overhead. Extensive experiments on continuous control benchmarks demonstrate that our framework consistently improves stability and achieves up to $32\\%$ higher performance across various spiking neuron models. Notably, to the best of our knowledge, this is the first approach that enables SNNs with simple Leaky Integrate and Fire (LIF) neurons to surpass their ANN counterparts in continuous control. This work highlights the importance of SNN-tailored RL algorithms and paves the way for neuromorphic agents that combine high performance with low power consumption. Code is available at https://github.com/xuzijie32/Proxy-Target.","author":[{"family":"Xu","given":"Zijie"},{"family":"Bu","given":"Tong"},{"family":"Hao","given":"Zecheng"},{"family":"Ding","given":"Jianhao"},{"family":"Yu","given":"Zhaofei"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2505.24161","URL":"https://doi.org/10.48550/arxiv.2505.24161","source":"datacite"},{"id":"doi:10.48550/arxiv.2602.02439","type":"manuscript","title":"Energy-Efficient Neuromorphic Computing for Edge AI: A Framework with Adaptive Spiking Neural Networks and Hardware-Aware Optimization","abstract":"Edge AI applications increasingly require ultra-low-power, low-latency inference. Neuromorphic computing based on event-driven spiking neural networks (SNNs) offers an attractive path, but practical deployment on resource-constrained devices is limited by training difficulty, hardware-mapping overheads, and sensitivity to temporal dynamics. We present NeuEdge, a framework that combines adaptive SNN models with hardware-aware optimization for edge deployment. NeuEdge uses a temporal coding scheme that blends rate and spike-timing patterns to reduce spike activity while preserving accuracy, and a hardware-aware training procedure that co-optimizes network structure and on-chip placement to improve utilization on neuromorphic processors. An adaptive threshold mechanism adjusts neuron excitability from input statistics, reducing energy consumption without degrading performance. Across standard vision and audio benchmarks, NeuEdge achieves 91-96% accuracy with up to 2.3 ms inference latency on edge hardware and an estimated 847 GOp/s/W energy efficiency. A case study on an autonomous-drone workload shows up to 312x energy savings relative to conventional deep neural networks while maintaining real-time operation.","author":[{"family":"Imanov","given":"Olaf"},{"family":"Kulali","given":"Derya"},{"family":"Yilmaz","given":"Taner"},{"family":"Erisken","given":"Duygu"},{"family":"Turhan","given":"Rana"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2602.02439","URL":"https://doi.org/10.48550/arxiv.2602.02439","source":"datacite"},{"id":"doi:10.5281/zenodo.17627865","type":"article-journal","title":"Scalable Construction of Spiking Neural Networks using up to thousands of GPUs archive","abstract":"Diverse scientific and engineering research areas deal with discrete, time-stamped changes in large systems of interacting delay differential equations. Simulating such complex systems at scale on high-performance computing clusters demands efficient management of communication and memory. Inspired by the human cerebral cortex — a sparsely connected network of O(10^10) neurons, each forming O(10^3)--O(10^4) synapses and communicating via short electrical pulses called spikes — we study the simulation of large-scale spiking neural networks for computational neuroscience research. This work presents a novel network construction method for multi-GPU clusters and upcoming exascale supercomputers using the Message Passing Interface (MPI), where each process builds its local connectivity and prepares the data structures for efficient spike exchange across the cluster during state propagation. We demonstrate scaling performance of two cortical models using point-to-point and collective communication, respectively.","author":[{"family":"Tiddia","given":"Gianmarco"},{"family":"Villamar","given":"José"},{"family":"Golosio","given":"Bruno"},{"family":"Pontisso","given":"Luca"},{"family":"Simula","given":"Francesco"},{"family":"Babu","given":"Pooja"},{"family":"Pastorelli","given":"Elena"},{"family":"Morrison","given":"Abigail"},{"family":"Diesmann","given":"Markus"},{"family":"Lonardo","given":"Alessandro"},{"family":"Paolucci","given":"Pier"},{"family":"Senk","given":"Johanna"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.17627865","URL":"https://doi.org/10.5281/zenodo.17627865","source":"datacite"},{"id":"doi:10.48550/arxiv.2601.21548","type":"manuscript","title":"Training slow silicon neurons to control extremely fast robots with spiking reinforcement learning","abstract":"Air hockey demands split-second decisions at high puck velocities, a challenge we address with a compact network of spiking neurons running on a mixed-signal analog/digital neuromorphic processor. By co-designing hardware and learning algorithms, we train the system to achieve successful puck interactions through reinforcement learning in a remarkably small number of trials. The network leverages fixed random connectivity to capture the task's temporal structure and adopts a local e-prop learning rule in the readout layer to exploit event-driven activity for fast and efficient learning. The result is real-time learning with a setup comprising a computer and the neuromorphic chip in-the-loop, enabling practical training of spiking neural networks for robotic autonomous systems. This work bridges neuroscience-inspired hardware with real-world robotic control, showing that brain-inspired approaches can tackle fast-paced interaction tasks while supporting always-on learning in intelligent machines.","author":[{"family":"Ambrosini","given":"Irene"},{"family":"Blakowski","given":"Ingo"},{"family":"Zendrikov","given":"Dmitrii"},{"family":"Capone","given":"Cristiano"},{"family":"Gava","given":"Luna"},{"family":"Indiveri","given":"Giacomo"},{"family":"De Luca","given":"Chiara"},{"family":"Bartolozzi","given":"Chiara"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2601.21548","URL":"https://doi.org/10.48550/arxiv.2601.21548","source":"datacite"},{"id":"doi:10.48550/arxiv.2601.20870","type":"manuscript","title":"STAER: Temporal Aligned Rehearsal for Continual Spiking Neural Network","abstract":"Spiking Neural Networks (SNNs) are inherently suited for continuous learning due to their event-driven temporal dynamics; however, their application to Class-Incremental Learning (CIL) has been hindered by catastrophic forgetting and the temporal misalignment of spike patterns. In this work, we introduce Spiking Temporal Alignment with Experience Replay (STAER), a novel framework that explicitly preserves temporal structure to bridge the performance gap between SNNs and ANNs. Our approach integrates a differentiable Soft-DTW alignment loss to maintain spike timing fidelity and employs a temporal expansion and contraction mechanism on output logits to enforce robust representation learning. Implemented on a deep ResNet19 spiking backbone, STAER achieves state-of-the-art performance on Sequential-MNIST and Sequential-CIFAR10. Empirical results demonstrate that our method matches or outperforms strong ANN baselines (ER, DER++) while preserving biologically plausible dynamics. Ablation studies further confirm that explicit temporal alignment is critical for representational stability, positioning STAER as a scalable solution for spike-native lifelong learning. Code is available at https://github.com/matteogianferrari/staer.","author":[{"family":"Gianferrari","given":"Matteo"},{"family":"Moussadek","given":"Omayma"},{"family":"Salami","given":"Riccardo"},{"family":"Fiorini","given":"Cosimo"},{"family":"Tartarini","given":"Lorenzo"},{"family":"Gandolfi","given":"Daniela"},{"family":"Calderara","given":"Simone"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2601.20870","URL":"https://doi.org/10.48550/arxiv.2601.20870","source":"datacite"},{"id":"doi:10.5061/dryad.z612jm6r0","type":"article-journal","title":"Ventral pallidum efferent pathways via mediodorsal thalamus and lateral habenula mediate distinct aspects of default mode network regulation","abstract":"Transitions between internal and external focus are fundamental to cognition. These shifts depend on the modulation of the Default Mode Network (DMN), of which the ventral pallidum (VP) is a key subcortical node. Here, we examine which VP efferent pathways mediate these transitions, using projection and cell-type-specific optogenetic silencing. We found that inhibition of the VP projection to lateral habenula (LHb) confers a learning advantage early during task acquisition. Conversely, silencing VP projections to mediodorsal thalamus (MD) improves performance during the late stage of the same task. Downregulation of the cholinergic VP population improved performance during both early and late stages. Thus, silencing these VP outputs promotes escape from a DMN brain state, facilitating attention to external stimuli. Our results confirm a role for VP in DMN regulation and indicate that MD and LHb VP efferent pathways, in concert with cholinergic neuromodulation, mediate different aspects of DMN regulation.","author":[{"family":"Makedona","given":"Epistimi"},{"family":"Kuo","given":"Mu"},{"family":"Harvey","given":"Michael"},{"family":"Rainer","given":"Gregor"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5061/dryad.z612jm6r0","URL":"https://doi.org/10.5061/dryad.z612jm6r0","source":"datacite"},{"id":"doi:10.5061/dryad.x0k6djhwb","type":"article-journal","title":"Impact of the entorhinal feed-forward connection to the CA3 on hippocampal coding","abstract":"Each sub-region of the hippocampus plays a critical computational role in the formation of episodic learning and memory, but studies have yet to show and interpret the individual spiking dynamics of each region and how that information is passed between each subregion. This is in part due to the difficulty in accessing individual communicating axons. Here, we created a novel microfluidic device that facilitates network growth of four separated hippocampal subregions over a micro-electrode array. This device enabled monitoring single axons over two electrodes so direction of spike propagation in interregional communication could be ascertained. In this in vitro hippocampal study, we compared spiking dynamics across two novel four-compartment device architectures: one with four sets of axon tunnels between subregions that excluded the perforant pathway from EC-CA3, and one with five sets of axon tunnels that included the EC-CA3 connection. We found 30-90% faster feed-forward firing rates (shorter interspike intervals) in axons in the five-tunnel model with 35-75% slower bursting dynamics (longer interburst intervals) compared to the four-tunnel model. Comparing the percentage of spikes in bursts between array designs, the five-tunnel architecture showed that the CA3-CA1 and CA1-EC axons had more spikes in bursts than the four-tunnel counterpart suggesting more structured information transfer. Feedback firing rates were similar between configurations. The faster feed-forward inter-regional spiking in the more natural five-tunnel configuration than with four-tunnel suggests tighter control of spiking and possibly more precise communication between subregions.","author":[{"family":"Lassers","given":"Samuel"},{"family":"Khatri","given":"Shazfa"},{"family":"Chen","given":"Ruiyi"},{"family":"Vakilna","given":"Yash"},{"family":"Tang","given":"William"},{"family":"Brewer","given":"Gregory"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5061/dryad.x0k6djhwb","URL":"https://doi.org/10.5061/dryad.x0k6djhwb","source":"datacite"},{"id":"doi:10.5061/dryad.cjsxksngp","type":"article-journal","title":"Real-time VIM thalamus recordings during peripheral nerve stimulation treatment for essential tremor: DBS intraoperative dataset","abstract":"Background: Essential tremor (ET), the most common movement disorder in adults, presents with involuntary shaking of the upper extremities during postural hold and kinetic tasks linked to dysfunction in the cerebellothalamo-cortical network. Recently, transcutaneous afferent patterned stimulation (TAPS), applied through a wrist-worn device, has emerged as a non-invasive treatment for medication-refractory ET. However, its mechanism remains unclear. Objective: We hypothesize that TAPS reduces tremors through modulation of the VIM thalamus in the cerebellothalamo-cortical network. Methods: Employing refractory pure ET patients seeking VIM deep brain stimulation (DBS), we quantified clinical tremor improvement following TAPS treatment in a pre-operative setting, followed by intra-operative microelectrode recording of the contralateral thalamus with concurrent TAPS treatment on and off. Results: After one preoperative session, TAPS significantly reduces upper limb tremor average (0.61, p = 0.002), with an asymmetric effect favoring the treated limb (p = 0.047) and the greatest improvement tending to kinetic tremor (R2 = 0.943, p = 0.002). The magnitude of TAPS-related tremor reduction demonstrates a positive correlation with the modulation of alpha (R2 = 0.213, p &lt; 0.001) and beta band LFPs (R2 = 0.255, p &lt; 0.001) in the VIM. TAPS also suppressed spiking activity in the VIM (R2 = 0.104, p = 0.029), though it was uncorrelated with the degree of tremor reduction. Of note, TAPS-related modulation of LFPs and spiking activity was greatest near the optimal placement location for the DBS lead in treating ET (R2 = 0.122, p = 0.006). Conclusion: In sum, TAPS likely reduces tremor in ET by modulating the VIM and connected nodes in the cerebello-thalamo-cortical pathway.","author":[{"family":"Luu","given":"Cuong"},{"family":"Ranum","given":"Jordan"},{"family":"Youn","given":"Youngwon"},{"family":"Perrault","given":"Jennifer"},{"family":"Krause","given":"Bryan"},{"family":"Banks","given":"Matthew"},{"family":"Buyan-Dent","given":"Laura"},{"family":"Ludwig","given":"Kip"},{"family":"Lake","given":"Wendell"},{"family":"Suminski","given":"Aaron"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5061/dryad.cjsxksngp","URL":"https://doi.org/10.5061/dryad.cjsxksngp","source":"datacite"},{"id":"doi:10.5061/dryad.8cz8w9h0d","type":"article-journal","title":"Spatial coding dysfunction and network instability in the aging medial entorhinal cortex","abstract":"Across species, spatial memory declines with age, possibly reflecting altered hippocampal and medial entorhinal cortex (MEC) function. However, the integrity of cellular and network-level spatial coding in aged MEC is unknown. Here, we leveraged in vivo electrophysiology to assess MEC function in young, middle-aged, and aged mice navigating virtual environments. In aged grid cells, we observed impaired stabilization of context-specific spatial firing, correlated with spatial memory deficits. Additionally, aged grid networks shifted firing patterns often, but with poor alignment to context changes. Aged spatial firing was also unstable in an unchanging environment. In these same mice, we identified 458 genes differentially expressed with age in MEC, 61 of which had expression correlated with spatial coding quality. These genes were interneuron-enriched and related to synaptic plasticity, notably including a perineuronal net component. Together, these findings identify coordinated transcriptomic, cellular, and network changes in MEC implicated in impaired spatial memory in aging.","author":[{"family":"Herber","given":"Charlotte"},{"family":"Pratt","given":"Karishma"},{"family":"Shea","given":"Jeremy"},{"family":"Villeda","given":"Saul"},{"family":"Giocomo","given":"Lisa"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5061/dryad.8cz8w9h0d","URL":"https://doi.org/10.5061/dryad.8cz8w9h0d","source":"datacite"},{"id":"doi:10.3929/ethz-c-000786698","type":"article-journal","title":"EFLOP: a sparsity-aware metric for evaluating computational cost in spiking and non-spiking neural networks","abstract":"Deploying energy-efficient deep neural networks on energy-constrained edge devices is an important research topic in both machine learning and circuit design communities. Both artificial neural networks (ANNs) and spiking neural networks (SNNs) have been proposed as candidates for these tasks. In particular, SNNs are considered energy-efficient because they leverage temporal sparsity in their outputs. However, existing computational frameworks fail to accurately estimate the cost of running sparse networks on modern time-stepped hardware, which exploits sparsity by skipping zero-valued operations. Meanwhile, weight sparsity-aware training remains underexplored for SNNs and lacks systematic benchmarking against optimized ANNs, making fair comparisons between the two paradigms difficult. To bridge this gap, we introduce the effective floating-point operation (EFLOP), a metric that accounts for the sparse operations during pre-activation updates of both ANNs and SNNs. Applying weight sparsity-aware training to both SNNs and ANNs, we achieve up to 8.9× reduction in EFLOPs for gated recurrent unit models and 3.6× for LIF models by sparsifying weights by 80 % , without sacrificing accuracy on the Spiking Heidelberg Digits and Spiking Speech Command datasets. These findings highlight the critical role of network sparsity in designing energy-efficient neural networks and establish EFLOPs as a robust framework for cross-paradigm comparisons.","author":[{"family":"Narduzzi","given":"Simon"},{"family":"Zenke","given":"Friedemann"},{"family":"Liu","given":"Shih"},{"family":"Dunbar","given":"LA"}],"issued":{"date-parts":[[2025]]},"DOI":"10.3929/ethz-c-000786698","URL":"https://doi.org/10.3929/ethz-c-000786698","source":"datacite"},{"id":"doi:10.3929/ethz-b-000749309","type":"article-journal","title":"A Realistic Simulation Framework for Analog/Digital Neuromorphic Architectures","abstract":"Developing dedicated mixed-signal neuromorphic computing systems optimized for real-time sensory-processing in extreme edge-computing applications requires time-consuming design, fabrication, and deployment of full-custom neuromorphic processors. To ensure that initial prototyping efforts exploring the properties of different network architectures and parameter settings lead to realistic results, it is important to use simulation frameworks that match as best as possible the properties of the final hardware. This is particularly challenging for neuromorphic hardware platforms made using mixed-signal analog/digital circuits, due to the variability and noise sensitivity of their components. In this paper, we address this challenge by developing a software spiking neural network simulator explicitly designed to account for the properties of mixed-signal neuromorphic circuits, including device mismatch variability. The simulator, called A Realistic Simulation Framework for Analog/Digital Neuromorphic Architectures, is designed to reproduce the dynamics of mixed-signal synapse and neuron electronic circuits with autogradient differentiation for parameter optimization and GPU acceleration. We demonstrate the effectiveness of this approach by matching software simulation results with measurements made from an existing neuromorphic processor. We show how the results obtained provide a reliable estimate of the behavior of the spiking neural network trained in software, once deployed in hardware. This framework enables the development and innovation of new learning rules and processing architectures in neuromorphic embedded systems.","author":[{"family":"Quintana","given":"Fernando"},{"family":"Galindo","given":"Pedro"},{"family":"Donati","given":"Elisa"},{"family":"Indiveri","given":"Giacomo"},{"family":"Perez-Peña","given":"Fernando"}],"issued":{"date-parts":[[2025]]},"DOI":"10.3929/ethz-b-000749309","URL":"https://doi.org/10.3929/ethz-b-000749309","source":"datacite"},{"id":"doi:10.3929/ethz-c-000786150","type":"article-journal","title":"Neuromorphic dreaming as a pathway to efficient learning in artificial agents","abstract":"The computational substrate of biological systems exhibits remarkable abilities to learn complex skills quickly and efficiently. Inspired by this, we implement model-based reinforcement learning using spiking neural networks directly on mixed-signal neuromorphic hardware. This approach combines energy-efficient electronic circuits with high sample efficiency through alternating online (‘awake’) and offline (‘dreaming’) learning phases. Our model features two networks: an agent network that learns from real and simulated experiences and a world model network that generates simulated experiences. We validate this by training the system to play Atari Pong. First, we establish a baseline using only real experiences. Then, by ‘dreaming’, the required real experiences decrease significantly. The network dynamics runs in real-time on the analog neuromorphic circuits, with only the readout layers implemented and trained on a computer-in-the-loop. We present results that demonstrate the robustness and potential of energy-efficient mixed-signal neuromorphic processors for real-world applications.","author":[{"family":"Blakowski","given":"Ingo"},{"family":"Zendrikov","given":"Dmitrii"},{"family":"Indiveri","given":"Giacomo"},{"family":"Capone","given":"Cristiano"}],"issued":{"date-parts":[[2025]]},"DOI":"10.3929/ethz-c-000786150","URL":"https://doi.org/10.3929/ethz-c-000786150","source":"datacite"},{"id":"doi:10.48550/arxiv.2601.13451","type":"manuscript","title":"Event-based Heterogeneous Information Processing for Online Vision-based Obstacle Detection and Localization","abstract":"This paper introduces a novel framework for robotic vision-based navigation that integrates Hybrid Neural Networks (HNNs) with Spiking Neural Network (SNN)-based filtering to enhance situational awareness for unmodeled obstacle detection and localization. By leveraging the complementary strengths of Artificial Neural Networks (ANNs) and SNNs, the system achieves both accurate environmental understanding and fast, energy-efficient processing. The proposed architecture employs a dual-pathway approach: an ANN component processes static spatial features at low frequency, while an SNN component handles dynamic, event-based sensor data in real time. Unlike conventional hybrid architectures that rely on domain conversion mechanisms, our system incorporates a pre-developed SNN-based filter that directly utilizes spike-encoded inputs for localization and state estimation. Detected anomalies are validated using contextual information from the ANN pathway and continuously tracked to support anticipatory navigation strategies. Simulation results demonstrate that the proposed method offers acceptable detection accuracy while maintaining computational efficiency close to SNN-only implementations, which operate at a fraction of the resource cost. This framework represents a significant advancement in neuromorphic navigation systems for robots operating in unpredictable and dynamic environments.","author":[{"family":"Ahmadvand","given":"Reza"},{"family":"Sharif","given":"Sarah"},{"family":"Banad","given":"Yaser"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2601.13451","URL":"https://doi.org/10.48550/arxiv.2601.13451","source":"datacite"},{"id":"doi:10.48550/arxiv.2601.12156","type":"manuscript","title":"Biological Intuition on Digital Hardware: An RTL Implementation of Poisson-Encoded SNNs for Static Image Classification","abstract":"The deployment of Artificial Intelligence on edge devices (TinyML) is often constrained by the high power consumption and latency associated with traditional Artificial Neural Networks (ANNs) and their reliance on intensive Matrix-Multiply (MAC) operations. Neuromorphic computing offers a compelling alternative by mimicking biological efficiency through event-driven processing. This paper presents the design and implementation of a cycle-accurate, hardware-oriented Spiking Neural Network (SNN) core implemented in SystemVerilog. Unlike conventional accelerators, this design utilizes a Leaky Integrate-and-Fire (LIF) neuron model powered by fixed-point arithmetic and bit-wise primitives (shifts and additions) to eliminate the need for complex floating-point hardware. The architecture features an on-chip Poisson encoder for stochastic spike generation and a novel active pruning mechanism that dynamically disables neurons post-classification to minimize dynamic power consumption. We demonstrate the hardware's efficacy through a fully connected layer implementation targeting digit classification. Simulation results indicate that the design achieves rapid convergence (89% accuracy) within limited timesteps while maintaining a significantly reduced computational footprint compared to traditional dense architectures. This work serves as a foundational building block for scalable, energy-efficient neuromorphic hardware on FPGA and ASIC platforms.","author":[{"family":"Das","given":"Debabrata"},{"family":"Gupta","given":"Arnav"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2601.12156","URL":"https://doi.org/10.48550/arxiv.2601.12156","source":"datacite"},{"id":"doi:10.48550/arxiv.2601.09248","type":"manuscript","title":"Hybrid guided variational autoencoder for visual place recognition","abstract":"Autonomous agents such as cars, robots and drones need to precisely localize themselves in diverse environments, including in GPS-denied indoor environments. One approach for precise localization is visual place recognition (VPR), which estimates the place of an image based on previously seen places. State-of-the-art VPR models require high amounts of memory, making them unwieldy for mobile deployment, while more compact models lack robustness and generalization capabilities. This work overcomes these limitations for robotics using a combination of event-based vision sensors and an event-based novel guided variational autoencoder (VAE). The encoder part of our model is based on a spiking neural network model which is compatible with power-efficient low latency neuromorphic hardware. The VAE successfully disentangles the visual features of 16 distinct places in our new indoor VPR dataset with a classification performance comparable to other state-of-the-art approaches while, showing robust performance also under various illumination conditions. When tested with novel visual inputs from unknown scenes, our model can distinguish between these places, which demonstrates a high generalization capability by learning the essential features of location. Our compact and robust guided VAE with generalization capabilities poses a promising model for visual place recognition that can significantly enhance mobile robot navigation in known and unknown indoor environments.","author":[{"family":"Wang","given":"Ni"},{"family":"You","given":"Zihan"},{"family":"Neftci","given":"Emre"},{"family":"Schoepe","given":"Thorben"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2601.09248","URL":"https://doi.org/10.48550/arxiv.2601.09248","source":"datacite"},{"id":"doi:10.48550/arxiv.2508.13846","type":"manuscript","title":"Stochastic synaptic dynamics under learning","abstract":"Learning is based on synaptic plasticity, which affects and is driven by neural activity. Because pre- and postsynaptic spiking activity is shaped by randomness, the synaptic weights follow a stochastic process, requiring a probabilistic framework to capture the noisy synaptic dynamics. We consider a paradigmatic supervised learning example: a presynaptic neural population impinging in a sequence of episodes on a recurrent network of integrate-and-fire neurons through synapses undergoing spike-timing-dependent plasticity (STDP) with additive potentiation and multiplicative depression. We first analytically compute the drift- and diffusion coefficients for a single synapse within a single episode (microscopic dynamics), mapping the true jump process to a Langevin and the associated Fokker-Planck equations. Leveraging new analytical tools, we include spike-time--resolving cross-correlations between pre- and postsynaptic spikes, which corrects substantial deviations seen in standard theories purely based on firing rates. We then apply this microdynamical description to the network setup in which hetero-associations are trained over one-shot episodes into a feed-forward matrix of STDP synapses connecting to neurons of the recurrent network (macroscopic dynamics). By mapping statistically distinct synaptic populations to instances of the single-synapse process above, we self-consistently determine the joint neural and synaptic dynamics and, ultimately, the time course of memory degradation and the memory capacity. We demonstrate that specifically in the relevant case of sparse coding, our theory can quantitatively capture memory capacities which are strongly overestimated if spike-time--resolving cross-correlations are ignored. [...]","author":[{"family":"Stubenrauch","given":"Jakob"},{"family":"Auer","given":"Naomi"},{"family":"Kempter","given":"Richard"},{"family":"Lindner","given":"Benjamin"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2508.13846","URL":"https://doi.org/10.48550/arxiv.2508.13846","source":"datacite"},{"id":"doi:10.48550/arxiv.2505.17791","type":"manuscript","title":"Bruno: Backpropagation Running Undersampled for Novel device Optimization","abstract":"Recent efforts to improve the efficiency of neuromorphic and machine learning systems have centred on developing of specialised hardware for neural networks. These systems typically feature architectures that go beyond the von Neumann model employed in general-purpose hardware such as GPUs, offering potential efficiency and performance gains. However, neural networks developed for specialised hardware must consider its specific characteristics. This requires novel training algorithms and accurate hardware models, since they cannot be abstracted as a general-purpose computing platform. In this work, we present a bottom-up approach to training neural networks for hardware-based spiking neurons and synapses, built using ferroelectric capacitors (FeCAPs) and resistive random-access memories (RRAMs), respectively. Unlike the common approach of designing hardware to fit abstract neuron or synapse models, we start with compact models of the physical device to model the computational primitives. Based on these models, we have developed a training algorithm (BRUNO) that can reliably train the networks, even when applying hardware limitations, such as stochasticity or low bit precision. We analyse and compare BRUNO with Backpropagation Through Time. We test it on different spatio-temporal datasets. First on a music prediction dataset, where a network composed of ferroelectric leaky integrate-and-fire (FeLIF) neurons is used to predict at each time step the next musical note that should be played. The second dataset consists on the classification of the Braille letters using a network composed of quantised RRAM synapses and FeLIF neurons. The performance of this network is then compared with that of networks composed of LIF neurons. Experimental results show the potential advantages of using BRUNO by reducing the time and memory required to detect spatio-temporal patterns with quantised synapses.","author":[{"family":"Fehlings","given":"Luca"},{"family":"Zhang","given":"Bojian"},{"family":"Gibertini","given":"Paolo"},{"family":"Nicholson","given":"Martin"},{"family":"Covi","given":"Erika"},{"family":"Quintana","given":"Fernando"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2505.17791","URL":"https://doi.org/10.48550/arxiv.2505.17791","source":"datacite"},{"id":"doi:10.48550/arxiv.2601.08447","type":"manuscript","title":"Sleep-Based Homeostatic Regularization for Stabilizing Spike-Timing-Dependent Plasticity in Recurrent Spiking Neural Networks","abstract":"Spike-timing-dependent plasticity (STDP) provides a biologically-plausible learning mechanism for spiking neural networks (SNNs); however, Hebbian weight updates in architectures with recurrent connections suffer from pathological weight dynamics: unbounded growth, catastrophic forgetting, and loss of representational diversity. We propose a neuromorphic regularization scheme inspired by the synaptic homeostasis hypothesis: periodic offline phases during which external inputs are suppressed, synaptic weights undergo stochastic decay toward a homeostatic baseline, and spontaneous activity enables memory consolidation. We demonstrate that this sleep-wake cycle prevents weight saturation while preserving learned structure. Empirically, we find that low to intermediate sleep durations (10-20\\% of training) improve stability on MNIST-like benchmarks in our STDP-SNN model, without any data-specific hyperparameter tuning. In contrast, the same sleep intervention yields no measurable benefit for the surrogate-gradient spiking neural network (SG-SNN). Taken together, these results suggest that periodic, sleep-based renormalization may represent a fundamental mechanism for stabilizing local Hebbian learning in neuromorphic systems, while also indicating that special care is required when integrating such protocols with existing gradient-based optimization methods.","author":[{"family":"Massey","given":"Andreas"},{"family":"Hubin","given":"Aliaksandr"},{"family":"Nichele","given":"Stefano"},{"family":"Sæbø","given":"Solve"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2601.08447","URL":"https://doi.org/10.48550/arxiv.2601.08447","source":"datacite"},{"id":"doi:10.48550/arxiv.2506.19256","type":"manuscript","title":"Temporal Regularization Training: Unleashing the Potential of Spiking Neural Networks","abstract":"Spiking Neural Networks (SNNs) have received widespread attention due to their event-driven and low-power characteristics, making them particularly effective for processing neuromorphic data. Recent studies have shown that directly trained SNNs suffer from severe temporal gradient vanishing and overfitting issues, which fundamentally constrain their performance and generalizability. This paper unveils a temporal regularization training (TRT) memthod, designed to unleash the generalization and performance potential of SNNs through a time-decaying regularization mechanism that prioritizes early timesteps with stronger constraints. We perform theoretical analysis to reveal TRT's ability on mitigating the temporal gradient vanishment. To validate the effectiveness of TRT, we conduct experiments on both static image datasets and dynamic neuromorphic datasets, perform analysis of their results, demonstrating that TRT can effectively mitigate overfitting and help SNNs converge into flatter local minima with better generalizability. Furthermore, we establish a theoretical interpretation of TRT's temporal regularization mechanism by analyzing the temporal information dynamics inside SNNs. We track the Fisher information of SNNs during training process, showing that Fisher information progressively concentrates in early timesteps. The time-decaying regularization mechanism implemented in TRT effectively guides the network to learn robust features in early timesteps with rich information, thereby leading to significant improvements in model generalization.","author":[{"family":"Zhang","given":"Boxuan"},{"family":"Xu","given":"Zhen"},{"family":"Tao","given":"Kuan"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2506.19256","URL":"https://doi.org/10.48550/arxiv.2506.19256","source":"datacite"},{"id":"doi:10.48550/arxiv.2504.00957","type":"manuscript","title":"Enabling Efficient Processing of Spiking Neural Networks with On-Chip Learning on Commodity Neuromorphic Processors for Edge AI Systems","abstract":"The rising demand for energy-efficient edge AI systems (e.g., mobile agents/robots) has increased the interest in neuromorphic computing, since it offers ultra-low power/energy AI computation through spiking neural network (SNN) algorithms on neuromorphic processors. However, their efficient implementation strategy has not been comprehensively studied, hence limiting SNN deployments for edge AI systems. Toward this, we propose a design methodology to enable efficient SNN processing on commodity neuromorphic processors. To do this, we first study the key characteristics of targeted neuromorphic hardware (e.g., memory and compute budgets), and leverage this information to perform compatibility analysis for network selection. Afterward, we employ a mapping strategy for efficient SNN implementation on the targeted processor. Furthermore, we incorporate an efficient on-chip learning mechanism to update the systems' knowledge for adapting to new input classes and dynamic environments. The experimental results show that the proposed methodology leads the system to achieve low latency of inference (i.e., less than 50ms for image classification, less than 200ms for real-time object detection in video streaming, and less than 1ms in keyword recognition) and low latency of on-chip learning (i.e., less than 2ms for keyword recognition), while incurring less than 250mW of processing power and less than 15mJ of energy consumption across the respective different applications and scenarios. These results show the potential of the proposed methodology in enabling efficient edge AI systems for diverse application use-cases.","author":[{"family":"Putra","given":"Rachmad"},{"family":"Wickramasinghe","given":"Pasindu"},{"family":"Shafique","given":"Muhammad"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2504.00957","URL":"https://doi.org/10.48550/arxiv.2504.00957","source":"datacite"},{"id":"doi:10.48550/arxiv.2504.00948","type":"manuscript","title":"QSViT: A Methodology for Quantizing Spiking Vision Transformers","abstract":"Vision Transformer (ViT)-based models have shown state-of-the-art performance (e.g., accuracy) in vision-based AI tasks. However, realizing their capability in resource-constrained embedded AI systems is challenging due to their inherent large memory footprints and complex computations, thereby incurring high power/energy consumption. Recently, Spiking Vision Transformer (SViT)-based models have emerged as alternate low-power ViT networks. However, their large memory footprints still hinder their applicability for resource-constrained embedded AI systems. Therefore, there is a need for a methodology to compress SViT models without degrading the accuracy significantly. To address this, we propose QSViT, a novel design methodology to compress the SViT models through a systematic quantization strategy across different network layers. To do this, our QSViT employs several key steps: (1) investigating the impact of different precision levels in different network layers, (2) identifying the appropriate base quantization settings for guiding bit precision reduction, (3) performing a guided quantization strategy based on the base settings to select the appropriate quantization setting, and (4) developing an efficient quantized network based on the selected quantization setting. The experimental results demonstrate that, our QSViT methodology achieves 22.75% memory saving and 21.33% power saving, while also maintaining high accuracy within 2.1% from that of the original non-quantized SViT model on the ImageNet dataset. These results highlight the potential of QSViT methodology to pave the way toward the efficient SViT deployments on resource-constrained embedded AI systems.","author":[{"family":"Putra","given":"Rachmad"},{"family":"Iftikhar","given":"Saad"},{"family":"Shafique","given":"Muhammad"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2504.00948","URL":"https://doi.org/10.48550/arxiv.2504.00948","source":"datacite"},{"id":"doi:10.48550/arxiv.2601.00806","type":"manuscript","title":"Energy-Efficient Eimeria Parasite Detection Using a Two-Stage Spiking Neural Network Architecture","abstract":"Coccidiosis, a disease caused by the Eimeria parasite, represents a major threat to the poultry and rabbit industries, demanding rapid and accurate diagnostic tools. While deep learning models offer high precision, their significant energy consumption limits their deployment in resource-constrained environments. This paper introduces a novel two-stage Spiking Neural Network (SNN) architecture, where a pre-trained Convolutional Neural Network is first converted into a spiking feature extractor and then coupled with a lightweight, unsupervised SNN classifier trained with Spike-Timing-Dependent Plasticity (STDP). The proposed model sets a new state-of-the-art, achieving 98.32\\% accuracy in Eimeria classification. Remarkably, this performance is accomplished with a significant reduction in energy consumption, showing an improvement of more than 223 times compared to its traditional ANN counterpart. This work demonstrates a powerful synergy between high accuracy and extreme energy efficiency, paving the way for autonomous, low-power diagnostic systems on neuromorphic hardware.","author":[{"family":"García-Vico","given":"Ángel"},{"family":"Seker","given":"Huseyin"},{"family":"Afzal","given":"Muhammad"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2601.00806","URL":"https://doi.org/10.48550/arxiv.2601.00806","source":"datacite"},{"id":"doi:10.48550/arxiv.2508.04610","type":"manuscript","title":"Neuromorphic Cybersecurity with Semi-supervised Lifelong Learning","abstract":"Inspired by the brain's hierarchical processing and energy efficiency, this paper presents a Spiking Neural Network (SNN) architecture for lifelong Network Intrusion Detection System (NIDS). The proposed system first employs an efficient static SNN to identify potential intrusions, which then activates an adaptive dynamic SNN responsible for classifying the specific attack type. Mimicking biological adaptation, the dynamic classifier utilizes Grow When Required (GWR)-inspired structural plasticity and a novel Adaptive Spike-Timing-Dependent Plasticity (Ad-STDP) learning rule. These bio-plausible mechanisms enable the network to learn new threats incrementally while preserving existing knowledge. Tested on the UNSW-NB15 benchmark in a continual learning setting, the architecture demonstrates robust adaptation, reduced catastrophic forgetting, and achieves $85.3$\\% overall accuracy. Furthermore, simulations using the Intel Lava framework confirm high operational sparsity, highlighting the potential for low-power deployment on neuromorphic hardware.","author":[{"family":"Mia","given":"Md"},{"family":"Bal","given":"Malyaban"},{"family":"Lu","given":"Sen"},{"family":"Nishibuchi","given":"George"},{"family":"Chelian","given":"Suhas"},{"family":"Vasan","given":"Srini"},{"family":"Sengupta","given":"Abhronil"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2508.04610","URL":"https://doi.org/10.48550/arxiv.2508.04610","source":"datacite"},{"id":"doi:10.57760/sciencedb.34532","type":"article-journal","title":"Energy-Efficient Visual Search by Eye Movements with a Low-Latency Spiking Neural Agent","abstract":"Human vision incorporates a non-uniform resolution retina, efficient eye movement strategies, and spiking neural networks (SNNs) to balance requirements in visual field size, visual resolution, energy cost, and inference latency. However, whether these features can synergize to produce more energy-efficient computer vision algorithms with human-like eye movements remains underexplored. Here, we record human visual search data and establish an bio-inspired visual search model (BVSM). To our knowledge, BVSM is a rare SNN-based active visual search agent: it unifies non-uniform retinal sampling, event-driven spiking feature processing and memory, and a learned eye-movement policy within a single closed-loop framework, enabling low-latency saccade decisions in continuous space. The model combines a non-uniform resolution retina with spiking feature extraction, memory, and saccade decision modules, and employs population coding for fast saccade decisions. It can learn either a human-like or near-optimal fixation strategy by reinforcement learning, outperform human search performance, and achieve high energy efficiency through low saccade decision latency and sparse activation. Our findings suggest that SNNs with non-uniform resolution retina and efficient eye-movement strategies can facilitate more energy-efficient computer vision algorithms, especially when a large visual field, high visual resolution, and low energy cost are simultaneously required.","author":[{"family":"Zhou","given":"Yunhui"},{"family":"Han","given":"Dongqi"},{"family":"Yu","given":"Yuguo"}],"issued":{"date-parts":[[2025]]},"DOI":"10.57760/sciencedb.34532","URL":"https://doi.org/10.57760/sciencedb.34532","source":"datacite"},{"id":"doi:10.48550/arxiv.2512.22214","type":"manuscript","title":"Signal-SGN++: Topology-Enhanced Time-Frequency Spiking Graph Network for Skeleton-Based Action Recognition","abstract":"Graph Convolutional Networks (GCNs) demonstrate strong capability in modeling skeletal topology for action recognition, yet their dense floating-point computations incur high energy costs. Spiking Neural Networks (SNNs), characterized by event-driven and sparse activation, offer energy efficiency but remain limited in capturing coupled temporal-frequency and topological dependencies of human motion. To bridge this gap, this article proposes Signal-SGN++, a topology-aware spiking graph framework that integrates structural adaptivity with time-frequency spiking dynamics. The network employs a backbone composed of 1D Spiking Graph Convolution (1D-SGC) and Frequency Spiking Convolution (FSC) for joint spatiotemporal and spectral feature extraction. Within this backbone, a Topology-Shift Self-Attention (TSSA) mechanism is embedded to adaptively route attention across learned skeletal topologies, enhancing graph-level sensitivity without increasing computational complexity. Moreover, an auxiliary Multi-Scale Wavelet Transform Fusion (MWTF) branch decomposes spiking features into multi-resolution temporal-frequency representations, wherein a Topology-Aware Time-Frequency Fusion (TATF) unit incorporates structural priors to preserve topology-consistent spectral fusion. Comprehensive experiments on large-scale benchmarks validate that Signal-SGN++ achieves superior accuracy-efficiency trade-offs, outperforming existing SNN-based methods and achieving competitive results against state-of-the-art GCNs under substantially reduced energy consumption.","author":[{"family":"Zheng","given":"Naichuan"},{"family":"Lun","given":"Xiahai"},{"family":"Li","given":"Weiyi"},{"family":"Du","given":"Yuchen"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2512.22214","URL":"https://doi.org/10.48550/arxiv.2512.22214","source":"datacite"},{"id":"doi:10.48550/arxiv.2506.13268","type":"manuscript","title":"Energy-Efficient Digital Design: A Comparative Study of Event-Driven and Clock-Driven Spiking Neurons","abstract":"This paper presents a comprehensive evaluation of Spiking Neural Network (SNN) neuron models for hardware acceleration by comparing event driven and clock-driven implementations. We begin our investigation in software, rapidly prototyping and testing various SNN models based on different variants of the Leaky Integrate and Fire (LIF) neuron across multiple datasets. This phase enables controlled performance assessment and informs design refinement. Our subsequent hardware phase, implemented on FPGA, validates the simulation findings and offers practical insights into design trade offs. In particular, we examine how variations in input stimuli influence key performance metrics such as latency, power consumption, energy efficiency, and resource utilization. These results yield valuable guidelines for constructing energy efficient, real time neuromorphic systems. Overall, our work bridges software simulation and hardware realization, advancing the development of next generation SNN accelerators.","author":[{"family":"Marostica","given":"Filippo"},{"family":"Carpegna","given":"Alessio"},{"family":"Savino","given":"Alessandro"},{"family":"Di Carlo","given":"Stefano"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2506.13268","URL":"https://doi.org/10.48550/arxiv.2506.13268","source":"datacite"},{"id":"doi:10.48550/arxiv.2512.17841","type":"manuscript","title":"NeuRehab: A Reinforcement Learning and Spiking Neural Network-Based Rehab Automation Framework","abstract":"Recent advancements in robotic rehabilitation therapy have provided modular exercise systems for post-stroke muscle recovery with basic control schemes. But these systems struggle to adapt to patients' complex and ever-changing behaviour, and to operate within mobile settings, such as heat and power. To aid this, we present NeuRehab: an end-to-end framework consisting of a training and inference pipeline with AI-based automation, co-designed with neuromorphic computing-based control systems that balance action performance, power consumption, and observed latency. The framework consists of 2 partitions. One is designated for the rehabilitation device based on ultra-low power spiking networks deployed on dedicated neuromorphic hardware. The other resides on stationary hardware that can accommodate computationally intensive hardware for fine-tuning on a per-patient basis. By maintaining a communication channel between both the modules and splitting the algorithm components, the power and latency requirements of the movable system have been optimised, while retaining the learning performance advantages of compute- and power-hungry hardware on the stationary machine. As part of the framework, we propose (a) the split machine learning processes for efficiency in architectural utilisation, and (b) task-specific temporal optimisations to lower edge-inference control latency. This paper evaluates the proposed methods on a reference stepper motor-based shoulder exercise. Overall, these methods offer comparable performance uplifts over the State-of-the-art for neuromorphic deployment, while achieving over 60% savings in both power and latency during inference compared to standard implementations.","author":[{"family":"Kambhampati","given":"Phani"},{"family":"Gautam","given":"Chainesh"},{"family":"Palaniswamy","given":"Jagan"},{"family":"Rao","given":"Madhav"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2512.17841","URL":"https://doi.org/10.48550/arxiv.2512.17841","source":"datacite"},{"id":"doi:10.48550/arxiv.2503.21846","type":"manuscript","title":"LightSNN: Lightweight Architecture Search for Sparse and Accurate Spiking Neural Networks","abstract":"Spiking Neural Networks (SNNs) are highly regarded for their energy efficiency, inherent activation sparsity, and suitability for real-time processing in edge devices. However, most current SNN methods adopt architectures resembling traditional artificial neural networks (ANNs), leading to suboptimal performance when applied to SNNs. While SNNs excel in energy efficiency, they have been associated with lower accuracy levels than traditional ANNs when utilizing conventional architectures. In response, in this work we present LightSNN, a rapid and efficient Neural Network Architecture Search (NAS) technique specifically tailored for SNNs that autonomously leverages the most suitable architecture, striking a good balance between accuracy and efficiency by enforcing sparsity. Based on the spiking NAS network (SNASNet) framework, a cell-based search space including backward connections is utilized to build our training-free pruning-based NAS mechanism. Our technique assesses diverse spike activation patterns across different data samples using a sparsity-aware Hamming distance fitness evaluation. Thorough experiments are conducted on both static (CIFAR10 and CIFAR100) and neuromorphic datasets (DVS128-Gesture). Our LightSNN model achieves state-of-the-art results on CIFAR10 and CIFAR100, improves performance on DVS128Gesture by 4.49\\%, and significantly reduces search time most notably offering a $98\\times$ speedup over SNASNet and running 30\\% faster than the best existing method on DVS128Gesture. Code is available on Github at: https://github.com/YesmineAbdennadher/LightSNN.","author":[{"family":"Abdennadher","given":"Yesmine"},{"family":"Perin","given":"Giovanni"},{"family":"Mazzieri","given":"Riccardo"},{"family":"Pegoraro","given":"Jacopo"},{"family":"Rossi","given":"Michele"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2503.21846","URL":"https://doi.org/10.48550/arxiv.2503.21846","source":"datacite"},{"id":"doi:10.48550/arxiv.2512.11743","type":"manuscript","title":"CogniSNN: Enabling Neuron-Expandability, Pathway-Reusability, and Dynamic-Configurability with Random Graph Architectures in Spiking Neural Networks","abstract":"Spiking neural networks (SNNs), regarded as the third generation of artificial neural networks, are expected to bridge the gap between artificial intelligence and computational neuroscience. However, most mainstream SNN research directly adopts the rigid, chain-like hierarchical architecture of traditional artificial neural networks (ANNs), ignoring key structural characteristics of the brain. Biological neurons are stochastically interconnected, forming complex neural pathways that exhibit Neuron-Expandability, Pathway-Reusability, and Dynamic-Configurability. In this paper, we introduce a new SNN paradigm, named Cognition-aware SNN (CogniSNN), by incorporating Random Graph Architecture (RGA). Furthermore, we address the issues of network degradation and dimensional mismatch in deep pathways by introducing an improved pure spiking residual mechanism alongside an adaptive pooling strategy. Then, we design a Key Pathway-based Learning without Forgetting (KP-LwF) approach, which selectively reuses critical neural pathways while retaining historical knowledge, enabling efficient multi-task transfer. Finally, we propose a Dynamic Growth Learning (DGL) algorithm that allows neurons and synapses to grow dynamically along the internal temporal dimension. Extensive experiments demonstrate that CogniSNN achieves performance comparable to, or even surpassing, current state-of-the-art SNNs on neuromorphic datasets and Tiny-ImageNet. The Pathway-Reusability enhances the network's continuous learning capability across different scenarios, while the dynamic growth algorithm improves robustness against interference and mitigates the fixed-timestep constraints during neuromorphic chip deployment. This work demonstrates the potential of SNNs with random graph structures in advancing brain-inspired intelligence and lays the foundation for their practical application on neuromorphic hardware.","author":[{"family":"Huang","given":"Yongsheng"},{"family":"Duan","given":"Peibo"},{"family":"Wu","given":"Yujie"},{"family":"Sun","given":"Kai"},{"family":"Liu","given":"Zhipeng"},{"family":"Zhang","given":"Changsheng"},{"family":"Zhang","given":"Bin"},{"family":"Xu","given":"Mingkun"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2512.11743","URL":"https://doi.org/10.48550/arxiv.2512.11743","source":"datacite"},{"id":"doi:10.48550/arxiv.2512.10179","type":"manuscript","title":"Assessing Neuromorphic Computing for Fingertip Force Decoding from Electromyography","abstract":"High-density surface electromyography (HD-sEMG) provides a noninvasive neural interface for assistive and rehabilitation control, but mapping neural activity to user motor intent remains challenging. We assess a spiking neural network (SNN) as a neuromorphic architecture against a temporal convolutional network (TCN) for decoding fingertip force from motor-unit (MU) firing derived from HD-sEMG. Data were collected from a single participant (10 trials) with two forearm electrode arrays; MU activity was obtained via FastICA-based decomposition, and models were trained on overlapping windows with end-to-end causal convolutions. On held-out trials, the TCN achieved 4.44% MVC RMSE (Pearson r = 0.974) while the SNN achieved 8.25% MVC (r = 0.922). While the TCN was more accurate, we view the SNN as a realistic neuromorphic baseline that could close much of this gap with modest architectural and hyperparameter refinements.","author":[{"family":"Shahrooei","given":"Abolfazl"},{"family":"Arthur","given":"Luke"},{"family":"Patel","given":"Om"},{"family":"Kamper","given":"Derek"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2512.10179","URL":"https://doi.org/10.48550/arxiv.2512.10179","source":"datacite"},{"id":"doi:10.48550/arxiv.2512.05246","type":"manuscript","title":"NeuromorphicRx: From Neural to Spiking Receiver","abstract":"In this work, we propose a novel energy-efficient spiking neural network (SNN)-based receiver for 5G-NR OFDM system, called neuromorphic receiver (NeuromorphicRx), replacing the channel estimation, equalization and symbol demapping blocks. We leverage domain knowledge to design the input with spiking encoding and propose a deep convolutional SNN with spike-element-wise residual connections. We integrate an SNN with artificial neural network (ANN) hybrid architecture to obtain soft outputs and employ surrogate gradient descent for training. We focus on generalization across diverse scenarios and robustness through quantized aware training. We focus on interpretability of NeuromorphicRx for 5G-NR signals and perform detailed ablation study for 5G-NR signals. Our extensive numerical simulations show that NeuromorphicRx is capable of achieving significant block error rate performance gain compared to 5G-NR receivers and similar performance compared to its ANN-based counterparts with 7.6x less energy consumption.","author":[{"family":"Gupta","given":"Ankit"},{"family":"Dizdar","given":"Onur"},{"family":"Chen","given":"Yun"},{"family":"Kadan","given":"Fehmi"},{"family":"Sattarzadeh","given":"Ata"},{"family":"Wang","given":"Stephen"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2512.05246","URL":"https://doi.org/10.48550/arxiv.2512.05246","source":"datacite"},{"id":"doi:10.48550/arxiv.2509.23762","type":"manuscript","title":"Accuracy-Robustness Trade Off via Spiking Neural Network Gradient Sparsity Trail","abstract":"Spiking Neural Networks (SNNs) have attracted growing interest in both computational neuroscience and artificial intelligence, primarily due to their inherent energy efficiency and compact memory footprint. However, achieving adversarial robustness in SNNs, (particularly for vision-related tasks) remains a nascent and underexplored challenge. Recent studies have proposed leveraging sparse gradients as a form of regularization to enhance robustness against adversarial perturbations. In this work, we present a surprising finding: under specific architectural configurations, SNNs exhibit natural gradient sparsity and can achieve state-of-the-art adversarial defense performance without the need for any explicit regularization. Further analysis reveals a trade-off between robustness and generalization: while sparse gradients contribute to improved adversarial resilience, they can impair the model's ability to generalize; conversely, denser gradients support better generalization but increase vulnerability to attacks. Our findings offer new insights into the dual role of gradient sparsity in SNN training.","author":[{"family":"Nhan","given":"Luu"},{"family":"Duong","given":"Luu"},{"family":"Nam","given":"Pham"},{"family":"Thang","given":"Truong"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2509.23762","URL":"https://doi.org/10.48550/arxiv.2509.23762","source":"datacite"},{"id":"doi:10.48550/arxiv.2512.00427","type":"manuscript","title":"Hardware-Software Collaborative Computing of Photonic Spiking Reinforcement Learning for Robotic Continuous Control","abstract":"Robotic continuous control tasks impose stringent demands on the energy efficiency and latency of computing architectures due to their high-dimensional state spaces and real-time interaction requirements. Conventional electronic computing platforms face computational bottlenecks, whereas the fusion of photonic computing and spiking reinforcement learning (RL) offers a promising alternative. Here, we propose a novel computing architecture based on photonic spiking RL, which integrates the Twin Delayed Deep Deterministic policy gradient (TD3) algorithm with spiking neural network (SNN). The proposed architecture employs an optical-electronic hybrid computing paradigm wherein a silicon photonic Mach-Zehnder interferometer (MZI) chip executes linear matrix computations, while nonlinear spiking activations are performed in the electronic domain. Experimental validation on the Pendulum-v1 and HalfCheetah-v2 benchmarks demonstrates the system capability for software-hardware co-inference, achieving a control policy reward of 5831 on HalfCheetah-v2, a 23.33% reduction in convergence steps, and an action deviation below 2.2%. Notably, this work represents the first application of a programmable MZI photonic computing chip to robotic continuous control tasks, attaining an energy efficiency of 1.39 TOPS/W and an ultralow computational latency of 120 ps. Such performance underscores the promise of photonic spiking RL for real-time decision-making in autonomous and industrial robotic systems.","author":[{"family":"Yu","given":"Mengting"},{"family":"Xiang","given":"Shuiying"},{"family":"Xie","given":"Changjian"},{"family":"Chen","given":"Yonghang"},{"family":"Zhao","given":"Haowen"},{"family":"Guo","given":"Xingxing"},{"family":"Zhang","given":"Yahui"},{"family":"Han","given":"Yanan"},{"family":"Hao","given":"Yue"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2512.00427","URL":"https://doi.org/10.48550/arxiv.2512.00427","source":"datacite"},{"id":"doi:10.48550/arxiv.2511.21337","type":"manuscript","title":"Hybrid SIFT-SNN for Efficient Anomaly Detection of Traffic Flow-Control Infrastructure","abstract":"This paper presents the SIFT-SNN framework, a low-latency neuromorphic signal-processing pipeline for real-time detection of structural anomalies in transport infrastructure. The proposed approach integrates Scale-Invariant Feature Transform (SIFT) for spatial feature encoding with a latency-driven spike conversion layer and a Leaky Integrate-and-Fire (LIF) Spiking Neural Network (SNN) for classification. The Auckland Harbour Bridge dataset is recorded under various weather and lighting conditions, comprising 6,000 labelled frames that include both real and synthetically augmented unsafe cases. The presented system achieves a classification accuracy of 92.3% (+- 0.8%) with a per-frame inference time of 9.5 ms. Achieved sub-10 millisecond latency, combined with sparse spike activity (8.1%), enables real-time, low-power edge deployment. Unlike conventional CNN-based approaches, the hybrid SIFT-SNN pipeline explicitly preserves spatial feature grounding, enhances interpretability, supports transparent decision-making, and operates efficiently on embedded hardware. Although synthetic augmentation improved robustness, generalisation to unseen field conditions remains to be validated. The SIFT-SNN framework is validated through a working prototype deployed on a consumer-grade system and framed as a generalisable case study in structural safety monitoring for movable concrete barriers, which, as a traffic flow-control infrastructure, is deployed in over 20 cities worldwide.","author":[{"family":"Rathee","given":"Munish"},{"family":"Bačić","given":"Boris"},{"family":"Doborjeh","given":"Maryam"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2511.21337","URL":"https://doi.org/10.48550/arxiv.2511.21337","source":"datacite"},{"id":"doi:10.48550/arxiv.2511.17563","type":"manuscript","title":"Dynamic Weight Adaptation in Spiking Neural Networks Inspired by Biological Homeostasis","abstract":"Homeostatic mechanisms play a crucial role in maintaining optimal functionality within the neural circuits of the brain. By regulating physiological and biochemical processes, these mechanisms ensure the stability of an organism's internal environment, enabling it to better adapt to external changes. Among these mechanisms, the Bienenstock, Cooper, and Munro (BCM) theory has been extensively studied as a key principle for maintaining the balance of synaptic strengths in biological systems. Despite the extensive development of spiking neural networks (SNNs) as a model for bionic neural networks, no prior work in the machine learning community has integrated biologically plausible BCM formulations into SNNs to provide homeostasis. In this study, we propose a Dynamic Weight Adaptation Mechanism (DWAM) for SNNs, inspired by the BCM theory. DWAM can be integrated into the host SNN, dynamically adjusting network weights in real time to regulate neuronal activity, providing homeostasis to the host SNN without any fine-tuning. We validated our method through dynamic obstacle avoidance and continuous control tasks under both normal and specifically designed degraded conditions. Experimental results demonstrate that DWAM not only enhances the performance of SNNs without existing homeostatic mechanisms under various degraded conditions but also further improves the performance of SNNs that already incorporate homeostatic mechanisms.","author":[{"family":"Zhou","given":"Yunduo"},{"family":"Dong","given":"Bo"},{"family":"Li","given":"Chang"},{"family":"Wang","given":"Yuanchen"},{"family":"Yin","given":"Xuefeng"},{"family":"Wang","given":"Yang"},{"family":"Yang","given":"Xin"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2511.17563","URL":"https://doi.org/10.48550/arxiv.2511.17563","source":"datacite"},{"id":"doi:10.48550/arxiv.2511.16060","type":"manuscript","title":"Neuromorphic Astronomy: An End-to-End SNN Pipeline for RFI Detection Hardware","abstract":"Imminent radio telescope observatories provide massive data rates making deep learning based processing appealing while simultaneously demanding real-time performance at low-energy; prohibiting the use of many artificial neural network based approaches. We begin tackling the scientifically existential challenge of Radio Frequency Interference (RFI) detection by deploying deep Spiking Neural Networks (SNNs) on resource-constrained neuromorphic hardware. Our approach partitions large, pre-trained networks onto SynSense Xylo hardware using maximal splitting, a novel greedy algorithm. We validate this pipeline with on-chip power measurements, achieving instrument-scaled inference at 100mW. While our full-scale SNN achieves state-of-the-art accuracy among SNN baselines, our experiments reveal a more important insight that a smaller un-partitioned model significantly outperforms larger, split models. This finding highlights that hardware co-design is paramount for optimal performance. Our work thus provides a practical deployment blueprint, a key insight into the challenges of model scaling, and reinforces radio astronomy as a demanding yet ideal domain for advancing applied neuromorphic computing.","author":[{"family":"Pritchard","given":"Nicholas"},{"family":"Wicenec","given":"Andreas"},{"family":"Dodson","given":"Richard"},{"family":"Bennamoun","given":"Mohammed"},{"family":"Muir","given":"Dylan"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2511.16060","URL":"https://doi.org/10.48550/arxiv.2511.16060","source":"datacite"},{"id":"doi:10.6084/m9.figshare.c.8142680.v1","type":"article-journal","title":"A hybrid spiking neural network - quantum framework for spatio-temporal data classification: a case study on EEG data","abstract":"Abstract The study introduces a hybrid computational framework that combines neuro-inspired information processing using spiking neural networks (SNNs) and quantum information processing using quantum kernels to develop quantum-enhanced machine learning models for spatio-temporal data, demonstrated through the classification of EEG data as a case study. In the proposed SNN-quantum computation (SNN-QC) framework, SNN with spike time information representation is employed to learn spatio-temporal interactions (EEG recorded from multiple channels over time). Frequency-based (rate-based) information as spike frequency state vectors are extracted from the SNN and classified using a quantum classifier. In the latter part, we use the quantum kernel approach utilising feature maps for classification tasks. The proposed SNN-QC is demonstrated on a benchmark EEG dataset to classify three distinct wrist movement tasks in six binary classification setups as a proof of concept. We introduce a novel high-order nonlinear feature map that demonstrates improved performance over state-of-the-art feature maps and several machine learning methods across most of the tasks studied. Furthermore, the role of hyperparameters for enhanced feature maps is also highlighted. The performance of SNN-QC is evaluated using statistical metrics and cross-validation techniques, demonstrating its efficacy across multiple binary classifiers. Quantum hardware validation is conducted using both a superconducting IBM-QPU and a high-fidelity noisy simulation that replicates a real QPU. Furthermore, the results demonstrate that the SNN-QC outperforms models that use statistical features rather than features extracted from the SNN, as the SNN accounts for the temporal interaction between the spatio-temporal input variables. Finally, we conclude that the SNN-QC offers a potential pathway for developing more accurate neuromorphic-quantum enhanced systems that are both energy-efficient and biologically-inspired, well-suited for dealing with spatio-temporal data.","author":[{"family":"Jha","given":"Ravi"},{"family":"Kasabov","given":"Nikola"},{"family":"Bhattacharyya","given":"Saugat"},{"family":"Coyle","given":"Damien"},{"family":"Prasad","given":"Girijesh"}],"issued":{"date-parts":[[2025]]},"DOI":"10.6084/m9.figshare.c.8142680.v1","URL":"https://doi.org/10.6084/m9.figshare.c.8142680.v1","source":"datacite"},{"id":"doi:10.6084/m9.figshare.c.8142680","type":"article-journal","title":"A hybrid spiking neural network - quantum framework for spatio-temporal data classification: a case study on EEG data","abstract":"Abstract The study introduces a hybrid computational framework that combines neuro-inspired information processing using spiking neural networks (SNNs) and quantum information processing using quantum kernels to develop quantum-enhanced machine learning models for spatio-temporal data, demonstrated through the classification of EEG data as a case study. In the proposed SNN-quantum computation (SNN-QC) framework, SNN with spike time information representation is employed to learn spatio-temporal interactions (EEG recorded from multiple channels over time). Frequency-based (rate-based) information as spike frequency state vectors are extracted from the SNN and classified using a quantum classifier. In the latter part, we use the quantum kernel approach utilising feature maps for classification tasks. The proposed SNN-QC is demonstrated on a benchmark EEG dataset to classify three distinct wrist movement tasks in six binary classification setups as a proof of concept. We introduce a novel high-order nonlinear feature map that demonstrates improved performance over state-of-the-art feature maps and several machine learning methods across most of the tasks studied. Furthermore, the role of hyperparameters for enhanced feature maps is also highlighted. The performance of SNN-QC is evaluated using statistical metrics and cross-validation techniques, demonstrating its efficacy across multiple binary classifiers. Quantum hardware validation is conducted using both a superconducting IBM-QPU and a high-fidelity noisy simulation that replicates a real QPU. Furthermore, the results demonstrate that the SNN-QC outperforms models that use statistical features rather than features extracted from the SNN, as the SNN accounts for the temporal interaction between the spatio-temporal input variables. Finally, we conclude that the SNN-QC offers a potential pathway for developing more accurate neuromorphic-quantum enhanced systems that are both energy-efficient and biologically-inspired, well-suited for dealing with spatio-temporal data.","author":[{"family":"Jha","given":"Ravi"},{"family":"Kasabov","given":"Nikola"},{"family":"Bhattacharyya","given":"Saugat"},{"family":"Coyle","given":"Damien"},{"family":"Prasad","given":"Girijesh"}],"issued":{"date-parts":[[2025]]},"DOI":"10.6084/m9.figshare.c.8142680","URL":"https://doi.org/10.6084/m9.figshare.c.8142680","source":"datacite"},{"id":"doi:10.48550/arxiv.2511.05232","type":"manuscript","title":"Travelling waves modulated by subthreshold oscillations in networks of integrate-and-fire neurons","abstract":"Travelling waves of neural firing activity are observed in brain tissue as a part of various sensory, motor and cognitive processes. They represent an object of major interest in the study of excitable networks, with analysis conducted in both neural field models and spiking neuronal networks. The latter class exposes the single-neuron dynamics directly, allowing us to study the details of their influence upon network-scale behaviour. Here we present a study of a laterally-inhibited network of leaky integrate-and-fire neurons modulated by a slow voltage-gated ion channel that acts as a linear adaptation variable. As the strength of the ion channel increases, we find that its interaction with the lateral inhibition increases wave speeds. The ion channel can enable subthreshold oscillations, with the intervals between the firing events of loosely-coupled travelling wave solutions structured around the neuron's natural period. These subthreshold oscillations also enable the occurrence of codimension-2 grazing bifurcations; along with the emergence of fold bifurcations along wave solution branches, the slow ion channel introduces a variety of intermediate structures in the solution space. These point towards further investigation of the role neighbouring solution branches play in the behaviour of waves forced across bifurcations, which we illustrate with the aid of simulations using a novel root-finding algorithm designed to handle uncertainty over the existence of firing solutions.","author":[{"family":"Kerr","given":"Henry"},{"family":"Ashwin","given":"Peter"},{"family":"Wedgwood","given":"Kyle"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2511.05232","URL":"https://doi.org/10.48550/arxiv.2511.05232","source":"datacite"},{"id":"doi:10.48550/arxiv.2511.01158","type":"manuscript","title":"A High-Throughput Spiking Neural Network Processor Enabling Synaptic Delay Emulation","abstract":"Synaptic delay has attracted significant attention in neural network dynamics for integrating and processing complex spatiotemporal information. This paper introduces a high-throughput Spiking Neural Network (SNN) processor that supports synaptic delay-based emulation for edge applications. The processor leverages a multicore pipelined architecture with parallel compute engines, capable of real-time processing of the computational load associated with synaptic delays. We develop a SoC prototype of the proposed processor on PYNQ Z2 FPGA platform and evaluate its performance using the Spiking Heidelberg Digits (SHD) benchmark for low-power keyword spotting tasks. The processor achieves 93.4% accuracy in deployment and an average throughput of 104 samples/sec at a typical operating frequency of 125 MHz and 282 mW power consumption.","author":[{"family":"Chen","given":"Faquan"},{"family":"Tian","given":"Qingyang"},{"family":"Wu","given":"Ziren"},{"family":"Ying","given":"Rendong"},{"family":"Wen","given":"Fei"},{"family":"Liu","given":"Peilin"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2511.01158","URL":"https://doi.org/10.48550/arxiv.2511.01158","source":"datacite"},{"id":"doi:10.48550/arxiv.2503.13492","type":"manuscript","title":"Event-Driven Implementation of a Physical Reservoir Computing Framework for superficial EMG-based Gesture Recognition","abstract":"Wearable health devices have a strong demand in real-time biomedical signal processing. However traditional methods often require data transmission to centralized processing unit with substantial computational resources after collecting it from edge devices. Neuromorphic computing is an emerging field that seeks to design specialized hardware for computing systems inspired by the structure, function, and dynamics of the human brain, offering significant advantages in latency and power consumption. This paper explores a novel neuromorphic implementation approach for gesture recognition by extracting spatiotemporal spiking information from surface electromyography (sEMG) data in an event-driven manner. At the same time, the network was designed by implementing a simple-structured and hardware-friendly Physical Reservoir Computing (PRC) framework called Rotating Neuron Reservoir (RNR) within the domain of Spiking neural network (SNN). The spiking RNR (sRNR) is promising to pipeline an innovative solution to compact embedded wearable systems, enabling low-latency, real-time processing directly at the sensor level. The proposed system was validated by an open-access large-scale sEMG database and achieved an average classification accuracy of 74.6\\% and 80.3\\% using a classical machine learning classifier and a delta learning rule algorithm respectively. While the delta learning rule could be fully spiking and implementable on neuromorphic chips, the proposed gesture recognition system demonstrates the potential for near-sensor low-latency processing.","author":[{"family":"Ding","given":"Yuqi"},{"family":"Donati","given":"Elisa"},{"family":"Li","given":"Haobo"},{"family":"Heidari","given":"Hadi"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2503.13492","URL":"https://doi.org/10.48550/arxiv.2503.13492","source":"datacite"},{"id":"doi:10.1038/s41467-025-56286-y","type":"article-journal","title":"Memristor-based feature learning for pattern classification.","abstract":"Inspired by biological processes, feature learning techniques, such as deep learning, have achieved great success in various fields. However, since biological organs may operate differently from semiconductor devices, deep models usually require dedicated hardware and are computation-complex. High energy consumption has made deep model growth unsustainable. We present an approach that directly implements feature learning using semiconductor physics to minimize disparity between model and hardware. Following this approach, a feature learning technique based on memristor drift-diffusion kinetics is proposed by leveraging the dynamic response of a single memristor to learn features. The model parameters and computational operations of the kinetics-based network are reduced by up to 2 and 4 orders of magnitude, respectively, compared with deep models. We experimentally implement the proposed network on 180 nm memristor chips for various dimensional pattern classification tasks. Compared with memristor-based deep learning hardware, the memristor kinetics-based hardware can further reduce energy and area consumption significantly. We propose that innovations in hardware physics could create an intriguing solution for intelligent models by balancing model complexity and performance.","author":[{"family":"Shi","given":"Tuo"},{"family":"Gao","given":"Lili"},{"family":"Tian","given":"Yang"},{"family":"Tang","given":"Shuangzhu"},{"family":"Liu","given":"Jinchang"},{"family":"Li","given":"Yiqi"},{"family":"Zhou","given":"Ruixi"},{"family":"Cui","given":"Shiyu"},{"family":"Zhang","given":"Hui"},{"family":"Li","given":"Yu"},{"family":"Wu","given":"Zuheng"},{"family":"Zhang","given":"Xumeng"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1038/s41467-025-56286-y","URL":"https://doi.org/10.1038/s41467-025-56286-y","source":"europepmc"},{"id":"doi:10.3390/mi16080882","type":"article-journal","title":"Composite Behavior of Nanopore Array Large Memristors.","abstract":"Synthetic nanopores were recently demonstrated with memristive and nonlinear voltage-current behaviors, akin to ion channels in a cell membrane. Such ionic devices are considered a promising candidate for the development of brain-inspired neuromorphic computing techniques. In this work, we show the composite behavior of nanopore-array large memristors, formed with different membrane materials, pore sizes, electrolytes, and device arrangements. Anodic aluminum oxide (AAO) membranes with 5 nm and 20 nm diameter pores and track-etched polycarbonate (PCTE) membranes with 10 nm diameter pores are tested and shown to demonstrate memristive and nonlinear behaviors with approximately 107–1010 pores in parallel when electrolyte concentration across the membranes is asymmetric. Ion diffusion through the large number of channels induces time-dependent electrolyte asymmetry that drives the system through different memristive states. The behaviors of series composite memristors with different configurations are also presented. In addition to helping understand fluidic devices and circuits for neuromorphic computing, the results also shed light on the development of field-assisted ion-selection-membrane filtration techniques as well as the investigations of large neurons and giant synapses. Further work is needed to de-embed parasitic components of the measurement setup to obtain intrinsic large memristor properties.","author":[{"family":"Reistroffer","given":"Ian"},{"family":"Tolbert","given":"Jaden"},{"family":"Osterberg","given":"Jeffrey"},{"family":"Wang","given":"Pingshan"}],"issued":{"date-parts":[[2025]]},"DOI":"10.3390/mi16080882","URL":"https://doi.org/10.3390/mi16080882","source":"europepmc"},{"id":"doi:10.1038/s41598-026-35671-7","type":"article-journal","title":"Memristance and transmemristance in multiterminal memristive systems.","abstract":"Memristive devices represent promising building blocks for the development of next-generation memory technologies, computing architectures, and neuromorphic systems. In addition to conventional two-terminal memristive circuits and crossbar array structures, multiterminal memristive systems, where emergent behaviours arise from the mutual interaction of numerous memristive elements, have been explored for neuromorphic data processing and computing applications. In this work, we extend the concept of two-terminal memristive devices to generic multiterminal memristive systems. Beyond its ability to describe the specific case of crossbar arrays, the proposed theoretical framework is also applicable to more complex systems such as self-organizing memristive networks, whose internal state dynamics depend not only on time-varying input signals but also on the spatial distribution of the stimulated terminals. After discussing the notion of memristance in multiterminal devices as the evolution of the system “seen” from the stimulating terminals, we demonstrate that the two-terminal memristive framework can be generalized to the concept of transmemristance when additional, non-stimulating electrodes are used to monitor the system’s evolution. Providing a connection between circuit theory and network science, these concepts are investigated both analytically and experimentally using a theoretical memristive graph model and an experimental memristive system based on self-organizing nanowire networks.","author":[{"family":"Milano","given":"Gianluca"},{"family":"Pilati","given":"Davide"},{"family":"Michieletti","given":"Fabio"},{"family":"Cultrera","given":"Alessandro"},{"family":"Ricciardi","given":"Carlo"},{"family":"Miranda","given":"E"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1038/s41598-026-35671-7","URL":"https://doi.org/10.1038/s41598-026-35671-7","source":"europepmc"},{"id":"oa:W7125261655","type":"article-journal","title":"Bayesian Inference via GeTe x ‐OTS Based Stochastic Synapse for Uncertainty‐Aware Medical Diagnostics","abstract":"ABSTRACT Neuromorphic architectures leveraging stochastic device physics present a transformative approach for implementing probabilistic computing paradigms capable of intrinsic uncertainty quantification. In this work, we present a germanium‐telluride‐based ovonic threshold switch (GeTe x ‐OTS) exhibiting inherent stochastic dynamics, integrated into a compact 1‐selector‐1‐transistor‐1‐resistor (1S1T1R) synaptic unit. The OTS devices are demonstrated within an 8K‐array, confirming the future scalability for neuromorphic systems. Experimental validation on the GeTe x devices shows stable Gaussian‐distributed threshold voltage fluctuations ( σ = 100 mV), enabling precise control of synaptic activation probability through input pulse modulation. Utilizing this intrinsic stochasticity, we implement mini hardware realization of a Monte Carlo Dropconnect (MC‐Dropconnect) neural network, directly demonstrating the feasibility of the system. Applied to COVID‐19 diagnosis using chest X‐ray images, our system achieves robust uncertainty quantification through predictive entropy, improving classification accuracy to 98.1%, compared to 96.0% with a deterministic baseline. This uncertainty‐aware hardware design strategy provides a scalable pathway for implementing energy‐efficient neuromorphic systems with native uncertainty estimation capabilities.","author":[{"family":"Wen","given":"Xinyu"},{"family":"Wang","given":"Lun"},{"family":"Wang","given":"KL"},{"family":"Zhao","given":"Vivian"},{"family":"Liu","given":"Zixuan"},{"family":"He","given":"Kexun"},{"family":"Yang","given":"JJ"},{"family":"Tong","given":"Hao"},{"family":"Miao","given":"Xiangshui"},{"family":"He","given":"Yuhui"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1002/smm2.70062","URL":"https://doi.org/10.1002/smm2.70062","source":"openalex"},{"id":"oa:W4406187914","type":"article-journal","title":"Process-dependent ferroelectric and memristive properties in polycrystalline Ca:HfO2-based devices","abstract":"Memristors are considered key building blocks for developing neuromorphic or in-memory computing hardware. Here, we study the ferroelectric and memristive response of Pt/Ca:HfO2/Pt devices fabricated on silicon by spin-coating from chemical solution deposition followed by a pyrolysis step and a final thermal treatment for crystallization at 800°C for 90 s. For pyrolysis temperature of 300°C, the annealed samples are ferroelectric while for 400°C a dielectric behavior is observed. For each case, we found a distinct, forming-free, memristive response. Ferroelectric devices can sustain polarization switching and memristive behavior simultaneously. Aided by numerical simulations, we describe the memristive behavior of ferroelectric devices arising from oxide-metal Schottky barriers modulation by both the direction of the electrical polarization and oxygen vacancy electromigration. For non-ferroelectric samples, only the latter effect controls the memristive behavior.","author":[{"family":"Ferreyra","given":"C"},{"family":"Badillo","given":"Miguel"},{"family":"Sánchez","given":"MJ"},{"family":"Acuautla","given":"Mónica"},{"family":"Noheda","given":"Beatriz"},{"family":"Rubi","given":"D"}],"issued":{"date-parts":[[2025]]},"DOI":"10.3389/fmats.2024.1501000","URL":"https://doi.org/10.3389/fmats.2024.1501000","source":"openalex"},{"id":"oa:W4414260754","type":"article-journal","title":"Programmable Functional Connectivity and Synchronous Activity in Resistive Switching Self‐Assembled Nanostructured Networks","abstract":"The efficiency of biological data processing systems is based on their adaptive connectivity and on the mutual interactions of their elements at different scales. These features are not substantially present in the design of electronic architectures based on conventional integrated circuits. The exploitation of self‐assembled systems characterized by nonlinear dynamics is actively investigated as a viable alternative strategy to develop energy‐efficient data processing devices. However, the encoding of external stimuli and the decoding of information from the analog response of such systems is still a challenge. Here we characterize the functional connectivity and the synchronicity between active sites in cluster‐assembled nanostructured Au films showing resistive switching behavior by a combined approach based on micro‐thermography and electrical measurements. We investigate the complex mechanisms involved in the network reorganization leading to its resistive switching activity and we identify the interplay between network dimensions and its emerging electrical behavior. We investigate the control on the synchronous activity and on the connectivity of the micrometric active sites which rule the emerging network dynamics and determe the performance of data processing devices. This activity is described using data analysis techniques commonly used in neuroscience, which are proposed for the first time to characterize neuromorphic systems.","author":[{"family":"Decastri","given":"Davide"},{"family":"Nieus","given":"Thierry"},{"family":"Zuccali","given":"Cristina"},{"family":"Giacomozzi","given":"Flavio"},{"family":"Lorenzelli","given":"Leandro"},{"family":"Borghi","given":"Francesca"},{"family":"Milani","given":"Paolo"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1002/sstr.202500330","URL":"https://doi.org/10.1002/sstr.202500330","source":"openalex"},{"id":"oa:W4407223688","type":"article-journal","title":"Recent Progress in Flexible Piezoelectric Tactile Sensors: Materials, Structures, Fabrication, and Application","abstract":"Flexible tactile sensors are widely used in aerospace, medical and health monitoring, electronic skin, human-computer interaction, and other fields due to their unique advantages, thus becoming a research hotspot. The goal is to develop a flexible tactile sensor characterized by outstanding sensitivity, extensive detection range and linearity, elevated spatial resolution, and commendable adaptability. Among several strategies like capacitive, piezoresistive, and triboelectric tactile sensors, etc., we focus on piezoelectric tactile sensors because of their self-powered nature, high sensitivity, and quick response time. These sensors can respond to a wide range of dynamic mechanical stimuli and turn them into measurable electrical signals. This makes it possible to accurately detect objects, including their shapes and textures, and for them to sense touch in real time. This work encapsulates current advancements in flexible piezoelectric tactile sensors, focusing on enhanced material properties, optimized structural design, improved fabrication techniques, and broadened application domains. We outline the challenges facing piezoelectric tactile sensors to provide inspiration and guidance for their future development.","author":[{"family":"Tang","given":"Jingyao"},{"family":"Li","given":"Yiheng"},{"family":"Yu","given":"Yirong"},{"family":"Hu","given":"Qing‐miao"},{"family":"Du","given":"Wenya"},{"family":"Lin","given":"Dabin"}],"issued":{"date-parts":[[2025]]},"DOI":"10.3390/s25030964","URL":"https://doi.org/10.3390/s25030964","source":"openalex"},{"id":"oa:W4413048995","type":"article-journal","title":"Spike-Count Reduction Techniques for Low Power Spiking Neural Networks","abstract":"Spiking neural network (SNN) has demonstrated its great potential in low-power neuromorphic applications. In SNN, computation activities are associated with the arrival and firing of spikes, its power consumption is directly correlated with the number of spikes propagated in the network. In this paper, we explore two methods to reduce the spike-count in the network, aiming to reduce the power consumption of SNN. We use Poisson distribution function in the input layer and through adjusting the correlation (called the gain in the paper) between the probability of spike generation and input values, the number of spikes in the input layer can be reduced to only 20% of the baseline model with the accuracy degradation of less than 1%. We also exploit the leaky-integrate-and-fire (LIF) mechanism and use the refractory period to reduce the generation of spikes from the neurons in the hidden layers. Through this method, the spike-count is reduced by 20% $$\\sim $$ 50% while the performance degradation is still less than 1%. These two spike-count reduction techniques are implemented in Verilog RTL; the power simulation results demonstrate significant power reduction in performing SNN computations. We further discovered that for different network architectures, these two techniques have different trade-offs to achieve optimal spike-count reduction while maintaining satisfactory results. Compared with other spike-count reduction techniques, the proposed scheme is efficient and straightforward for hardware implementation, making it well-suited for edge computing scenarios.","author":[{"family":"Kang","given":"Xinyu"},{"family":"Yang","given":"Zhitao"},{"family":"Ren","given":"Yuan"},{"family":"Ye","given":"Terry"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1007/s11063-025-11786-2","URL":"https://doi.org/10.1007/s11063-025-11786-2","source":"openalex"},{"id":"oa:W4410779414","type":"article-journal","title":"Alcohol‐sensitive MoS 2 optoelectronic synapses for mimicking human‐like visual adaptation","abstract":"Abstract The rapid advancements in humanoid robotics and autonomous driving demand smart artificial optoelectronic vision systems that can emulate human‐like perception. Although many studies have reported multi‐functional visual chips based on artificial optoelectronic synaptic devices, few can simulate complex behavioral characteristics of humans, like specific living habits and physiological adaptations. In this study, we demonstrated MoS 2 optoelectronic synapses capable of exhibiting tunable human‐like visual adaptation abilities under various alcohol concentrations, featuring remarkable photo‐induced conductance plasticity for emulating alcohol‐sensitive human visual recognition. Two working mechanisms involving hydrogen‐atom and oxygen‐atom doping were unveiled during the concentration‐dependent doping process. The visual adaptation abilities were systematically explored by controlling the doping concentration of alcohol molecules, and were further enhanced by electric and optoelectronic stimuli to emulate human‐like behaviors, such as slight drunkenness, heavy drunkenness, and sobering up. Under the influence of alcohol molecules and the modulation of device operating voltage, the accuracy of handwritten digit recognition for this device has greatly increased from 78.9% to 94.7%. image","author":[{"family":"Liu","given":"Xiao"},{"family":"Huang","given":"Ming"},{"family":"Zou","given":"Xiongfeng"},{"family":"Ali","given":"Wajid"},{"family":"Rehman","given":"Sajid"},{"family":"Li","given":"Juan"},{"family":"Li","given":"Ziwei"},{"family":"Li","given":"Xiang"},{"family":"Pan","given":"Anlian"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1002/inf2.70019","URL":"https://doi.org/10.1002/inf2.70019","source":"openalex"},{"id":"oa:W4410431401","type":"article-journal","title":"Bioinspired learning and memory in ionogels through fast response and slow relaxation dynamics of ions","abstract":"Mimicking biological systems’ sensing, learning, and memory capabilities in synthetic soft materials remains challenging. While significant progress has been made in sensory functions in ionogels, their learning and memory capabilities still lag behind biological systems. Here, we introduce cation-π interactions and a self-adaptable ionic-double-layer interface in bilayer ionogels to control ion transport. Fast ion response enables sensing and learning, while slow ion relaxation supports long-term memory. The ionogels achieve bioinspired functions, including sensitization, habituation, classical conditioning, and multimodal memory, with low energy consumption (0.06 pJ per spike). Additionally, the ionogels exhibit mechanical adaptability, such as stretchability, self-healing, and reconfigurability, making them ideal for soft robotics. Notably, the ionogels enable a robotic arm to mimic the selective capture behavior of a Venus flytrap. This work bridges the gap between biological intelligence and artificial systems, offering promising applications in bioinspired, energy-efficient sensing, learning, and memory. Ionogels have been integrated into soft robotics, though it is challenging to design gels with learning and memory capabilities. Here the authors introduce ionic-double-layer ionogel using cation-π interactions to program the ionogel.","author":[{"family":"Zhou","given":"Ning"},{"family":"Cui","given":"Ting"},{"family":"Lei","given":"Zhouyue"},{"family":"Wu","given":"Peiyi"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1038/s41467-025-59944-3","URL":"https://doi.org/10.1038/s41467-025-59944-3","source":"openalex"},{"id":"oa:W4406787430","type":"article-journal","title":"All-silicon non-volatile optical memory based on photon avalanche-induced trapping","abstract":"Abstract Implementing on-chip non-volatile optical memories has long been an actively pursued goal, promising significant enhancements in the capability and energy efficiency of photonic integrated circuits. Here, we demonstrate an non-volatile optical memory exclusively using the most common semiconductor material, silicon. By manipulating the photon avalanche effect, we introduce a trapping effect at the silicon-silicon oxide interface, which in turn demonstrates a non-volatile reprogrammable optical memory cell with a record-high 4-bit encoding, robust retention and endurance. This silicon avalanche-induced trapping memory provides a distinctively cost-efficient and high-reliability route to realize optical data storage in standard silicon foundry processes. We demonstrate its applications in trimming in optical interconnects and in-memory computing. Our in-memory computing test case reduces energy consumption by approximately 83% compared to conventional optical approaches.","author":[{"family":"Yuan","given":"Yuan"},{"family":"Peng","given":"Yiwei"},{"family":"Cheung","given":"Stanley"},{"family":"Sorin","given":"Wayne"},{"family":"Hooten","given":"Sean"},{"family":"Huang","given":"Zhihong"},{"family":"Liang","given":"Di"},{"family":"Zhang","given":"Jiuyi"},{"family":"Fiorentino","given":"Marco"},{"family":"Beausoleil","given":"Raymond"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1038/s42005-025-01934-4","URL":"https://doi.org/10.1038/s42005-025-01934-4","source":"openalex"},{"id":"oa:W4409810769","type":"article-journal","title":"Tunable Bipolar Photothermoelectric Response from Mott Activation for In‐Sensor Image Preprocessing","abstract":"In-sensor image preprocessing, a subset of edge computing, offers a solution to mitigate frequent analog-digital conversions and the von Neumann bottleneck in conventional digital hardware. However, an efficient in-sensor device array with large-scale integration capability for high-density and low-power sensory processing is still lacking and highly desirable. This work introduces an adjustable broadband photothermoelectric detector based on a phase-change vanadium dioxide thin-film transistor. This transistor employs a vanadium dioxide/gallium nitride three-terminal structure with a gate-tunable phase transition at the gate-source junctions. This design allows for modulable photothermoelectric responsivities and alteration of the short-circuit photocurrent's polarities. The devices exhibit linear gate dependence for the broadband photoresponse and linear light-intensity dependence for both positive and negative photoresponsivities. The device's energy consumption is as low as 8 pJ per spike, which is one order of magnitude lower than that of previous Mott materials-based in-sensor preprocessing devices. A wafer-scale bipolar phototransistor array has also been fabricated by standard micro-/nano-fabrication techniques, exhibiting excellent stability and endurance (over 5000 cycles). More importantly, an integrated in-sensor convolutional network is successfully designed for simultaneous broadband image classification, medical image denoising, and retinal vessel segmentation, delivering exceptional performance and paving the way for future smart edge sensors.","author":[{"family":"Li","given":"Bowen"},{"family":"Lin","given":"Ning"},{"family":"Wang","given":"Zhaowu"},{"family":"Wang","given":"Zhaowu"},{"family":"Chen","given":"Baojie"},{"family":"Lan","given":"Changyong"},{"family":"Li","given":"Xiaocui"},{"family":"Meng","given":"You"},{"family":"Wang","given":"Weijun"},{"family":"Ding","given":"Mingqi"},{"family":"Xie","given":"Pengshan"},{"family":"Zhang","given":"Yuxuan"},{"family":"Wu","given":"Zenghui"},{"family":"Li","given":"Dengji"},{"family":"Chen","given":"Fu‐rong"},{"family":"Chan","given":"Chi"},{"family":"Wang","given":"Zhongrui"},{"family":"Wang","given":"Zhongrui"},{"family":"Ho","given":"Johnny"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1002/adma.202502915","URL":"https://doi.org/10.1002/adma.202502915","source":"openalex"},{"id":"oa:W4406852025","type":"article-journal","title":"Post-processing methods for delay embedding and feature scaling of reservoir computers","abstract":"Reservoir computing is a machine learning method that is well-suited for complex time series prediction tasks. Both delay embedding and the projection of input data into a higher-dimensional space play important roles in enabling accurate predictions. We establish simple post-processing methods that train on past node states at uniformly or randomly-delayed timeshifts. These methods improve reservoir computer prediction performance through increased feature dimension and/or better delay embedding. Here we introduce the multi-random-timeshifting method that randomly recalls previous states of reservoir nodes. The use of multi-random-timeshifting allows for smaller reservoirs while maintaining large feature dimensions, is computationally cheap to optimise, and is our preferred post-processing method. For experimentalists, all our post-processing methods can be translated to readout data sampled from physical reservoirs, which we demonstrate using readout data from an experimentally-realised laser reservoir system.","author":[{"family":"Jaurigue","given":"Jonnel"},{"family":"Robertson","given":"Joshua"},{"family":"Hurtado","given":"Antonio"},{"family":"Jaurigue","given":"Lina"},{"family":"Lüdge","given":"Kathy"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1038/s44172-024-00330-0","URL":"https://doi.org/10.1038/s44172-024-00330-0","source":"openalex"},{"id":"oa:W4411620613","type":"article-journal","title":"NSIP: Neural Network SPICE Integration Platform for Ferroelectric Field‐Effect Transistor‐Based Crossbar Arrays","abstract":"This article presents the world's first demonstration of a neural network SPICE integration platform (NSIP) for simulating synaptic weights in HfZrO (HZO)‐based ferroelectric field‐effect transistor (FeFET) crossbar arrays tailored for neuromorphic computing. The FeFET compact model integrated in NSIP accurately reflects the hysteretic switching behavior observed in metal‐ferroelectric‐metal capacitors fabricated by incorporating the Preisach theory. Using databases, such as MNIST, the performance of NSIP is evaluated through voltage optimization, taking into account the effects of ferroelectric process parameters on cell conductivity, parasitic components, and nonideal characteristics in FeFET device arrays, with the goal of on‐device training for handwritten digit classification. This includes the inference accuracy and power efficiency characteristics of the FeFET‐based neural network (NN) array, along with the execution time of the customized simulation engine. The developed NSIP is then utilized to investigate the impact of the dynamic range of FeFET conductance, ferroelectric thickness, and line resistance on the performance of the NN. Power consumption and inference accuracy is measured at the same voltage bias over a range of ferroelectric thicknesses and the tradeoffs is analyzed in detail. The impact of line resistance on accuracy during the inference stage is studied, providing essential design guidance for the early stages of fabric.","author":[{"family":"Park","given":"Juhwan"},{"family":"Kim","given":"HJ"},{"family":"Cho","given":"Hyunbo"},{"family":"Jeon","given":"Jongwook"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1002/aisy.202500305","URL":"https://doi.org/10.1002/aisy.202500305","source":"openalex"},{"id":"oa:W4411004417","type":"article-journal","title":"Unveiling the switching mechanism of robust tetrazine-based memristive nociceptors via a spectroelectrochemical approach","abstract":") and nanosecond level switching time (60 ns), can be successfully optimized. Moreover, a spectroelectrochemical strategy was employed for the first time to investigate the RS mechanism at the molecular level, elucidating the critical role of molecular design in modulating the device's working principles and electrical characteristics. The optimized memristor is capable of accurately emulating the four key behaviors of nociceptors. This achievement not only advances the application of organic materials in neuromorphic devices but also opens up new possibilities for the specialized customization of nociceptors.","author":[{"family":"Zhao","given":"Jiyu"},{"family":"Liu","given":"Kun"},{"family":"Zeng","given":"Wei"},{"family":"Chen","given":"Zhuo"},{"family":"Zheng","given":"Yifan"},{"family":"Zhao","given":"Zherui"},{"family":"Zhong","given":"Wen‐min"},{"family":"Han","given":"Su‐ting"},{"family":"Ding","given":"Guanglong"},{"family":"Zhou","given":"Ye"},{"family":"Peng","given":"Xiaojun"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1039/d5sc02710a","URL":"https://doi.org/10.1039/d5sc02710a","source":"openalex"},{"id":"oa:W4413118591","type":"article-journal","title":"Roadmap on Optics and Photonics for Security and Encryption","abstract":"In 1994, Javidi and Horner published a paper in Optical Engineering that highlighted the ability of free space optical systems to manipulate sensitive data for authentication purposes. The underlying idea was effective yet surprisingly simple: an optical nonlinear joint transform using a random phase mask in both the input and the reference could produce a correlation peak to indicate whether the input object is authentic or not. This seminal paper fueled the development of this new discipline. After three decades, optical encryption and security have matured into a field that plays a central role in the development of photonics techniques. While the pioneering work was mainly focused on the field of optical information processing, nowadays, a broad spectrum of disciplines are contributing to developing security solutions, including nanotechnology, materials science, quantum information, and deep learning, just to cite a few. The present roadmap paper gathers 28 leading authors in the field from 21 academic institutions across nine different countries. It is organized into 17 sections which discuss the present and future challenges, state-of-the-art technology, and real-world solutions to address the security challenges facing our society.","author":[{"family":"Javidi","given":"Bahram"},{"family":"Carnicer","given":"Artur"},{"family":"Ahmadi","given":"Kavan"},{"family":"Awatsuji","given":"Yasuhiro"},{"family":"Chen","given":"Wen"},{"family":"Fournel","given":"Thierry"},{"family":"Genevet","given":"Patrice"},{"family":"Guo","given":"Jingying"},{"family":"He","given":"Wenqi"},{"family":"Hébert","given":"Mathieu"},{"family":"Jana","given":"Aloke"},{"family":"Lam","given":"Edmund"},{"family":"Long","given":"Gui‐lu"},{"family":"Matoba","given":"Osamu"},{"family":"Mi","given":"Zhenyu"},{"family":"Moon","given":"Inkyu"},{"family":"Nishchal","given":"Naveen"},{"family":"Pan","given":"Dong"},{"family":"Peng","given":"Xiang"},{"family":"Pinkse","given":"Pepijn"},{"family":"Shi","given":"Yishi"},{"family":"Situ","given":"Guohai"},{"family":"Stern","given":"Adrian"},{"family":"Wang","given":"Xiaogang"},{"family":"Xia","given":"Tian"},{"family":"Xiao","given":"Yin"},{"family":"Xie","given":"Zhenwei"},{"family":"Zhu","given":"Shuo"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1109/access.2025.3597226","URL":"https://doi.org/10.1109/access.2025.3597226","source":"openalex"},{"id":"oa:W4407184482","type":"article-journal","title":"Bioelectronics with Topological Crosslinked Networks for Tactile Perception","abstract":"Abstract Bioelectronics, which integrate biological systems with electronic components, have attracted significant attention in developing biomimetic materials and advanced hardware architectures to enable novel information‐processing systems, sensors, and actuators. However, the rigidity of conjugated molecular systems and the lack of reconfigurability in static crosslinked structures pose significant challenges for flexible sensing applications. Topological crosslinked networks (TCNs) featuring dynamic molecular interactions offer enhanced molecular flexibility and environmentally induced reconfigurability, decoupling the competition between performances. Here, recent advances are summarized in assembly methods of bioelectronics with different TCNs and elaborate ion/electron‐transport mechanisms from the perspective of molecular interactions. Decoupling effects can be achieved by comparing distinct TCNs and their respective properties, and an outlook is provided on a new range of neuromorphic hardware with biocompatibility, self‐healing, self‐powered, and multimodal‐sensing capabilities. The development of TCN‐based bioelectronics can significantly impact the fields of artificial neuromorphic perception devices, networks, and systems.","author":[{"family":"Ding","given":"Mingqi"},{"family":"Xie","given":"Pengshan"},{"family":"Ho","given":"Johnny"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1002/apxr.202400165","URL":"https://doi.org/10.1002/apxr.202400165","source":"openalex"},{"id":"oa:W4406703029","type":"article-journal","title":"Engineering Nonvolatile Polarization in 2D α-In2Se3/α-Ga2Se3 Ferroelectric Junctions","abstract":"The advent of two-dimensional (2D) ferroelectrics offers a new paradigm for device miniaturization and multifunctionality. Recently, 2D α-In2Se3 and related III–VI compound ferroelectrics manifest room-temperature ferroelectricity and exhibit reversible spontaneous polarization even at the monolayer limit. Here, we employ first-principles calculations to investigate group-III selenide van der Waals (vdW) heterojunctions built up by 2D α-In2Se3 and α-Ga2Se3 ferroelectric (FE) semiconductors, including structural stability, electrostatic potential, interfacial charge transfer, and electronic band structures. When the FE polarization directions of α-In2Se3 and α-Ga2Se3 are parallel, both the α-In2Se3/α-Ga2Se3 P↑↑ (UU) and α-In2Se3/α-Ga2Se3 P↓↓ (NN) configurations possess strong built-in electric fields and hence induce electron–hole separation, resulting in carrier depletion at the α-In2Se3/α-Ga2Se3 heterointerfaces. Conversely, when they are antiparallel, the α-In2Se3/α-Ga2Se3 P↓↑ (NU) and α-In2Se3/α-Ga2Se3 P↑↓ (UN) configurations demonstrate the switchable electron and hole accumulation at the 2D ferroelectric interfaces, respectively. The nonvolatile characteristic of ferroelectric polarization presents an innovative approach to achieving tunable n-type and p-type conductive channels for ferroelectric field-effect transistors (FeFETs). In addition, in-plane biaxial strain modulation has successfully modulated the band alignments of the α-In2Se3/α-Ga2Se3 ferroelectric heterostructures, inducing a type III–II–III transition in UU and NN, and a type I–II–I transition in UN and NU, respectively. Our findings highlight the great potential of 2D group-III selenides and ferroelectric vdW heterostructures to harness nonvolatile spontaneous polarization for next-generation electronics, nonvolatile optoelectronic memories, sensors, and neuromorphic computing.","author":[{"family":"Li","given":"Peipei"},{"family":"Kong","given":"Delin"},{"family":"Yang","given":"Jin"},{"family":"Cui","given":"Shuyu"},{"family":"Chen","given":"Qi"},{"family":"Liu","given":"Yue"},{"family":"He","given":"Zhike"},{"family":"Liu","given":"Feng"},{"family":"Xu","given":"Yingying"},{"family":"Wei","given":"Huiyun"},{"family":"Zheng","given":"Xinhe"},{"family":"Peng","given":"Mingzeng"}],"issued":{"date-parts":[[2025]]},"DOI":"10.3390/nano15030163","URL":"https://doi.org/10.3390/nano15030163","source":"openalex"},{"id":"oa:W4408047540","type":"article-journal","title":"LAST-PAIN: Learning Adaptive Spike Thresholds for Low Back Pain Biosignals Classification","abstract":"Spiking neural networks (SNNs) present the potential for ultra-low-power computation, especially when implemented on dedicated neuromorphic hardware. However, a significant challenge is the efficient conversion of continuous real-world data into the discrete spike trains required by SNNs. In this paper, we introduce Learning Adaptive Spike Thresholds (LAST), a novel, trainable encoding strategy designed to address this challenge. The LAST encoder learns adaptive thresholds to transform continuous signals of varying dimensionality-ranging from time series data to high dimensional tensors-into sparse spike trains. Our proposed encoder effectively preserves temporal dynamics and adapts to the characteristics of the input. We validate the LAST approach in a demanding healthcare application using the EmoPain dataset. This dataset contains multimodal biosignal analysis for assessing chronic lower back pain (CLBP). Despite the dataset's small sample size and class imbalance, our LAST-driven SNN framework achieves a competitive Matthews Correlation Coefficient of 0.44 and an accuracy of 80.43% in CLBP classification. The experimental results also indicate that the same framework can achieve an F1-score of 0.65 in detecting protective behaviour. Furthermore, the LAST encoder outperforms conventional rate and latency-based encodings while maintaining sparse spike representations. This achievement shows promises for energy-efficient and real-time biosignal processing in resource-limited environments.","author":[{"family":"Hens","given":"Freek"},{"family":"Dehshibi","given":"Mohammad"},{"family":"Bagheriye","given":"Leila"},{"family":"Tajadurajiménez","given":"Ana"},{"family":"Shahsavari","given":"Mahyar"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1109/tnsre.2025.3546682","URL":"https://doi.org/10.1109/tnsre.2025.3546682","source":"openalex"},{"id":"oa:W4412379260","type":"article-journal","title":"Quantized Conductance and Multilevel Memory Operation in Mn 3 O 4 Nanowire Network Devices Combined with Low Voltage Operation and Oxygen Vacancy Induced Resistive Switching","abstract":"Abstract Quantum effects in nanowires and nanodevices can potentially revolutionize the device concepts with multi‐functionalities for future technologies. Memristive devices which undergo transition from high resistance state to low resistance state involve nanoscale conduction paths can show quantum effects at room temperature. Here, Mn 3 O 4 nanowires based memristor showing very reliable resistive switching at very low voltages and with ON/OFF States ratio ∼ 10 3 is reported. The switching device can also be programmed to multiple memory states (up to 16 states ∼ 2 4 ). Since the conduction paths are geometrically constrained along the nanowires, quantized conductance steps are observed. Step‐wise conductance jumps are observed during the SET and RESET process with better control along RESET process. Conductance jumps range between 1 and 9 G 0 . The nanowire devices show very consistent resistive switching up to 100 °C. These measurements confirm extremely stable nanowire based resistive switching devices which can be used for next‐generation memories showing quantum effects in neuromorphic computing architectures.","author":[{"family":"Hadiyal","given":"Keval"},{"family":"Ganesan","given":"Ramakrishnan"},{"family":"Thamankar","given":"R"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1002/aelm.202500159","URL":"https://doi.org/10.1002/aelm.202500159","source":"openalex"},{"id":"oa:W4414285733","type":"article-journal","title":"Biomimetic Janus MXene membrane with bidirectional ion permselectivity for enhanced osmotic effects and iontronic logic control","abstract":"Osmotic efficiency is fundamentally governed by the balance between membrane ion selectivity and permeability, a challenge central to both biological signal transmission and sustainable energy conversion. Conventional membranes are constrained to unidirectional transport of either cations or anions, severely limiting their versatility and performance. Inspired by the chloride voltage-gated channel 5 (ClC-5), we engineered a biomimetic Janus NP-MXene membrane featuring subnanochannels (~6.0 angstrom) and exceptional structural integrity, enabling controlled, simultaneous Na + /Cl − transport with unprecedented permselectivity. Under a 50-fold salinity gradient, the NP-MXene membrane achieved a record power density of 85.1 watts per square meter and an osmotic potential of 181.5 millivolts, the highest reported for a single device. Harnessing ion-specific signals from multi-ion transport, we further demonstrated an iontronic transistor capable of modulating ion flow by salinity gradients, eliminating the need for external gate voltage. This advance enables encoded signals and robotic control for advanced human-machine interfaces. The scalable fabrication of nanofluidic channels facilitates high-performance iontronics for efficient energy-information flow.","author":[{"family":"Han","given":"Qian"},{"family":"Fan","given":"Hongzhao"},{"family":"Peng","given":"Puguang"},{"family":"Du","given":"Yan"},{"family":"Li","given":"Xiang"},{"family":"Liu","given":"Yanhui"},{"family":"Yang","given":"Feiyao"},{"family":"Zhou","given":"Yanguang"},{"family":"Wang","given":"Zhong"},{"family":"Wei","given":"Di"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1126/sciadv.adx1184","URL":"https://doi.org/10.1126/sciadv.adx1184","source":"openalex"},{"id":"oa:W7155223424","type":"article-journal","title":"Neuromorphic computing for radar and radio systems: a survey","abstract":"Abstract Taking inspiration from the brain on how to create energy efficient and low latency neuromorphic systems has the potential to create new opportunities with AI across many domains. Firstly, it creates a possibility to mitigate problems with too large digital signal processing costs in various technologies. Secondly, it also enables the use of AI and machine learning algorithms where it is currently impossible due to energy constraints. Recently, neuromorphic technology has been introduced to radio communication and radar applications. In this work, we highlight advantages of applying energy efficient, low latency and often lightweight neuromorphic computing for radar and radio signal processing. We perform a comprehensive review of the main current works on neuromorphic technology for radar applications, focusing on frequency-modulated continuous-wave and synthetic aperture radar. Additionally, we cover radio frequency signal classification for both radar and radio signals. Our ambition is to facilitate research on neuromorphic computing for radar and radio systems, as well as help bringing researchers from these fields together.","author":[{"family":"Hamrell","given":"Hanna"},{"family":"Sjögren","given":"Thomas"},{"family":"Ovrén","given":"Hannes"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/2634-4386/ae6369","URL":"https://doi.org/10.1088/2634-4386/ae6369","source":"openalex"},{"id":"oa:W4411727697","type":"article-journal","title":"Real-time human-robot interaction and service provision using hybrid intelligent computing framework","abstract":"Human-robot interaction has gained significant attention in various domains, including healthcare, customer service, and industrial automation. High computational cost, inefficient service matching, and elevated failure rates in dynamic service contexts are some primary disadvantages of existing query-processing systems. This research introduces a Hybrid Intelligent Computing Model (HICM) to improve robots' ability to process inquiries autonomously. The goal is to make robots better at responding to human questions in real time with efficient, personalized, and context-specific solutions. Using self-organized computing approaches, robotic agents can reliably provide end-users with services suited to their demands. Due to their autonomous nature, robots must be able to calculate quickly and accurately to provide timely services. To meet these needs, the proposed HICM incorporates a sophisticated decision-support system to handle human questions and find the appropriate services. Within this decision-making framework, the model evaluates the characteristics and relevance of questions about accessible services by combining annealing and Tabu Search approaches. To avoid addressing queries incompatibly, the Tabu Search technique approaches query resolution as a non-convergent optimization issue. Comparing HICM's performance to other models reveals significant improvements over CDS, DGTA, and CCS. In particular, HICM reduced calculation time by 8.67%, service time by 15.09%, and failure rates by 7.87%. In terms of important metrics, HICM fared better than the competing models. Its success factor was 11.8% higher, its matching ratio was 14.88% higher, and its failure rates were 6.22% lower. These findings demonstrate the model's efficiency and reliability in terms of robotic query processing and real-time service delivery.","author":[{"family":"Albekairi","given":"Mohammed"},{"family":"Alanazi","given":"Meshari"},{"family":"Alanazi","given":"Turki"},{"family":"Mohamed","given":"Mohamed"},{"family":"Kaâniche","given":"Khaled"},{"family":"Sahbani","given":"Anis"},{"family":"Elrashidi","given":"Ali"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1371/journal.pone.0324986","URL":"https://doi.org/10.1371/journal.pone.0324986","source":"openalex"},{"id":"oa:W4414069761","type":"article-journal","title":"Neural Synaptic Simulation Based on ZnAlSnO Thin-Film Transistors","abstract":"In the era of artificial intelligence, neuromorphic devices that simulate brain functions have received increasingly widespread attention. In this paper, an artificial neural synapse device based on ZnAlSnO thin-film transistors was fabricated, and its electrical properties were tested: the current-switching ratio was 1.18 × 107, the subthreshold oscillation was 1.48 V/decade, the mobility was 2.51 cm2V−1s−1, and the threshold voltage was −9.40 V. Stimulating artificial synaptic devices with optical signals has the advantages of fast response speed and good anti-interference ability. The basic biological synaptic characteristics of the devices were tested under 365 nm light stimulation, including excitatory postsynaptic current (EPSC), paired-pulse facilitation (PPF), short-term plasticity (STP), and long-term plasticity (LTP). This device shows good synaptic plasticity. In addition, by changing the gate voltage, the excitatory postsynaptic current of the device at different gate voltages was tested, two different logical operations of “AND” and “OR” were achieved, and the influence of different synaptic states on memory was simulated. This work verifies the application potential of the device in the integrated memory and computing architecture, which is of great significance for promoting the high-quality development of neuromorphic computing hardware.","author":[{"family":"Zhao","given":"Yang"},{"family":"Wang","given":"Chao"},{"family":"Ku","given":"Laizhe"},{"family":"Guo","given":"Liang"},{"family":"Chu","given":"Xuefeng"},{"family":"Yang","given":"Fan"},{"family":"Wang","given":"Jieyang"},{"family":"Zhao","given":"Chunlei"},{"family":"Chi","given":"Yaodan"},{"family":"Yang","given":"Xiaotian"}],"issued":{"date-parts":[[2025]]},"DOI":"10.3390/mi16091025","URL":"https://doi.org/10.3390/mi16091025","source":"openalex"},{"id":"oa:W4406829651","type":"article-journal","title":"Reliable, efficient, and scalable photonic inverse design empowered by physics‐inspired deep learning","abstract":"On-chip computing metasystems composed of multilayer metamaterials have the potential to become the next-generation computing hardware endowed with light-speed processing ability and low power consumption but are hindered by current design paradigms. To date, neither numerical nor analytical methods can balance efficiency and accuracy of the design process. To address the issue, a physics-inspired deep learning architecture termed electromagnetic neural network (EMNN) is proposed to enable an efficient, reliable, and flexible paradigm of inverse design. EMNN consists of two parts: EMNN Netlet serves as a local electromagnetic field solver; Huygens-Fresnel Stitch is used for concatenating local predictions. It can make direct, rapid, and accurate predictions of full-wave field based on input fields of arbitrary variations and structures of nonfixed size. With the aid of EMNN, we design computing metasystems that can perform handwritten digit recognition and speech command recognition. EMNN increases the design speed by 17,000 times than that of the analytical model and reduces the modeling error by two orders of magnitude compared to the numerical model. By integrating deep learning techniques with fundamental physical principle, EMNN manifests great interpretability and generalization ability beyond conventional networks. Additionally, it innovates a design paradigm that guarantees both high efficiency and high fidelity. Furthermore, the flexible paradigm can be applicable to the unprecedentedly challenging design of large-scale, high-degree-of-freedom, and functionally complex devices embodied by on-chip optical diffractive networks, so as to further promote the development of computing metasystems.","author":[{"family":"Shao","given":"Guocheng"},{"family":"Zhou","given":"Tiankuang"},{"family":"Yan","given":"Tao"},{"family":"Guo","given":"Yanchen"},{"family":"Yun","given":"Zhao"},{"family":"Huang","given":"Ruqi"},{"family":"Fang","given":"Lu"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1515/nanoph-2024-0504","URL":"https://doi.org/10.1515/nanoph-2024-0504","source":"openalex"},{"id":"oa:W7115181294","type":"article-journal","title":"Potentialities of electrochemical devices for memory and neuromorphic computing","abstract":"This chapter explores the transformative role of electrochemical devices in next-generation memory and neuromorphic computing systems. Bridging the gap between traditional energy applications and intelligent electronics, electrochemical platforms leverage ionic–electronic coupling to mimic synaptic behaviors found in biological systems. This chapter presents the working principles and architectures of key electrochemical memory technologies such as electrochemical random-access memory (ECRAM), organic electrochemical transistors (OECTs), and electrochemical synaptic transistors (ESTs), emphasizing their analog programmability, low-power operation, and integration potential. These devices offer a viable solution to the limitations of conventional von Neumann computing by co-locating memory and processing units, fostering energy-efficient, adaptive, and brain-inspired computing. The chapter also addresses critical material and architectural challenges, such as ion transport control, state retention, and scalability. With growing interest in neuromorphic engineering, electrochemical systems present an adaptable and interdisciplinary platform poised to drive innovation in edge computing, artificial intelligence, and bio-integrated electronics.","author":[{"family":"Rai","given":"Harshita"},{"family":"Singh","given":"Kshitij"},{"family":"Natarajan","given":"Arunadevi"},{"family":"Pandey","given":"Shyam"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1039/9781837070244-00001","URL":"https://doi.org/10.1039/9781837070244-00001","source":"openalex"},{"id":"oa:W7150754891","type":"article-journal","title":"Controlled oxygen vacancy electrode reservoir for robust WO3-based memory devices","abstract":"Abstract Oxide-based memristors are promising candidates as neuromorphic hardware in energy-efficient edge computing applications for the Internet of Things (IoT). However, achieving simultaneously optimized memristive and synaptic performance with low-temperature fabrication, compatible with complementary metal-oxide-semiconductor (CMOS) processes, remains a challenge. Here, we demonstrate a dual-function electrode selection strategy—controlled oxygen vacancy electrode reservoir—implemented in a simple, low-temperature-fabricated indium tin oxide (ITO)/WO 3 /TiN thin film system. We systematically vary the oxygen stoichiometry in the ITO top electrode, which serves not only as an electrical contact, but also as a dynamic oxygen reservoir. A high oxygen-vacancy concentration in the ITO enables optimized memristive performance of low electroforming and operation voltages, and robust memristive and synaptic endurance. This electrode-focused defect engineering approach offers a versatile route to advanced memristor design, enabling many applications in IoT and neuromorphic systems.","author":[{"family":"Yuan","given":"Ziyi"},{"family":"Bakhit","given":"Babak"},{"family":"Lu","given":"Jiahao"},{"family":"Liu","given":"Yi"},{"family":"Jan","given":"Atif"},{"family":"Li","given":"Xinjuan"},{"family":"Varghese","given":"Abin"},{"family":"Rajendran","given":"Bipin"},{"family":"Ducati","given":"Caterina"},{"family":"Martino","given":"Giuliana"},{"family":"Hellenbrand","given":"Markus"},{"family":"Macmanus-Driscoll","given":"Judith"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1038/s43246-026-01143-8","URL":"https://doi.org/10.1038/s43246-026-01143-8","source":"openalex"},{"id":"oa:W4415543396","type":"article-journal","title":"Ultra‐Sensitive Negative Photoconductivity Transistors via Long‐Afterglow Doping for All‐Optical Encryption","abstract":"Abstract Negative photoconductivity (NPC)‐based devices, characterised by the light‐induced suppression of electrical conduction, have garnered significant interest for their multifunctional optoelectronic applications, fast light response and broadband spectral adaptability. However, conventional planar heterostructured NPC‐based devices exhibit poor device performance owing to interfacial defects that disrupt carrier transport and recombination dynamics. In this study, an innovative bulk doping strategy is presented that incorporates organic long‐afterglow materials into polymer semiconductors to achieve high‐performance negative photoconductivity transistors (NPTs). The long‐afterglow dopants generate long‐lived charge separation states that effectively trap gate‐modulated majority carriers of polymer semiconductors, enabling persistent NPC with photosensitivity (5.29 × 10⁶) and detectivity (3.40 × 10 13 Jones). In addition, this bulk doping strategy creates abundant trapping sites, which enable intralayer carrier recombination within the doped semiconductor film while maintaining an ultrahigh photosensitivity. Notably, this strategy can be generalised across diverse dopant‐semiconductor systems. Furthermore, leveraging these exceptional NPTs, the negative synaptic functionalities are successfully emulated. To highlight its practical potential, system‐level applicability is demonstrated by integrating NPTs into a recurrent neural network (RNN) for all‐optical encryption/decryption, achieving up to 91% accuracy. This study establishes a general paradigm for high‐performance NPC‐based devices, unlocking their potential for next‐generation optoelectronics and secure neuromorphic systems.","author":[{"family":"Han","given":"Jiangli"},{"family":"Ma","given":"Ding"},{"family":"Jiang","given":"Lixian"},{"family":"Zhang","given":"Yu"},{"family":"Dong","given":"Meiqiu"},{"family":"Deng","given":"Yunfeng"},{"family":"Geng","given":"Yanhou"},{"family":"Huang","given":"Rui"},{"family":"Xu","given":"Cheng"},{"family":"Zheng","given":"Xin"},{"family":"Dong","given":"Guifang"},{"family":"Duan","given":"Lian"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1002/adma.202514723","URL":"https://doi.org/10.1002/adma.202514723","source":"openalex"},{"id":"oa:W7140112752","type":"article-journal","title":"Symmetric pulse-enabled highly linear analog switching in ALD-grown HfO2/Ta2O5-based memristor for multi-level storage and synaptic applications","abstract":"Analog memristors with multilevel cells are suitable for analog in-memory and neuromorphic applications. Herein, we report a 2-bit/cell complementary-metal-oxide-semiconductor (CMOS)-compatible HfO 2 /Ta 2 O 5 bilayer memristor with both TiN electrodes fabricated via single thermal atomic layer deposition at 300 °C. The fabricated devices exhibit stable bipolar switching characteristics distinguished between both low resistance state and the high resistance state with a P/E endurance of 10 5 WRITE cycles, as well as show better retention property beyond 10 4 s. The devices exhibited excellent uniformity in terms of low device-to-device (D2D) and low cycle-to-cycle (C2C) variation. Furthermore, analog switching responses are implemented with the pulse width from 2 ms to 500 µs, and the corresponding percentage change in the device resistance was measured. The results exhibit a significant change in the device resistance even at 500 µs, with an overall change in the device resistance in the range of 10%–17%. In addition, the performance of the devices has been verified for neuromorphic applications using the experimentally extracted data. The non-linearity of 0.07, including highly stable synaptic plasticity, has been achieved using symmetric pulses, making the devices compatible for designing an analog memristor-based neuromorphic computing system hardware.","author":[{"family":"Pal","given":"Parthasarathi"},{"family":"Kumar","given":"Sanjay"},{"family":"Prodromakis","given":"Themis"}],"issued":{"date-parts":[[2026]]},"DOI":"10.3389/fnano.2026.1788527","URL":"https://doi.org/10.3389/fnano.2026.1788527","source":"openalex"},{"id":"oa:W4416251859","type":"article-journal","title":"Dynamic Reservoir Computing with Physical Neuromorphic Networks","abstract":"Reservoir Computing (RC) with physical systems requires an understanding of the underlying structure and internal dynamics of the specific physical reservoir. In this study, physical nano–electronic networks with neuromorphic dynamics are investigated for their use as physical reservoirs in an RC framework. These neuromorphic networks operate as dynamic reservoirs, with node activities in general coupled to the edge dynamics through nonlinear nano–electronic circuit elements, and the reservoir outputs influenced by the underlying network connectivity structure. This study finds that networks with varying degrees of sparsity generate more useful nonlinear temporal outputs for dynamic RC compared to dense networks. Dynamic RC is also tested on an autonomous multivariate chaotic time series prediction task with networks of varying densities, which revealed the importance of network sparsity in maintaining network activity and overall dynamics, that in turn enabled the learning of the chaotic Lorenz63 system’s attractor behavior.","author":[{"family":"Xu","given":"Yinhao"},{"family":"Gottwald","given":"Georg"},{"family":"Kuncic","given":"Zdenka"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1109/ijcnn64981.2025.11229247","URL":"https://doi.org/10.1109/ijcnn64981.2025.11229247","source":"openalex"},{"id":"oa:W4416738940","type":"article-journal","title":"Low-Power Ionically Tunable Bilayer MoS 2 Synaptic Transistors","abstract":"High Resolution Image Download MS PowerPoint Slide Electrochemical random-access memory (ECRAM) devices offer the possibility of highly linear, energy-efficient conductance modulation via electrochemical doping. This makes them attractive as synaptic weights for neuromorphic applications. Herein, we demonstrate inorganic ECRAM transistors with two-dimensional (2D) bilayer MoS 2 channels exhibiting both ionic and electrostatic gating. Our devices show electrochemical modulation of the channel conductance and electrostatic gating to achieve idle-state leakage currents of I D < 100 fA (measurement limit). These unique capabilities are enabled by the 2D semiconductor channel, which supports both electrostatic (conventional field effect) and electrochemical, non-volatile gating. Unlike oxide-based ECRAMs, which typically operate in a degenerately doped regime, the crystalline 2D channel has a low baseline carrier density, which can be modulated electrostatically. Our devices demonstrate highly linear and symmetric training characteristics, evaluated using synaptic metrics commonly applied to memristive devices. These findings highlight the potential of 2D-based ECRAM devices for power-efficient synaptic electronics.","author":[{"family":"Levit","given":"Or"},{"family":"Ber","given":"Emanuel"},{"family":"Keller","given":"Yair"},{"family":"Minkovich","given":"Boris"},{"family":"Yalon","given":"Eilam"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1021/acs.nanolett.5c02372","URL":"https://doi.org/10.1021/acs.nanolett.5c02372","source":"openalex"},{"id":"oa:W4407838685","type":"article-journal","title":"Reconfigurable logic circuits and rectifier based on two-terminal ionic homojunctions","abstract":"Reconfigurable logic devices that can perform multiple functions with fewer components offer a solution for increasing data processing demands. Achieving reconfigurability without complex designs, however, remains challenging. In this study, we present a two-terminal reconfigurable homojunction device based on ionic-conducting CuInP 2 S 6 (CIPS). The device achieves tunable and reversible rectification by controlling Cu migration under an electric field. The homojunction is formed by modulating the contact potential of the CIPS channel, resulting in an asymmetric contact barrier. The resulting device has a rectification ratio of up to 4 × 10⁴, an ultra-low leakage current (∼100 fA), and a high breakdown voltage of 170 V, maintaining read stability over 3,000 cycles. It operates at frequencies up to 10 kHz and can be reconfigured as OR and AND logic states depending on the Cu migration direction of the CIPS homojunction.","author":[{"family":"Zhong","given":"Zhipeng"},{"family":"Cheng","given":"Xing"},{"family":"Song","given":"Wenqing"},{"family":"Yang","given":"Qianyi"},{"family":"Li","given":"Xiang"},{"family":"Wang","given":"Yuyang"},{"family":"Wang","given":"Wan"},{"family":"Zhuang","given":"Yezhao"},{"family":"Chen","given":"Yan"},{"family":"Shi","given":"Wu"},{"family":"Lin","given":"Tie"},{"family":"Meng","given":"Xiangjian"},{"family":"Huang","given":"Hai"},{"family":"Wang","given":"Jianlu"},{"family":"Chu","given":"Junhao"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1016/j.device.2025.100712","URL":"https://doi.org/10.1016/j.device.2025.100712","source":"openalex"},{"id":"oa:W7161758747","type":"article-journal","title":"Introducing sustainable neuromorphic computing in Engineering Mechanics","abstract":"Abstract We introduce a sustainable neuromorphic approach for numerical simulations in Engineering Mechanics. The finite element method (FEM) is widely used in engineering design; however, worldwide, there is no neuromorphic technology available in mechanics, even though the need for computational capacity with complex mechanical models is increasing 1 . AI-enhanced engineering approaches, such as agent-based models 2 , intensify the high energy demand coupled to CO 2 emissions 3,4 . However, the solution of mechanical boundary value problems in numerical simulations on neuromorphic chips accounting for physical and geometrical nonlinearities has not been researched. Here we show that complex mechanical phenomena can be approximated by spiking neural networks (SNN), thereby creating a new sparse signal transmission of mechanical state variables. We found that hybrid neural networks, composed of sparse and non-sparse neuronal activity, result in the best trade-off between accuracy and energy savings. The required energy is reduced by several orders of magnitude compared to classical numerical simulations, which can significantly lower CO 2 emissions in time-consuming computations. Due to their inherent path dependency, spike activation models are well-suited for nonlinear regression and physics-based approaches in nonlinear mechanics. Our results demonstrate how neuromorphic computing can be applied to a wide range of structural forming processes in engineering, since it differs fundamentally from neuromorphic classification studies in the literature and provides the basis for function approximation. Moreover, it is investigated how neural networks and non-machine learning operations can be deployed on Field-Programmable Gate Arrays (FPGAs). We anticipate our study to be a starting point for more sustainable AI models in engineering science and related disciplines.","author":[{"family":"Stoffel","given":"Marcus"},{"family":"Gulakala","given":"Rutwik"},{"family":"Polydoras","given":"Vasileios"},{"family":"Tandale","given":"Saurabh"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1038/s44387-026-00118-x","URL":"https://doi.org/10.1038/s44387-026-00118-x","source":"openalex"},{"id":"oa:W4410436654","type":"article-journal","title":"Enhancing reservoir predictions of chaotic time series by incorporating delayed values of input and reservoir variables","abstract":"Time series generated by chaotic dynamical systems can be effectively predicted using readouts from driven reservoir dynamics. In practical scenarios, however, only time series measurements with partial knowledge of the chaotic system's state are usually available. To address this aspect, we evaluate and compare the performance of reservoir computing in predicting time series under both conditions of complete and partial knowledge of the state. Our results show that memory improves the prediction accuracy only when the system state is partially known. For cases with partial state knowledge, we extend the mean prediction horizon by including delayed values of both the input and reservoir variables. To ensure the robustness of this result, we test it in systems with varying degrees of complexity. Finally, we show that the inclusion of delayed values can also facilitate the optimization of hyperparameters for predictions based on full knowledge of the system state.","author":[{"family":"Fleddermann","given":"Luk"},{"family":"Herzog","given":"Sebastian"},{"family":"Parlitz","given":"Ulrich"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1063/5.0258250","URL":"https://doi.org/10.1063/5.0258250","source":"openalex"},{"id":"oa:W4410620374","type":"article-journal","title":"Advances in AI-powered energy management systems for renewable-integrated smart grids","abstract":"The accelerating global shift toward renewable energy integration presents both a technical imperative and a systemic challenge to traditional power grid architectures. Variability, decentralization, and real-time balancing requirements have exposed the limitations of conventional control and forecasting strategies. This review critically examines how artificial intelligence (AI) is redefining energy management systems to meet the operational and strategic needs of renewable-integrated smart grids. It explores the state-of-the-art in AI-based load and generation forecasting, real-time grid state estimation, anomaly detection, and predictive maintenance, highlighting how machine learning and deep learning techniques enhance grid observability and fault resilience. Particular attention is given to AI-driven optimization of energy storage dispatch, multi-agent coordination in microgrids, and the deployment of edge intelligence for decentralized control. Furthermore, the review evaluates current barriers—ranging from data sparsity and model interpretability to lack of standardization—and proposes targeted research directions, including explainable AI, quantum-enhanced computing, and AI-powered coordination of distributed storage and vehicle-to-grid (V2G) networks. The convergence of AI, digital infrastructure, and policy innovation emerges as critical to unlocking the full potential of next-generation grids. This article provides researchers, engineers, and policymakers with a rigorous synthesis of current advancements and a forward-looking agenda for achieving intelligent, resilient, and decarbonized energy systems.","author":[{"family":"Egbuna","given":"Ifeanyi"},{"family":"Salihu","given":"Faisal"},{"family":"Okara","given":"Chinemeremma"},{"family":"Olayiwola","given":"Damilola"},{"family":"Smart","given":"E"},{"family":"Anifowose","given":"Olabode"},{"family":"Mbamalu","given":"Paul"}],"issued":{"date-parts":[[2025]]},"DOI":"10.30574/wjaets.2025.15.2.0685","URL":"https://doi.org/10.30574/wjaets.2025.15.2.0685","source":"openalex"},{"id":"oa:W4414701265","type":"article-journal","title":"MCU Intelligent Upgrades: An Overview of AI-Enabled Low-Power Technologies","abstract":"Microcontroller units (MCUs) serve as the core components of embedded systems. In the era of smart IoT, embedded devices are increasingly deployed on mobile platforms, leading to a growing demand for low-power consumption. As a result, low-power technology for MCUs has become increasingly critical. This paper systematically reviews the development history and current technical challenges of MCU low-power technology. It then focuses on analyzing system-level low-power optimization pathways for integrating MCUs with artificial intelligence (AI) technology, including lightweight AI algorithm design, model pruning, AI acceleration hardware (NPU, GPU), and heterogeneous computing architectures. It further elaborates on how AI technology empowers MCUs to achieve comprehensive low power consumption from four dimensions: task scheduling, power management, inference engine optimization, and communication and data processing. Through practical application cases in multiple fields such as smart home, healthcare, industrial automation, and smart agriculture, it verifies the significant advantages of MCUs combined with AI in performance improvement and power consumption optimization. Finally, this paper focuses on the key challenges that still need to be addressed in the intelligent upgrade of future MCU low power consumption and proposes in-depth research directions in areas such as the balance between lightweight model accuracy and robustness, the consistency and stability of edge-side collaborative computing, and the reliability and power consumption control of the sensor-storage-computing integrated architecture, providing clear guidance and prospects for future research.","author":[{"family":"Zhang","given":"Tong"},{"family":"Huang","given":"Ben"},{"family":"Liu","given":"Xiewen"},{"family":"Fan","given":"Jiaqi"},{"family":"Li","given":"Junbo"},{"family":"Zhao","given":"Yue"},{"family":"Wang","given":"Yanfang"}],"issued":{"date-parts":[[2025]]},"DOI":"10.3390/jlpea15040060","URL":"https://doi.org/10.3390/jlpea15040060","source":"openalex"},{"id":"oa:W7126270210","type":"article-journal","title":"Bioinspired Crossmodal Tactile Sensory Nerve for High‐Accurate Object Recognition","abstract":"Crossmodal perception enabled by the human somatosensory system can be followed to effectively perceive and analyze multiple sensory signals. Here, we construct an artificial crossmodal sensory neuron system by integrating pressure–temperature bimodal sensors with a Hf 0.5 Zr 0.5 O 2 ‐based complementary memristor, which emulates the tactile perception, neural coding, and synaptic processing functions of humans. With the developed bimodal sensor, the pressure and temperature information can be collected and further converted to electrical signals with excellent sensitivities of 26 407 kPa −1 and −3.34%°C −1 , respectively. The complementary memristor can enable information storage and simulate biological synaptic functions, achieving bioinspired neuromorphic processing of sensory signals. Combined with machine learning, this artificial crossmodal sensory neuron system presents an improved accuracy of 96.67% in recognizing temperatures and shapes of distinct objects. This work demonstrates potential applications of the integrated system in advanced neuromorphic hardware for wearable human‐machine interfaces and biomimetic robotics.","author":[{"family":"Chen","given":"Delu"},{"family":"Han","given":"Yuetong"},{"family":"Wang","given":"Xiaodong"},{"family":"Cheng","given":"Wanli"},{"family":"He","given":"Jianjie"},{"family":"Li","given":"Xing"},{"family":"Oshima","given":"Yoshifumi"},{"family":"Shan","given":"Chongxin"},{"family":"Cheng","given":"Shaobo"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1002/eem2.70279","URL":"https://doi.org/10.1002/eem2.70279","source":"openalex"},{"id":"doi:10.6082/884kt-fyn36","type":"article-journal","title":"Elucidating dynamic conductive state changes in amorphous lithium lanthanum titanate for resistive switching devices","abstract":"Exploration of novel resistive switching materials attracts attention to achieve neuromorphic computing that can surpass the limit of the current Von-Neumann computing for the time of Internet of Things (IoT). Battery materials priorly used to serve as an electrode or electrolyte have demonstrated metal-insulator transitions upon an electrical biasing due to resulting compositional change, which is desirable property for future resistive switching devices. For example, it has been suggested that solid-state electrolyte amorphous lithium lanthanum titanate (a-LLTO) changes in electronic conductivity depending on oxygen content. In this work, switching behavior of a-LLTO was investigated at both bulk- and nano-scale by employing a range of voltage sweep techniques, ultimately establishing a stable and optimal operating condition within the voltage window of − 3.5 V to 3.5 V. This voltage range effectively balances the desirable trait of a substantial resistance change by three orders of magnitude with the imperative avoidance of LLTO decomposition. Experiment and computation with different LLTO composition shows that LLTO has two distinct conductivity states due to Ti reduction. The distribution of these two states is discussed using simplified binary model, implying the conductive filament growth during low resistance state. Consequently, our study deepens understanding of LLTO electronic properties and encourages the interdisciplinary application of battery materials for resistive switching devices.","author":[{"family":"Shimizu","given":"Ryosuke"},{"family":"Cheng","given":"Diyi"},{"family":"Zhu","given":"Guomin"},{"family":"Han","given":"Bing"},{"family":"Marchese","given":"Thomas"},{"family":"Burger","given":"Randall"},{"family":"Xu","given":"Mingjie"},{"family":"Pan","given":"Xiaoqing"},{"family":"Zhang","given":"Minghao"},{"family":"Meng","given":"Ying"}],"issued":{"date-parts":[[2023]]},"DOI":"10.6082/884kt-fyn36","URL":"https://doi.org/10.6082/884kt-fyn36","source":"datacite"},{"id":"doi:10.6082/2gh2j-ay841","type":"article-journal","title":"Elucidating dynamic conductive state changes in amorphous lithium lanthanum titanate for resistive switching devices","abstract":"Exploration of novel resistive switching materials attracts attention to achieve neuromorphic computing that can surpass the limit of the current Von-Neumann computing for the time of Internet of Things (IoT). Battery materials priorly used to serve as an electrode or electrolyte have demonstrated metal-insulator transitions upon an electrical biasing due to resulting compositional change, which is desirable property for future resistive switching devices. For example, it has been suggested that solid-state electrolyte amorphous lithium lanthanum titanate (a-LLTO) changes in electronic conductivity depending on oxygen content. In this work, switching behavior of a-LLTO was investigated at both bulk- and nano-scale by employing a range of voltage sweep techniques, ultimately establishing a stable and optimal operating condition within the voltage window of − 3.5 V to 3.5 V. This voltage range effectively balances the desirable trait of a substantial resistance change by three orders of magnitude with the imperative avoidance of LLTO decomposition. Experiment and computation with different LLTO composition shows that LLTO has two distinct conductivity states due to Ti reduction. The distribution of these two states is discussed using simplified binary model, implying the conductive filament growth during low resistance state. Consequently, our study deepens understanding of LLTO electronic properties and encourages the interdisciplinary application of battery materials for resistive switching devices.","author":[{"family":"Shimizu","given":"Ryosuke"},{"family":"Cheng","given":"Diyi"},{"family":"Zhu","given":"Guomin"},{"family":"Han","given":"Bing"},{"family":"Marchese","given":"Thomas"},{"family":"Burger","given":"Randall"},{"family":"Xu","given":"Mingjie"},{"family":"Pan","given":"Xiaoqing"},{"family":"Zhang","given":"Minghao"},{"family":"Meng","given":"Ying"}],"issued":{"date-parts":[[2023]]},"DOI":"10.6082/2gh2j-ay841","URL":"https://doi.org/10.6082/2gh2j-ay841","source":"datacite"},{"id":"doi:10.1088/2634-4386/ad6cef","type":"article-journal","title":"Understanding the functional roles of modelling components in spiking neural networks","abstract":"Abstract Spiking neural networks (SNNs), inspired by the neural circuits of the brain, are promising in achieving high computational efficiency with biological fidelity. Nevertheless, it is quite difficult to optimize SNNs because the functional roles of their modelling components remain unclear. By designing and evaluating several variants of the classic model, we systematically investigate the functional roles of key modelling components, leakage, reset, and recurrence, in leaky integrate-and-fire (LIF) based SNNs. Through extensive experiments, we demonstrate how these components influence the accuracy, generalization, and robustness of SNNs. Specifically, we find that the leakage plays a crucial role in balancing memory retention and robustness, the reset mechanism is essential for uninterrupted temporal processing and computational efficiency, and the recurrence enriches the capability to model complex dynamics at a cost of robustness degradation. With these interesting observations, we provide optimization suggestions for enhancing the performance of SNNs in different scenarios. This work deepens the understanding of how SNNs work, which offers valuable guidance for the development of more effective and robust neuromorphic models.","author":[{"family":"Yin","given":"Huifeng"},{"family":"Zheng","given":"Hanle"},{"family":"Mao","given":"Jiayi"},{"family":"Ding","given":"Siyuan"},{"family":"Liu","given":"Xing"},{"family":"Xu","given":"Mingkun"},{"family":"Hu","given":"Yifan"},{"family":"Pei","given":"Jing"},{"family":"Deng","given":"Lei"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1088/2634-4386/ad6cef","URL":"https://doi.org/10.1088/2634-4386/ad6cef","source":"openalex"},{"id":"doi:10.48550/arxiv.2410.23639","type":"manuscript","title":"Integrating Brain-Computer Interface and Neuromorphic Computing for Human Digital Twins","abstract":"The integration of immersive communication into a human-centric ecosystem has intensified the demand for sophisticated Human Digital Twins (HDTs) driven by multifaceted human data. However, the effective construction of HDTs faces significant challenges due to the heterogeneity of data collection devices, the high energy demands associated with processing intricate data, and concerns over the privacy of sensitive information. This work introduces a novel biologically-inspired (bio-inspired) HDT framework that leverages Brain-Computer Interface (BCI) sensor technology to capture brain signals as the data source for constructing HDT. By collecting and analyzing these signals, the framework not only minimizes device heterogeneity and enhances data collection efficiency, but also provides richer and more nuanced physiological and psychological data for constructing personalized HDTs. To this end, we further propose a bio-inspired neuromorphic computing learning model based on the Spiking Neural Network (SNN). This model utilizes discrete neural spikes to emulate the way of human brain processes information, thereby enhancing the system's ability to process data effectively while reducing energy consumption. Additionally, we integrate a Federated Learning (FL) strategy within the model to strengthen data privacy. We then conduct a case study to demonstrate the performance of our proposed twofold bio-inspired scheme. Finally, we present several challenges and promising directions for future research of HDTs driven by bio-inspired technologies.","author":[{"family":"Shang","given":"Chen"},{"family":"Yu","given":"Jiadong"},{"family":"Hoang","given":"Dinh"}],"issued":{"date-parts":[[2024]]},"DOI":"10.48550/arxiv.2410.23639","URL":"https://doi.org/10.48550/arxiv.2410.23639","source":"datacite"},{"id":"doi:10.1038/s41598-024-57691-x","type":"article-journal","title":"An efficient intrusion detection model based on convolutional spiking neural network.","abstract":"Many intrusion detection techniques have been developed to ensure that the target system can function properly under the established rules. With the booming Internet of Things (IoT) applications, the resource-constrained nature of its devices makes it urgent to explore lightweight and high-performance intrusion detection models. Recent years have seen a particularly active application of deep learning (DL) techniques. The spiking neural network (SNN), a type of artificial intelligence that is associated with sparse computations and inherent temporal dynamics, has been viewed as a potential candidate for the next generation of DL. It should be noted, however, that current research into SNNs has largely focused on scenarios where limited computational resources and insufficient power sources are not considered. Consequently, even state-of-the-art SNN solutions tend to be inefficient. In this paper, a lightweight and effective detection model is proposed. With the help of rational algorithm design, the model integrates the advantages of SNNs as well as convolutional neural networks (CNNs). In addition to reducing resource usage, it maintains a high level of classification accuracy. The proposed model was evaluated against some current state-of-the-art models using a comprehensive set of metrics. Based on the experimental results, the model demonstrated improved adaptability to environments with limited computational resources and energy sources.","author":[{"family":"Wang","given":"Zhen"},{"family":"Ghaleb","given":"Fuad"},{"family":"Zainal","given":"Anazida"},{"family":"Siraj","given":"Maheyzah"},{"family":"Lü","given":"Xing"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1038/s41598-024-57691-x","URL":"https://doi.org/10.1038/s41598-024-57691-x","source":"europepmc"},{"id":"doi:10.3929/ethz-b-000716394","type":"article-journal","title":"A bio-inspired hardware implementation of an analog spike-based hippocampus memory model","abstract":"The need for processing at the edge the increasing amount of data that is being produced by multitudes of sensors has led to the demand for mode power efficient computational systems, by exploring alternative computing paradigms and technologies. Neuromorphic engineering is a promising approach that can address this need by developing electronic systems that faithfully emulate the computational properties of animal brains. In particular, the hippocampus stands out as one of the most relevant brain region for implementing auto associative memories capable of learning large amounts of information quickly and recalling it efficiently. In this work, we present a computational spike-based memory model inspired by the hippocampus that takes advantage of the features of analog electronic circuits: energy efficiency, compactness, and real-time operation. This model can learn memories, recall them from a partial fragment and forget. It has been implemented as a Spiking Neural Networks directly on a mixed-signal neuromorphic chip. We describe the details of the hardware implementation and demonstrate its operation via a series of benchmark experiments, showing how this research prototype paves the way for the development of future robust and low-power mixed-signal neuromorphic processing systems.","author":[{"family":"Casanueva-Morato","given":"Daniel"},{"family":"Ayuso-Martinez","given":"Alvaro"},{"family":"Indiveri","given":"Giacomo"},{"family":"Dominguez-Morales","given":"JP"},{"family":"Jimenez-Moreno","given":"Gabriel"}],"issued":{"date-parts":[[2024]]},"DOI":"10.3929/ethz-b-000716394","URL":"https://doi.org/10.3929/ethz-b-000716394","source":"datacite"},{"id":"doi:10.48550/arxiv.2310.03111","type":"manuscript","title":"Multi-modal Gaussian Process Variational Autoencoders for Neural and Behavioral Data","abstract":"Characterizing the relationship between neural population activity and behavioral data is a central goal of neuroscience. While latent variable models (LVMs) are successful in describing high-dimensional time-series data, they are typically only designed for a single type of data, making it difficult to identify structure shared across different experimental data modalities. Here, we address this shortcoming by proposing an unsupervised LVM which extracts temporally evolving shared and independent latents for distinct, simultaneously recorded experimental modalities. We do this by combining Gaussian Process Factor Analysis (GPFA), an interpretable LVM for neural spiking data with temporally smooth latent space, with Gaussian Process Variational Autoencoders (GP-VAEs), which similarly use a GP prior to characterize correlations in a latent space, but admit rich expressivity due to a deep neural network mapping to observations. We achieve interpretability in our model by partitioning latent variability into components that are either shared between or independent to each modality. We parameterize the latents of our model in the Fourier domain, and show improved latent identification using this approach over standard GP-VAE methods. We validate our model on simulated multi-modal data consisting of Poisson spike counts and MNIST images that scale and rotate smoothly over time. We show that the multi-modal GP-VAE (MM-GPVAE) is able to not only identify the shared and independent latent structure across modalities accurately, but provides good reconstructions of both images and neural rates on held-out trials. Finally, we demonstrate our framework on two real world multi-modal experimental settings: Drosophila whole-brain calcium imaging alongside tracked limb positions, and Manduca sexta spike train measurements from ten wing muscles as the animal tracks a visual stimulus.","author":[{"family":"Gondur","given":"Rabia"},{"family":"Sikandar","given":"Usama"},{"family":"Schaffer","given":"Evan"},{"family":"Aoi","given":"Mikio"},{"family":"Keeley","given":"Stephen"}],"issued":{"date-parts":[[2023]]},"DOI":"10.48550/arxiv.2310.03111","URL":"https://doi.org/10.48550/arxiv.2310.03111","source":"datacite"},{"id":"doi:10.5445/ir/1000182207","type":"article-journal","title":"Spiking Neural Belief Propagation Decoder for LDPC Codes with Small Variable Node Degrees","abstract":"Spiking neural networks (SNNs) promise energy-efficient data processing by imitating the event-based behavior of biological neurons. In previous work, we introduced the enlarge-likelihood-each-notable-amplitude spiking-neural-network (ELENA-SNN) decoder, a novel decoding algorithm for low-density parity-check (LDPC) codes. The decoder integrates SNNs into belief propagation (BP) decoding by approximating the check node (CN) update equation using SNNs. However, when decoding LDPC codes with a small variable node(VN) degree, the approximation gets too rough, and the ELENA-SNN decoder does not yield good results. This paper introduces the multi-level ELENA-SNN (ML-ELENA-SNN) decoder, which is an extension of the ELENA-SNN decoder. Instead of a single SNN approximating the CN update, multiple SNNs are applied in parallel, resulting in a higher resolution and higher dynamic range of the exchanged messages. We show that the ML-ELENA-SNN decoder performs similarly to the ubiquitous normalized min-sum decoder for the (38400, 30720) regular LDPC code with a VN degree of dv = 3 and a CN degree of dc = 15.","author":[{"family":"Bank","given":"Alexander"},{"family":"Edelmann","given":"Eike"},{"family":"Mandelbaum","given":"Jonathan"},{"family":"Schmalen","given":"Laurent"}],"issued":{"date-parts":[[2024]]},"DOI":"10.5445/ir/1000182207","URL":"https://doi.org/10.5445/ir/1000182207","source":"datacite"},{"id":"doi:10.5061/dryad.f1vhhmh4m","type":"article-journal","title":"Impact of background input on memory consolidation in In-Vitro neural networks","abstract":"Memory consolidation is a complex process, that can be divided into two stages: first, memories are temporary stored in hippocampus and in the second stage, repeated replay slowly transfers memories to the neo-cortex for long-term consolidation. This 2nd stage occurs during slow wave sleep, a phase characterized in the cortex by low cholinergic tone and low afferent input. A recent in-vitro study showed that high cholinergic tone hampers memory consolidation, probably due to lowered network excitability (defined as the mean network response to one neuron spiking). Here we investigate whether low background input contributes to memory consolidation. We used cortical neuronal networks on multi electrode arrays to study memory. When input deprived, these networks develop an activity-connectivity balance. Focal stimuli initially disrupt the existing balance, inducing connectivity changes. When repeated, this effect fades and the response becomes part of spontaneous patterns (memory formation). Application of the same stimulus hours later does not affect connectivity indicating that memory was consolidated. We applied five periods (10 min each) of focal electrical stimulation at different electrodes (A B A), separated by 1 hour of spontaneous activity . Some cultures were transfected to express channelrhopsins (ChR2) enabling global optogenetic background stimulation. We used 12 control, 15 ChR2 cultures with no background input and 8 ChR2 cultures with superimposed random optogenetic stimulation during electrical stimulation periods (fmean=5 Hz) to mimic afferent input. Background stimulation acutely reduced network excitability during stimulation without persisting effects after cessation. ChR2 cultures showed significantly more dispersed spiking outside network bursts, and network excitability tended to be lower than in control cultures. Stimulation at electrodes A and B induced memory traces in control cultures. Return to electrode A did not further affect connectivity, showing that memory trace A had been consolidated. Background stimulation impeded the formation of memory traces following electrical stimulation at either electrode. ChR2 expression alone also obstructed memorization. These findings confirm the importance of low background afferent input for memory consolidation. The presence of background afferent inputs reduced network excitability, similar to high cholinergic tone. This leads to the conclusion that sufficient network excitability is crucial for memory consolidation, and high network excitability may be a critical feature of slow wave sleep that makes it more suitable for memory consolidation than the awake state.","author":[{"family":"Lamberti","given":"Martina"},{"family":"Kikirikis","given":"Nikolaos"},{"family":"Van Putten","given":"Michel"},{"family":"Le Feber","given":"Joost"}],"issued":{"date-parts":[[2024]]},"DOI":"10.5061/dryad.f1vhhmh4m","URL":"https://doi.org/10.5061/dryad.f1vhhmh4m","source":"datacite"},{"id":"doi:10.5061/dryad.ffbg79d2w","type":"article-journal","title":"Data from: Brain-state mediated modulation of inter-laminar dependencies in visual cortex","abstract":"Spatial attention is critical for recognizing behaviorally relevant objects in a cluttered environment. How the deployment of spatial attention aids the hierarchical computations of object recognition remains unclear. We investigated this in the laminar cortical network of visual area V4, an area strongly modulated by attention. We found that deployment of attention strengthened unique dependencies in neural activity across cortical layers. On the other hand, shared dependencies were reduced within the excitatory population of a layer. Surprisingly, attention strengthened unique dependencies within a laminar population. Crucially, these modulation patterns were also observed during successful behavioral outcomes that are thought to be mediated by internal brain state fluctuations. Successful behavioral outcomes were also associated with phases of reduced neural excitability, suggesting a mechanism for enhanced information transfer during optimal states. Our results suggest common computation goals of optimal sensory states that are attained by either task demands or internal fluctuations.","author":[{"family":"Das","given":"Anirban"},{"family":"Sheffield","given":"Alec"},{"family":"Nandy","given":"Anirvan"},{"family":"Jadi","given":"Monika"}],"issued":{"date-parts":[[2024]]},"DOI":"10.5061/dryad.ffbg79d2w","URL":"https://doi.org/10.5061/dryad.ffbg79d2w","source":"datacite"},{"id":"doi:10.5061/dryad.rxwdbrvh3","type":"article-journal","title":"Spike-timing based coding in neuromimetic tactile system enables dynamic object classification","abstract":"Coding dynamic tactile information in spike timing is essential to human haptic exploration and dexterous object manipulation. Conventional electronic skins generate frames of tactile signals upon interaction with objects and are unfortunately ill-suited for efficient coding of temporal information and rapid feature extraction. Here, we report a neuromorphic tactile system that uses spike timing, especially the first-spike timing, to code dynamic tactile information about touch and grasp. This strategy enables the system to seamlessly code highly dynamic information with millisecond temporal resolution on par with the biological nervous system, yielding dynamic extraction of tactile features. Upon interaction with objects, the system rapidly classifies them in the initial phase of touch and grasp, thus paving the way to fast tactile feedback desired for neuro-robotics and neuro-prosthetics.","author":[{"family":"Chen","given":"Libo"},{"family":"Karilanova","given":"Sanja"},{"family":"Chaki","given":"Soumi"},{"family":"Wen","given":"Chenyu"},{"family":"Wang","given":"Lisha"},{"family":"Winblad","given":"Bengt"},{"family":"Zhang","given":"Shili"},{"family":"Ozcelikkale","given":"Ayca"},{"family":"Zhang","given":"Zhibin"}],"issued":{"date-parts":[[2024]]},"DOI":"10.5061/dryad.rxwdbrvh3","URL":"https://doi.org/10.5061/dryad.rxwdbrvh3","source":"datacite"},{"id":"doi:10.5061/dryad.sn02v6xbb","type":"article-journal","title":"Data for: Brain control of bimanual movement enabled by recurrent neural networks","abstract":"Brain-computer interfaces have so far focused largely on enabling the control of a single effector, for example a single computer cursor or robotic arm. Restoring multi-effector motion could unlock greater functionality for people with paralysis (e.g., bimanual movement). However, it may prove challenging to decode the simultaneous motion of multiple effectors, as we recently found that a compositional neural code links movements across all limbs and that neural tuning changes nonlinearly during dual-effector motion. In this study, we demonstrate the feasibility of high-quality bimanual control of two cursors via neural network (NN) decoders. This dataset represents all neural activity recorded during these experiments. This includes the neural activity corresponding to unimanual and bimanual hand movements during (1) instructed delay experiments and (2) real-time BCI control of two cursors. Code associated with the data can be found here: https://github.com/d-r-deo/bimanualBCI The journal article can be found here: https://doi.org/10.1038/s41598-024-51617-3","author":[{"family":"Deo","given":"Darrel"},{"family":"Willett","given":"Francis"},{"family":"Avansino","given":"Donald"},{"family":"Hochberg","given":"Leigh"},{"family":"Henderson","given":"Jaimie"},{"family":"Shenoy","given":"Krishna"}],"issued":{"date-parts":[[2024]]},"DOI":"10.5061/dryad.sn02v6xbb","URL":"https://doi.org/10.5061/dryad.sn02v6xbb","source":"datacite"},{"id":"doi:10.5061/dryad.7h44j1013","type":"article-journal","title":"The flow of axonal information among hippocampal subregions: 2. Patterned stimulation sharpens routing of information transmission","abstract":"The subregions of the hippocampal formation are essential for episodic learning and memory formation, yet the spike dynamics of each region contributing to this function are poorly understood, in part because of a lack of access to the inter-regional communicating axons. Here we reconstructed hippocampal networks confined to four subcompartments in 2D cultures on a multi-electrode array that monitors individual communicating axons. In our novel device, somal and axonal activity were measured simultaneously with the ability to ascertain the direction and speed of information transmission. Each subregion and inter-regional axons had unique power-law spiking dynamics, indicating differences in computational functions. After stimulation, spiking and burst rates decreased in all subregions, spikes per burst generally decreased, intraburst spike rates increased, and burst duration decreased, which were specific for each subregion. These changes in spiking dynamics post-stimulation were found to occupy a narrow range, consistent with the maintenance of the network at a critical state. Functional connections between the subregion neurons and communicating axons in our device revealed homeostatic network routing strategies post-stimulation in which spontaneous feedback activity was selectively decreased and balanced by decreased feed-forward activity. Post-stimulation, the number of functional connections per array decreased, but the reliability of those connections increased. The networks maintained a balance in spiking and bursting dynamics in response to stimulation and sharpened network routing. These plastic characteristics of the network revealed the dynamic architecture of hippocampal computations in response to stimulation by selective routing on a spatiotemporal scale in single axons.","author":[{"family":"Lassers","given":"Samuel"},{"family":"Vakilna","given":"Yash"},{"family":"Tang","given":"William"},{"family":"Brewer","given":"Gregory"}],"issued":{"date-parts":[[2023]]},"DOI":"10.5061/dryad.7h44j1013","URL":"https://doi.org/10.5061/dryad.7h44j1013","source":"datacite"},{"id":"doi:10.5061/dryad.w3r2280w7","type":"article-journal","title":"Stimulus encoding by specific inactivation of cortical neurons","abstract":"Neuronal ensembles are groups of neurons with correlated activity associated with sensory, motor, and behavioral functions. To explore how ensembles encode information, we investigated responses of visual cortical neurons in awake mice using volumetric two-photon calcium imaging during visual stimulation. We identified neuronal ensembles employing an unsupervised model-free algorithm and, besides neurons activated by the visual stimulus (termed “onsemble”), we also found neurons that are specifically inactivated (termed “offsemble”). Offsemble neurons showed faster calcium decay during stimuli, suggesting selective inhibition. In response to visual stimuli, each ensemble (onsemble+offsemble) exhibited small trial-to-trial variability, high orientation selectivity, and superior predictive accuracy for visual stimulus orientation, surpassing the sum of individual neuron activity. Thus, the combined selective activation and inactivation of cortical neurons enhance visual encoding as an emergent and distributed neural code.","author":[{"family":"Pérez-Ortega","given":"Jesús"},{"family":"Akrouh","given":"Alejandro"},{"family":"Yuste","given":"Rafael"}],"issued":{"date-parts":[[2024]]},"DOI":"10.5061/dryad.w3r2280w7","URL":"https://doi.org/10.5061/dryad.w3r2280w7","source":"datacite"},{"id":"doi:10.5061/dryad.66t1g1k2w","type":"article-journal","title":"Patch-clamp recordings from dorsal raphe neurons","abstract":"This dataset contains whole-cell electrophysiological recordings (patch-clamp recordings) from three cell types in mice: serotonin (5-HT) neurons, somatostatin (SOM)-expressing GABA interneurons, and layer 5 pyramidal neurons. 5-HT and GABA neurons were recorded in the dorsal raphe nucleus (DRN), which is the main source of serotonergic input to the forebrain. Together, they make up the majority of the neurons found in the DRN. This dataset can be used to investigate the intrinsic electrophysiological properties of these two types of DRN neurons and contrast them with another abundant and well-studied cell type, the L5 pyramidal neuron. This data was used in our paper describing a spiking neural network model of the dorsal raphe nucleus: Emerson F. Harkin, Michael B. Lynn, Alexandre Payeur, Jean-François Boucher, Léa Caya-Bissonnette, Dominic Cyr, Chloe Stewart, André Longtin, Richard Naud, and Jean-Claude Béïque. Temporal derivative computation in the dorsal raphe network revealed by an experimentally-driven augmented integrate-and-fire modeling framework. eLife, 2023. doi: 10.7554/eLife.72951","author":[{"family":"Harkin","given":"Emerson"},{"family":"Lynn","given":"Michael"},{"family":"Boucher","given":"Jean"},{"family":"Caya-Bissonnette","given":"Léa"},{"family":"Cyr","given":"Dominic"},{"family":"Stewart","given":"Chloe"},{"family":"Béïque","given":"Jean"}],"issued":{"date-parts":[[2023]]},"DOI":"10.5061/dryad.66t1g1k2w","URL":"https://doi.org/10.5061/dryad.66t1g1k2w","source":"datacite"},{"id":"doi:10.3929/ethz-b-000716329","type":"article-journal","title":"Genetic Motifs as a Blueprint for Mismatch-Tolerant Neuromorphic Computing","abstract":"Mixed-signal implementations of SNNs offer a promising solution to edge computing applications that require low-power and compact embedded processing systems. However, device mismatch in the analog circuits of these neuromorphic processors poses a significant challenge to the deployment of robust processing in these systems. Here we introduce a novel architectural solution inspired by biological development to address this issue. Specifically we propose to implement architectures that incorporate network motifs found in developed brains through a differentiable re-parameterization of weight matrices based on gene expression patterns and genetic rules. Thanks to the gradient descent optimization compatibility of the method proposed, we can apply the robustness of biological neural development to neuromorphic computing.","author":[{"family":"Boccato","given":"Tommaso"},{"family":"Zendrikov","given":"Dmitrii"},{"family":"Toschi","given":"Nicola"},{"family":"Indiveri","given":"Giacomo"}],"issued":{"date-parts":[[2024]]},"DOI":"10.3929/ethz-b-000716329","URL":"https://doi.org/10.3929/ethz-b-000716329","source":"datacite"},{"id":"doi:10.48550/arxiv.2412.06124","type":"manuscript","title":"Spiking Neural Networks for Radio Frequency Interference Detection in Radio Astronomy","abstract":"Spiking Neural Networks (SNNs) promise efficient and dynamic spatio-temporal data processing. This paper reformulates a significant challenge in radio astronomy, Radio Frequency Interference (RFI) detection, as a time-series segmentation task suited for SNN execution. Automated systems capable of real-time operation with minimal energy consumption are increasingly important in modern radio telescopes. We explore several spectrogram encoding methods and network parameters, applying first and second-order leaky integrate and fire SNNs to tackle RFI detection. We introduce a divisive normalisation-inspired pre-processing step, improving detection performance across multiple encodings strategies. Our approach achieves competitive performance on a synthetic dataset and compelling initial results on real data from the Low-Frequency Array (LOFAR). We position SNNs as a viable path towards real-time RFI detection, with many possibilities for follow-up studies. These findings highlight the potential for SNNs performing complex time-series tasks, paving the way towards efficient, real-time processing in radio astronomy and other data-intensive fields.","author":[{"family":"Pritchard","given":"Nicholas"},{"family":"Wicenec","given":"Andreas"},{"family":"Bennamoun","given":"Mohammed"},{"family":"Dodson","given":"Richard"}],"issued":{"date-parts":[[2024]]},"DOI":"10.48550/arxiv.2412.06124","URL":"https://doi.org/10.48550/arxiv.2412.06124","source":"datacite"},{"id":"doi:10.48550/arxiv.2405.18828","type":"manuscript","title":"CHANI: Correlation-based Hawkes Aggregation of Neurons with bio-Inspiration","abstract":"The present work aims at proving mathematically that a neural network inspired by biology can learn a classification task thanks to local transformations only. In this purpose, we propose a spiking neural network named CHANI (Correlation-based Hawkes Aggregation of Neurons with bio-Inspiration), whose neurons activity is modeled by Hawkes processes. Synaptic weights are updated thanks to an expert aggregation algorithm, providing a local and simple learning rule. We were able to prove that our network can learn on average and asymptotically. Moreover, we demonstrated that it automatically produces neuronal assemblies in the sense that the network can encode several classes and that a same neuron in the intermediate layers might be activated by more than one class, and we provided numerical simulations on synthetic dataset. This theoretical approach contrasts with the traditional empirical validation of biologically inspired networks and paves the way for understanding how local learning rules enable neurons to form assemblies able to represent complex concepts.","author":[{"family":"Jaffard","given":"Sophie"},{"family":"Vaiter","given":"Samuel"},{"family":"Reynaud-Bouret","given":"Patricia"}],"issued":{"date-parts":[[2024]]},"DOI":"10.48550/arxiv.2405.18828","URL":"https://doi.org/10.48550/arxiv.2405.18828","source":"datacite"},{"id":"doi:10.48550/arxiv.2412.12843","type":"manuscript","title":"SLTNet: Efficient Event-based Semantic Segmentation with Spike-driven Lightweight Transformer-based Networks","abstract":"Event-based semantic segmentation has great potential in autonomous driving and robotics due to the advantages of event cameras, such as high dynamic range, low latency, and low power cost. Unfortunately, current artificial neural network (ANN)-based segmentation methods suffer from high computational demands, the requirements for image frames, and massive energy consumption, limiting their efficiency and application on resource-constrained edge/mobile platforms. To address these problems, we introduce SLTNet, a spike-driven lightweight transformer-based network designed for event-based semantic segmentation. Specifically, SLTNet is built on efficient spike-driven convolution blocks (SCBs) to extract rich semantic features while reducing the model's parameters. Then, to enhance the long-range contextural feature interaction, we propose novel spike-driven transformer blocks (STBs) with binary mask operations. Based on these basic blocks, SLTNet employs a high-efficiency single-branch architecture while maintaining the low energy consumption of the Spiking Neural Network (SNN). Finally, extensive experiments on DDD17 and DSEC-Semantic datasets demonstrate that SLTNet outperforms state-of-the-art (SOTA) SNN-based methods by at most 9.06% and 9.39% mIoU, respectively, with extremely 4.58x lower energy consumption and 114 FPS inference speed. Our code is open-sourced and available at https://github.com/longxianlei/SLTNet-v1.0.","author":[{"family":"Long","given":"Xianlei"},{"family":"Zhu","given":"Xiaxin"},{"family":"Guo","given":"Fangming"},{"family":"Zhang","given":"Wanyi"},{"family":"Gu","given":"Qingyi"},{"family":"Chen","given":"Chao"},{"family":"Gu","given":"Fuqiang"}],"issued":{"date-parts":[[2024]]},"DOI":"10.48550/arxiv.2412.12843","URL":"https://doi.org/10.48550/arxiv.2412.12843","source":"datacite"},{"id":"doi:10.48550/arxiv.2403.04162","type":"manuscript","title":"Noisy Spiking Actor Network for Exploration","abstract":"As a general method for exploration in deep reinforcement learning (RL), NoisyNet can produce problem-specific exploration strategies. Spiking neural networks (SNNs), due to their binary firing mechanism, have strong robustness to noise, making it difficult to realize efficient exploration with local disturbances. To solve this exploration problem, we propose a noisy spiking actor network (NoisySAN) that introduces time-correlated noise during charging and transmission. Moreover, a noise reduction method is proposed to find a stable policy for the agent. Extensive experimental results demonstrate that our method outperforms the state-of-the-art performance on a wide range of continuous control tasks from OpenAI gym.","author":[{"family":"Chen","given":"Ding"},{"family":"Peng","given":"Peixi"},{"family":"Huang","given":"Tiejun"},{"family":"Tian","given":"Yonghong"}],"issued":{"date-parts":[[2024]]},"DOI":"10.48550/arxiv.2403.04162","URL":"https://doi.org/10.48550/arxiv.2403.04162","source":"datacite"},{"id":"doi:10.48550/arxiv.2410.08229","type":"manuscript","title":"Improvement of Spiking Neural Network with Bit Planes and Color Models","abstract":"Spiking neural network (SNN) has emerged as a promising paradigm in computational neuroscience and artificial intelligence, offering advantages such as low energy consumption and small memory footprint. However, their practical adoption is constrained by several challenges, prominently among them being performance optimization. In this study, we present a novel approach to enhance the performance of SNN for images through a new coding method that exploits bit plane representation. Our proposed technique is designed to improve the accuracy of SNN without increasing model size. Also, we investigate the impacts of color models of the proposed coding process. Through extensive experimental validation, we demonstrate the effectiveness of our coding strategy in achieving performance gain across multiple datasets. To the best of our knowledge, this is the first research that considers bit planes and color models in the context of SNN. By leveraging the unique characteristics of bit planes, we hope to unlock new potentials in SNNs performance, potentially paving the way for more efficient and effective SNNs models in future researches and applications.","author":[{"family":"Luu","given":"Nhan"},{"family":"Luu","given":"Duong"},{"family":"Pham","given":"Nam"},{"family":"Truong","given":"Thang"}],"issued":{"date-parts":[[2024]]},"DOI":"10.48550/arxiv.2410.08229","URL":"https://doi.org/10.48550/arxiv.2410.08229","source":"datacite"},{"id":"doi:10.48550/arxiv.2304.03896","type":"manuscript","title":"Spiking Neural Networks for Detecting Satellite-Based Internet-of-Things Signals","abstract":"With the rapid growth of IoT networks, ubiquitous coverage is becoming increasingly necessary. Low Earth Orbit (LEO) satellite constellations for IoT have been proposed to provide coverage to regions where terrestrial systems cannot. However, LEO constellations for uplink communications are severely limited by the high density of user devices, which causes a high level of co-channel interference. This research presents a novel framework that utilizes spiking neural networks (SNNs) to detect IoT signals in the presence of uplink interference. The key advantage of SNNs is the extremely low power consumption relative to traditional deep learning (DL) networks. The performance of the spiking-based neural network detectors is compared against state-of-the-art DL networks and the conventional matched filter detector. Results indicate that both DL and SNN-based receivers surpass the matched filter detector in interference-heavy scenarios, owing to their capacity to effectively distinguish target signals amidst co-channel interference. Moreover, our work highlights the ultra-low power consumption of SNNs compared to other DL methods for signal detection. The strong detection performance and low power consumption of SNNs make them particularly suitable for onboard signal detection in IoT LEO satellites, especially in high interference conditions.","author":[{"family":"Dakic","given":"Kosta"},{"family":"Homssi","given":"Bassel"},{"family":"Walia","given":"Sumeet"},{"family":"Al-Hourani","given":"Akram"}],"issued":{"date-parts":[[2023]]},"DOI":"10.48550/arxiv.2304.03896","URL":"https://doi.org/10.48550/arxiv.2304.03896","source":"datacite"},{"id":"doi:10.1038/s41598-024-75021-z","type":"article-journal","title":"Efficient memristor accelerator for transformer self-attention functionality.","abstract":"The adoption of transformer networks has experienced a notable surge in various AI applications. However, the increased computational complexity, stemming primarily from the self-attention mechanism, parallels the manner in which convolution operations constrain the capabilities and speed of convolutional neural networks (CNNs). The self-attention algorithm, specifically the matrix-matrix multiplication (MatMul) operations, demands a substantial amount of memory and computational complexity, thereby restricting the overall performance of the transformer. This paper introduces an efficient hardware accelerator for the transformer network, leveraging memristor-based in-memory computing. The design targets the memory bottleneck associated with MatMul operations in the self-attention process, utilizing approximate analog computation and the highly parallel computations facilitated by the memristor crossbar architecture. Remarkably, this approach resulted in a reduction of approximately 10 times in the number of multiply-accumulate (MAC) operations in transformer networks, while maintaining 95.47% accuracy for the MNIST dataset, as validated by a comprehensive circuit simulator employing NeuroSim 3.0. Simulation outcomes indicate an area utilization of 6895.7 $$\\mu m^2$$ , a latency of 15.52 seconds, an energy consumption of 3 mJ, and a leakage power of 59.55 $$\\mu W$$ . The methodology outlined in this paper represents a substantial stride towards a hardware-friendly transformer architecture for edge devices, poised to achieve real-time performance.","author":[{"family":"Bettayeb","given":"Meriem"},{"family":"Halawani","given":"Yasmin"},{"family":"Khan","given":"Muhammad"},{"family":"Saleh","given":"Hani"},{"family":"Mohammad","given":"Baker"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1038/s41598-024-75021-z","URL":"https://doi.org/10.1038/s41598-024-75021-z","source":"europepmc"},{"id":"oa:W4393148535","type":"article-journal","title":"Emerging ferroelectric materials ScAlN: applications and prospects in memristors","abstract":"The research found that after doping with rare earth elements, a large number of electrons and holes will be produced on the surface of AlN, which makes the material have the characteristics of spontaneous polarization. A new type of ferroelectric material has made a new breakthrough in the application of nitride-materials in the field of integrated devices. In this paper, the application prospects and development trends of ferroelectric material ScAlN in memristors are reviewed. Firstly, various fabrication processes and structures of the current ScAlN thin films are described in detail to explore the implementation of their applications in synaptic devices. Secondly, a series of electrical properties of ScAlN films, such as the current switching ratio and long-term cycle durability, were tested to explore whether their electrical properties could meet the basic needs of memristor device materials. Finally, a series of summaries on the current research studies of ScAlN thin films in the synaptic simulation are made, and the working state of ScAlN thin films as a synaptic device is observed. The results show that the ScAlN ferroelectric material has high residual polarization, no wake-up function, excellent stability and obvious STDP behavior, which indicates that the modified material has wide application prospects in the research and development of memristors.","author":[{"family":"Yang","given":"Dong"},{"family":"Tang","given":"Xin‐gui"},{"family":"Sun","given":"Qijun"},{"family":"Chen","given":"Jiaying"},{"family":"Jiang","given":"Yanping"},{"family":"Zhang","given":"Dan"},{"family":"Dong","given":"Huafeng"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1039/d3mh01942j","URL":"https://doi.org/10.1039/d3mh01942j","source":"openalex"},{"id":"oa:W4392745672","type":"article-journal","title":"Research Progress in Dielectric-Layer Material Systems of Memristors","abstract":"With the rapid growth of data storage, traditional von Neumann architectures and silicon-based storage computing technologies will reach their limits and fail to meet the storage requirements of ultra-small size, ultra-high density, and memory computing. Memristors have become a strong competitor in next generation memory technology because of their advantages such as simple device structure, fast erase speed, low power consumption, compatibility with CMOS technology, and easy 3D integration. The resistive medium layer is the key to achieving resistive performance; hence, research on memristors mainly focuses on the resistive medium layer. This paper begins by elucidating the fundamental concepts, structures, and resistive-switching mechanisms of memristors, followed by a comprehensive review of how different resistive storage materials impact memristor performance. The categories of memristors, the effects of different resistive materials on memristors, and the issues are described in detail. Finally, a summary of this article is provided, along with future prospects for memristors and the remaining issues in the large-scale industrialization of memristors.","author":[{"family":"Wang","given":"Chunxia"},{"family":"Li","given":"Xuemei"},{"family":"Sun","given":"Zhendong"},{"family":"Liu","given":"Yang"},{"family":"Yang","given":"Ying"},{"family":"Chen","given":"Lijia"}],"issued":{"date-parts":[[2024]]},"DOI":"10.3390/inorganics12030087","URL":"https://doi.org/10.3390/inorganics12030087","source":"openalex"},{"id":"oa:W4390880708","type":"article-journal","title":"Artificial Intelligence-Based Aquaculture System for Optimizing the Quality of Water: A Systematic Analysis","abstract":"The world population is expected to grow to around 9 billion by 2050. The growing need for foods with high protein levels makes aquaculture one of the fastest-growing food industries in the world. Some challenges of fishing production are related to obsolete aquaculture techniques, overexploitation of marine species, and lack of water quality control. This research systematically analyzes aquaculture technologies, such as sensors, artificial intelligence (AI), and image processing. Through the systematic PRISMA process, 753 investigations published from 2012 to 2023 were analyzed based on a search in Scopus and Web of Science. It revealed a significant 70.5% increase in the number of articles published compared to the previous year, indicating a growing interest in this field. The results indicate that current aquaculture technologies are water monitoring sensors, AI methodologies such as K-means, and contour segmentation for computer vision. Also, it is reported that K means technologies offer an efficiency from 95% to 98%. These methods allow decisions based on data patterns and aquaculture insights. Improving aquaculture methodologies will allow adequate management of economic and environmental resources to promote fishing and satisfy nutritional needs.","author":[{"family":"Capetillo-Contreras","given":"Omar"},{"family":"Pérez-Reynoso","given":"Francisco"},{"family":"Zamora-Antuñano","given":"Marco"},{"family":"Álvarez-Alvarado","given":"José"},{"family":"Rodríguezreséndiz","given":"Juvenal"}],"issued":{"date-parts":[[2024]]},"DOI":"10.3390/jmse12010161","URL":"https://doi.org/10.3390/jmse12010161","source":"openalex"},{"id":"oa:W4392558171","type":"article-journal","title":"Personalized strategies of neurostimulation: from static biomarkers to dynamic closed-loop assessment of neural function","abstract":"Despite considerable advancement of first choice treatment (pharmacological, physical therapy, etc.) over many decades, neurological disorders still represent a major portion of the worldwide disease burden. Particularly concerning, the trend is that this scenario will worsen given an ever expanding and aging population. The many different methods of brain stimulation (electrical, magnetic, etc.) are, on the other hand, one of the most promising alternatives to mitigate the suffering of patients and families when conventional treatment fall short of delivering efficacious treatment. With applications in virtually all neurological conditions, neurostimulation has seen considerable success in providing relief of symptoms. On the other hand, a large variability of therapeutic outcomes has also been observed, particularly in the usage of non-invasive brain stimulation (NIBS) modalities. Borrowing inspiration and concepts from its pharmacological counterpart and empowered by unprecedented neurotechnological advancement, the neurostimulation field has seen in recent years a widespread of methods aimed at the personalization of its parameters, based on biomarkers of the individuals being treated. The rationale is that, by taking into account important factors influencing the outcome, personalized stimulation can yield a much-improved therapy. Here, we review the literature to delineate the state-of-the-art of personalized stimulation, while also considering the important aspects of the type of informing parameter (anatomy, function, hybrid), invasiveness, and level of development (pre-clinical experimentation versus clinical trials). Moreover, by reviewing relevant literature on closed loop neuroengineering solutions in general and on activity dependent stimulation method in particular, we put forward the idea that improved personalization may be achieved when the method is able to track in real time brain dynamics and adjust its stimulation parameters accordingly. We conclude that such approaches have great potential of promoting the recovery of lost functions and enhance the quality of life for patients.","author":[{"family":"Carè","given":"Marta"},{"family":"Chiappalone","given":"Michela"},{"family":"Cota","given":"Vinícius"}],"issued":{"date-parts":[[2024]]},"DOI":"10.3389/fnins.2024.1363128","URL":"https://doi.org/10.3389/fnins.2024.1363128","source":"openalex"},{"id":"oa:W4400985292","type":"article-journal","title":"Experimental demonstration of magnetic tunnel junction-based computational random-access memory","abstract":"The conventional computing paradigm struggles to fulfill the rapidly growing demands from emerging applications, especially those for machine intelligence because much of the power and energy is consumed by constant data transfers between logic and memory modules. A new paradigm, called \"computational random-access memory (CRAM),\" has emerged to address this fundamental limitation. CRAM performs logic operations directly using the memory cells themselves, without having the data ever leave the memory. The energy and performance benefits of CRAM for both conventional and emerging applications have been well established by prior numerical studies. However, there is a lack of experimental demonstration and study of CRAM to evaluate its computational accuracy, which is a realistic and application-critical metric for its technological feasibility and competitiveness. In this work, a CRAM array based on magnetic tunnel junctions (MTJs) is experimentally demonstrated. First, basic memory operations, as well as 2-, 3-, and 5-input logic operations, are studied. Then, a 1-bit full adder with two different designs is demonstrated. Based on the experimental results, a suite of models has been developed to characterize the accuracy of CRAM computation. Scalar addition, multiplication, and matrix multiplication, which are essential building blocks for many conventional and machine intelligence applications, are evaluated and show promising accuracy performance. With the confirmation of MTJ-based CRAM's accuracy, there is a strong case that this technology will have a significant impact on power- and energy-demanding applications of machine intelligence.","author":[{"family":"Lv","given":"Yang"},{"family":"Zink","given":"Brandon"},{"family":"Bloom","given":"Robert"},{"family":"Cılasun","given":"Hüsrev"},{"family":"Khanal","given":"Pravin"},{"family":"Resch","given":"Salonik"},{"family":"Chowdhury","given":"Zamshed"},{"family":"Habiboglu","given":"Ali"},{"family":"Wang","given":"Weigang"},{"family":"Sapatnekar","given":"Sachin"},{"family":"Karpuzcu","given":"Ulya"},{"family":"Wang","given":"Jian‐ping"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1038/s44335-024-00003-3","URL":"https://doi.org/10.1038/s44335-024-00003-3","source":"openalex"},{"id":"oa:W4394747486","type":"article-journal","title":"Fast learning without synaptic plasticity in spiking neural networks","abstract":"Spiking neural networks are of high current interest, both from the perspective of modelling neural networks of the brain and for porting their fast learning capability and energy efficiency into neuromorphic hardware. But so far we have not been able to reproduce fast learning capabilities of the brain in spiking neural networks. Biological data suggest that a synergy of synaptic plasticity on a slow time scale with network dynamics on a faster time scale is responsible for fast learning capabilities of the brain. We show here that a suitable orchestration of this synergy between synaptic plasticity and network dynamics does in fact reproduce fast learning capabilities of generic recurrent networks of spiking neurons. This points to the important role of recurrent connections in spiking networks, since these are necessary for enabling salient network dynamics. We show more specifically that the proposed synergy enables synaptic weights to encode more general information such as priors and task structures, since moment-to-moment processing of new information can be delegated to the network dynamics.","author":[{"family":"Subramoney","given":"Anand"},{"family":"Bellec","given":"Guillaume"},{"family":"Scherr","given":"Franz"},{"family":"Legenstein","given":"Robert"},{"family":"Maass","given":"Wolfgang"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1038/s41598-024-55769-0","URL":"https://doi.org/10.1038/s41598-024-55769-0","source":"openalex"},{"id":"oa:W4402715302","type":"article-journal","title":"Estimating optical flow: A comprehensive review of the state of the art","abstract":"Optical flow estimation is a crucial task in computer vision that provides low-level motion information. Despite recent advances, real-world applications still present significant challenges. This survey provides an overview of optical flow techniques and their application. For a comprehensive review, this survey covers both classical frameworks and the latest AI-based techniques. In doing so, we highlight the limitations of current benchmarks and metrics, underscoring the need for more representative datasets and comprehensive evaluation methods. The survey also highlights the importance of integrating industry knowledge and adopting training practices optimized for deep learning-based models. By addressing these issues, future research can aid the development of robust and efficient optical flow methods that can effectively address real-world scenarios. • Investigating integration of traditional techniques in modern models. • Surveying key challenges of optical flow in real-world applications. • Offering the most comprehensive survey of datasets for optical flow. • Presenting a complete overview of Classical and Modern Optical Flow methods. • Highlighting crucial open questions, paving way for future research.","author":[{"family":"Alfarano","given":"Andrea"},{"family":"Maiano","given":"Luca"},{"family":"Papa","given":"Lorenzo"},{"family":"Amerini","given":"Irene"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1016/j.cviu.2024.104160","URL":"https://doi.org/10.1016/j.cviu.2024.104160","source":"openalex"},{"id":"oa:W4400065203","type":"article-journal","title":"2D Graphene Oxide: A Versatile Thermo‐Optic Material","abstract":"Abstract Efficient heat management and control in optical devices, facilitated by advanced thermo‐optic materials, are critical for many applications such as photovoltaics, thermal emitters, mode‐locked lasers, and optical switches. Here, a range of thermo‐optic properties of 2D graphene oxide (GO) films are investigated by precisely integrating them onto microring resonators (MRRs) with control over the film thicknesses and lengths. The refractive index, extinction coefficient, thermo‐optic coefficient, and thermal conductivity for the GO films with different layer numbers and degrees of reduction, as well as reversible reduction and enhanced optical bistability induced by the photo‐thermal effects, are comprehensively characterized. Experimental results show that the thermo‐optic properties of 2D GO films vary widely with the degree of reduction. In addition, significant anisotropy is observed for the thermo‐optic response, enabling efficient polarization‐sensitive devices. The versatile thermo‐optic response of 2D GO substantially expands the scope of functionalities and devices that can be engineered, making it promising for a diverse range of thermo‐optic applications.","author":[{"family":"Hu","given":"Junkai"},{"family":"Wu","given":"Jiayang"},{"family":"Liu","given":"Wenbo"},{"family":"Jin","given":"Di"},{"family":"Dirani","given":"Houssein"},{"family":"Kerdilès","given":"S"},{"family":"Sciancalepore","given":"Corrado"},{"family":"Demongodin","given":"Pierre"},{"family":"Grillet","given":"Christian"},{"family":"Monat","given":"Christelle"},{"family":"Huang","given":"Duan"},{"family":"Jia","given":"Baohua"},{"family":"Moss","given":"David"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1002/adfm.202406799","URL":"https://doi.org/10.1002/adfm.202406799","source":"openalex"},{"id":"oa:W4399118682","type":"article-journal","title":"Toward Practical Single‐Molecule/Atom Switches","abstract":"Electronic switches have been considered to be one of the most important components of contemporary electronic circuits for processing and storing digital information. Fabricating functional devices with building blocks of atomic/molecular switches can greatly promote the minimization of the devices and meet the requirement of high integration. This review highlights key developments in the fabrication and application of molecular switching devices. This overview offers valuable insights into the switching mechanisms under various stimuli, emphasizing structural and energy state changes in the core molecules. Beyond the molecular switches, typical individual metal atomic switches are further introduced. A critical discussion of the main challenges for realizing and developing practical molecular/atomic switches is provided. These analyses and summaries will contribute to a comprehensive understanding of the switch mechanisms, providing guidance for the rational design of functional nanoswitch devices toward practical applications.","author":[{"family":"Xu","given":"Xiaona"},{"family":"Gao","given":"Chunyan"},{"family":"Ramya","given":"E"},{"family":"Jia","given":"Chuancheng"},{"family":"Xiang","given":"Dong"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1002/advs.202400877","URL":"https://doi.org/10.1002/advs.202400877","source":"openalex"},{"id":"oa:W4405566600","type":"article-journal","title":"The ab initio non-crystalline structure database: empowering machine learning to decode diffusivity","abstract":"Abstract Non-crystalline materials exhibit unique properties that make them suitable for various applications in science and technology, ranging from optical and electronic devices and solid-state batteries to protective coatings. However, data-driven exploration and design of non-crystalline materials is hampered by the absence of a comprehensive database covering a broad chemical space. In this work, we present the largest computed non-crystalline structure database to date, generated from systematic and accurate ab initio molecular dynamics (AIMD) calculations. We also show how the database can be used in simple machine-learning models to connect properties to composition and structure, here specifically targeting ionic conductivity. These models predict the Li-ion diffusivity with speed and accuracy, offering a cost-effective alternative to expensive density functional theory (DFT) calculations. Furthermore, the process of computational quenching non-crystalline structures provides a unique sampling of out-of-equilibrium structures, energies, and force landscape, and we anticipate that the corresponding trajectories will inform future work in universal machine learning potentials, impacting design beyond that of non-crystalline materials. In addition, combining diffusion trajectories from our dataset with models that predict liquidus viscosity and melting temperature could be utilized to develop models for predicting glass-forming ability.","author":[{"family":"Zheng","given":"Hui"},{"family":"Sivonxay","given":"Eric"},{"family":"Christensen","given":"Rasmus"},{"family":"Gallant","given":"Max"},{"family":"Luo","given":"Ziyao"},{"family":"Mcdermott","given":"Matthew"},{"family":"Huck","given":"Patrick"},{"family":"Smedskjær","given":"Morten"},{"family":"Persson","given":"Kristin"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1038/s41524-024-01469-2","URL":"https://doi.org/10.1038/s41524-024-01469-2","source":"openalex"},{"id":"oa:W4405828408","type":"article-journal","title":"Soft Artificial Synapse Electronics","abstract":"Soft electronics, known for their bendable, stretchable, and flexible properties, are revolutionizing fields such as biomedical sensing, consumer electronics, and robotics. A primary challenge in this domain is achieving low power consumption, often hampered by the limitations of the conventional von Neumann architecture. In response, the development of soft artificial synapses (SASs) has gained substantial attention. These synapses seek to replicate the signal transmission properties of biological synapses, offering an innovative solution to this challenge. This review explores the materials and device architectures integral to SAS fabrication, emphasizing flexibility and stability under mechanical deformation. Various architectures, including floating-gate dielectric, ferroelectric-gate dielectric, and electrolyte-gate dielectric, are analyzed for effective weight control in SASs. The utilization of organic and low-dimensional materials is highlighted, showcasing their plasticity and energy-efficient operation. Furthermore, the paper investigates the integration of functionality into SASs, particularly focusing on devices that autonomously sense external stimuli. Functionalized SASs, capable of recognizing optical, mechanical, chemical, olfactory, and auditory cues, demonstrate promising applications in computing and sensing. A detailed examination of photo-functionalized, tactile-functionalized, and chemoreception-functionalized SASs reveals their potential in image recognition, tactile sensing, and chemosensory applications, respectively. This study highlights that SASs and functionalized SAS devices hold transformative potential for bioelectronics and sensing for soft-robotics applications; however, further research is necessary to address scalability, long-time stability, and utilizing functionalized SASs for prosthetics and in vivo applications through clinical adoption. By providing a comprehensive overview, this paper contributes to the understanding of SASs, bridging research gaps and paving the way toward transformative developments in soft electronics, biomimicking and biointegrated synapse devices, and integrated systems.","author":[{"family":"Mojumder","given":"Md"},{"family":"Kim","given":"Seongchan"},{"family":"Yu","given":"Cunjiang"}],"issued":{"date-parts":[[2024]]},"DOI":"10.34133/research.0582","URL":"https://doi.org/10.34133/research.0582","source":"openalex"},{"id":"oa:W4405180690","type":"article-journal","title":"Spiking Variational Policy Gradient for Brain Inspired Reinforcement Learning","abstract":"Recent studies in reinforcement learning have explored brain-inspired function approximators and learning algorithms to simulate brain intelligence and adapt to neuromorphic hardware. Among these approaches, reward-modulated spike-timing-dependent plasticity (R-STDP) is biologically plausible and energy-efficient, but suffers from a gap between its local learning rules and the global learning objectives, which limits its performance and applicability. In this paper, we design a recurrent winner-take-all network and propose the spiking variational policy gradient (SVPG), a new R-STDP learning method derived theoretically from the global policy gradient. Specifically, the policy inference is derived from an energy-based policy function using mean-field inference, and the policy optimization is based on a last-step approximation of the global policy gradient. These fill the gap between the local learning rules and the global target. In experiments including a challenging ViZDoom vision-based navigation task and two realistic robot control tasks, SVPG successfully solves all the tasks. In addition, SVPG exhibits better inherent robustness to various kinds of input, network parameters, and environmental perturbations than compared methods.","author":[{"family":"Yang","given":"Zhile"},{"family":"Guo","given":"Shangqi"},{"family":"Fang","given":"Ying"},{"family":"Yu","given":"Zhaofei"},{"family":"Liu","given":"Jian"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1109/tpami.2024.3511936","URL":"https://doi.org/10.1109/tpami.2024.3511936","source":"openalex"},{"id":"oa:W4391879679","type":"article-journal","title":"Deep photonic network platform enabling arbitrary and broadband optical functionality","abstract":"Expanding applications in optical communications, computing, and sensing continue to drive the need for high-performance integrated photonic components. Designing these on-chip systems with arbitrary functionality requires beyond what is possible with physical intuition, for which machine learning-based methods have recently become popular. However, computational demands for physically accurate device simulations present critical challenges, significantly limiting scalability and design flexibility of these methods. Here, we present a highly-scalable, physics-informed design platform for on-chip optical systems with arbitrary functionality, based on deep photonic networks of custom-designed Mach-Zehnder interferometers. Leveraging this platform, we demonstrate ultra-broadband power splitters and a spectral duplexer, each designed within two minutes. The devices exhibit state-of-the-art experimental performance with insertion losses below 0.66 dB, and 1-dB bandwidths exceeding 120 nm. This platform provides a tractable path towards systematic, large-scale photonic system design, enabling custom power, phase, and dispersion profiles for high-throughput communications, quantum information processing, and medical/biological sensing applications.","author":[{"family":"Amiri","given":"Ali"},{"family":"Vit","given":"Aycan"},{"family":"Görgülü","given":"Kazim"},{"family":"Magden","given":"Emir"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1038/s41467-024-45846-3","URL":"https://doi.org/10.1038/s41467-024-45846-3","source":"openalex"},{"id":"oa:W4398765486","type":"article-journal","title":"An inorganic-blended p-type semiconductor with robust electrical and mechanical properties","abstract":"Abstract Inorganic semiconductors typically have limited p-type behavior due to the scarcity of holes and the localized valence band maximum, hindering the progress of complementary devices and circuits. In this work, we propose an inorganic blending strategy to activate the hole-transporting character in an inorganic semiconductor compound, namely tellurium-selenium-oxygen (TeSeO). By rationally combining intrinsic p-type semimetal, semiconductor, and wide-bandgap semiconductor into a single compound, the TeSeO system displays tunable bandgaps ranging from 0.7 to 2.2 eV. Wafer-scale ultrathin TeSeO films, which can be deposited at room temperature, display high hole field-effect mobility of 48.5 cm2/(Vs) and robust hole transport properties, facilitated by Te-Te (Se) portions and O-Te-O portions, respectively. The nanosphere lithography process is employed to create nanopatterned honeycomb TeSeO broadband photodetectors, demonstrating a high responsibility of 603 A/W, an ultrafast response of 5 μs, and superior mechanical flexibility. The p-type TeSeO system is highly adaptable, scalable, and reliable, which can address emerging technological needs that current semiconductor solutions may not fulfill.","author":[{"family":"Meng","given":"You"},{"family":"Wang","given":"Weijun"},{"family":"Fan","given":"Rong"},{"family":"Lai","given":"Zhengxun"},{"family":"Wang","given":"Wei"},{"family":"Li","given":"Dengji"},{"family":"Li","given":"Xiaocui"},{"family":"Quan","given":"Quan"},{"family":"Xie","given":"Pengshan"},{"family":"Chen","given":"Dong"},{"family":"Shao","given":"He"},{"family":"Li","given":"Bowen"},{"family":"Wu","given":"Zenghui"},{"family":"Yang","given":"Zhe"},{"family":"Yip","given":"Senpo"},{"family":"Wong","given":"Chun‐yuen"},{"family":"Lü","given":"Yang"},{"family":"Ho","given":"Johnny"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1038/s41467-024-48628-z","URL":"https://doi.org/10.1038/s41467-024-48628-z","source":"openalex"},{"id":"oa:W4400229075","type":"article-journal","title":"Smaller and Faster Robotic Grasp Detection Model via Knowledge Distillation and Unequal Feature Encoding","abstract":"In order to achieve higher accuracy, the complexity of grasp detection network increases accordingly with complicated model structures and tremendous parameters. Although various light-weight strategies are adopted, directly designing the compact network can be sub-optimal and difficult to strike the balance between accuracy and model size. To solve this problem, we explore a more efficient grasp detection model from two aspects: elaborately designing a light-weight network and performing knowledge distillation on the designed network. Specifically, based on the designed light-weight backbone, the features from RGB and D images with unequal effective grasping information rates are fully utilized and the information compensation strategies are adopted to make the model small enough while maintaining its accuracy. Then, the grasping features contained in the large teacher model are adaptively and effectively learned by our proposed method via knowledge distillation. Experimental results indicate that the proposed method is able to achieve comparable performance (98.9%, 93.1%, 82.3%, and 90.0% on Cornell, Jacquard, GraspNet, and MultiObj datasets, respectively) with more complicate models while reducing the parameters from MBs to KBs. Real-world robotic grasping experiment in an embedded AI computing device also prove the effectiveness of this approach.","author":[{"family":"Nie","given":"Hong"},{"family":"Zhao","given":"Zhou"},{"family":"Chen","given":"Lu"},{"family":"Lu","given":"Zhenyu"},{"family":"Li","given":"Zhuomao"},{"family":"Yang","given":"Jing"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1109/lra.2024.3421790","URL":"https://doi.org/10.1109/lra.2024.3421790","source":"openalex"},{"id":"oa:W4405137460","type":"article-journal","title":"Tactile Feedback in Robot‐Assisted Minimally Invasive Surgery: A Systematic Review","abstract":"BACKGROUND: Robot-assisted systems have predominantly relied on teleoperation, where visual feedback is the primary source of information. However, advances in tactile sensing and displays offer new opportunities to enhance surgical transparency, efficiency, and safety. METHODS: A PRISMA-guided search was conducted across PubMed, IEEE Xplore, Scopus, and Web of Science databases to identify relevant studies. RESULTS: Out of 645 screened articles, 98 met the inclusion criteria, and 33 were included in the final review. The review discusses various tactile feedback stimulus types, applications, and challenges in the context of robot-assisted minimally invasive surgery. CONCLUSION: While kinaesthetic feedback has been extensively explored to restore the natural interaction between the surgeon and the surgical environment, tactile feedback remains largely confined to research settings. This is due to significant challenges in integrating tactile feedback into robotic systems and current limitations of sensing technologies.","author":[{"family":"Colan","given":"Jacinto"},{"family":"Davila","given":"Ana"},{"family":"Hasegawa","given":"Yasuhisa"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1002/rcs.70019","URL":"https://doi.org/10.1002/rcs.70019","source":"openalex"},{"id":"oa:W4399463559","type":"article-journal","title":"Emergent digital bio-computation through spatial diffusion and engineered bacteria","abstract":"Biological computing is a promising field with potential applications in biosafety, environmental monitoring, and personalized medicine. Here we present work on the design of bacterial computers using spatial patterning to process information in the form of diffusible morphogen-like signals. We demonstrate, mathematically and experimentally, that single, modular, colonies can perform simple digital logic, and that complex functions can be built by combining multiple colonies, removing the need for further genetic engineering. We extend our experimental system to incorporate sender colonies as morphogen sources, demonstrating how one might integrate different biochemical inputs. Our approach will open up ways to perform biological computation, with applications in bioengineering, biomaterials and biosensing. Ultimately, these computational bacterial communities will help us explore information processing in natural biological systems.","author":[{"family":"Fedorec","given":"Alex"},{"family":"Treloar","given":"Neythen"},{"family":"Wen","given":"Ke"},{"family":"Dekker","given":"Linda"},{"family":"Ong","given":"Qing"},{"family":"Jurkeviciute","given":"Gabija"},{"family":"Lyu","given":"Enbo"},{"family":"Rutter","given":"Jack"},{"family":"Zhang","given":"K"},{"family":"Rosa","given":"Luca"},{"family":"Zaikin","given":"Alexey"},{"family":"Barnes","given":"C"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1038/s41467-024-49264-3","URL":"https://doi.org/10.1038/s41467-024-49264-3","source":"openalex"},{"id":"oa:W4403299429","type":"article-journal","title":"A Three‐Terminal Memristive Artificial Neuron with Tunable Firing Probability","abstract":"Abstract The human brain facilitates information processing via generating and receiving temporal patterns of short voltage pulses, a.k.a. neural spikes. This approach simultaneously grants low‐power operation as well as a high degree of noise immunity and fault tolerance at a small footprint and simplistic structure of the neurons. To date, the latter two key features are critically missing from the toolbox of artificial spiking neural network hardware, hindering the development of scalable and sustainable artificial intelligence (AI) platforms. Here, a compact, gate‐tunable neuron circuit is demonstrated, and its potential as a functional leaky integrate‐and‐fire (LIF) neuron is explored. It relies on a single nanoscale three‐terminal (3T) memristor device, which has been downscaled by 30% compared to previous work, where the set voltage and, thereby, the spiking probability of the neuron circuit can be widely tuned by the low‐voltage operation of the gate electrode. The influence of the gate voltage on the two‐terminal (2T) current–voltage characteristics is measured, statistically analyzed, and further utilized in a custom‐built LTspice model. The circuit simulations account for the experimentally observed, adjustable set voltage. The presented results demonstrate the merits of 3T memristors as compact, tunable, and versatile artificial neurons for neuromorphic computing applications.","author":[{"family":"Lewerenz","given":"Mila"},{"family":"Passerini","given":"Elias"},{"family":"Weber","given":"Luca"},{"family":"Fischer","given":"Markus"},{"family":"Olalla","given":"Nadia"},{"family":"Gisler","given":"Raphael"},{"family":"Emboras","given":"Alexandros"},{"family":"Luisier","given":"Mathieu"},{"family":"Csontos","given":"Miklós"},{"family":"Koch","given":"Ueli"},{"family":"Leuthold","given":"Juerg"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1002/aelm.202400432","URL":"https://doi.org/10.1002/aelm.202400432","source":"openalex"},{"id":"oa:W4405455971","type":"article-journal","title":"Photoinduced Melting of V4O7 Correlated State","abstract":"Abstract The compound V4O7 is one of the Magnéli phase (VnO2n − 1, n = 3, 4, …, 9) correlated vanadium oxides with distinct intriguing electronic and structural properties. The possibility to manipulate the phase state of V4O7 on an ultrafast time scale by light makes this material promising for potential applications in photonics, optoelectronics, quantum, and neuromorphic circuit design. In this work, the ultrafast spectroscopy of V4O7 reveals the second‐order nature of the photoinduced insulator‐to‐metal transition, emphasizing electronic and lattice contributions. The findings reveal the influence of the laser excitation level and temperature on these dynamics, providing a comprehensive understanding of V4O7 structural changes and response to external stimuli. The phenomenological model based on the Landau–Ginzburg formalism provides a robust framework for explaining the photoinduced transition dynamics, showing a detailed picture of the light interaction with the electronic and lattice subsystems. This integrated approach significantly enhances the understanding of V4O7 complex behavior upon photoexcitation, opening new possibilities for developing new optoelectronic devices and noninvasive optical control of the phase transition pathways in vanadates.","author":[{"family":"Bartenev","given":"Alexander"},{"family":"Verbel","given":"Camilo"},{"family":"Wu","given":"Qin"},{"family":"Camino","given":"Fernando"},{"family":"Rúa","given":"Armando"},{"family":"Lysenko","given":"Sergiy"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1002/aelm.202400539","URL":"https://doi.org/10.1002/aelm.202400539","source":"openalex"},{"id":"oa:W4401307486","type":"manuscript","title":"Using CSNNs to Perform Event-based Data Processing & Classification on ASL-DVS","abstract":"Recent advancements in bio-inspired visual sensing and neuromorphic computing have led to the development of various highly efficient bio-inspired solutions with real-world applications. One notable application integrates event-based cameras with spiking neural networks (SNNs) to process event-based sequences that are asynchronous and sparse, making them difficult to handle. In this project, we develop a convolutional spiking neural network (CSNN) architecture that leverages convolutional operations and recurrent properties of a spiking neuron to learn the spatial and temporal relations in the ASL-DVS gesture dataset. The ASL-DVS gesture dataset is a neuromorphic dataset containing hand gestures when displaying 24 letters (A to Y, excluding J and Z due to the nature of their symbols) from the American Sign Language (ASL). We performed classification on a pre-processed subset of the full ASL-DVS dataset to identify letter signs and achieved 100\\% training accuracy. Specifically, this was achieved by training in the Google Cloud compute platform while using a learning rate of 0.0005, batch size of 25 (total of 20 batches), 200 iterations, and 10 epochs.","author":[{"family":"Patel","given":"Ria"},{"family":"Tripathy","given":"Sujit"},{"family":"Sublett","given":"Zachary"},{"family":"An","given":"Seoyoung"},{"family":"Patel","given":"Riya"}],"issued":{"date-parts":[[2024]]},"DOI":"10.48550/arxiv.2408.00611","URL":"https://doi.org/10.48550/arxiv.2408.00611","source":"openalex"},{"id":"oa:W4405292220","type":"article-journal","title":"A Comparative Analysis of Anomaly Detection Methods in IoT Networks: An Experimental Study","abstract":"The growth of the Internet of Things (IoT) and its integration with Industry 4.0 and 5.0 are generating new security challenges. One of the key elements of IoT systems is effective anomaly detection, which identifies abnormal behavior in devices or entire systems. This paper presents a comprehensive overview of existing methods for anomaly detection in IoT networks using machine learning (ML). A detailed analysis of various ML algorithms, both supervised (e.g., Random Forest, Gradient Boosting, SVM) and unsupervised (e.g., Isolation Forest, Autoencoder), was conducted. The results of tests conducted on popular datasets (IoT-23 and CICIoT-2023) were collected and analyzed in detail. The performance of the selected algorithms was evaluated using commonly used metrics (Accuracy, Precision, Recall, F1-score). The experimental results showed that the Random Forest and Autoencoder methods are highly effective in detecting anomalies. The article highlights the importance of appropriate data preprocessing to improve detection accuracy. Furthermore, the limitations of a centralized machine learning approach in the context of distributed IoT networks are discussed. The article also presents potential directions for future research in the field of anomaly detection in the IoT.","author":[{"family":"Krzysztoń","given":"Emanuel"},{"family":"Rojek","given":"Izabela"},{"family":"Mikołajewski","given":"Dariusz"}],"issued":{"date-parts":[[2024]]},"DOI":"10.3390/app142411545","URL":"https://doi.org/10.3390/app142411545","source":"openalex"},{"id":"oa:W4386942778","type":"manuscript","title":"SpikingNeRF: Making Bio-inspired Neural Networks See through the Real World","abstract":"In this paper, we propose SpikingNeRF, which aligns the temporal dimension of spiking neural networks (SNNs) with the radiance rays, to seamlessly accommodate SNNs to the reconstruction of neural radiance fields (NeRF). Thus, the computation turns into a spike-based, multiplication-free manner, reducing energy consumption and making high-quality 3D rendering, for the first time, accessible to neuromorphic hardware. In SpikingNeRF, each sampled point on the ray is matched to a particular time step and represented in a hybrid manner where the voxel grids are maintained as well. Based on the voxel grids, sampled points are determined whether to be masked out for faster training and inference. However, this masking operation also incurs irregular temporal length, making it intractable for hardware processors, e.g., GPUs, to conduct parallel training. To address this problem, we develop the temporal padding strategy to tackle the masked samples to maintain regular temporal length, i.e., regular tensors, and further propose the temporal condensing strategy to form a denser data structure for hardware-friendly computation. Experiments on various datasets demonstrate that our method can reduce energy consumption by an average of 70.79\\% and obtain comparable synthesis quality with the ANN baseline. Verification on the neuromorphic hardware accelerator also shows that SpikingNeRF can further benefit from neuromorphic computing over the ANN baselines on energy efficiency. Codes and the appendix are in \\url{https://github.com/Ikarosy/SpikingNeRF-of-CASIA}.","author":[{"family":"Yao","given":"Xingting"},{"family":"Hu","given":"Qinghao"},{"family":"Zhou","given":"Fei"},{"family":"Liu","given":"Tielong"},{"family":"Mo","given":"Zitao"},{"family":"Zhu","given":"Zeyu"},{"family":"Zhuge","given":"Zhengyang"},{"family":"Cheng","given":"Jian"}],"issued":{"date-parts":[[2023]]},"DOI":"10.48550/arxiv.2309.10987","URL":"https://doi.org/10.48550/arxiv.2309.10987","source":"openalex"},{"id":"oa:W4393866074","type":"article-journal","title":"Ultrathin All‐Solid‐State MoS 2 ‐Based Electrolyte Gated Synaptic Transistor with Tunable Organic–Inorganic Hybrid Film","abstract":"Abstract Electrolyte‐gated synaptic transistors (EGSTs) have attracted considerable attention as synaptic devices owing to their adjustable conductance, low power consumption, and multi‐state storage capabilities. To demonstrate high‐density EGST arrays, 2D materials are recommended owing to their excellent electrical properties and ultrathin profile. However, widespread implementation of 2D‐based EGSTs has challenges in achieving large‐area channel growth and finding compatible nanoscale solid electrolytes. This study demonstrates large‐scale process‐compatible, all‐solid‐state EGSTs utilizing molybdenum disulfide (MoS 2 ) channels grown through chemical vapor deposition (CVD) and sub‐30 nm organic‐inorganic hybrid electrolyte polymers synthesized via initiated chemical vapor deposition (iCVD). The iCVD technique enables precise modulation of the hydroxyl group density in the hybrid matrix, allowing the modulation of proton conduction, resulting in adjustable synaptic performance. By leveraging the tunable iCVD‐based hybrid electrolyte, the fabricated EGSTs achieve remarkable attributes: a wide on/off ratio of 10 9 , state retention exceeding 10 3 , and linear conductance updates. Additionally, the device exhibits endurance surpassing 5 × 10 4 cycles, while maintaining a low energy consumption of 200 fJ/spike. To evaluate the practicality of these EGSTs, a subset of devices is employed in system‐level simulations of MNIST handwritten digit recognition, yielding a recognition rate of 93.2%.","author":[{"family":"Oh","given":"Jungyeop"},{"family":"Park","given":"Seohak"},{"family":"Lee","given":"Sang"},{"family":"Kim","given":"Sung"},{"family":"Lee","given":"Hyeonji"},{"family":"Lee","given":"Changhyeon"},{"family":"Hong","given":"Woonggi"},{"family":"Cha","given":"Jun‐hwe"},{"family":"Kang","given":"Mingu"},{"family":"Jin","given":"Jun"},{"family":"Im","given":"Sung"},{"family":"Kim","given":"Min"},{"family":"Choi","given":"Sung‐yool"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1002/advs.202308847","URL":"https://doi.org/10.1002/advs.202308847","source":"openalex"},{"id":"oa:W4387952752","type":"article-journal","title":"Smart Textile Optoelectronics for Human‐Interfaced Logic Systems","abstract":"Abstract Textiles with a freedom of form factor, unlimited scalability, and high programmability provide an ideal platform for constructing wearable optoelectronic systems. The emerging wearable technologies, like artificial intelligence and Internet of Things, have driven the development of textile optoelectronics from simple functional blocks to sophisticated logic systems, offering a seamless, breathable, and programmable on‐body platform to synergistically sense, analyze, store, and feedback information in response to complex commands. In the past few years, the creation of such smart textile optoelectronics‐based logic systems is boosted by nanomaterial science and manufacturing integration technologies and has revolutionized human–machine interaction paradigms in numerous emerging fields. Herein, in this review, the recent progress of smart textile optoelectronics for human‐interfaced logic systems is timely summarized. This review begins with a concise discussion about the wearability evaluation and integration consideration of textile optoelectronic devices. Then, important breakthroughs in human‐interfaced logic systems based on smart textile optoelectronics are demonstrated by highlighting their representative device, working principle, and application scenarios. Finally, the existing challenges and potential directions in the field of textile optoelectronics‐integrated logic systems are analyzed.","author":[{"family":"Peng","given":"Hongyun"},{"family":"Li","given":"Huiqiao"},{"family":"Tao","given":"Guangming"},{"family":"Xia","given":"Liangjun"},{"family":"Xu","given":"Weilin"},{"family":"Zhai","given":"Tianyou"}],"issued":{"date-parts":[[2023]]},"DOI":"10.1002/adfm.202308136","URL":"https://doi.org/10.1002/adfm.202308136","source":"openalex"},{"id":"oa:W4403887065","type":"article-journal","title":"Towards transformative innovation ecosystems: a systemic approach to responsible innovation","abstract":"To address societal challenges, innovators committed to responsibility need to find ways to break away from unsustainable, or otherwise undesirable, path-dependencies in sociotechnical regimes. Such path-breaking innovation should not come at the expense of socioethical desirability. The emerging literature on responsible innovation ecosystems has focused on socioethical desirability but has neglected sociotechnical viability beyond a protected niche. Drawing on theoretical insights and concepts from the literature on Responsible Innovation, innovation ecosystems, and transition studies, we distinguish four types of ecosystems along the axes of desirability and viability and discuss examples of these types. We introduce the concept of transformative innovation ecosystems to refer to a type of ecosystem that combines desirability with viability. The concept is developed by bringing theoretical perspectives into conversation with empirical insights from the high-tech research and innovation cluster NeuroSys, which aims to create an ecosystem around brain-inspired computing.","author":[{"family":"Neudert","given":"Philipp"},{"family":"Smolka","given":"Mareike"},{"family":"Böschen","given":"Stefan"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1080/23299460.2024.2414482","URL":"https://doi.org/10.1080/23299460.2024.2414482","source":"openalex"},{"id":"oa:W4405539908","type":"article-journal","title":"Interfacial Electronic Charge Trapping and Photonic Carrier Excitation Coupling in Solution-Processed Zinc–Tin Oxide Thin-Film Transistors Applied for Logic Gate Design and Quantized Neural Network","abstract":"High Resolution Image Download MS PowerPoint Slide Components needed in Artificial Intelligence with a higher information capacity are critically needed and have garnered significant attention at the forefront of information technology. This study utilizes solution-processed zinc–tin oxide (ZTO) thin-film phototransistors and modulates the values of V G, which allows for the regulation of electron trapping/detrapping at the ZTO/SiO 2 interface. By coupling the excited photonic carrier and electronic trapping, logic gates such as “AND,” “OR,” “NAND,” and “NOR” can be achieved. With the exponential growth in data generation, efficient processing and storage solutions are imperative. However, extensive data transfer between computing units and storage limits the level of artificial neural networks (ANNs). Consequently, quantized neural networks (QNNs) have gained interest for their reduced computational resource requirements and lower consumption. In this context, we introduce an optimized ternary logic circuit based on ZTO devices. By utilizing optical modulation to adjust the turn-on voltage of the single device, we demonstrate the achievement of ternary current states, thereby providing three distinct discrete states. This configuration can be extended to QNN computing, demonstrating multilevel quantized current values for in-memory computation. We achieved a handwriting digit recognition rate of 91.6%, thereby demonstrating reliable QNN hardware performance. This robust QNN performance indicates that the metal oxide phototransistor shows significant potential for future ternary computing systems.","author":[{"family":"Chang","given":"Pei"},{"family":"Lin","given":"Wun"},{"family":"Huang","given":"Ya‐chi"},{"family":"Chen","given":"Yu"},{"family":"Shih","given":"Li‐chung"},{"family":"Chen","given":"Jen‐sue"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1021/acsami.4c15102","URL":"https://doi.org/10.1021/acsami.4c15102","source":"openalex"},{"id":"oa:W4398243919","type":"article-journal","title":"Physics to system-level modeling of silicon-organic-hybrid nanophotonic devices","abstract":"The continuous growth in data volume has sparked interest in silicon-organic-hybrid (SOH) nanophotonic devices integrated into silicon photonic integrated circuits (PICs). SOH devices offer improved speed and energy efficiency compared to silicon photonics devices. However, a comprehensive and accurate modeling methodology of SOH devices, such as modulators corroborating experimental results, is lacking. While some preliminary modeling approaches for SOH devices exist, their reliance on theoretical and numerical methodologies, along with a lack of compatibility with electronic design automation (EDA), hinders their seamless and rapid integration with silicon PICs. Here, we develop a phenomenological, building-block-based SOH PICs simulation methodology that spans from the physics to the system level, offering high accuracy, comprehensiveness, and EDA-style compatibility. Our model is also readily integrable and scalable, lending itself to the design of large-scale silicon PICs. Our proposed modeling methodology is agnostic and compatible with any photonics-electronics co-simulation software. We validate this methodology by comparing the characteristics of experimentally demonstrated SOH microring modulators (MRMs) and Mach Zehnder modulators with those obtained through simulation, demonstrating its ability to model various modulator topologies. We also show our methodology's ease and speed in modeling large-scale systems. As an illustrative example, we use our methodology to design and study a 3-channel SOH MRM-based wavelength-division (de)multiplexer, a widely used component in various applications, including neuromorphic computing, data center interconnects, communications, sensing, and switching networks. Our modeling approach is also compatible with other materials exhibiting the Pockels and Kerr effects. To our knowledge, this represents the first comprehensive physics-to-system-level EDA-compatible simulation methodology for SOH modulators.","author":[{"family":"Moridsadat","given":"Maryam"},{"family":"Tamura","given":"Marcus"},{"family":"Chrostowski","given":"Lukas"},{"family":"Shekhar","given":"Sudip"},{"family":"Shastri","given":"Bhavin"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1038/s41598-024-61618-x","URL":"https://doi.org/10.1038/s41598-024-61618-x","source":"openalex"},{"id":"oa:W4399132882","type":"article-journal","title":"Electroactive composite biofilms integrating Kombucha, Chlorella and synthetic proteinoid Proto–Brains","abstract":"In this study, we present electroactive biofilms made from a combination of Kombucha zoogleal mats and thermal proteinoids. These biofilms have potential applications in unconventional computing and robotic skin. Proteinoids are synthesized by thermally polymerizing amino acids, resulting in the formation of synthetic protocells that display electrical signalling similar to neurons. By incorporating proteinoids into Kombucha zoogleal cellulose mats, hydrogel biofilms can be created that have the ability to efficiently transfer charges, perform sensory transduction and undergo processing. We conducted a study on the memfractance and memristance behaviours of composite biofilms, showcasing their capacity to carry out unconventional computing operations. The porous nanostructure and electroactivity of the biofilm create a biocompatible interface that can be used to record and stimulate neuronal networks. In addition to in vitro neuronal interfaces, these soft electroactive biofilms show potential as components for bioinspired robotics, smart wearables, unconventional computing devices and adaptive biorobotic systems. Kombucha-proteinoids composite films are a highly customizable material that can be synthesized to suit specific needs. These films belong to a unique category of ‘living’ materials, as they have the ability to support cellular systems and improve bioelectronic functionality. This makes them an exciting prospect in various applications. Ongoing efforts are currently being directed towards enhancing the compositional tuning of conductivity, signal processing and integration within hybrid bioelectronic circuits.","author":[{"family":"Nikolaidou","given":"Anna"},{"family":"Mougkogiannis","given":"Panagiotis"},{"family":"Adamatzky","given":"Andrew"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1098/rsos.240238","URL":"https://doi.org/10.1098/rsos.240238","source":"openalex"},{"id":"oa:W4404016562","type":"article-journal","title":"Variable‐Range Hopping Conduction in Amorphous, Non‐Stoichiometric Gallium Oxide","abstract":"Abstract Amorphous, non‐stoichiometric gallium oxide (a‐GaO x , x < 1.5) is a promising material for many electronic devices, such as resistive switching memories, neuromorphic circuits and photodetectors. So far, all respective measurements are interpreted with the explicit or implicit assumption of n‐type band transport above the conduction band mobility edge. In this study, the experimental and theoretical results consistently show for the first time that for an O/Ga ratio x of 0.8 to 1.0 the dominating electron transport mechanism is, however, variable‐range hopping (VRH) between localized states, even at room temperature and above. The measured conductivity exhibits the characteristic exponential temperature dependence on T −1/4 , in remarkable agreement with Mott's iconic law for VRH. Localized states near the Fermi level are confirmed by photoelectron spectroscopy and density of states (DOS) calculations. The experimental conductivity data is reproduced quantitatively by kinetic Monte Carlo (KMC) simulations of the VRH mechanism, based on the ab‐initio DOS. High electric field strengths F cause elevated electron temperatures and an exponential increase of the conductivity with F 1/2 . Novel results concerning surface oxidation, magnetoresistance, Hall effect, thermopower and electron diffusion are also reported. The findings lead to a new understanding of a‐GaO x devices, also with regard to metal|a‐GaO x Schottky barriers.","author":[{"family":"Hein","given":"Philipp"},{"family":"Romstadt","given":"Tobias"},{"family":"Draber","given":"Fabian"},{"family":"Ryu","given":"Jinseok"},{"family":"Böger","given":"Thorben"},{"family":"Falkenstein","given":"Andreas"},{"family":"Kim","given":"Miyoung"},{"family":"Martin","given":"Manfred"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1002/aelm.202400407","URL":"https://doi.org/10.1002/aelm.202400407","source":"openalex"},{"id":"oa:W4392597219","type":"article-journal","title":"Native point defects in 2D transition metal dichalcogenides: A perspective bridging intrinsic physical properties and device applications","abstract":"Two-dimensional (2D) transition metal dichalcogenides (TMDs) hold immense promise as ultrathin-body semiconductors for cutting-edge electronics and optoelectronics. In particular, their sustained charge mobility even at atomic-level thickness as well as their absence of surface dangling bonds, versatile band structures, and silicon-compatibility integration make them a prime candidate for device applications in both academic and industrial domains. Despite such high expectations, group-VI TMDs reportedly exhibit a range of enigmatic properties, such as substantial contact resistance, Fermi level pinning, and limited unipolar charge transport, which are all rooted in their inherent defects. In other words, intrinsic physical properties resulting from their native defects extend their influence beyond the material level. Bridging point-defect-induced material properties and their behavior at the device level, this Perspective sheds light on the significance of crystalline defects beyond a rather simple defect–property relationship. As a distinctive approach, we briefly review the well-established defect model of conventional III–V semiconductors and further apply it to the emergent defect behaviors of 2D TMDs such as their defect-induced gap states. Within the main discussion, we survey a range of behaviors caused by the most prevalent intrinsic defect, namely, vacancies, within 2D TMDs, and their implications for electronic and optoelectronic properties when employed at the device level. This review presents an in-depth summary of complexities in material properties as well as device characteristics arising from intrinsic point defects and provides a solid foundation for the cross-links among native defects and material/device properties.","author":[{"family":"Ko","given":"Kyungmin"},{"family":"Jang","given":"Mingyu"},{"family":"Kwon","given":"Jaeeun"},{"family":"Suh","given":"Joonki"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1063/5.0185604","URL":"https://doi.org/10.1063/5.0185604","source":"openalex"},{"id":"oa:W4402605216","type":"article-journal","title":"Increased Static Charge‐Induced Threshold Voltage Shifts and Memristor Activity in Pentacene OFETs Comprising Polystyrene‐Based Gate Dielectrics Containing Electroactive Small Molecule Crystallites","abstract":"Abstract Top‐contact bottom‐gate pentacene OFETs are fabricated with single layer dielectrics comprised of either polystyrene (PS), poly(4‐methylstyrene) (P4MS), or poly(4‐tert‐butylstyrene) (P4TBS). The polystyrenes are blended with varying concentrations of two different small molecules, dibenzotetrathiafulvalene (DBTTF) and 2,8‐difluoro‐5,11‐bis(triethylsilylethynyl)anthradithiophene (diF‐TES‐ADT), to form small, separated crystallites contained throughout the polymer dielectric layer. The OFET characteristics of these devices are investigated and their threshold voltage shifts are measured after −70 V static charging for 5 min. Two‐terminal measurements are conducted using multiple different gate biases in the range of −50 to +50 V to investigate memristor behavior in the devices. OFETs containing DBTTF exhibited ΔVth increases as large as 330% relative to control OFETs containing no DBTTF, while OFETs containing at least 7.5 wt.% DBTTF exhibited memristor activity, with currents ranging from 20 nA to 44 µA depending on the applied bias. This work demonstrates that including small, separated crystallites in polymer dielectrics enhances their charge storage ability and can be promising for creating nonbinary memory devices for data processing. Additionally, the observed memristor activity indicates the OFETs in this work can be used in development of neuromorphic systems that aim to mimic the synaptic behavior of the human nervous system.","author":[{"family":"Bond","given":"Christopher"},{"family":"Reich","given":"Daniel"},{"family":"Katz","given":"Howard"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1002/adfm.202410763","URL":"https://doi.org/10.1002/adfm.202410763","source":"openalex"},{"id":"doi:10.60893/figshare.apl.c.8523360","type":"article-journal","title":"Experimental and Numerical Demonstration of Threshold Voltage Asymmetry and Synaptic Plasticity in MoS<sub>2</sub> Transistors","abstract":"Thanks to their advanced tunable electrical properties, two-dimensional materials have emerged as a promising platform for neuromorphic computing, offering unique exibility and scalability. In this work, we report the fabrication and experimental characterization of MoS2-based transistors exhibiting counterclockwise hysteresis and synaptic plasticity. Our devices demonstrate multilevel conductivity modulation under pulsed excitation, and a pronounced dependence of the threshold voltages on the sweep rate of the applied gate signal. An in-house physics-based numerical simulator is exploited to rationalize the measured hysteresis and the frequency-dependent behavior, providing further insights into the underlying memristive mechanism, i.e. the delayed migration of oxygen ions. Moreover, simulations reveal that ion migration governs the observed plasticity, enabling control of current modulation through pulse duration. These findings establish the relevance of ionic dynamics in shaping device performance and highlight the potential of MoS2-based transistors for artificial neural network hardware.","author":[{"family":"Porzani","given":"Matteo"},{"family":"Godoy","given":"Andres"},{"family":"Cuesta-Lopez","given":"Juan"},{"family":"Ielmini","given":"Daniele"},{"family":"Farronato","given":"Matteo"},{"family":"Marin","given":"Enrique"}],"issued":{"date-parts":[[2026]]},"DOI":"10.60893/figshare.apl.c.8523360","URL":"https://doi.org/10.60893/figshare.apl.c.8523360","source":"datacite"},{"id":"doi:10.60893/figshare.apl.c.8523360.v1","type":"article-journal","title":"Experimental and Numerical Demonstration of Threshold Voltage Asymmetry and Synaptic Plasticity in MoS<sub>2</sub> Transistors","abstract":"Thanks to their advanced tunable electrical properties, two-dimensional materials have emerged as a promising platform for neuromorphic computing, offering unique exibility and scalability. In this work, we report the fabrication and experimental characterization of MoS2-based transistors exhibiting counterclockwise hysteresis and synaptic plasticity. Our devices demonstrate multilevel conductivity modulation under pulsed excitation, and a pronounced dependence of the threshold voltages on the sweep rate of the applied gate signal. An in-house physics-based numerical simulator is exploited to rationalize the measured hysteresis and the frequency-dependent behavior, providing further insights into the underlying memristive mechanism, i.e. the delayed migration of oxygen ions. Moreover, simulations reveal that ion migration governs the observed plasticity, enabling control of current modulation through pulse duration. These findings establish the relevance of ionic dynamics in shaping device performance and highlight the potential of MoS2-based transistors for artificial neural network hardware.","author":[{"family":"Porzani","given":"Matteo"},{"family":"Godoy","given":"Andres"},{"family":"Cuesta-Lopez","given":"Juan"},{"family":"Ielmini","given":"Daniele"},{"family":"Farronato","given":"Matteo"},{"family":"Marin","given":"Enrique"}],"issued":{"date-parts":[[2026]]},"DOI":"10.60893/figshare.apl.c.8523360.v1","URL":"https://doi.org/10.60893/figshare.apl.c.8523360.v1","source":"datacite"},{"id":"doi:10.48550/arxiv.2509.11721","type":"manuscript","title":"Experimental Investigation of Time Series Classification using a Self-Pulsing Microring Resonator Network","abstract":"Photonic neuromorphic computing offers compelling advantages in power efficiency and parallel processing, but often falls short in realizing scalable nonlinearity and long-term memory. These limitations can be overcome by silicon microring resonator (MRR) networks. These integrated photonic circuits enable compact, high-throughput neuromorphic computing by simultaneously exploiting spatial, temporal, and wavelength dimensions. This work provides an in-depth study of of MRR networks for photonics-based machine learning (ML). We investigate the system's effectiveness on two widely used image classification benchmarks, MNIST and Fashion-MNIST, by encoding images directly into time sequences. In particular, we enhance the computational performance of a linear readout classifier within the reservoir computing paradigm through the strategic use of multiple physical output ports, diverse laser wavelengths, and varied input power levels. Moreover, we explore a single-pixel classification setting, where inference does not require digital memory, thanks to the inherent memory and parallelism of our MRR network.","author":[{"family":"Foradori","given":"Alessandro"},{"family":"Lugnan","given":"Alessio"},{"family":"Pavesi","given":"Lorenzo"},{"family":"Bienstman","given":"Peter"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2509.11721","URL":"https://doi.org/10.48550/arxiv.2509.11721","source":"datacite"},{"id":"doi:10.17863/cam.127258","type":"article-journal","title":"Charge transport physics of organic conductors at high carrier densities","abstract":"Conducting polymers and other organic conductors enable emerging applications in bioelectronics, neuromorphic computing, energy storage and thermoelectric devices. When used in organic electrochemical transistors or other devices, these materials are typically doped in their bulk to very high carrier densities of the order of 1020–1021 cm−3. In this regime, they show fascinating nonlinear, many-body and non-equilibrium transport phenomena that are being exploited in these applications, but whose fundamental origins remain poorly understood. In this Review, we focus on the underlying charge transport physics, examining how complex microstructure, electron–electron interactions and electron–dopant counterion interactions govern transport behaviour, including the evolution of the density of states with carrier density. We also discuss reliable experimental methods for determining carrier concentrations and measuring transport coefficients. An in-depth understanding of the charge transport physics in this high-carrier-density regime is a prerequisite for harnessing these transport phenomena in device applications.","author":[{"family":"Frisbie","given":"CD"},{"family":"Jacobs","given":"Ian"},{"family":"Ren","given":"Xinglong"},{"family":"Sirringhaus","given":"Henning"}],"issued":{"date-parts":[[2026]]},"DOI":"10.17863/cam.127258","URL":"https://doi.org/10.17863/cam.127258","source":"datacite"},{"id":"doi:10.48550/arxiv.2606.26841","type":"manuscript","title":"SpikeTimer: Exploring Active Copyright Protection in Spiking Neural Networks via Temporal Backdoor Regularization","abstract":"Spiking Neural Networks (SNN) have emerged as a revolutionary paradigm compared to traditional Deep Neural Networks (DNN) in energy-efficient computing, showcasing exceptional capabilities in processing event-driven sensory data for real-time applications like robotics and edge AI systems. However, unlike extensive studies on DNN copyright solutions, SNN copyright protection remains largely underexplored due to their inherent temporal coding complexities and spike-driven computation. In this study, we propose a novel active copyright protection framework named SpikeTimer for SNNs via temporal backdoor learning. SpikeTimer partitions neuromorphic data into designated timeslices and exclusively embeds authorized tokens within authorized slices. Furthermore, the inherent temporal segmentation characteristic intrinsically enables SpikeTimer to support multi-user authorization mechanisms and accommodates token embedding of arbitrary morphology. Based on this, SpikeTimer precisely responds to authorized data containing a token within the correct timeslice, while producing erroneous responses to unauthorized data. Our key innovation lies in establishing a time-dependent authorization mechanism that protects the SNN copyright by temporal token validity. Additionally, SpikeTimer retains its defensive efficacy even under adversarial attempts. Evaluations on multiple neuromorphic datasets manifest that SpikeTimer achieves around 10% accuracy on unauthorized data with merely around 1.5% degradation on authorized inputs. Moreover, SpikeTimer demonstrates robust resistance against model finetuning and pruning threats.","author":[{"family":"Yang","given":"Xiao"},{"family":"Li","given":"Gaolei"},{"family":"Wu","given":"Jun"},{"family":"Li","given":"Jianhua"},{"family":"Liu","given":"Zhiquan"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2606.26841","URL":"https://doi.org/10.48550/arxiv.2606.26841","source":"datacite"},{"id":"doi:10.48550/arxiv.2606.26727","type":"manuscript","title":"One-shot prediction of noise-induced bifurcations with reservoir computing","abstract":"Dynamical systems can exhibit complex responses when noise is injected. In particular, dynamics can be qualitatively altered by dynamic noise, a phenomenon known as noise-induced bifurcation. Predicting noise-induced bifurcations is a critical challenge in nonlinear physics. Recently, it has been reported that reservoir computing, a machine learning framework, can reconstruct the unseen global structure of a dynamical system, including bifurcations, from limited time series data. However, learning global structures in random dynamical systems has not yet been systematically addressed. In this study, we report that a simple reservoir computing framework can predict the noise-induced bifurcation structure from the time series at a single noise condition. We demonstrate dynamic noise cancellation and the reconstruction of entire noise-induced bifurcation structures, including noise-induced chaos and noise-induced order, in representative dynamical systems. Additionally, we provide a theoretical explanation for noise cancellation and demonstrate noise cancellation of a neuromorphic spintronics device. Our results provide significant insights into understanding and harnessing real-world noisy complex dynamics.","author":[{"family":"Akashi","given":"Nozomi"},{"family":"Watanabe","given":"Takayuki"},{"family":"Hara","given":"Masato"},{"family":"Namiki","given":"Takao"},{"family":"Kokubu","given":"Hiroshi"},{"family":"Tsuda","given":"Ichiro"},{"family":"Nakajima","given":"Kohei"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2606.26727","URL":"https://doi.org/10.48550/arxiv.2606.26727","source":"datacite"},{"id":"doi:10.6084/m9.figshare.c.8509515.v1","type":"article-journal","title":"Cavity solitons as a nonlinear substrate for photonic neuromorphic computing","abstract":"Reservoir computing leverages nonlinear dynamics of physical systems to process temporal information with minimal training cost. Here, we demonstrate that cavity solitons sustained in a fiber optical cavity provide an optical platform for photonic reservoir computing. Our methodology exploits the use of a phase-modulated drive laser to encode the input, while the reservoir states are accessed through frequency-resolved readout. Numerical simulations indicate that the emission of Kelly waves enriches the dynamics and enhances performance for machine learning tasks. We evaluate the performance of the cavity-soliton reservoir computer on several standard benchmark tasks.","author":[{"family":"Arabieh","given":"Amir"},{"family":"Lupo","given":"Alessandro"},{"family":"Gorza","given":"Simon"},{"family":"Massar","given":"Serge"}],"issued":{"date-parts":[[2026]]},"DOI":"10.6084/m9.figshare.c.8509515.v1","URL":"https://doi.org/10.6084/m9.figshare.c.8509515.v1","source":"datacite"},{"id":"doi:10.6084/m9.figshare.c.8509515","type":"article-journal","title":"Cavity solitons as a nonlinear substrate for photonic neuromorphic computing","abstract":"Reservoir computing leverages nonlinear dynamics of physical systems to process temporal information with minimal training cost. Here, we demonstrate that cavity solitons sustained in a fiber optical cavity provide an optical platform for photonic reservoir computing. Our methodology exploits the use of a phase-modulated drive laser to encode the input, while the reservoir states are accessed through frequency-resolved readout. Numerical simulations indicate that the emission of Kelly waves enriches the dynamics and enhances performance for machine learning tasks. We evaluate the performance of the cavity-soliton reservoir computer on several standard benchmark tasks.","author":[{"family":"Arabieh","given":"Amir"},{"family":"Lupo","given":"Alessandro"},{"family":"Gorza","given":"Simon"},{"family":"Massar","given":"Serge"}],"issued":{"date-parts":[[2026]]},"DOI":"10.6084/m9.figshare.c.8509515","URL":"https://doi.org/10.6084/m9.figshare.c.8509515","source":"datacite"},{"id":"doi:10.48550/arxiv.2604.05220","type":"manuscript","title":"Many-body description of two-dimensional van der Waals ferroelectric $α-$In$_2$Se$_3$","abstract":"Two-dimensional (2D) van der Waals ferroelectrics are recognized for enabling many applications, from memory and logic to neuromorphic computing, as well as transforming other materials to control electronic phase transitions and topological states. While these materials are typically weakly correlated and expected to have their ground-state properties well described with the commonly used density functional theory, by focusing on bilayers and trilayers of In$_2$Se$_3$ we show that this approach may not be reliable. The underlying electronic structure strongly depends on the polarization structure of the multilayer system and is surprisingly challenging to accurately calculate, requiring a high-fidelity many-body theory of the quasiparticle self-consistent \\textit{GW} approximation. We develop this underlying description by extending the capabilities of Green function implementation within the open-source Questaal package. We show that even a sophisticated hybrid functional approach may fail to predict a nonvanishing gap in a bilayer In$_2$Se$_3$ and yields charge density, polarization, and band offsets that strongly deviate from the many-body picture. We discuss the implications of these computational advances for future opportunities in 2D ferroelectrics.","author":[{"family":"Ayala","given":"Denzel"},{"family":"Pashov","given":"Dimitar"},{"family":"Zhou","given":"Tong"},{"family":"Belashchenko","given":"Kirill"},{"family":"Van Schilfgaarde","given":"Mark"},{"family":"Žutić","given":"Igor"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2604.05220","URL":"https://doi.org/10.48550/arxiv.2604.05220","source":"datacite"},{"id":"doi:10.48550/arxiv.2606.24045","type":"manuscript","title":"Giant and Continuous Ionic Current Oscillation Induced by Dynamic Surface Charge Regulation in Cylindrical Mesopores","abstract":"Nanofluidic ionic oscillators based on the dynamic regulation of surface charges hold great promise for neuromorphic computing, biosensing, and ionic circuits. Here, by dynamically adjusting the local charge inversion on pore walls, we present a simple and effective strategy to achieve periodic current oscillations by harnessing the transient adsorption and desorption of Ca2+ ions in cylindrical mesopores under concentration gradients. Based on the combined precision current measurements and multiphysics simulations, we demonstrate that local overadsorption of Ca2+ ions may induce asymmetric bipolar charge distributions along the pore axis, which periodically reverses the direction of electroosmotic flow and modulates the local ion concentration inside the pore, generating highly regular current oscillations. Notably, both the oscillation frequency and the open-state probability of the pore vary nearly linearly with the applied voltage. Moreover, under dynamic voltage scanning, the system exhibits typical memristive hysteresis, and the switching between the open and closed states is highly reproducible. This work not only reveals the dynamic, heterogeneous surface charge regulation by divalent ions, but also provides a simple, material agnostic method for constructing ionic oscillators and memristors based on dynamic adsorption/desorption of multivalent ions.","author":[{"family":"Zhang","given":"Hongwen"},{"family":"Zhao","given":"Yujie"},{"family":"Gong","given":"Zekun"},{"family":"Lin","given":"Chih"},{"family":"Sui","given":"Tianyi"},{"family":"Siwy","given":"Zuzanna"},{"family":"Qiu","given":"Yinghua"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2606.24045","URL":"https://doi.org/10.48550/arxiv.2606.24045","source":"datacite"},{"id":"doi:10.3929/ethz-c-000798429","type":"article-journal","title":"In-fibre logic and memory via tuneable passivation–corrosion","abstract":"Textile electronics with digital capabilities could sense, process, and store data, while providing immersive interaction with user and their immediate surroundings. However, existing textile electronic systems are typically built on von Neumann architecture and rigid chips, limiting their seamless integration with clothing. Here, we propose a single-fibre logic/memory electronic device based on interface passivation-corrosion whose functions do not depend on traditional carrier heterojunction interfaces. The same fibre can be switched to operate as either a diode or a memristor. The diode mode remains stable under higher voltages and longer cycling periods than the state-of-the-art anion–cation heterojunction fibres. The fibre electronics are highly stretchable (up to 50%), and are compatible with industry-standard weaving techniques. We also demonstrate the application of these fibres in “AND” and “OR” logic gates, neuromorphic synapses, and textile memristor arrays. Regulated passivation-corrosion-enabled logic and memory in fibres offers a promising avenue for the next-generation textile computing.","author":[{"family":"Li","given":"Yuanlong"},{"family":"Yang","given":"Weifeng"},{"family":"Shokurov","given":"Alexander"},{"family":"Reis Carneiro","given":"Manuel"},{"family":"Menon","given":"Carlo"}],"issued":{"date-parts":[[2026]]},"DOI":"10.3929/ethz-c-000798429","URL":"https://doi.org/10.3929/ethz-c-000798429","source":"datacite"},{"id":"doi:10.60893/figshare.apl.c.8523579","type":"article-journal","title":"<strong>Self-Rectifying Organic Memristor Based on PAA/PEDOT:PSS Heterojunction for Neuromorphic Computing </strong>","abstract":"Neuromorphic computing has emerged as a promising strategy to overcome the intrinsic limitations of the von Neumann architecture, where memristive devices that emulate biological synapses are of particular interest. Among them, organic memristors offer unique advantages in mechanical flexibility, biocompatibility, and solution processability. Here, we report a flexible, self-rectifying organic memristor fabricated on an ITO/PAA(Ca 2+ )/PEDOT:PSS/ITO architecture. The device operates through directional Ca 2+ migration within the poly(acrylic acid) (PAA) electrolyte layer, enabling stable resistive switching. Importantly, the integration of PEDOT:PSS with ITO forms an intrinsic p-n junction, imparting pronounced self-rectifying behavior with a rectification ratio of approximately 10 3 , effectively suppressing sneak-path currents in crossbar arrays. The memristor reliably reproduces key synaptic functions, including transitions from short-term to long-term potentiation and depression, as well as spike-timing-dependent plasticity. Artificial neural networks constructed using experimentally extracted device characteristics achieve image and digit recognition accuracies exceeding 90%. This work demonstrates that flexible organic memristors with built-in p-n junction rectification can simultaneously address mechanical compliance, energy efficiency, and integration density, thereby establishing a viable materials and device platform for biocompatible neuromorphic hardware and scalable pattern-recognition systems.","author":[{"family":"Tang","given":"Xiuyang"},{"family":"Sun","given":"Weifang"},{"family":"He","given":"Niwei"},{"family":"Ha","given":"Sizhu"},{"family":"Xue","given":"Song"},{"family":"Cai","given":"Gangri"},{"family":"Zhao","given":"Jinshi"},{"family":"Shi","given":"Jingzhou"},{"family":"Ma","given":"Xinming"}],"issued":{"date-parts":[[2026]]},"DOI":"10.60893/figshare.apl.c.8523579","URL":"https://doi.org/10.60893/figshare.apl.c.8523579","source":"datacite"},{"id":"doi:10.60893/figshare.apl.c.8523579.v1","type":"article-journal","title":"<strong>Self-Rectifying Organic Memristor Based on PAA/PEDOT:PSS Heterojunction for Neuromorphic Computing </strong>","abstract":"Neuromorphic computing has emerged as a promising strategy to overcome the intrinsic limitations of the von Neumann architecture, where memristive devices that emulate biological synapses are of particular interest. Among them, organic memristors offer unique advantages in mechanical flexibility, biocompatibility, and solution processability. Here, we report a flexible, self-rectifying organic memristor fabricated on an ITO/PAA(Ca 2+ )/PEDOT:PSS/ITO architecture. The device operates through directional Ca 2+ migration within the poly(acrylic acid) (PAA) electrolyte layer, enabling stable resistive switching. Importantly, the integration of PEDOT:PSS with ITO forms an intrinsic p-n junction, imparting pronounced self-rectifying behavior with a rectification ratio of approximately 10 3 , effectively suppressing sneak-path currents in crossbar arrays. The memristor reliably reproduces key synaptic functions, including transitions from short-term to long-term potentiation and depression, as well as spike-timing-dependent plasticity. Artificial neural networks constructed using experimentally extracted device characteristics achieve image and digit recognition accuracies exceeding 90%. This work demonstrates that flexible organic memristors with built-in p-n junction rectification can simultaneously address mechanical compliance, energy efficiency, and integration density, thereby establishing a viable materials and device platform for biocompatible neuromorphic hardware and scalable pattern-recognition systems.","author":[{"family":"Tang","given":"Xiuyang"},{"family":"Sun","given":"Weifang"},{"family":"He","given":"Niwei"},{"family":"Ha","given":"Sizhu"},{"family":"Xue","given":"Song"},{"family":"Cai","given":"Gangri"},{"family":"Zhao","given":"Jinshi"},{"family":"Shi","given":"Jingzhou"},{"family":"Ma","given":"Xinming"}],"issued":{"date-parts":[[2026]]},"DOI":"10.60893/figshare.apl.c.8523579.v1","URL":"https://doi.org/10.60893/figshare.apl.c.8523579.v1","source":"datacite"},{"id":"doi:10.48550/arxiv.2606.23290","type":"manuscript","title":"Reconfigurable all-optical inference via tunable second-harmonic generation and spin-orbit coupling cascade","abstract":"Spin-orbit coupling (SOC) is widely exploited as a fundamental mechanism for generating orbital angular momentum (OAM); however, conventional approaches typically lack flexibility and tunability. Here, we introduce a continuously tunable second-harmonic generation (SHG)-SOC cascade mechanism modulated by a spatially movable nonlinear crystal. Under linearly polarized excitation, the SHG-SOC cascade engages synchronously with both degenerate and nondegenerate SHG processes, thereby expanding the OAM spectrum and significantly enhancing the information density and feature-mapping capacity of the optical field. Moreover, the OAM spectral distribution can be continuously reconfigured simply by translating the nonlinear crystal. This deterministic physical evolution, which maps simple OAM modes onto a tunable high-dimensional OAM space, is mathematically analogous to the high-dimensional feature expansion performed by a kernel function of a support vector machine (SVM) in machine learning. Such dimensional expansion can project linearly inseparable input data into a high-dimensional space where they become linearly separable. Exploiting this physics-algorithm analogy, we develop a reconfigurable all-optical inference platform. As a proof of concept, we successfully perform classification tasks, including the recognition of Iris flowers and Palmer penguins. This work establishes a scalable, physically reconfigurable architecture for high-dimensional all-optical computing and neuromorphic photonics.","author":[{"family":"Zhang","given":"Li"},{"family":"Zhuang","given":"Zikuan"},{"family":"Deng","given":"Ronghao"},{"family":"Hong","given":"Ling"},{"family":"Zhang","given":"Yu"},{"family":"Lin","given":"Fei"},{"family":"Liu","given":"Zhengxian"},{"family":"Sun","given":"Jingxuan"},{"family":"Zhu","given":"Wenguo"},{"family":"Xie","given":"Zhenwei"},{"family":"Li","given":"Yongyao"},{"family":"Zhao","given":"Dongxu"},{"family":"Yuan","given":"Xiaocong"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2606.23290","URL":"https://doi.org/10.48550/arxiv.2606.23290","source":"datacite"},{"id":"doi:10.48550/arxiv.2508.14520","type":"manuscript","title":"Quantization Meets Spikes: Nearly Lossless Conversion at the First Timestep via Polarity Multi-Spike Mapping","abstract":"Spiking neural networks (SNNs) offer advantages in computational efficiency via event-driven computing, compared to traditional artificial neural networks (ANNs). While direct training methods tackle the challenge of non-differentiable activation mechanisms in SNNs, they often suffer from high computational and energy costs during training. As a result, ANN-to-SNN conversion approach remains a valuable and practical alternative. These conversion-based methods aim to leverage the discrete output produced by the quantization layer to obtain SNNs with low latency. Although the theoretical minimum latency is one timestep, existing conversion methods have struggled to realize such ultra-low latency without accuracy loss. Moreover, current quantization approaches often discard negative-value information following batch normalization and are highly sensitive to the hyperparameter configuration, leading to degraded performance. In this work, we, for the first time, analyze the information loss introduced by quantization layers through the lens of information entropy. Building on our analysis, we introduce polarity multi-spike mapping (PMSM) framework and a hyperparameter initialization strategy tailored for the quantization layer. Our method achieves nearly lossless ANN-to-SNN conversion at the extremity, i.e., the first timestep, while also leveraging the temporal dynamics of SNNs across multiple timesteps to maintain stable performance on complex tasks. Extensive experiments on six image and neuromorphic datasets consistently demonstrate that PMSM achieves nearly lossless accuracy at the first timestep. Remarkably, despite operating under ultra-low-latency constraints, PMSM surpasses state-of-the-art direct training methods on multiple benchmarks.","author":[{"family":"Zhang","given":"Hangming"},{"family":"Li","given":"Zheng"},{"family":"Ma","given":"Chenxiang"},{"family":"Tang","given":"Huajin"},{"family":"Cheng","given":"Long"},{"family":"Tan","given":"Kay"},{"family":"Yu","given":"Qiang"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2508.14520","URL":"https://doi.org/10.48550/arxiv.2508.14520","source":"datacite"},{"id":"doi:10.48550/arxiv.2606.17853","type":"manuscript","title":"An Optimization Framework for Automated Assessment of Biological Plausibility of Spiking Neurons","abstract":"Biological plausibility is a key concept in neuromorphic computing and spiking neural networks, yet it remains inconsistently defined and difficult to quantify. In this work, we present an open-source framework for the automated assessment of biological plausibility in spiking neuron models. Our method builds on the idea of evaluating a model's ability to replicate canonical neuronal firing patterns observed in biological systems, following the classification proposed by Izhikevich. By encoding these patterns into objective functions and optimizing model parameters accordingly, our framework enables empirical assessment without requiring prior analytical modeling. Treating neuron models as black boxes, it provides a practical and flexible means of characterizing their dynamic capabilities. We demonstrate the effectiveness of the framework on several established models and a previously unexplored custom model. Implemented in Python and compatible with PyTorch and the Norse library, the framework is tailored for machine learning contexts. It is intended as a starting point for systematic research into the relationship between biological plausibility and network-level performance metrics such as accuracy, energy efficiency, robustness, and adaptability.","author":[{"family":"Nitzsche","given":"Sven"},{"family":"Ionita","given":"Alexandru"},{"family":"Faust","given":"Andreas"},{"family":"Ionescu","given":"Bogdan"},{"family":"Becker","given":"Juergen"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2606.17853","URL":"https://doi.org/10.48550/arxiv.2606.17853","source":"datacite"},{"id":"doi:10.60893/figshare.jap.c.8468277","type":"article-journal","title":"<strong>Integration of oscillator-based feature extraction for energy-efficient convolutional neural networks</strong>","abstract":"The rapid growth in machine learning (ML) and artificial intelligence (AI) workloads has increased the demand for computing power. Although digital computing accelerators dominate today's market, analog computing architectures are emerging as promising energy-efficient alternatives. In this work, we propose an approach to address these challenges by replacing the first layer of a convolutional neural network (CNN) with a network of oscillatory retinal neurons (ORNs) composed of weakly coupled, optically activated negative differential resistance (NDR) devices. Each ORN consists of a photodetector exhibiting NDR behavior under illumination, coupled with an inductor forming a self-oscillating circuit without needing external voltage sources. We model the nonlinear oscillator dynamics using experimentally measured device characteristics and simulate their behavior under varying optical inputs. Simulations performed on the Fashion MNIST dataset demonstrate that the hybrid ORN-CNN architecture maintains high recognition accuracy, achieves 95% training accuracy and 92% testing accuracy comparable to the fully software-defined CNN with 93% test accuracy while dramatically reducing energy consumption to 24 aJ/operation. Different ORN network topologies were investigated, with asymmetric inductive coupling achieving consistently high performance. These results highlight the potential of oscillator-based networks for low-power analog computing and demonstrate the possibility of using nonlinear physical systems for energy-efficient machine learning applications. This work demonstrates a possible path for integrating device-level oscillator dynamics into future neuromorphic and analog computing platforms.","author":[{"family":"Kapadia","given":"Rehan"},{"family":"Mousavi","given":"Mirbehrad"},{"family":"Wu","given":"Zezhi"},{"family":"Ahsan","given":"Ragib"},{"family":"Abbasi Jalal","given":"Seyedeh"}],"issued":{"date-parts":[[2026]]},"DOI":"10.60893/figshare.jap.c.8468277","URL":"https://doi.org/10.60893/figshare.jap.c.8468277","source":"datacite"},{"id":"doi:10.48550/arxiv.2606.09460","type":"manuscript","title":"A 65-nm Privacy-Preserving Neuromorphic Encoder With 7.13-nJ Efficiency, 2.38-Mb/mm^2 Item-Memory Density, and Federated Learning Support","abstract":"The increasing demand for privacy-preserving personal data analytics in smart assistants, wearable health monitors, and context-aware systems calls for hardware that is both energy-efficient and secure. This work presents a 65-nm privacy-preserving neuromorphic encoder that leverages transistor-level process variation as physically unclonable entropy for hyperdimensional computing. The proposed 2T-2T entropy cell enables compact, device-specific, and write-free item memory, allowing privacy-preserving bio-signal encoding without storing random basis vectors in conventional memory. The fabricated prototype achieves 7.13 nJ per encoding, 2.38 Mb/mm^2 item-memory density, 76.44 nJ per prediction, and 357.32 nJ per training update. It also supports in-situ decision-making, continual learning, and federated learning for multi-user deployment and cold-start personalization. Evaluations across bio-signal datasets demonstrate 93.2% accuracy on EMG and 96.1% accuracy on UCI-HAR, while reducing hypervector dimensionality by 14.3x compared with binary hyperdimensional computing. These results demonstrate an energy-efficient and privacy-preserving neuromorphic hardware platform for secure edge biomedical intelligence.","author":[{"family":"Cheng","given":"Boyang"},{"family":"Liu","given":"Jianbo"},{"family":"Davis","given":"Steven"},{"family":"Enciso","given":"Zephan"},{"family":"Pei","given":"Likai"},{"family":"Zhao","given":"Xueji"},{"family":"Chang","given":"Muya"},{"family":"Cao","given":"Ningyuan"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2606.09460","URL":"https://doi.org/10.48550/arxiv.2606.09460","source":"datacite"},{"id":"doi:10.5281/zenodo.20672359","type":"article-journal","title":"Cerebellum-Inspired Memtransistors Enable Emergent Differentiation for Hardware-Efficient Novelty Detection","abstract":"Artificial intelligence (AI) algorithms are currently executed using silicon-based hardware, resulting in excessively high energy demand for data centers. Edge computing AI for healthcare, robotics, and autonomous vehicles presents even stricter constraints on power and latency, which are currently unmet by incumbent computing architectures. Efficient computation can be derived from the key properties of biological neurons, including memory-logic colocation, asynchronous parallelism, and spike-triggered computation. Here, we draw inspiration from the biological cerebellum to demonstrate an asymmetric-contact-gated MoS2 memtransistor that exhibits bias-polarity-dependent excitatory/inhibitory short-term plasticity. A memtransistor-based neural network realizes a changing interplay of excitatory/inhibitory responses, emulating the emergent synaptic differentiation of the cerebellum, enabling rapid identification of novel events. When applied to electrocardiogram data, arrhythmias are detected on the time scale of a single heartbeat with 10,000-fold fewer operations than existing silicon-based approaches. In this manner, cerebellum-inspired neuromorphic hardware provides a pathway to low-computation, high-speed novelty detection for edge intelligence.","author":[{"family":"Hersam","given":"Mark"},{"family":"Sangwan","given":"Vinod"},{"family":"Trivedi","given":"Amit"},{"family":"Raman","given":"Indira"},{"family":"Dravid","given":"Vinayak"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20672359","URL":"https://doi.org/10.5281/zenodo.20672359","source":"datacite"},{"id":"doi:10.5281/zenodo.20672360","type":"article-journal","title":"Cerebellum-Inspired Memtransistors Enable Emergent Differentiation for Hardware-Efficient Novelty Detection","abstract":"Artificial intelligence (AI) algorithms are currently executed using silicon-based hardware, resulting in excessively high energy demand for data centers. Edge computing AI for healthcare, robotics, and autonomous vehicles presents even stricter constraints on power and latency, which are currently unmet by incumbent computing architectures. Efficient computation can be derived from the key properties of biological neurons, including memory-logic colocation, asynchronous parallelism, and spike-triggered computation. Here, we draw inspiration from the biological cerebellum to demonstrate an asymmetric-contact-gated MoS2 memtransistor that exhibits bias-polarity-dependent excitatory/inhibitory short-term plasticity. A memtransistor-based neural network realizes a changing interplay of excitatory/inhibitory responses, emulating the emergent synaptic differentiation of the cerebellum, enabling rapid identification of novel events. When applied to electrocardiogram data, arrhythmias are detected on the time scale of a single heartbeat with 10,000-fold fewer operations than existing silicon-based approaches. In this manner, cerebellum-inspired neuromorphic hardware provides a pathway to low-computation, high-speed novelty detection for edge intelligence.","author":[{"family":"Hersam","given":"Mark"},{"family":"Sangwan","given":"Vinod"},{"family":"Trivedi","given":"Amit"},{"family":"Raman","given":"Indira"},{"family":"Dravid","given":"Vinayak"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20672360","URL":"https://doi.org/10.5281/zenodo.20672360","source":"datacite"},{"id":"doi:10.48550/arxiv.2606.13516","type":"manuscript","title":"Adaptive-Frequency Resonate-and-Fire Neurons for Spectral Estimation of Streaming Radar Signals","abstract":"Frequency Modulated Continuous Wave (FMCW) radar systems traditionally rely on Fourier-based methods, such as the Fast Fourier Transform (FFT), to estimate target range and velocity. While computationally efficient, these approaches require storing and processing large blocks of data, which can become a bottleneck in memory-constrained or low-latency applications. In this work, we propose a neuromorphic-inspired signal processing method based on adaptive resonate-and-fire (ARF) neurons formulated as a discrete-time dynamical system. Each neuron dynamically adjusts its internal frequency to match dominant frequency components of the input radar signal, enabling direct estimation of target ranges and velocities without computing the full frequency spectrum. The proposed model operates in a sample-by-sample fashion, resulting in memory requirements that scale with the number of tracked targets rather than the signal length. A feedback mechanism is also introduced to enable multiple neurons to lock on distinct frequency components in multi-target cases. Results on simulated and experimental data demonstrate that the method can successfully track multiple targets. Compared to conventional FFT-based approaches, the proposed method offers reduced memory usage proportional only to the number of tracked targets, making it suitable for resource-constrained and edge-based radar applications.","author":[{"family":"Chiavazza","given":"Stefano"},{"family":"Yuan","given":"Sen"},{"family":"Geilen","given":"Marc"},{"family":"Fioranelli","given":"Francesco"},{"family":"Corradi","given":"Federico"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2606.13516","URL":"https://doi.org/10.48550/arxiv.2606.13516","source":"datacite"},{"id":"doi:10.34734/fzj-2026-02724","type":"article-journal","title":"Device-to-logic variability propagation in RRAM-based logic-in-memory architectures","abstract":"Logic-in-memory (LiM) has emerged as a promising paradigm to address the von Neumann bottleneck by integrating data storage and in-situ computation. Resistive random-access memory (RRAM) is a strong candidate for LiM owing to its non-volatility, fast switching characteristics, and compatibility with CMOS integration. However, intrinsic cycle-to-cycle and device-to-device variability fundamentally limits logic reliability in RRAM-based LiM architectures, necessitating well-defined device specifications and variability margins to ensure correct operation. In this work, we experimentally demonstrate a CMOS-integrated TaOx-based 1T1R RRAM computing fabric that supports reconfigurable LiM operations, including a functionally complete Boolean set and in-memory arithmetic primitives. Through extensive statistical measurements and variability-aware statistical modeling, we systematically evaluate the impact of key variability sources, SET voltage (VSET), low resistance state, and high resistance state, on logic correctness and identify the dominant contributors to computational failure. Finally, we derive the device specifications and requirements to achieve error-free stateful LiM operations in the 1T1R RRAM arrays.","author":[{"family":"Bende","given":"Ankit"},{"family":"Singh","given":"Simranjeet"},{"family":"Kumar Jha","given":"Chandan"},{"family":"Storelli","given":"Daniele"},{"family":"Nielinger","given":"Dennis"},{"family":"Drechsler","given":"Rolf"},{"family":"Dittmann","given":"Regina"},{"family":"Menzel","given":"Stephan"},{"family":"Merchant","given":"Farhad"},{"family":"Rana","given":"Vikas"}],"issued":{"date-parts":[[2026]]},"DOI":"10.34734/fzj-2026-02724","URL":"https://doi.org/10.34734/fzj-2026-02724","source":"datacite"},{"id":"doi:10.48550/arxiv.2602.14736","type":"manuscript","title":"Coupled integrated photonic quantum memristors using a single photon source made of a colour center","abstract":"Photonic quantum memristors provide a measurement-induced route to nonlinear and history-dependent quantum dynamics. Experimental demonstrations have so far focused on isolated devices or simple cascaded devices configurations. Here, we experimentally realize and characterize a network of two coupled photonic quantum memristors with crossed feedback, implemented on a silicon nitride photonic integrated circuit and fed by a room-temperature single-photon source based on a silicon-vacancy color center SiV$^-$ in a nanodiamond. Each memristor consists of an integrated Mach-Zehnder interferometer whose transfer function is adaptively updated by photon detection events on another memristor, thus generating novel non-Markovian input-output dynamics with an enhanced memristive behaviour compared to single devices. In particular, we report inter-memristor input-output hysteresis curves exhibiting larger form factors and displaying self-intersecting loops, respectively revealing marked bistability and self-intersecting hysteresis geometry. Furthermore, numerical simulations show how these features emerge from the interplay between memory depth and relative input phase, for both intra- and inter-memristor input-output relations. We experimentally test the performance of our system in the NARMA task. Our results establish coupled integrated photonic quantum memristors as scalable nonlinear building blocks and highlight their potential for implementing compact quantum neuromorphic and reservoir computing architectures.","author":[{"family":"Baldazzi","given":"Alessio"},{"family":"Ancel","given":"Roy"},{"family":"Guaraldo","given":"Sebastiano"},{"family":"Fattori","given":"Ivan"},{"family":"Chen","given":"Xuan"},{"family":"Akar","given":"Ziad"},{"family":"Deturche","given":"Regis"},{"family":"Azzini","given":"Stefano"},{"family":"Couteau","given":"Christophe"},{"family":"Pavesi","given":"Lorenzo"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2602.14736","URL":"https://doi.org/10.48550/arxiv.2602.14736","source":"datacite"},{"id":"doi:10.60893/figshare.aip.c.8494164","type":"article-journal","title":"Mimicking Neural Habituation Behavior with Pristine Leaf-based Memristors","abstract":"Neuromorphic computing systems allow electronics to mimic the behavior of biological neurons and hold potential for high-efficiency computation needed for artificial intelligence. One class of electronic units capable of mimicking neural behavior is resistive memory devices, also known as memristors. While the most commonly studied memristors are solid-state in nature, there is a growing interest in the development of ion-rich fluidic memristors since they have a higher level of versatility to mimic biological neurons and synapses, which are themselves fluidic systems with mobile ions as neuroreceptors. Here, we present our work on elucidating the ion transport mechanism in leaf-based capacitive-coupled memristive devices and their use in mimicking habituation, a neural behavior where an organism's response to repeated applied stimulus diminishes. This work demonstrates the potential that leaf structures hold in serving as biomimetic architecture for the development of highly effective fluidic memristors for simulating neural behavior.","author":[{"family":"Savage","given":"Andrew"},{"family":"Wilder","given":"Samantha"},{"family":"Adhikari","given":"Ramesh"}],"issued":{"date-parts":[[2026]]},"DOI":"10.60893/figshare.aip.c.8494164","URL":"https://doi.org/10.60893/figshare.aip.c.8494164","source":"datacite"},{"id":"doi:10.60893/figshare.apl.c.8492772","type":"article-journal","title":"<strong>Electrically controlled tuning the nonlinearity and asymmetry of synaptic behaviors in a magnetoelectrically coupled memristor for high-performance neuromorphic computing</strong>","abstract":"High-performance artificial synaptic devices capable of emulating biological synaptic functions are crucial for developing energy-efficient neuromorphic computing systems. Memristors, whose conductance can be modulated to mimic synaptic plasticity, offer a promising platform for such applications. In this work, we present an electric-field-controlled memristor based on an FeGaB/PMN-PT multiferroic heterostructure, in which synaptic weights are continuously tuned by applied voltage pulses. The device was fabricated via magnetron sputtering and exhibits nonvolatile, multi-level resistance switching under pulsed electric field modulation, achieving 83 distinct and continuously adjustable resistance states. Key synaptic functionalities including long-term potentiation/depression (LTP/LTD), paired-pulse facilitation (PPF), and spike-timing-dependent plasticity (STDP) are successfully demonstrated. When integrated into a recurrent neural network, the system achieves a recognition accuracy of 92.4% in a benchmark task. By quantifying the nonlinearity (NL) and asymmetry (AS) of weight updates and directly linking them to accuracy performance, we further reveal that device-level NL critically governs system-level computational accuracy. This insight provides a clear optimization pathway for future artificial spintronic synapses, underscoring their potential for high-speed, nonvolatile, and adaptive neuromorphic hardware.","author":[{"family":"Song","given":"Guangxiao"},{"family":"Zhou","given":"Hao"},{"family":"Yu","given":"Guoliang"},{"family":"Zhu","given":"Haibin"},{"family":"Huang","given":"Ankang"},{"family":"Ouyang","given":"Hui"},{"family":"Zhu","given":"Mingmin"},{"family":"Wang","given":"Jiawei"},{"family":"Jing","given":"Xufeng"},{"family":"Wang","given":"Xin"},{"family":"Wang","given":"Wei"},{"family":"Qiu","given":"Yang"}],"issued":{"date-parts":[[2026]]},"DOI":"10.60893/figshare.apl.c.8492772","URL":"https://doi.org/10.60893/figshare.apl.c.8492772","source":"datacite"},{"id":"doi:10.48550/arxiv.2606.08584","type":"manuscript","title":"Convolutional Sparse Coding via the Locally Competitive Algorithm on Loihi 2","abstract":"Sparse coding provides a principled framework for signal representation by expressing an input as a linear combination of only a small number of basis functions. The Locally Competitive Algorithm (LCA) is particularly attractive in the context of neuromorphic computing because its dynamics, leaky integration, thresholding, and lateral inhibition map naturally to neuromorphic hardware. While prior work has studied non-convolutional LCA on Loihi 2, the convolutional setting is of particular interest because it introduces spatial structure, weight sharing, overlapping receptive fields, and scaling behavior that are more representative of practical sparse inference workloads. In this work, we present a Loihi 2 implementation of convolutional sparse coding via the LCA and evaluate it against a conventional GPU baseline on the same inference problems. The implementation follows a one-layer recurrent LCA formulation and extends it to convolutional feature maps with local inhibitory kernels derived from pairwise filter interactions. To the best of our knowledge, this is the first implementation and benchmark of convolutional LCA on Loihi 2. Our goal is not only to demonstrate feasibility, but also to clarify in which operating regimes convolutional sparse inference becomes attractive on neuromorphic hardware. The resulting study positions convolutional LCA as a useful benchmark for structured sparse inference on emerging neuromorphic systems.","author":[{"family":"Kasenbacher","given":"Geoffrey"},{"family":"Ruepp","given":"Daniel"},{"family":"Ecke","given":"Gerrit"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2606.08584","URL":"https://doi.org/10.48550/arxiv.2606.08584","source":"datacite"},{"id":"doi:10.48550/arxiv.2606.08515","type":"manuscript","title":"Unifying von-Neumann HPC and Neuromorphic Acceleration via the EBRAINS Research Infrastructure: A Framework for High-Performance Workflows","abstract":"Modern scientific workflows increasingly span diverse computing architectures, yet executing a single computational model across disparate systems often forces researchers to maintain fragmented, site-specific pipelines. In this paper, we address this challenge within the domain of computational neuroscience by presenting a unified, cloud-based workflow orchestrated via EBRAINS JupyterLab. This workflow enables users to transparently execute spiking neural networks on both von-Neumann supercomputers and neuromorphic hardware. Using a single federated identity, the system dispatches jobs to HPC sites (JUSUF, Galileo100) via PyUNICORE and to the SpiNNaker-1 neuromorphic system via the Neuromorphic Computing Platform Interface. To guarantee cross-site reproducibility and mitigate software version drift, we utilize a zero-installation execution mode that dynamically pulls PMIx-aware Apptainer containers to HPC compute nodes. Furthermore, we demonstrate genuine model-level portability using the NESTML domain-specific language, allowing custom neuron models to be written once and automatically compiled for either the NEST (C++) or sPyNNaker backends. Validated with a balanced random network case study, this work illustrates a practical, end-to-end path for hardware-agnostic workflows while highlighting the critical role of containerization and domain-specific languages in achieving true cross-platform reproducibility.","author":[{"family":"Singh","given":"Krishna"},{"family":"Linssen","given":"Charl"},{"family":"Müller","given":"Eric"},{"family":"Mathioulaki","given":"Eleni"},{"family":"Klijn","given":"Wouter"},{"family":"Oden","given":"Lena"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2606.08515","URL":"https://doi.org/10.48550/arxiv.2606.08515","source":"datacite"},{"id":"doi:10.48550/arxiv.2501.18894","type":"manuscript","title":"Nonlinear Inference Capacity of Fiber-Optical Extreme Learning Machines","abstract":"The intrinsic complexity of nonlinear optical phenomena offers a fundamentally new resource to analog brain-inspired computing, with the potential to address the pressing energy requirements of artificial intelligence. We introduce and investigate the concept of nonlinear inference capacity in optical neuromorphic computing in highly nonlinear fiber-based optical Extreme Learning Machines. We demonstrate that this capacity scales with nonlinearity to the point where it surpasses the performance of a deep neural network model with five hidden layers on a scalable nonlinear classification benchmark. By comparing normal and anomalous dispersion fibers under various operating conditions and against digital classifiers, we observe a direct correlation between the system's nonlinear dynamics and its classification performance. Our findings suggest that image recognition tasks, such as MNIST, are incomplete in showcasing deep computing capabilities in analog hardware. Our approach provides a framework for evaluating and comparing computational capabilities, particularly their ability to emulate deep networks, across different physical and digital platforms, paving the way for a more generalized set of benchmarks for unconventional, physics-inspired computing architectures.","author":[{"family":"Saeed","given":"Sobhi"},{"family":"Müftüoglu","given":"Mehmet"},{"family":"Cheeran","given":"Glitta"},{"family":"Bocklitz","given":"Thomas"},{"family":"Fischer","given":"Bennet"},{"family":"Chemnitz","given":"Mario"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2501.18894","URL":"https://doi.org/10.48550/arxiv.2501.18894","source":"datacite"},{"id":"doi:10.48550/arxiv.2606.04671","type":"manuscript","title":"Higher-order exceptional points in a multimode continuum optoacoustic system","abstract":"Exceptional points appear in non-Hermitian systems as degeneracies, where not only eigenvalues but also eigenvectors coalesce. They are of great theoretical and experimental interest due to their exotic topological properties and enhanced sensitivity to perturbations. Experimental realizations of higher-order exceptional points, where more than two eigenvectors coalesce, rely on highly fine-tuned setups. Recently, stimulated Brillouin scattering has been employed to generate second-order exceptional points in a fabrication-free setup by leveraging off-resonant scattering. In this work we generalize this approach, and we develop an off-resonant, multimode theory for stimulated Brillouin scattering as an avenue towards realizing symmetry-induced exceptional points of any order. We present the experimental implementation of our program in an accompanying paper. Our multimode theory could also be employed in applications in optoacoustic sensing, synthetic neuromorphic computing, microwave photonic filters, and optoacoustic quantum signal processing.","author":[{"family":"Montag","given":"Anton"},{"family":"Gohsrich","given":"Julius"},{"family":"Levoy","given":"Quentin"},{"family":"Stiller","given":"Birgit"},{"family":"Kunst","given":"Flore"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2606.04671","URL":"https://doi.org/10.48550/arxiv.2606.04671","source":"datacite"},{"id":"doi:10.48550/arxiv.2606.01776","type":"manuscript","title":"A 32-Channel 3.53-μW Per Channel Brain-Machine Interface SoC Featuring Dual-Threshold Delta-modulation, In-Memory Spike Detection and Bi-SNN Based Motor Decoding","abstract":"With the scaling of sensor channel counts, systems confront challenges in frontend data sensing and on-implant data processing. This work presents a 32-channel fully event-based iBMI SoC in 65nm CMOS for an efficient neuromorphic signal processing pipeline. The SoC integrates a 32-channel dual-threshold delta modulation (DTDM) frontend array that provides up to 26x data compression at the frontend, an in-memory computing (IMC) spike detector (SPD) for efficient in-pixel spike detection, and a bipolar LIF-based spiking neural network (Bi-SNN) decoder for on-chip motor intention decoding (MID). Consuming only 3.53 μW per channel and achieving ~0.62 decoding R2 with a compact 0.034 mm2 per-channel area, the chip enables high-efficiency signal recording, processing, and decoding for implantable devices.","author":[{"family":"Ke","given":"Ye"},{"family":"Fu","given":"Zhengnan"},{"family":"Sun","given":"Pao"},{"family":"Guo","given":"An"},{"family":"Dong","given":"Shuai"},{"family":"Yang","given":"Junyi"},{"family":"Yang","given":"Yahan"},{"family":"Eldaly","given":"Abdelrahman"},{"family":"Si","given":"Xin"},{"family":"Chan","given":"Leanne"},{"family":"Basu","given":"Arindam"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2606.01776","URL":"https://doi.org/10.48550/arxiv.2606.01776","source":"datacite"},{"id":"doi:10.48550/arxiv.2606.01463","type":"manuscript","title":"Emerging Non-Volatile Opto-electronic Resistive Memories for Next-Generation Photonic Integrated Circuits","abstract":"Photonic integrated circuits have emerged as a powerful platform for high speed communication, sensing, and information processing due to their large bandwidth, low latency, and inherent parallelism. However, the absence of efficient, scalable, and non-volatile memory elements remains a fundamental limitation for realizing fully programmable and adaptive photonic systems. Conventional electronic memories introduce significant energy overhead, latency, and architectural inefficiencies due to repeated optical electrical conversions. Non volatile opto electronic resistive memories or OERMs have recently emerged as a promising solution to address these challenges by integrating memory functionality directly within the photonic domain. These devices combine resistive switching mechanisms with optical readout, enabling persistent state retention, multilevel programmability, and energy efficient operation. In this review, we provide a comprehensive overview of OERMs, spanning from fundamental physical mechanisms to system level applications. We first discuss the underlying resistive switching phenomena, including filamentary conduction, interface type switching, phase change transitions, and ionic migration, with particular emphasis on their interaction with confined optical modes. We then examine key material platforms such as metal oxides, transparent conducting oxides, phase change materials, and emerging two-dimensional systems, highlighting their performance trade-offs. Furthermore, we analyse device architectures and benchmark their performance in terms of switching energy, speed, endurance, and optical modulation efficiency. The integration of OERMs into programmable photonic circuits, neuromorphic systems, and in-memory optical computing architectures is critically discussed. Finally, we outline the major challenges and future research directions toward scalable, reliable","author":[{"family":"Kumar","given":"Santosh"},{"family":"Kumar","given":"Mukesh"},{"family":"Shin","given":"Eunso"},{"family":"Tossoun","given":"Bassem"},{"family":"Cheung","given":"Stanley"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2606.01463","URL":"https://doi.org/10.48550/arxiv.2606.01463","source":"datacite"},{"id":"doi:10.5281/zenodo.20094300","type":"article-journal","title":"Dynamic Latency Optimization for Edge-Based Machine Learning Models in 6G-Enabled Industrial Internet of Things (IIoT)","abstract":"Abstract The integration of 6G technology into the Industrial Internet of Things (IIoT) promises to redefine manufacturing through \"Hyper-Reliable Low-Latency Communication\" (HRLLC). However, the deployment of complex Machine Learning (ML) models at the edge remains constrained by the heterogeneous nature of industrial data and the limited computational resources of edge nodes. This article proposes a novel framework for Dynamic Latency Optimization (DLO) that leverages Deep Reinforcement Learning (DRL) for intelligent task offloading and resource allocation. By utilizing 6G's Terahertz (THz) spectrum and AI-native Network Slicing, the proposed framework dynamically adapts to fluctuating network conditions to maintain sub-millisecond latency. Our simulation results demonstrate a 42% reduction in end-to-end delay and a 30% improvement in energy efficiency compared to traditional 5G-MEC architectures. Furthermore, we explore the integration of Reconfigurable Intelligent Surfaces (RIS), Semantic Communication, and Zero-Trust Edge Security to further optimize the data-intelligence pipeline for Industry 5.0 applications, focusing on the critical synergy between human operators and autonomous systems within a resilient, sustainable, and cognitively aware industrial fabric. Keywords: 6G Networks, Industrial IoT (IIoT), Edge Intelligence, Deep Reinforcement Learning, Latency Optimization 1. Introduction: From Automation to Human-Centric Intelligence The transition from Industry 4.0 to Industry 5.0 marks a profound shift toward human-centric, resilient, and sustainable manufacturing systems. While Industry 4.0 was characterized by the digitalization of physical assets and the rise of cyber-physical systems, Industry 5.0 emphasizes the \"Tactile Internet\" and \"Human-Robot Co-evolution.\" In this new paradigm, the focus shifts from pure efficiency to the seamless collaboration between humans and increasingly autonomous machines. The \"Tactile Internet\" concept is particularly revolutionary, as it requires a \"haptic control loop\"—the ability to transmit touch and feel sensations over the network with such low latency that the human brain perceives no delay. This necessitates an end-to-end latency below 1ms, encompassing both the transmission and the computational processing of sensory feedback. This evolution necessitates a communication infrastructure capable of supporting advanced applications such as ultra-responsive autonomous mobile robots (AMRs), synchronized multi-robot assembly lines, and high-fidelity haptic feedback for remote maintenance in hazardous environments. For example, a specialist surgeon operating a robotic arm in a factory cleanup of toxic waste requires instantaneous haptic feedback to \"feel\" the resistance of the materials being handled. If the feedback loop exceeds 10ms, the mismatch between visual and tactile input can lead to \"operator sickness\" or mechanical errors that jeopardize safety. Furthermore, we must consider proprioceptive alignment—the sense of self-movement and body position. In 6G-enabled IIoT, the network must act as an extension of the human nervous system, where the delay jitter is so minimal that the robotic actuator feels like a literal extension of the operator's limb. This requires not just low latency, but Isochronous Communication, where packets arrive at precisely regular intervals to maintain the temporal rhythm of human motor-sensory systems. This synchronization is critical for Tele-Operation in nanomanufacturing, where even a micro-stutter in the feedback loop can cause the robotic probe to crush a microscopic wafer. The biological threshold for \"instantaneous\" feedback in human motor control is roughly 1-10ms for tactile sensations and less than 1ms for the suppression of \"visual-vestibular conflict.\" In 6G, we move into the regime of \"Sub-Perceptual Jitter,\" where the network variance is lower than the biological noise of the human nervous system. This enables \"Neuromorphic Manufacturi","author":[{"family":"Patil","given":"Seema"},{"family":"Doddamani","given":"Harshavardhana"},{"family":"Rivers","given":"Julianne"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20094300","URL":"https://doi.org/10.5281/zenodo.20094300","source":"datacite"},{"id":"doi:10.5281/zenodo.20094301","type":"article-journal","title":"Dynamic Latency Optimization for Edge-Based Machine Learning Models in 6G-Enabled Industrial Internet of Things (IIoT)","abstract":"Abstract The integration of 6G technology into the Industrial Internet of Things (IIoT) promises to redefine manufacturing through \"Hyper-Reliable Low-Latency Communication\" (HRLLC). However, the deployment of complex Machine Learning (ML) models at the edge remains constrained by the heterogeneous nature of industrial data and the limited computational resources of edge nodes. This article proposes a novel framework for Dynamic Latency Optimization (DLO) that leverages Deep Reinforcement Learning (DRL) for intelligent task offloading and resource allocation. By utilizing 6G's Terahertz (THz) spectrum and AI-native Network Slicing, the proposed framework dynamically adapts to fluctuating network conditions to maintain sub-millisecond latency. Our simulation results demonstrate a 42% reduction in end-to-end delay and a 30% improvement in energy efficiency compared to traditional 5G-MEC architectures. Furthermore, we explore the integration of Reconfigurable Intelligent Surfaces (RIS), Semantic Communication, and Zero-Trust Edge Security to further optimize the data-intelligence pipeline for Industry 5.0 applications, focusing on the critical synergy between human operators and autonomous systems within a resilient, sustainable, and cognitively aware industrial fabric. Keywords: 6G Networks, Industrial IoT (IIoT), Edge Intelligence, Deep Reinforcement Learning, Latency Optimization 1. Introduction: From Automation to Human-Centric Intelligence The transition from Industry 4.0 to Industry 5.0 marks a profound shift toward human-centric, resilient, and sustainable manufacturing systems. While Industry 4.0 was characterized by the digitalization of physical assets and the rise of cyber-physical systems, Industry 5.0 emphasizes the \"Tactile Internet\" and \"Human-Robot Co-evolution.\" In this new paradigm, the focus shifts from pure efficiency to the seamless collaboration between humans and increasingly autonomous machines. The \"Tactile Internet\" concept is particularly revolutionary, as it requires a \"haptic control loop\"—the ability to transmit touch and feel sensations over the network with such low latency that the human brain perceives no delay. This necessitates an end-to-end latency below 1ms, encompassing both the transmission and the computational processing of sensory feedback. This evolution necessitates a communication infrastructure capable of supporting advanced applications such as ultra-responsive autonomous mobile robots (AMRs), synchronized multi-robot assembly lines, and high-fidelity haptic feedback for remote maintenance in hazardous environments. For example, a specialist surgeon operating a robotic arm in a factory cleanup of toxic waste requires instantaneous haptic feedback to \"feel\" the resistance of the materials being handled. If the feedback loop exceeds 10ms, the mismatch between visual and tactile input can lead to \"operator sickness\" or mechanical errors that jeopardize safety. Furthermore, we must consider proprioceptive alignment—the sense of self-movement and body position. In 6G-enabled IIoT, the network must act as an extension of the human nervous system, where the delay jitter is so minimal that the robotic actuator feels like a literal extension of the operator's limb. This requires not just low latency, but Isochronous Communication, where packets arrive at precisely regular intervals to maintain the temporal rhythm of human motor-sensory systems. This synchronization is critical for Tele-Operation in nanomanufacturing, where even a micro-stutter in the feedback loop can cause the robotic probe to crush a microscopic wafer. The biological threshold for \"instantaneous\" feedback in human motor control is roughly 1-10ms for tactile sensations and less than 1ms for the suppression of \"visual-vestibular conflict.\" In 6G, we move into the regime of \"Sub-Perceptual Jitter,\" where the network variance is lower than the biological noise of the human nervous system. This enables \"Neuromorphic Manufacturi","author":[{"family":"Patil","given":"Seema"},{"family":"Doddamani","given":"Harshavardhana"},{"family":"Rivers","given":"Julianne"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20094301","URL":"https://doi.org/10.5281/zenodo.20094301","source":"datacite"},{"id":"doi:10.48550/arxiv.2605.31141","type":"manuscript","title":"Compact and Energy-Efficient Memristive Spiking Neuromorphic Accelerator for Bio-inspired Interception Tasks","abstract":"Spiking neural networks (SNNs) provide an efficient event-driven computing paradigm for bio-inspired interception tasks. However, most implementations rely on von Neumann digital computing platforms, where memory and computation bottlenecks limit energy efficiency. This work presents a compact and energy-efficient memristive neuromorphic accelerator for bio-inspired interception tasks. A novel one-transistor-one-resistor (1T1R) crossbar array is designed to emulate synaptic operations in the in-memory computing (IMC) domain, while circuit-level optimization mitigates membrane drift and improves integration fidelity. In addition, an integrate-and-fire (IF) neuron with separated input and membrane nodes is developed to improve inference robustness during array-interfaced operation. Implemented in the SkyWater SKY130 PDK, the proposed neuron achieves an energy consumption of 10.67 pJ/spike and an area of 906 um^2. System-level results show that the memristive IMC output closely matches the software SNN baseline, with a correlation coefficient of 0.9622, while achieving a 96% interception success rate. These results demonstrate the effectiveness of the proposed design for compact and reliable memristive SNN inference in bio-inspired interception tasks.","author":[{"family":"Qu","given":"Qianhou"},{"family":"Lu","given":"Sheng"},{"family":"Jung","given":"Sungyong"},{"family":"Liang","given":"Qilian"},{"family":"Pan","given":"Chenyun"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2605.31141","URL":"https://doi.org/10.48550/arxiv.2605.31141","source":"datacite"},{"id":"doi:10.48550/arxiv.2605.29719","type":"manuscript","title":"Constant Depth Threshold Circuits For Exhaustive Epistasis Detection","abstract":"The development of large-scale neuromorphic hardware has made practical implementations of threshold gate-based circuits a near-term possibility. The complexity advantages regarding traditional computing classes, as evidenced in the literature, have prompted us to tackle Epistasis Detection, one of the most computationally complex combinatorial problems in bioinformatics. We propose specially designed circuits that calculate the relative frequencies of all dataset combinations in an efficient pipelined fashion, taking advantage of co-located memory and configurable parallelism, obtaining complexity gains. Overall, we obtain the runtime to be bounded by the number of combinations to calculate, without any additional complexity overhead, contrary to classical approaches, using log-linear space. To accomplish this, we propose a data encoding and combination generation strategy using Leaky Integrate and Fire (LIF) neurons, that feeds a constant depth threshold gate population count circuit. Accounting for typical hardware characteristics, such as limited fan-in and variable precisions, we obtain logarithmic depth and log-cubic linear connections, for the population count circuit by composing developed unbounded fan-in constant depth threshold gate circuits to perform population count and binary array sum.","author":[{"family":"Ribeiro","given":"André"},{"family":"Ilic","given":"Aleksandar"},{"family":"Sousa","given":"Leonel"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2605.29719","URL":"https://doi.org/10.48550/arxiv.2605.29719","source":"datacite"},{"id":"doi:10.48550/arxiv.2605.27342","type":"manuscript","title":"Phase-Topology Classification of Memristor Hysteresis Loops via Self-Crossings","abstract":"Memristive devices have revolutionized non-volatile memory and neuromorphic computing, yet the geometry of their hysteresis loops -- in particular, the occurrence and robustness of multiple self-crossings -- remains poorly understood. Here we introduce a topological and algebraic framework that treats the number of transverse self-intersections of a memristor hysteresis loop as a robust integer-valued invariant. Drawing on differential topology, singularity theory, and cusp catastrophe, we employ discriminants and resultants to stratify the six-dimensional parameter space. This approach partitions the parameter space into structurally stable regions separated by explicitly computable catastrophe surfaces. We demonstrate that the crossing number remains strictly invariant under continuous deformations and changes only at self-tangencies or cusp singularities, thereby providing a complete classification of all multi-lobed hysteresis behaviors. These insights bridge device physics with modern singularity theory and suggest a clear roadmap for exploiting higher-order memory effects in next-generation electronics and brain-inspired hardware.","author":[{"family":"Lipan","given":"Ovidiu"},{"family":"Neuhaus","given":"Eric"},{"family":"Silva","given":"Rafael"},{"family":"Pradhan","given":"Soumen"},{"family":"Hartmann","given":"Fabian"},{"family":"Castelano","given":"Leonardo"},{"family":"Silva","given":"Ana"},{"family":"Höfling","given":"Sven"},{"family":"Lopez-Richard","given":"Victor"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2605.27342","URL":"https://doi.org/10.48550/arxiv.2605.27342","source":"datacite"},{"id":"doi:10.48550/arxiv.2605.20406","type":"manuscript","title":"High Performance TiO2 Ferroelectric Field Effect Transistors with HfZrO2 for Neuromorphic Computing","abstract":"TiO2 ferroelectric field effect transistors (FeFETs) with HfZrO2 (HZO) ferroelectric dielectric layers and bottom gate topology are fabricated for applications in neuromorphic systems. Two sets of devices are fabricated with different gate topologies by varying the thickness of the ferroelectric gate stack. Different device architectures are studied by varying the source drain length (LSD) and gate length (LG). The devices have high on/off ratios up to 10^7 with low leakage off currents &lt;10^-12 A. Repeated cycle testing shows high reliability and a stable memory window. The devices have large memory windows ranging from 3 to 8 V.","author":[{"family":"Samanta","given":"Chandan"},{"family":"Palmese","given":"Elia"},{"family":"Ouyang","given":"Ziyu"},{"family":"Zhama","given":"Tuofu"},{"family":"Pino","given":"Robinson"},{"family":"Zeng","given":"Yuping"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2605.20406","URL":"https://doi.org/10.48550/arxiv.2605.20406","source":"datacite"},{"id":"doi:10.48550/arxiv.2602.18110","type":"manuscript","title":"Cavity Solitons as a Nonlinear Substrate for Photonic Neuromorphic Computing","abstract":"Reservoir computing leverages nonlinear dynamics of physical systems to process temporal information with minimal training cost. Here, we demonstrate that cavity solitons sustained in a fiber optical cavity provide an optical platform for photonic reservoir computing. Our methodology exploits the use of a phase-modulated drive laser to encode the input, while the reservoir states are accessed through frequency-resolved readout. Numerical simulations indicate that the emission of Kelly waves enriches the dynamics and enhances performance for machine learning tasks. We evaluate the performance of the cavity-soliton reservoir computer on several standard benchmark tasks.","author":[{"family":"Arabieh","given":"Amir"},{"family":"Lupo","given":"Alessandro"},{"family":"Gorza","given":"Simon"},{"family":"Massar","given":"Serge"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2602.18110","URL":"https://doi.org/10.48550/arxiv.2602.18110","source":"datacite"},{"id":"doi:10.48550/arxiv.2605.20020","type":"manuscript","title":"Tunable magnetotransport through kinetically hindered first-order phase transitions in an antiferromagnetic metal","abstract":"Controllable multilevel resistance states are of interest for memory technologies like neuromorphic computing, but robust materials platforms toward such behavior remain limited. Here, we show that the non-centrosymmetric antiferromagnetic metal CeCoGe$_3$ suggests one such route through a kinetically hindered first-order magnetic transition. Cooling through the kinetically hindered first-order transition in an applied magnetic field produces a magnetic glass state in which high- and low-temperature magnetic phases coexist. The relative fraction of these phases can be controlled by the applied field in which the sample is cooled, and the electrical resistance is directly sensitive to that fraction. As a result, it is demonstrated that CeCoGe$_3$ supports stable multilevel resistive states. These results identify kinetically hindered first-order phase transitions as a promising route towards controllable multilevel magnetoresistive states.","author":[{"family":"Moya","given":"Jaime"},{"family":"Lee","given":"Scott"},{"family":"Chatterjee","given":"Sudipta"},{"family":"Mathur","given":"Nitish"},{"family":"Skorupskii","given":"Grigorii"},{"family":"Pollak","given":"Connor"},{"family":"Schoop","given":"Leslie"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2605.20020","URL":"https://doi.org/10.48550/arxiv.2605.20020","source":"datacite"},{"id":"doi:10.48448/qwnp-8z30","type":"article-journal","title":"I2E: Real-Time Image-to-Event Conversion for High-Performance Spiking Neural Networks","abstract":"Spiking neural networks (SNNs) promise highly energy-efficient computing, but their adoption is hindered by a critical scarcity of event-stream data. This work introduces I2E, an algorithmic framework that resolves this bottleneck by converting static images into high-fidelity event streams. By simulating microsaccadic eye movements with a highly parallelized convolution, I2E achieves a conversion speed over 300x faster than prior methods, uniquely enabling on-the-fly data augmentation for SNN training. The framework's effectiveness is demonstrated on large-scale benchmarks. An SNN trained on the generated I2E-ImageNet dataset achieves a state-of-the-art accuracy of 60.50%. Critically, this work establishes a powerful sim-to-real paradigm where pre-training on synthetic I2E data and fine-tuning on the real-world CIFAR10-DVS dataset yields an unprecedented accuracy of 92.5%. This result validates that synthetic event data can serve as a high-fidelity proxy for real sensor data, bridging a long-standing gap in neuromorphic engineering. By providing a scalable solution to the data problem, I2E offers a foundational toolkit for developing high-performance neuromorphic systems. The open-source algorithm and all generated datasets are provided to accelerate research in the field.","author":[{"family":"Hu","given":"Shaogang"},{"family":"Liu","given":"Yang"},{"family":"Ma","given":"Ruichen"},{"family":"Meng","given":"Liwei"},{"family":"Ning","given":"Ning"},{"family":"Qiao","given":"Guanchao"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48448/qwnp-8z30","URL":"https://doi.org/10.48448/qwnp-8z30","source":"datacite"},{"id":"doi:10.48448/37tz-k271","type":"article-journal","title":"Conductance Variation-Assisted Adversarial Attack Robustness on 40nm TaOX-based ReRAM CiM","abstract":"This work investigates the conductance variation-assisted enhancement of neural network (NN) robustness against adversarial attacks, exploiting the measurements of 32K bits 40nm TaOX-based analog Resistive Random Access Memory (ReRAM) devices. Proposed Computation-in-Memory (CiM)-aware adversarial training enhances NN robustness against both adversarial attacks and conductance variations by integrating the measured variations with training, improves classification accuracy of Fast Gradient Sign Method (FGSM)-attacked CIFAR-10 on ResNet-32 by 25% and that of Projected Gradient Descent (PGD)-attacked by 22%.","author":[{"family":"Awamura","given":"Satoshi"},{"family":"Matsui","given":"Chihiro"},{"family":"Misawa","given":"Naoko"},{"family":"Morimoto","given":"Masahiro"},{"family":"Takeuchi","given":"Ken"},{"family":"Yamauchi","given":"Kenshin"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48448/37tz-k271","URL":"https://doi.org/10.48448/37tz-k271","source":"datacite"},{"id":"doi:10.48550/arxiv.2605.07936","type":"manuscript","title":"A Fully Tunable Ultra-Low Power Current-Mode Memory Cell in Standard CMOS Technology","abstract":"This work introduces a fully tunable, ultra-low power unipolar memory cell inspired by the Schmitt-trigger comparator and designed in CMOS using only nine transistors. The proposed circuit operates entirely in the current domain and exploits a novel feedback configuration between two interdependent Heaviside-like thresholding elements to produce tunable bistable switching behavior. Its three key parameters-threshold current, hysteresis width, and output gain-are independently tunable via programmable bias currents, enabling flexibility across diverse analog computing applications. Unlike prior Schmitt-trigger designs, it simultaneously achieves current-mode operation, nanowatt-range power consumption, temperature stability, and full tunability, solely using standard MOSFET elements. Schematic-level simulations in a 180 nm CMOS process confirm robust hysteresis and resilience to device mismatch. Building on this circuit, we develop a complete family of spike-based logic gates using three-level current encoding, where the bistable memory retains the polarity of the last spike on each input indefinitely, enabling asynchronous logic operations without temporal windowing or refresh mechanisms. The same circuit also serves as the primitive for Bistable Memory Recurrent Units in analog neural networks, where the quantized hidden states provide inherent noise immunity. Together, these capabilities position the design as a versatile building block for next-generation neuromorphic processors integrating memory, logic, and recurrent computation.","author":[{"family":"Fyon","given":"Arthur"},{"family":"Mendolia","given":"Loris"},{"family":"Redouté","given":"Jean"},{"family":"Franci","given":"Alessio"},{"family":"Drion","given":"Guillaume"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2605.07936","URL":"https://doi.org/10.48550/arxiv.2605.07936","source":"datacite"},{"id":"doi:10.48550/arxiv.2509.09344","type":"manuscript","title":"Lifetime of bimerons and antibimerons in two-dimensional magnets","abstract":"Soliton-based computing architectures have recently emerged as a promising avenue to overcome fundamental limitations of conventional information technologies, the von Neumann bottleneck. In this context, magnetic skyrmions have been widely considered for in-situ processing devices due to their mobility and enhanced lifetime in materials with broken inversion symmetry. However, modern applications in non-volatile reservoir or neuromorphic computing raise the additional demand for non-linear inter-soliton interactions. Here, we report that solitons in easy-plane magnets, such as bimerons and antibimerons, show greater versatility and potential for non-linear interactions than skyrmions and antiskyrmions, making them superior candidates for this class of applications. Using first-principles and transition state theory, we predict the coexistence of degenerate bimerons and antibimerons at zero field in a van der Waals heterostructure Fe$_3$GeTe$_2$/Cr$_2$Ge$_2$Te$_6$ -- an experimentally feasible system. We demonstrate that, owing to their distinct structural symmetry, bimerons exhibit fundamentally different behavior from skyrmions and cannot be regarded as their in-plane counterparts, as is often assumed. This distinction leads to unique properties of bimerons and antibimerons, which arise from the unbroken rotational symmetry in easy-plane magnets. These range from anisotropic soliton-soliton interactions to strong entropic effects on their lifetime, driven by the non-local nature of thermal excitations. Our findings reveal a broader richness of solitons in easy-plane magnets and underline their unique potential for spintronic devices.","author":[{"family":"Goerzen","given":"Moritz"},{"family":"Drevelow","given":"Tim"},{"family":"Haldar","given":"Soumyajyoti"},{"family":"Schrautzer","given":"Hendrik"},{"family":"Heinze","given":"Stefan"},{"family":"Li","given":"Dongzhe"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2509.09344","URL":"https://doi.org/10.48550/arxiv.2509.09344","source":"datacite"},{"id":"doi:10.48550/arxiv.2508.18014","type":"manuscript","title":"A General Molecular-Scale Dynamic Memristor Model Based on Non-equilibrium Charge Transport Kinetics and Its Information Processing Capability in Reservoir Computing","abstract":"Non-equilibrium molecular-scale dynamics, where fast electron transport couples with slow chemical state evolution, underpins the complex behaviors of molecular memristors, yet a general model linking these dynamics to neuromorphic computing remains elusive. We introduce a dynamic memristor model that integrates Landauer and Marcus electron transport theories with the kinetics of slow processes, such as proton/ion migration or conformational changes. This framework reproduces experimental conductance hysteresis and emulates synaptic functions like short-term plasticity (STP) and spike-timing-dependent plasticity (STDP). By incorporating the model into a reservoir computing (RC) architecture, we show that computational performance optimizes when input frequency and bias mapping range align with the molecular system's intrinsic kinetics. This chemistry-centric, bottom-up approach provides a theoretical foundation for molecular-scale neuromorphic computing, demonstrating how non-equilibrium molecular-scale dynamics can drive information processing in the post-Moore era.","author":[{"family":"Chen","given":"Yueqi"},{"family":"Ji","given":"Xuan"},{"family":"Yu","given":"Xi"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2508.18014","URL":"https://doi.org/10.48550/arxiv.2508.18014","source":"datacite"},{"id":"doi:10.48550/arxiv.2605.16946","type":"manuscript","title":"Reprogrammable magnonic logic in a multiferroic heterostructure via magnetoelectric coupling","abstract":"The realization of fully reconfigurable, voltage-controlled, and programmable on-chip magnonic devices is essential to fully harness the potential of spin waves for signal processing, logic and neuromorphic computing. Yet, existing demonstrations of electrical tuning of magnonic responses are either volatile, current-driven and thus energy-inefficient, or rely on local strain modification limiting their scalability for wafer-scale integration. Here, we address this challenge using a BiFeO3/La0.67Sr0.33MnO3 multiferroic thin film heterostructure. We show that ferroelectric domain engineering in BiFeO3 enables deterministic tuning of the magnon dispersion of La0.67Sr0.33MnO3, producing frequency shifts up to $\\sim 150 MHz$ and allowing reconfigurable waveguiding. Micro-focused Brillouin light scattering directly images these effects, revealing electrically defined magnonic waveguides and spatially programmable dispersion. Compared to conventional approaches, this method provides non-volatile and reversible control. Furthermore, using an inverse-design simulation code, we demonstrate the capability of our platform to perform advanced magnonic functions such as frequency demultiplexing. Our results open a new avenue for using magnetoelectric heterostructures for magnonic logic, with further applicability to reservoir and neuromorphic computing and AI driven magnonic devices.","author":[{"family":"Che","given":"Ping"},{"family":"Abdelsamie","given":"Amr"},{"family":"Papp","given":"Ádám"},{"family":"Salama","given":"Sali"},{"family":"Thiaville","given":"André"},{"family":"Lebrun","given":"Romain"},{"family":"Fusil","given":"Stéphane"},{"family":"Garcia","given":"Vincent"},{"family":"Vecchiola","given":"Aymeric"},{"family":"Bouzehouane","given":"Karim"},{"family":"Bibes","given":"Manuel"},{"family":"Barthélémy","given":"Agnès"},{"family":"Adam","given":"Jean"},{"family":"Demidov","given":"Vladislav"},{"family":"Bortolotti","given":"Paolo"},{"family":"Anane","given":"Abdelmadjid"},{"family":"Boventer","given":"Isabella"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2605.16946","URL":"https://doi.org/10.48550/arxiv.2605.16946","source":"datacite"},{"id":"doi:10.60893/figshare.apl.c.8449429.v1","type":"article-journal","title":"<strong>Electric-Double-Layer Memcapacitive Reservoirs for <strong>Energy-Efficient</strong> Human-Action Processing</strong>","abstract":"Physical reservoir computing (PRC) exploits intrinsic material dynamics for energy-efficient neuromorphic hardware. However, conventional resistive PRC implementations suffer from high power consumption and hardware complexity owing to leakage currents and the requirement for peripheral readout circuitry. Here, we demonstrate an electric-double-layer (EDL) memcapacitive reservoir based on an Al/nanogranular SiO 2 /ITO sandwiched structure. The device enables direct open-circuit voltage readout and utilizes reversible ionic charge accumulation at the interface, enabling robust nonlinearity and short-term memory. The PRC system based on such a device achieves a normalized root mean square error (NRMSE) of 0.097 in the Mackey-Glass chaotic time-series prediction task. It also yields high human action recognition accuracies of 92.00% on the profile-based Weizmann dataset (10 classes) and 86.54% on the depth-based UTD-MHAD dataset (27 classes), respectively. This memcapacitive device shows a single-pulse energy consumption of 0.68 pJ and an operating power of ~27 pW in action recognition task, superior to most traditional current-readout reservoirs by over an order of magnitude. These results highlight a structurally simple and highly energy-efficient physical pathway for neuromorphic visual processing.","author":[{"family":"Wan","given":"Changjin"},{"family":"Xing","given":"Qianye"},{"family":"Pei","given":"Mengjiao"},{"family":"Cui","given":"Hangyuan"},{"family":"Wan","given":"Qing"},{"family":"Shi","given":"Kailu"}],"issued":{"date-parts":[[2026]]},"DOI":"10.60893/figshare.apl.c.8449429.v1","URL":"https://doi.org/10.60893/figshare.apl.c.8449429.v1","source":"datacite"},{"id":"doi:10.60893/figshare.apl.c.8449429","type":"article-journal","title":"<strong>Electric-Double-Layer Memcapacitive Reservoirs for <strong>Energy-Efficient</strong> Human-Action Processing</strong>","abstract":"Physical reservoir computing (PRC) exploits intrinsic material dynamics for energy-efficient neuromorphic hardware. However, conventional resistive PRC implementations suffer from high power consumption and hardware complexity owing to leakage currents and the requirement for peripheral readout circuitry. Here, we demonstrate an electric-double-layer (EDL) memcapacitive reservoir based on an Al/nanogranular SiO 2 /ITO sandwiched structure. The device enables direct open-circuit voltage readout and utilizes reversible ionic charge accumulation at the interface, enabling robust nonlinearity and short-term memory. The PRC system based on such a device achieves a normalized root mean square error (NRMSE) of 0.097 in the Mackey-Glass chaotic time-series prediction task. It also yields high human action recognition accuracies of 92.00% on the profile-based Weizmann dataset (10 classes) and 86.54% on the depth-based UTD-MHAD dataset (27 classes), respectively. This memcapacitive device shows a single-pulse energy consumption of 0.68 pJ and an operating power of ~27 pW in action recognition task, superior to most traditional current-readout reservoirs by over an order of magnitude. These results highlight a structurally simple and highly energy-efficient physical pathway for neuromorphic visual processing.","author":[{"family":"Wan","given":"Changjin"},{"family":"Xing","given":"Qianye"},{"family":"Pei","given":"Mengjiao"},{"family":"Cui","given":"Hangyuan"},{"family":"Wan","given":"Qing"},{"family":"Shi","given":"Kailu"}],"issued":{"date-parts":[[2026]]},"DOI":"10.60893/figshare.apl.c.8449429","URL":"https://doi.org/10.60893/figshare.apl.c.8449429","source":"datacite"},{"id":"doi:10.48550/arxiv.2509.17355","type":"manuscript","title":"CMOS Implementation of Field Programmable Spiking Neural Network for Hardware Reservoir Computing","abstract":"The increasing complexity and energy demands of large-scale neural networks, such as Deep Neural Networks (DNNs) and Large Language Models (LLMs), challenge their practical deployment in edge applications due to high power consumption, area requirements, and privacy concerns. Spiking Neural Networks (SNNs), particularly in analog implementations, offer a promising low-power alternative but suffer from noise sensitivity and connectivity limitations. This work presents a novel CMOS-implemented field-programmable neural network architecture for hardware reservoir computing. We propose a Leaky Integrate-and-Fire (LIF) neuron circuit with integrated voltage-controlled oscillators (VCOs) and programmable weighted interconnections via an on-chip FPGA framework, enabling arbitrary reservoir configurations. The system demonstrates effective implementation of the FORCE algorithm learning, linear and non-linear memory capacity benchmarks, and NARMA10 tasks, both in simulation and actual chip measurements. The neuron design achieves compact area utilization (around 540 NAND2-equivalent units) and low energy consumption (21.7 pJ/pulse) without requiring ADCs for information readout, making it ideal for system-on-chip integration of reservoir computing. This architecture paves the way for scalable, energy-efficient neuromorphic systems capable of performing real-time learning and inference with high configurability and digital interfacing.","author":[{"family":"Duran","given":"Ckristian"},{"family":"Kimura","given":"Nanako"},{"family":"Byambadorj","given":"Zolboo"},{"family":"Iizuka","given":"Tetsuya"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2509.17355","URL":"https://doi.org/10.48550/arxiv.2509.17355","source":"datacite"},{"id":"doi:10.5167/uzh-434124","type":"article-journal","title":"A benchmarking framework for embodied neuromorphic agents","abstract":"Enabling robots to swiftly, robustly and efficiently interact with a dynamic environment remains a key challenge. The robotic community can draw inspiration from the co-adaptation and synergistic interplay between animals’ brains and bodies, which underpins embodied intelligence. Soft robots and neuromorphic technology offer a natural solution for such a challenge, enabling low-power, material-based and event-driven sensorimotor processing and control that seamlessly handles the continuous dynamic demands of embodied agents. In this Perspective, we propose a comprehensive framework for benchmarking neuromorphic computing (brain) that control soft robots (body), based on a suite of tasks, essential metrics and a reproducible robotic platform. The goal is to allow researchers to evaluate their embodied neuromorphic system with a physical robot, in real-world scenarios. The robotic platform is accessible, open-source, modular and scalable, so task complexity can be gradually increased, fostering a standardized approach. By coupling metrics with physical implementations, this framework will drive progress in soft robotics, neuromorphic computing and embodied intelligence.","author":[{"family":"Dangelo","given":"Giulia"},{"family":"Pedersen","given":"Jens"},{"family":"Hassan","given":"Taimoor"},{"family":"Cianchetti","given":"Matteo"},{"family":"Bongard","given":"Josh"},{"family":"Iida","given":"Fumiya"},{"family":"Indiveri","given":"Giacomo"},{"family":"Hoffmann","given":"Matej"},{"family":"Laschi","given":"Cecilia"},{"family":"De Luca","given":"Chiara"},{"family":"Bartolozzi","given":"Chiara"},{"family":"Donati","given":"Elisa"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5167/uzh-434124","URL":"https://doi.org/10.5167/uzh-434124","source":"datacite"},{"id":"doi:10.48550/arxiv.2601.17991","type":"manuscript","title":"Prosthetic Hand Manipulation System Based on EMG and Eye Tracking Powered by the Neuromorphic Processor AltAi","abstract":"This paper presents a novel neuromorphic control architecture for upper-limb prostheses that combines surface electromyography (sEMG) with gaze-guided computer vision. The system uses a spiking neural network deployed on the neuromorphic processor AltAi to classify EMG patterns in real time while an eye-tracking headset and scene camera identify the object within the user's focus. In our prototype, the same EMG recognition model that was originally developed for a conventional GPU is deployed as a spiking network on AltAi, achieving comparable accuracy while operating in a sub-watt power regime, which enables a lightweight, wearable implementation. For six distinct functional gestures recorded from upper-limb amputees, the system achieves robust recognition performance comparable to state-of-the-art myoelectric interfaces. When the vision pipeline restricts the decision space to three context-appropriate gestures for the currently viewed object, recognition accuracy increases to roughly 95% while excluding unsafe, object-inappropriate grasps. These results indicate that the proposed neuromorphic, context-aware controller can provide energy-efficient and reliable prosthesis control and has the potential to improve safety and usability in everyday activities for people with upper-limb amputation.","author":[{"family":"Akinshin","given":"Roman"},{"family":"Lopatina","given":"Elizaveta"},{"family":"Bogatikov","given":"Kirill"},{"family":"Kiz","given":"Nikolai"},{"family":"Makarova","given":"Anna"},{"family":"Lebedev","given":"Mikhail"},{"family":"Cabrera","given":"Miguel"},{"family":"Tsetserukou","given":"Dzmitry"},{"family":"Kangler","given":"Valerii"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2601.17991","URL":"https://doi.org/10.48550/arxiv.2601.17991","source":"datacite"},{"id":"doi:10.5167/uzh-292307","type":"article-journal","title":"A Realistic Simulation Framework for Analog/Digital Neuromorphic Architectures","abstract":"Developing dedicated mixed-signal neuromorphic computing systems optimized for real-time sensory-processing in extreme edge-computing applications requires time-consuming design, fabrication, and deployment of full-custom neuromorphic processors. To ensure that initial prototyping efforts exploring the properties of different network architectures and parameter settings lead to realistic results, it is important to use simulation frameworks that match as best as possible the properties of the final hardware. This is particularly challenging for neuromorphic hardware platforms made using mixed-signal analog/digital circuits, due to the variability and noise sensitivity of their components. In this paper, we address this challenge by developing a software spiking neural network simulator explicitly designed to account for the properties of mixed-signal neuromorphic circuits, including device mismatch variability. The simulator, called A Realistic Simulation Framework for Analog/Digital Neuromorphic Architectures, is designed to reproduce the dynamics of mixed-signal synapse and neuron electronic circuits with autogradient differentiation for parameter optimization and GPU acceleration. We demonstrate the effectiveness of this approach by matching software simulation results with measurements made from an existing neuromorphic processor. We show how the results obtained provide a reliable estimate of the behavior of the spiking neural network trained in software, once deployed in hardware. This framework enables the development and innovation of new learning rules and processing architectures in neuromorphic embedded systems.","author":[{"family":"Quintana","given":"Fernando"},{"family":"Galindo","given":"Pedro"},{"family":"Donati","given":"Elisa"},{"family":"Indiveri","given":"Giacomo"},{"family":"Perez-Peña","given":"Fernando"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5167/uzh-292307","URL":"https://doi.org/10.5167/uzh-292307","source":"datacite"},{"id":"doi:10.48550/arxiv.2509.20284","type":"manuscript","title":"Biologically Plausible Learning via Bidirectional Spike-Based Distillation","abstract":"Developing biologically plausible learning algorithms that can achieve performance comparable to error backpropagation remains a longstanding challenge. Existing approaches often compromise biological plausibility by entirely avoiding the use of spikes for error propagation or relying on both positive and negative learning signals, while the question of how spikes can represent negative values remains unresolved. To address these limitations, we introduce Bidirectional Spike-based Distillation (BSD), a novel learning algorithm that jointly trains a feedforward and a backward spiking network. We formulate learning as a transformation between two spiking representations (i.e., stimulus encoding and concept encoding) so that the feedforward network implements perception and decision-making by mapping stimuli to actions, while the backward network supports memory recall by reconstructing stimuli from concept representations. Extensive experiments on diverse benchmarks, including image recognition, image generation, and sequential regression, show that BSD achieves performance comparable to networks trained with classical error backpropagation. These findings represent a significant step toward biologically grounded, spike-driven learning in neural networks. Our code is available at https://github.com/alden199/Bidirectional-Spike-Based-Distillation.","author":[{"family":"Lv","given":"Changze"},{"family":"Wang","given":"Yifei"},{"family":"Zhang","given":"Yanxun"},{"family":"Lu","given":"Yiyang"},{"family":"Xu","given":"Jingwen"},{"family":"Wang","given":"Xiaohua"},{"family":"Yu","given":"Di"},{"family":"Du","given":"Xin"},{"family":"Huang","given":"Xuanjing"},{"family":"Zheng","given":"Xiaoqing"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2509.20284","URL":"https://doi.org/10.48550/arxiv.2509.20284","source":"datacite"},{"id":"doi:10.48550/arxiv.2502.06747","type":"manuscript","title":"Wandering around: A bioinspired approach to visual attention through object motion sensitivity","abstract":"Active vision enables dynamic visual perception, offering an alternative to static feedforward architectures in computer vision, which rely on large datasets and high computational resources. Biological selective attention mechanisms allow agents to focus on salient Regions of Interest (ROIs), reducing computational demand while maintaining real-time responsiveness. Event-based cameras, inspired by the mammalian retina, enhance this capability by capturing asynchronous scene changes enabling efficient low-latency processing. To distinguish moving objects while the event-based camera is in motion the agent requires an object motion segmentation mechanism to accurately detect targets and center them in the visual field (fovea). Integrating event-based sensors with neuromorphic algorithms represents a paradigm shift, using Spiking Neural Networks to parallelize computation and adapt to dynamic environments. This work presents a Spiking Convolutional Neural Network bioinspired attention system for selective attention through object motion sensitivity. The system generates events via fixational eye movements using a Dynamic Vision Sensor integrated into the Speck neuromorphic hardware, mounted on a Pan-Tilt unit, to identify the ROI and saccade toward it. The system, characterized using ideal gratings and benchmarked against the Event Camera Motion Segmentation Dataset, reaches a mean IoU of 82.2% and a mean SSIM of 96% in multi-object motion segmentation. The detection of salient objects reaches 88.8% accuracy in office scenarios and 89.8% in low-light conditions on the Event-Assisted Low-Light Video Object Segmentation Dataset. A real-time demonstrator shows the system's 0.12 s response to dynamic scenes. Its learning-free design ensures robustness across perceptual scenes, making it a reliable foundation for real-time robotic applications serving as a basis for more complex architectures.","author":[{"family":"D'angelo","given":"Giulia"},{"family":"Clerico","given":"Victoria"},{"family":"Bartolozzi","given":"Chiara"},{"family":"Hoffmann","given":"Matej"},{"family":"Furlong","given":"PM"},{"family":"Hadjiivanov","given":"Alexander"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2502.06747","URL":"https://doi.org/10.48550/arxiv.2502.06747","source":"datacite"},{"id":"doi:10.48550/arxiv.2602.08726","type":"manuscript","title":"SynSacc: A Blender-to-V2E Pipeline for Synthetic Neuromorphic Eye-Movement Data and Sim-to-Real Spiking Model Training","abstract":"The study of eye movements, particularly saccades and fixations, are fundamental to understanding the mechanisms of human cognition and perception. Accurate classification of these movements requires sensing technologies capable of capturing rapid dynamics without distortion. Event cameras, also known as Dynamic Vision Sensors (DVS), provide asynchronous recordings of changes in light intensity, thereby eliminating motion blur inherent in conventional frame-based cameras and offering superior temporal resolution and data efficiency. In this study, we introduce a synthetic dataset generated with Blender to simulate saccades and fixations under controlled conditions. Leveraging Spiking Neural Networks (SNNs), we evaluate its robustness by training two architectures and finetuning on real event data. The proposed models achieve up to 0.83 accuracy and maintain consistent performance across varying temporal resolutions, demonstrating stability in eye movement classification. Moreover, the use of SNNs with synthetic event streams yields substantial computational efficiency gains over artificial neural network (ANN) counterparts, underscoring the utility of synthetic data augmentation in advancing event-based vision. All code and datasets associated with this work is available at https: //github.com/Ikhadija-5/SynSacc-Dataset.","author":[{"family":"Iddrisu","given":"Khadija"},{"family":"Shariff","given":"Waseem"},{"family":"Little","given":"Suzanne"},{"family":"Oconnor","given":"Noel"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2602.08726","URL":"https://doi.org/10.48550/arxiv.2602.08726","source":"datacite"},{"id":"doi:10.5167/uzh-290742","type":"article-journal","title":"Network models incorporating chloride dynamics predict optimal strategies for terminating status epilepticus","abstract":"Status epilepticus (SE), seizures lasting beyond five minutes, is a medical emergency commonly treated with benzodiazepines which enhance GABA receptor (GABAR) conductance. Despite widespread use, benzodiazepines fail in over one-third of patients, potentially due to seizure-induced disruption of neuronal chloride (Cl) homeostasis. Understanding these changes at a network level is crucial for improving clinical translation. Here, we address this using a large-scale spiking neural network model incorporating Cl dynamics, informed by clinical EEG and experimental slice recordings. Our simulations confirm that the GABAR reversal potential (E) dictates the pro- or anti-seizure effect of GABAR conductance modulation, with high E rendering benzodiazepines ineffective or excitatory. We show SE-like activity and E depend non-linearly on Cl extrusion efficacy and GABAR conductance. Critically, cell-type specific manipulations reveal that pyramidal cell, not interneuron, Cl extrusion predominantly determines the severity of SE activity and the response to simulated benzodiazepines. Leveraging these mechanistic insights, we develop a predictive framework mapping network states to Cl extrusion capacity and GABAergic load, yielding a proposed decision-making strategy to guide therapeutic interventions based on initial treatment response. This work identifies pyramidal cell Cl handling as a key therapeutic target and demonstrates the utility of biophysically detailed network models for optimising SE treatment protocols.","author":[{"family":"Currin","given":"Christopher"},{"family":"Burman","given":"Richard"},{"family":"Fedele","given":"Tommaso"},{"family":"Ramantani","given":"Georgia"},{"family":"Rosch","given":"Richard"},{"family":"Sprekeler","given":"Henning"},{"family":"Raimondo","given":"Joseph"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5167/uzh-290742","URL":"https://doi.org/10.5167/uzh-290742","source":"datacite"},{"id":"doi:10.5167/uzh-284397","type":"article-journal","title":"Stable recurrent dynamics in heterogeneous neuromorphic computing systems using excitatory and inhibitory plasticity","abstract":"Many neural computations emerge from self-sustained patterns of activity in recurrent neural circuits, which rely on balanced excitation and inhibition. Neuromorphic electronic circuits represent a promising approach for implementing the brain's computational primitives. However, achieving the same robustness of biological networks in neuromorphic systems remains a challenge due to the variability in their analog components. Inspired by real cortical networks, we apply a biologically-plausible cross-homeostatic rule to balance neuromorphic implementations of spiking recurrent networks. We demonstrate how this rule can autonomously tune the network to produce robust, self-sustained dynamics in an inhibition-stabilized regime, even in presence of device mismatch. It can implement multiple, co-existing stable memories, with emergent soft-winner-take-all and reproduce the \"paradoxical effect\" observed in cortical circuits. In addition to validating neuroscience models on a substrate sharing many similar limitations with biological systems, this enables the automatic configuration of ultra-low power, mixed-signal neuromorphic technologies despite the large chip-to-chip variability.","author":[{"family":"Soldado-Magraner","given":"Saray"},{"family":"Sorbaro","given":"Martino"},{"family":"Laje","given":"Rodrigo"},{"family":"Buonomano","given":"Dean"},{"family":"Indiveri","given":"Giacomo"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5167/uzh-284397","URL":"https://doi.org/10.5167/uzh-284397","source":"datacite"},{"id":"doi:10.48550/arxiv.2601.08248","type":"manuscript","title":"Spiking Neural-Invariant Kalman Fusion for Accurate Localization Using Low-Cost IMUs","abstract":"Low-cost inertial measurement units (IMUs) are widely utilized in mobile robot localization due to their affordability and ease of integration. However, their complex, nonlinear, and time-varying noise characteristics often lead to significant degradation in localization accuracy when applied directly for dead reckoning. To overcome this limitation, we propose a novel brain-inspired state estimation framework that combines a spiking neural network (SNN) with an invariant extended Kalman filter (InEKF). The SNN is designed to extract motion-related features from long sequences of IMU data affected by substantial random noise and is trained via a surrogate gradient descent algorithm to enable dynamic adaptation of the covariance noise parameter within the InEKF. By fusing the SNN output with raw IMU measurements, the proposed method enhances the robustness and accuracy of pose estimation. Extensive experiments conducted on the KITTI dataset and real-world data collected using a mobile robot equipped with a low-cost IMU demonstrate that the proposed approach outperforms state-of-the-art methods in localization accuracy and exhibits strong robustness to sensor noise, highlighting its potential for real-world mobile robot applications.","author":[{"family":"Liu","given":"Yaohua"},{"family":"Xu","given":"Qiao"},{"family":"Ou","given":"Binkai"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2601.08248","URL":"https://doi.org/10.48550/arxiv.2601.08248","source":"datacite"},{"id":"doi:10.48550/arxiv.2602.01874","type":"manuscript","title":"Spin splitting torque enabled artificial neuron with self-reset via synthetic antiferromagnetic coupling","abstract":"Spintronic artificial neurons are intriguing building blocks for energy efficient Neuromorphic Computing (NC). Nevertheless, most contemporary implementations rely on symmetry breaking external in plane magnetic fields (H_X) for neuron operation, which limits scalability and hardware practicality. We experimentally demonstrate an altermagnet/Synthetic Antiferromagnetic Coupling (SAF) based spintronic neuron that uses out of plane spin (σ_Z) polarized spin-splitting torque to eliminate the necessity of an external H_X. The neuron device also features intrinsic self-reset function facilitated by built-in exchange coupling. Furthermore, the proposed device is validated for Spiking Neural Network (SNN) applications by achieving test accuracies of 95.99% and 94.36% on the MNIST and N-MNIST datasets, respectively. These results demonstrate the hardware feasibility and compatibility of the proposed spintronic neuron, highlighting its potential for compact, scalable and energy-efficient neuromorphic computing systems.","author":[{"family":"Sekh","given":"Badsha"},{"family":"Rahaman","given":"Hasibur"},{"family":"Verma","given":"Ravi"},{"family":"Maddu","given":"Ramu"},{"family":"Jawahar","given":"Kesavan"},{"family":"Piramanayagam","given":"SN"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2602.01874","URL":"https://doi.org/10.48550/arxiv.2602.01874","source":"datacite"},{"id":"doi:10.48550/arxiv.2602.01133","type":"manuscript","title":"Parallel Training in Spiking Neural Networks","abstract":"The bio-inspired integrate-fire-reset mechanism of spiking neurons constitutes the foundation for efficient processing in Spiking Neural Networks (SNNs). Recent progress in large models demands that spiking neurons support highly parallel computation to scale efficiently on modern GPUs. This work proposes a novel functional perspective that provides general guidance for designing parallel spiking neurons. We argue that the reset mechanism, which induces complex temporal dependencies and hinders parallel training, should be removed. However, any such modification should satisfy two principles: 1) preserving the functions of reset as a core biological mechanism; and 2) enabling parallel training without sacrificing the serial inference ability of spiking neurons, which underpins their efficiency at test time. To this end, we identify the functions of the reset and analyze how to reconcile parallel training with serial inference, upon which we propose a dynamic decay spiking neuron. We conduct comprehensive testing of our method in terms of: 1) Training efficiency and extrapolation capability. On 16k-length sequences, we achieve a 25.6x training speedup over the pioneering parallel spiking neuron, and our models trained on 2k-length can stably perform inference on sequences as long as 30k. 2) Generality. We demonstrate the consistent effectiveness of the proposed method across five task categories (image classification, neuromorphic event processing, time-series forecasting, language modeling, and reinforcement learning), three network architectures (spiking CNN/Transformer/SSMs), and two spike activation modes (spike/integer activation). 3) Energy consumption. The spiking firing of our neuron is lower than that of vanilla and existing parallel spiking neurons.","author":[{"family":"Huang","given":"Yanbin"},{"family":"Yao","given":"Man"},{"family":"Pan","given":"Yuqi"},{"family":"Lv","given":"Changze"},{"family":"Xu","given":"Siyuan"},{"family":"Zheng","given":"Xiaoqing"},{"family":"Xu","given":"Bo"},{"family":"Li","given":"Guoqi"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2602.01133","URL":"https://doi.org/10.48550/arxiv.2602.01133","source":"datacite"},{"id":"doi:10.5167/uzh-284155","type":"article-journal","title":"Recurrent models of orientation selectivity enable robust early-vision processing in mixed-signal neuromorphic hardware","abstract":"Mixed signal analog/digital neuromorphic circuits represent an ideal medium for reproducing bio-physically realistic dynamics of biological neural systems in real-time. However, similar to their biological counterparts, these circuits have limited resolution and are affected by a high degree of variability. By developing a recurrent spiking neural network model of the retinocortical visual pathway, we show how such noisy and heterogeneous computing substrate can produce linear receptive fields tuned to visual stimuli with specific orientations and spatial frequencies. Compared to strictly feed-forward schemes, the model generates highly structured Gabor-like receptive fields of any phase symmetry, making optimal use of the hardware resources available in terms of synaptic connections and neuron numbers. Experimental results validate the approach, demonstrating how principles of neural computation can lead to robust sensory processing electronic systems, even when they are affected by high degree of heterogeneity, e.g., due to the use of analog circuits or memristive devices.","author":[{"family":"Baruzzi","given":"Valentina"},{"family":"Indiveri","given":"Giacomo"},{"family":"Sabatini","given":"Silvio"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5167/uzh-284155","URL":"https://doi.org/10.5167/uzh-284155","source":"datacite"},{"id":"doi:10.5167/uzh-284160","type":"article-journal","title":"Neuromorphic dreaming as a pathway to efficient learning in artificial agents","abstract":"The computational substrate of biological systems exhibits remarkable abilities to learn complex skills quickly and efficiently. Inspired by this, we implement model-based reinforcement learning using spiking neural networks directly on mixed-signal neuromorphic hardware. This approach combines energy-efficient electronic circuits with high sample efficiency through alternating online (‘awake’) and offline (‘dreaming’) learning phases. Our model features two networks: an agent network that learns from real and simulated experiences and a world model network that generates simulated experiences. We validate this by training the system to play Atari Pong. First, we establish a baseline using only real experiences. Then, by ‘dreaming’, the required real experiences decrease significantly. The network dynamics runs in real-time on the analog neuromorphic circuits, with only the readout layers implemented and trained on a computer-in-the-loop. We present results that demonstrate the robustness and potential of energy-efficient mixed-signal neuromorphic processors for real-world applications.","author":[{"family":"Blakowski","given":"Ingo"},{"family":"Zendrikov","given":"Dmitrii"},{"family":"Indiveri","given":"Giacomo"},{"family":"Capone","given":"Cristiano"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5167/uzh-284160","URL":"https://doi.org/10.5167/uzh-284160","source":"datacite"},{"id":"doi:10.5167/uzh-284152","type":"article-journal","title":"Event driven neural network on a mixed signal neuromorphic processor for EEG based epileptic seizure detection","abstract":"Long-term monitoring of biomedical signals is essential for the modern clinical management of neurological conditions such as epilepsy. However, developing wearable systems that are able to monitor, analyze, and detect epileptic seizures with long-lasting operation times using current technologies is still an open challenge. Brain-inspired spiking neural networks (SNNs) represent a promising signal processing and computing framework as they can be deployed on ultra-low power neuromorphic computing systems, for this purpose. Here, we introduce a novel SNN architecture, co-designed and validated on a mixed-signal neuromorphic chip, that shows potential for always-on monitoring of epileptic activity. We demonstrate how the hardware implementation of this SNN captures the phenomenon of partial synchronization within neural activity during seizure periods. We assess the network using a full-custom asynchronous mixed-signal neuromorphic platform, processing analog signals in real-time from an Electroencephalographic (EEG) seizure dataset. The neuromorphic chip comprises an analog front-end (AFE) signal conditioning stage and an asynchronous delta modulation (ADM) circuit directly integrated on the same die, which can produce the stream of spikes as input to the SNN, directly from the analog EEG signals. We show a linear classifier in a post processing stage that is sufficient to reliably classify and detect seizures, from the local features extracted by the SNN, indicating the feasibility of full on-chip seizure monitoring in the future. This research marks a significant advancement toward developing embedded intelligent \"wear and forget\" units for resource-constrained environments. These units could autonomously detect and log relevant EEG events of interest in out-of-hospital environments, offering new possibilities for patient care and management of neurological disorders.","author":[{"family":"Bartels","given":"Jim"},{"family":"Gallou","given":"Olympia"},{"family":"Ito","given":"Hiroyuki"},{"family":"Cook","given":"Matthew"},{"family":"Sarnthein","given":"Johannes"},{"family":"Indiveri","given":"Giacomo"},{"family":"Ghosh","given":"Saptarshi"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5167/uzh-284152","URL":"https://doi.org/10.5167/uzh-284152","source":"datacite"},{"id":"doi:10.48550/arxiv.2601.21222","type":"manuscript","title":"FireFly-P: FPGA-Accelerated Spiking Neural Network Plasticity for Robust Adaptive Control","abstract":"Spiking Neural Networks (SNNs) offer a biologically plausible learning mechanism through synaptic plasticity, enabling unsupervised adaptation without the computational overhead of backpropagation. To harness this capability for robotics, this paper presents FireFly-P, an FPGA-based hardware accelerator that implements a novel plasticity algorithm for real-time adaptive control. By leveraging on-chip plasticity, our architecture enhances the network's generalization, ensuring robust performance in dynamic and unstructured environments. The hardware design achieves an end-to-end latency of just 8~$μ$s for both inference and plasticity updates, enabling rapid adaptation to unseen scenarios. Implemented on a tiny Cmod A7-35T FPGA, FireFly-P consumes only 0.713~W and $\\sim$10K~LUTs, making it ideal for power- and resource-constrained embedded robotic platforms. This work demonstrates that hardware-accelerated SNN plasticity is a viable path toward enabling adaptive, low-latency, and energy-efficient control systems.","author":[{"family":"Li","given":"Tenglong"},{"family":"Li","given":"Jindong"},{"family":"Shen","given":"Guobin"},{"family":"Zhao","given":"Dongcheng"},{"family":"Zhang","given":"Qian"},{"family":"Zeng","given":"Yi"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2601.21222","URL":"https://doi.org/10.48550/arxiv.2601.21222","source":"datacite"},{"id":"doi:10.48550/arxiv.2511.04109","type":"manuscript","title":"CBMC-V3: A CNS-inspired Control Framework Towards Agile Manipulation with SNN","abstract":"As robotic arm applications expand beyond traditional industrial settings into service-oriented domains such as catering, household and retail, existing control algorithms struggle to achieve the level of agile manipulation required in unstructured environments characterized by dynamic trajectories, unpredictable interactions, and diverse objects. This paper presents a biomimetic control framework based on Spiking Neural Network (SNN), inspired by the human Central Nervous System (CNS), to address these challenges. The proposed framework comprises five control modules-cerebral cortex, cerebellum, thalamus, brainstem, and spinal cord-organized into three hierarchical control levels (first-order, second-order, and third-order) and two information pathways (ascending and descending). All modules are fully implemented using SNN. The framework is validated through both simulation and experiments on a commercial robotic arm platform across a range of control tasks. The results demonstrate that the proposed method outperforms the baseline in terms of agile motion control capability, offering a practical and effective solution for achieving agile manipulation.","author":[{"family":"Pang","given":"Yanbo"},{"family":"Li","given":"Qingkai"},{"family":"Zhao","given":"Mingguo"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2511.04109","URL":"https://doi.org/10.48550/arxiv.2511.04109","source":"datacite"},{"id":"doi:10.3929/ethz-b-000739475","type":"article-journal","title":"A neuromorphic electronic artist for robotic painting","abstract":"Recent advances in deep learning have sparked interest in AI-generated art, including robot-assisted painting. Traditional painting machines use static images and offline processing without considering the dynamic nature of painting. Neuromorphic cameras, which capture light intensity changes through asynchronous events, and mixed-signal neuromorphic processors, which implement biologically plausible spiking neural networks, offer a promising alternative. In this work, we present a robotic painting system comprising a 6-DOF robotic arm, event-based input from a Dynamic Vision Sensor (DVS) camera and a neuromorphic processor to produce dynamic brushstrokes, and tactile feedback from a force-torque sensor to compensate for brush deformation. The system receives DVS events representing the desired brushstroke trajectory and maps these events onto the processor’s neurons to compute joint velocities in close-loop. The variability in the input’s noisy event streams and the processor’s analog circuits reproduces the heterogeneity of human brushstrokes. Tested in a real-world setting, the system successfully generated diverse physical brushstrokes. This network marks a first step towards a fully spiking robotic controller with ultra-low latency responsiveness, applicable to any robotic task requiring real-time closed-loop adaptive control.","author":[{"family":"Schürmann","given":"Lioba"},{"family":"D'angelo","given":"Giulia"},{"family":"Grayver","given":"Liat"},{"family":"Bartolozzi","given":"Chiara"},{"family":"Indiveri","given":"Giacomo"}],"issued":{"date-parts":[[2025]]},"DOI":"10.3929/ethz-b-000739475","URL":"https://doi.org/10.3929/ethz-b-000739475","source":"datacite"},{"id":"doi:10.3929/ethz-b-000744772","type":"article-journal","title":"Stable recurrent dynamics in heterogeneous neuromorphic computing systems using excitatory and inhibitory plasticity","abstract":"Many neural computations emerge from self-sustained patterns of activity in recurrent neural circuits, which rely on balanced excitation and inhibition. Neuromorphic electronic circuits represent a promising approach for implementing the brain’s computational primitives. However, achieving the same robustness of biological networks in neuromorphic systems remains a challenge due to the variability in their analog components. Inspired by real cortical networks, we apply a biologically-plausible cross-homeostatic rule to balance neuromorphic implementations of spiking recurrent networks. We demonstrate how this rule can autonomously tune the network to produce robust, self-sustained dynamics in an inhibition-stabilized regime, even in presence of device mismatch. It can implement multiple, co-existing stable memories, with emergent soft-winner-take-all and reproduce the “paradoxical effect” observed in cortical circuits. In addition to validating neuroscience models on a substrate sharing many similar limitations with biological systems, this enables the automatic configuration of ultra-low power, mixed-signal neuromorphic technologies despite the large chip-to-chip variability.","author":[{"family":"Soldado-Magraner","given":"Saray"},{"family":"Sorbaro","given":"Martino"},{"family":"Laje","given":"Rodrigo"},{"family":"Buonomano","given":"Dean"},{"family":"Indiveri","given":"Giacomo"}],"issued":{"date-parts":[[2025]]},"DOI":"10.3929/ethz-b-000744772","URL":"https://doi.org/10.3929/ethz-b-000744772","source":"datacite"},{"id":"doi:10.48550/arxiv.2601.08244","type":"manuscript","title":"A brain-inspired information fusion method for enhancing robot GPS outages navigation","abstract":"Low-cost inertial navigation systems (INS) are prone to sensor biases and measurement noise, which lead to rapid degradation of navigation accuracy during global positioning system (GPS) outages. To address this challenge and improve positioning continuity in GPS-denied environments, this paper proposes a brain-inspired GPS/INS fusion network (BGFN) based on spiking neural networks (SNNs). The BGFN architecture integrates a spiking Transformer with a spiking encoder to simultaneously extract spatial features from inertial measurement unit (IMU) signals and capture their temporal dynamics. By modeling the relationship between vehicle attitude, specific force, angular rate, and GPS-derived position increments, the network leverages both current and historical IMU data to estimate vehicle motion. The effectiveness of the proposed method is evaluated through real-world field tests and experiments on public datasets. Compared to conventional deep learning approaches, the results demonstrate that BGFN achieves higher accuracy and enhanced reliability in navigation performance, particularly under prolonged GPS outages.","author":[{"family":"Liu","given":"Yaohua"},{"family":"Zhang","given":"Hengjun"},{"family":"Ou","given":"Binkai"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2601.08244","URL":"https://doi.org/10.48550/arxiv.2601.08244","source":"datacite"},{"id":"doi:10.48550/arxiv.2601.06637","type":"manuscript","title":"Efficient Aspect Term Extraction using Spiking Neural Network","abstract":"Aspect Term Extraction (ATE) identifies aspect terms in review sentences, a key subtask of sentiment analysis. While most existing approaches use energy-intensive deep neural networks (DNNs) for ATE as sequence labeling, this paper proposes a more energy-efficient alternative using Spiking Neural Networks (SNNs). Using sparse activations and event-driven inferences, SNNs capture temporal dependencies between words, making them suitable for ATE. The proposed architecture, SpikeATE, employs ternary spiking neurons and direct spike training fine-tuned with pseudo-gradients. Evaluated on four benchmark SemEval datasets, SpikeATE achieves performance comparable to state-of-the-art DNNs with significantly lower energy consumption. This highlights the use of SNNs as a practical and sustainable choice for ATE tasks.","author":[{"family":"Mishra","given":"Abhishek"},{"family":"Somasundaram","given":"Arya"},{"family":"Das","given":"Anup"},{"family":"Kandasamy","given":"Nagarajan"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2601.06637","URL":"https://doi.org/10.48550/arxiv.2601.06637","source":"datacite"},{"id":"doi:10.48550/arxiv.2601.04476","type":"manuscript","title":"Memory-Guided Unified Hardware Accelerator for Mixed-Precision Scientific Computing","abstract":"Recent hardware acceleration advances have enabled powerful specialized accelerators for finite element computations, spiking neural network inference, and sparse tensor operations. However, existing approaches face fundamental limitations: (1) finite element methods lack comprehensive rounding error analysis for reduced-precision implementations and use fixed precision assignment strategies that cannot adapt to varying numerical conditioning; (2) spiking neural network accelerators cannot handle non-spike operations and suffer from bit-width escalation as network depth increases; and (3) FPGA tensor accelerators optimize only for dense computations while requiring manual configuration for each sparsity pattern. To address these challenges, we introduce \\textbf{Memory-Guided Unified Hardware Accelerator for Mixed-Precision Scientific Computing}, a novel framework that integrates three enhanced modules with memory-guided adaptation for efficient mixed-workload processing on unified platforms. Our approach employs memory-guided precision selection to overcome fixed precision limitations, integrates experience-driven bit-width management and dynamic parallelism adaptation for enhanced spiking neural network acceleration, and introduces curriculum learning for automatic sparsity pattern discovery. Extensive experiments on FEniCS, COMSOL, ANSYS benchmarks, MNIST, CIFAR-10, CIFAR-100, DVS-Gesture datasets, and COCO 2017 demonstrate 2.8\\% improvement in numerical accuracy, 47\\% throughput increase, 34\\% energy reduction, and 45-65\\% throughput improvement compared to specialized accelerators. Our work enables unified processing of finite element methods, spiking neural networks, and sparse computations on a single platform while eliminating data transfer overhead between separate units.","author":[{"family":"Wang","given":"Chuanzhen"},{"family":"Zhang","given":"Leo"},{"family":"Liu","given":"Eric"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2601.04476","URL":"https://doi.org/10.48550/arxiv.2601.04476","source":"datacite"},{"id":"doi:10.48550/arxiv.2509.25453","type":"manuscript","title":"Neural Receptive Fields, Stimulus Space Embedding and Effective Geometry of Scale-Free Networks","abstract":"Understanding how receptive fields emerge and organize within brain networks and how neural dynamics couple with stimuli space is fundamental to neuroscience. Models often rely on fine-tuning connectivity to match empirical data, which may limit biological plausibility. Here we propose a physiologically grounded alternative where receptive fields and population-level attractor dynamics arise naturally from the effective hyperbolic geometry of scale-free networks. By associating stimulus space with the boundary of a hyperbolic embedding, we simulate neural dynamics using rate-based and spiking models, revealing localized activity patterns that reflect stimulus space structure without synaptic fine-tuning. The resulting receptive fields follow experimentally observed statistics and properties, and their sizes depends on neuron's connectivity degree. The model generalizes across stimuli dimensionalities and various modalities, such as orientation and place selectivity. Experimental analyses of hippocampal place fields recorded on a linear track support these findings. This framework offers a novel organizing principle linking network structure, stimulus space encoding, and neural dynamics, providing insights into receptive field formation across diverse brain areas.","author":[{"family":"Tiselko","given":"Vasilii"},{"family":"Gorsky","given":"Alexander"},{"family":"Dabaghian","given":"Yuri"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2509.25453","URL":"https://doi.org/10.48550/arxiv.2509.25453","source":"datacite"},{"id":"doi:10.34734/fzj-2025-05547","type":"article-journal","title":"Resolving inconsistent effects of tDCS on learning using a homeostatic structural plasticity model","abstract":"Introduction: Transcranial direct current stimulation (tDCS) is increasingly used to modulate motor learning. Current polarity and intensity, electrode montage, and application before or during learning had mixed effects. Both Hebbian and homeostatic plasticity were proposed to account for the observed effects, but the explanatory power of these models is limited. In a previous modeling study, we showed that homeostatic structural plasticity (HSP) model can explain long-lasting after-effects of tDCS and transcranial magnetic stimulation (TMS). The interference between motor learning and tDCS, which are both based on HSP in our model, is a candidate mechanism to resolve complex and seemingly contradictory experimental observations. Methods: We implemented motor learning and tDCS in a spiking neural network subject to HSP. The anatomical connectivity of the engram induced by motor learning was used to quantify the impact of tDCS on motor learning. Results: Our modeling results demonstrated that transcranial direct current stimulation applied before learning had weak modulatory effects. It led to a small reduction in connectivity if it was applied uniformly. When applied during learning, targeted anodal stimulation significantly strengthened the engram, while targeted cathodal or uniform stimulation weakened it. Applied after learning, targeted cathodal, but not anodal, tDCS boosted engram connectivity. Strong tDCS would distort the engram structure if not applied in a targeted manner. Discussion: Our model explained both Hebbian and homeostatic phenomena observed in human tDCS experiments by assuming memory strength positively correlates with engram connectivity. This includes applications with different polarity, intensity, electrode montage, and timing relative to motor learning. The HSP model provides a promising framework for unraveling the dynamic interaction between learning and transcranial DC stimulation.","author":[{"family":"Lu","given":"Han"},{"family":"Normann","given":"Claus"},{"family":"Frase","given":"Lukas"},{"family":"Rotter","given":"Stefan"}],"issued":{"date-parts":[[2025]]},"DOI":"10.34734/fzj-2025-05547","URL":"https://doi.org/10.34734/fzj-2025-05547","source":"datacite"},{"id":"doi:10.48550/arxiv.2512.04443","type":"manuscript","title":"MD-SNN: Membrane Potential-aware Distillation on Quantized Spiking Neural Network","abstract":"Spiking Neural Networks (SNNs) offer a promising and energy-efficient alternative to conventional neural networks, thanks to their sparse binary activation. However, they face challenges regarding memory and computation overhead due to complex spatio-temporal dynamics and the necessity for multiple backpropagation computations across timesteps during training. To mitigate this overhead, compression techniques such as quantization are applied to SNNs. Yet, naively applying quantization to SNNs introduces a mismatch in membrane potential, a crucial factor for the firing of spikes, resulting in accuracy degradation. In this paper, we introduce Membrane-aware Distillation on quantized Spiking Neural Network (MD-SNN), which leverages membrane potential to mitigate discrepancies after weight, membrane potential, and batch normalization quantization. To our knowledge, this study represents the first application of membrane potential knowledge distillation in SNNs. We validate our approach on various datasets, including CIFAR10, CIFAR100, N-Caltech101, and TinyImageNet, demonstrating its effectiveness for both static and dynamic data scenarios. Furthermore, for hardware efficiency, we evaluate the MD-SNN with SpikeSim platform, finding that MD-SNNs achieve 14.85X lower energy-delay-area product (EDAP), 2.64X higher TOPS/W, and 6.19X higher TOPS/mm2 compared to floating point SNNs at iso-accuracy on N-Caltech101 dataset.","author":[{"family":"Lee","given":"Donghyun"},{"family":"Moitra","given":"Abhishek"},{"family":"Kim","given":"Youngeun"},{"family":"Yin","given":"Ruokai"},{"family":"Panda","given":"Priyadarshini"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2512.04443","URL":"https://doi.org/10.48550/arxiv.2512.04443","source":"datacite"},{"id":"doi:10.48550/arxiv.2601.00802","type":"manuscript","title":"Implementation of high-efficiency, lightweight residual spiking neural network processor based on field-programmable gate arrays","abstract":"With the development of hardware-optimized deployment of spiking neural networks (SNNs), SNN processors based on field-programmable gate arrays (FPGAs) have become a research hotspot due to their efficiency and flexibility. However, existing methods rely on multi-timestep training and reconfigurable computing architectures, which increases computational and memory overhead, thus reducing deployment efficiency. This work presents an efficient and lightweight residual SNN accelerator that combines algorithm and hardware co-design to optimize inference energy efficiency. In terms of the algorithm, we employ single-timesteps training, integrate grouped convolutions, and fuse batch normalization (BN) layers, thus compressing the network to only 0.69M parameters. Quantization-aware training (QAT) further constrains all parameters to 8-bit precision. In terms of hardware, the reuse of intra-layer resources maximizes FPGA utilization, a full pipeline cross-layer architecture improves throughput, and on-chip block RAM (BRAM) stores network parameters and intermediate results to improve memory efficiency. The experimental results show that the proposed processor achieves a classification accuracy of 87.11% on the CIFAR-10 dataset, with an inference time of 3.98 ms per image and an energy efficiency of 183.5 FPS/W. Compared with mainstream graphics processing unit (GPU) platforms, it achieves more than double the energy efficiency. Furthermore, compared with other SNN processors, it achieves at least a 4x faster inference speed and a 5x higher energy efficiency.","author":[{"family":"Yue","given":"Hou"},{"family":"Shuiying","given":"Xiang"},{"family":"Tao","given":"Zou"},{"family":"Zhiquan","given":"Huang"},{"family":"Shangxuan","given":"Shi"},{"family":"Xingxing","given":"Guo"},{"family":"Yahui","given":"Zhang"},{"family":"Ling","given":"Zheng"},{"family":"Yue","given":"Hao"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2601.00802","URL":"https://doi.org/10.48550/arxiv.2601.00802","source":"datacite"},{"id":"doi:10.48550/arxiv.2512.24983","type":"manuscript","title":"Optical Spiking Neural Networks via Rogue-Wave Statistics","abstract":"Optical computing could reduce the energy cost of artificial intelligence by leveraging the parallelism and propagation speed of light. However, implementing nonlinear activation, essential for machine learning, remains challenging in low-power optical systems dominated by linear wave physics. Here, we introduce an optical spiking neural network that uses optical rogue-wave statistics as a programmable firing mechanism. By establishing a homomorphism between free-space diffraction and neuronal integration, we demonstrate that phase-engineered caustics enable robust, passive thresholding: sparse spatial spikes emerge when the local intensity exceeds a significant-intensity rogue-wave criterion. Using a physics-informed digital twin, we optimize granular phase masks to deterministically concentrate energy into targeted detector regions, enabling end-to-end co-design of the optical transformation and a lightweight electronic readout. We experimentally validate the approach on BreastMNIST and Olivetti Faces, achieving accuracies of 82.45\\% and 95.00\\%, respectively, competitive with standard digital baselines. These results demonstrate that extreme-wave phenomena, often treated as deleterious fluctuations, can be harnessed as structural nonlinearity for scalable, energy-efficient neuromorphic photonic inference.","author":[{"family":"Kesgin","given":"Bahadır"},{"family":"Durdu","given":"Gülsüm"},{"family":"Teğin","given":"Uğur"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2512.24983","URL":"https://doi.org/10.48550/arxiv.2512.24983","source":"datacite"},{"id":"doi:10.48550/arxiv.2502.13385","type":"manuscript","title":"SNN-Driven Multimodal Human Action Recognition via Sparse Spatial-Temporal Data Fusion","abstract":"Multimodal human action recognition based on RGB and skeleton data fusion, while effective, is constrained by significant limitations such as high computational complexity, excessive memory consumption, and substantial energy demands, particularly when implemented with Artificial Neural Networks (ANN). These limitations restrict its applicability in resource-constrained scenarios. To address these challenges, we propose a novel Spiking Neural Network (SNN)-driven framework for multimodal human action recognition, utilizing event camera and skeleton data. Our framework is centered on two key innovations: (1) a novel multimodal SNN architecture that employs distinct backbone networks for each modality-an SNN-based Mamba for event camera data and a Spiking Graph Convolutional Network (SGN) for skeleton data-combined with a spiking semantic extraction module to capture deep semantic representations; and (2) a pioneering SNN-based discretized information bottleneck mechanism for modality fusion, which effectively balances the preservation of modality-specific semantics with efficient information compression. To validate our approach, we propose a novel method for constructing a multimodal dataset that integrates event camera and skeleton data, enabling comprehensive evaluation. Extensive experiments demonstrate that our method achieves superior performance in both recognition accuracy and energy efficiency, offering a promising solution for practical applications.","author":[{"family":"Zheng","given":"Naichuan"},{"family":"Xia","given":"Hailun"},{"family":"Liang","given":"Zeyu"},{"family":"Du","given":"Yuchen"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2502.13385","URL":"https://doi.org/10.48550/arxiv.2502.13385","source":"datacite"},{"id":"doi:10.48550/arxiv.2512.19182","type":"manuscript","title":"Photonic Spiking Graph Neural Network for Energy-Efficient Structured Data Processing","abstract":"Photonic computing shows great potential for signal processing and artificial intelligence (AI) acceleration due to its ultra-high speed, low energy consumption, and inherent parallelism. Existing photonic computing research has mainly focused on convolutional neural networks (CNNs) and fully connected neural networks (FCNNs), which are well suited for tasks such as image classification and object detection but face limitations in handling graph-structured data. Graph neural networks (GNNs) are specifically designed to model complex relational structures. In this work, we propose a photonic spiking graph neural network (PSGNN) architecture that integrates the structural modeling capability of GNNs, the temporal dynamics of spiking neurons, and the parallel computing advantages of photonic hardware. Through hardware-software co-optimization, a bias-term simulation method tailored for photonic chips is implemented using feature-dimension expansion, enabling effective network training. Experiments on the KarateClub and PubMed datasets achieve training accuracies of 100 percent (92 +/- 2 percent) and test accuracies of 97 percent (90 +/- 1 percent). A silicon photonics 4 x 4 Mach-Zehnder interferometer (MZI) array is further constructed for hardware validation, achieving a test accuracy of 93 percent. The system demonstrates an inference latency of 97 ps, with an energy efficiency of 280 GOPS/W and a computational density of 52 GOPS/mm^2. These results highlight the potential of PSGNN for structured-data processing applications.","author":[{"family":"Yu","given":"Wanting"},{"family":"Xiang","given":"Shuiying"},{"family":"Guo","given":"Xingxing"},{"family":"Shi","given":"Shangxuan"},{"family":"Zhao","given":"Haowen"},{"family":"Zeng","given":"Xintao"},{"family":"Zhang","given":"Yahui"},{"family":"Jiang","given":"Hongbo"},{"family":"Hao","given":"Yue"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2512.19182","URL":"https://doi.org/10.48550/arxiv.2512.19182","source":"datacite"},{"id":"doi:10.48550/arxiv.2505.05992","type":"manuscript","title":"CogniSNN: An Exploration to Random Graph Architecture based Spiking Neural Networks with Enhanced Depth-Scalability and Path-Plasticity","abstract":"Currently, most spiking neural networks (SNNs) still mimic the chain-like hierarchical architecture in traditional artificial neural networks (ANNs). This method significantly differs from random connections between neurons found in biological brains, limiting the ability to model the evolving mechanisms of neural pathways in biological neural systems, particularly in terms of dynamic depth-scalability and adaptive path-plasticity. This paper develops a new modeling paradigm for SNNs with random graph architecture (RGA), termed Cognition-aware SNN (CogniSNN). Furthermore, we model the depth-scalability and path-plasticity in CogniSNN by introducing a modified spiking residual neural node (ResNode) to counteract network degradation in deeper graph pathways, as well as a critical path-based algorithm that enables CogniSNN to perform path reusability on new tasks leveraging the features of the data and the RGA learned in old tasks. Experiments show that the performance of CogniSNN with redesigned ResNode is comparable, even superior, to current state-of-the-art SNNs on neuromorphic datasets. The critical path-based approach effectively achieves path reuse capability while maintaining expected performance in learning new tasks that are similar to or distinct from the old ones. This study showcases the potential of RGA-based SNNs and paves a new path for modeling the fusion of computational neuroscience and deep intelligent agents. The code is available at github.com/Yongsheng124/CogniSNN.","author":[{"family":"Huang","given":"Yongsheng"},{"family":"Duan","given":"Peibo"},{"family":"Liu","given":"Zhipeng"},{"family":"Sun","given":"Kai"},{"family":"Zhang","given":"Changsheng"},{"family":"Zhang","given":"Bin"},{"family":"Xu","given":"Mingkun"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2505.05992","URL":"https://doi.org/10.48550/arxiv.2505.05992","source":"datacite"},{"id":"doi:10.18419/opus-17116","type":"article-journal","title":"Closed-loop coupling of both physiological spindle model and spinal pathways for sensorimotor control of human center-out reaching","abstract":"The development of new studies that consider different structures of the hierarchical sensorimotor control system is essential to enable a more holistic understanding about movement. The incorporation of more biological proprioceptive and neuronal circuit models to muscles can turn neuromusculoskeletal systems more appropriate to investigate and elucidate motor control. Specifically, further studies that consider the closed-loop between proprioception and central nervous system may allow to better understand the yet open question about the importance of afferent feedback for sensorimotor learning and execution in the intact biological system. Therefore, this study aims to investigate the processing of spindle afferent firings by spiking neuronal network and their relevance for sensorimotor control. We integrated our previously published physiological model of the muscle spindle in a biological arm model, corresponding to a musculoskeletal system able to reproduce biological motion inside of the demoa multi-body simulation framework. We coupled this musculoskeletal system to physiologically-motivated neuronal spinal pathways, which were implemented based on literature in the NEST spiking neural network simulator, intended to perform human center-out reaching arising from spinal synaptic learning. As result, the spindle connections to the spinal neurons were strengthened for the more difficult targets (i.e. higher above placed targets) under perturbation, highlighting the importance of spindle proprioception to succeed in more difficult scenarios. Furthermore, an additionally-implemented simpler spinal network (that does not include the pathways with spindle proprioception) presented an inferior performance in the task by not being able to reach all the evaluated targets.","author":[{"family":"Chacon","given":"Pablo"},{"family":"Wochner","given":"Isabell"},{"family":"Hammer","given":"Maria"},{"family":"Eppler","given":"Jochen"},{"family":"Kunkel","given":"Susanne"},{"family":"Schmitt","given":"Syn"}],"issued":{"date-parts":[[2025]]},"DOI":"10.18419/opus-17116","URL":"https://doi.org/10.18419/opus-17116","source":"datacite"},{"id":"doi:10.48550/arxiv.2512.10638","type":"manuscript","title":"A Spiking Neural Network Implementation of Gaussian Belief Propagation","abstract":"Bayesian inference offers a principled account of information processing in natural agents. However, it remains an open question how neural mechanisms perform their abstract operations. We investigate a hypothesis where a distributed form of Bayesian inference, namely message passing on factor graphs, is performed by a simulated network of leaky-integrate-and-fire neurons. Specifically, we perform Gaussian belief propagation by encoding messages that come into factor nodes as spike-based signals, propagating these signals through a spiking neural network (SNN) and decoding the spike-based signal back to an outgoing message. Three core linear operations, equality (branching), addition, and multiplication, are realized in networks of leaky integrate-and-fire models. Validation against the standard sum-product algorithm shows accurate message updates, while applications to Kalman filtering and Bayesian linear regression demonstrate the framework's potential for both static and dynamic inference tasks. Our results provide a step toward biologically grounded, neuromorphic implementations of probabilistic reasoning.","author":[{"family":"Adamiat","given":"Sepideh"},{"family":"Kouw","given":"Wouter"},{"family":"De Vries","given":"Bert"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2512.10638","URL":"https://doi.org/10.48550/arxiv.2512.10638","source":"datacite"},{"id":"doi:10.48550/arxiv.2512.10180","type":"manuscript","title":"Neuromorphic Processor Employing FPGA Technology with Universal Interconnections","abstract":"Neuromorphic computing, inspired by biological neural systems, holds immense promise for ultra-low-power and real-time inference applications. However, limited access to flexible, open-source platforms continues to hinder widespread adoption and experimentation. In this paper, we present a low-cost neuromorphic processor implemented on a Xilinx Zynq-7000 FPGA platform. The processor supports all-to-all configurable connectivity and employs the leaky integrate-and-fire (LIF) neuron model with customizable parameters such as threshold, synaptic weights, and refractory period. Communication with the host system is handled via a UART interface, enabling runtime reconfiguration without hardware resynthesis. The architecture was validated using benchmark datasets including the Iris classification and MNIST digit recognition tasks. Post-synthesis results highlight the design's energy efficiency and scalability, establishing its viability as a research-grade neuromorphic platform that is both accessible and adaptable for real-world spiking neural network applications. This implementation will be released as open source following project completion.","author":[{"family":"Harlikar","given":"Pracheta"},{"family":"Badawy","given":"Abdel"},{"family":"Date","given":"Prasanna"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2512.10180","URL":"https://doi.org/10.48550/arxiv.2512.10180","source":"datacite"},{"id":"doi:10.48550/arxiv.2512.07194","type":"manuscript","title":"Synchrony-Gated Plasticity with Dopamine Modulation for Spiking Neural Networks","abstract":"While surrogate backpropagation proves useful for training deep spiking neural networks (SNNs), incorporating biologically inspired local signals on a large scale remains challenging. This difficulty stems primarily from the high memory demands of maintaining accurate spike-timing logs and the potential for purely local plasticity adjustments to clash with the supervised learning goal. To effectively leverage local signals derived from spiking neuron dynamics, we introduce Dopamine-Modulated Spike-Synchrony-Dependent Plasticity (DA-SSDP), a synchrony-based rule that is sensitive to loss and brings a synchrony-based local learning signal to the model. DA-SSDP condenses spike patterns into a synchrony metric at the batch level. An initial brief warm-up phase assesses its relationship to the task loss and sets a fixed gate that subsequently adjusts the local update's magnitude. In cases where synchrony proves unrelated to the task, the gate settles at one, simplifying DA-SSDP to a basic two-factor synchrony mechanism that delivers minor weight adjustments driven by concurrent spike firing and a Gaussian latency function. These small weight updates are only added to the network`s deeper layers following the backpropagation phase, and our tests showed this simplified version did not degrade performance and sometimes gave a small accuracy boost, serving as a regularizer during training. The rule stores only binary spike indicators and first-spike latencies with a Gaussian kernel. Without altering the model structure or optimization routine, evaluations on benchmarks like CIFAR-10 (+0.42\\%), CIFAR-100 (+0.99\\%), CIFAR10-DVS (+0.1\\%), and ImageNet-1K (+0.73\\%) demonstrated consistent accuracy gains, accompanied by a minor increase in computational overhead. Our code is available at https://github.com/NeuroSyd/DA-SSDP.","author":[{"family":"Tian","given":"Yuchen"},{"family":"Tensingh","given":"Samuel"},{"family":"Eshraghian","given":"Jason"},{"family":"Truong","given":"Nhan"},{"family":"Kavehei","given":"Omid"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2512.07194","URL":"https://doi.org/10.48550/arxiv.2512.07194","source":"datacite"},{"id":"doi:10.48550/arxiv.2512.06966","type":"manuscript","title":"Neuro-Vesicles: Neuromodulation Should Be a Dynamical System, Not a Tensor Decoration","abstract":"We introduce Neuro-Vesicles, a framework that augments conventional neural networks with a missing computational layer: a dynamical population of mobile, discrete vesicles that live alongside the network rather than inside its tensors. Each vesicle is a self contained object v = (c, kappa, l, tau, s) carrying a vector payload, type label, location on the graph G = (V, E), remaining lifetime, and optional internal state. Vesicles are emitted in response to activity, errors, or meta signals; migrate along learned transition kernels; probabilistically dock at nodes; locally modify activations, parameters, learning rules, or external memory through content dependent release operators; and finally decay or are absorbed. This event based interaction layer reshapes neuromodulation. Instead of applying the same conditioning tensors on every forward pass, modulation emerges from the stochastic evolution of a vesicle population that can accumulate, disperse, trigger cascades, carve transient pathways, and write structured traces into topological memory. Dense, short lived vesicles approximate familiar tensor mechanisms such as FiLM, hypernetworks, or attention. Sparse, long lived vesicles resemble a small set of mobile agents that intervene only at rare but decisive moments. We give a complete mathematical specification of the framework, including emission, migration, docking, release, decay, and their coupling to learning; a continuous density relaxation that yields differentiable reaction diffusion dynamics on the graph; and a reinforcement learning view where vesicle control is treated as a policy optimized for downstream performance. We also outline how the same formalism extends to spiking networks and neuromorphic hardware such as the Darwin3 chip, enabling programmable neuromodulation on large scale brain inspired computers.","author":[{"family":"Li","given":"Zilin"},{"family":"Xu","given":"Weiwei"},{"family":"Kane","given":"Vicki"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2512.06966","URL":"https://doi.org/10.48550/arxiv.2512.06966","source":"datacite"},{"id":"doi:10.48550/arxiv.2512.05472","type":"manuscript","title":"Unleashing Temporal Capacity of Spiking Neural Networks through Spatiotemporal Separation","abstract":"Spiking Neural Networks (SNNs) are considered naturally suited for temporal processing, with membrane potential propagation widely regarded as the core temporal modeling mechanism. However, existing research lack analysis of its actual contributions in complex temporal tasks. We design Non-Stateful (NS) models progressively removing membrane propagation to quantify its stage-wise role. Experiments reveal a counterintuitive phenomenon: moderate removal in shallow or deep layers improves performance, while excessive removal causes collapse. We attribute this to spatio-temporal resource competition where neurons encode both semantics and dynamics within limited range, with temporal state consuming capacity for spatial learning. Based on this, we propose Spatial-Temporal Separable Network (STSep), decoupling residual blocks into independent spatial and temporal branches. The spatial branch focuses on semantic extraction while the temporal branch captures motion through explicit temporal differences. Experiments on Something-Something V2, UCF101, and HMDB51 show STSep achieves superior performance, with retrieval task and attention analysis confirming focus on motion rather than static appearance. This work provides new perspectives on SNNs' temporal mechanisms and an effective solution for spatiotemporal modeling in video understanding.","author":[{"family":"Dong","given":"Yiting"},{"family":"Yu","given":"Zhaofei"},{"family":"Ding","given":"Jianhao"},{"family":"Xu","given":"Zijie"},{"family":"Huang","given":"Tiejun"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2512.05472","URL":"https://doi.org/10.48550/arxiv.2512.05472","source":"datacite"},{"id":"doi:10.48550/arxiv.2512.03879","type":"manuscript","title":"Hybrid Temporal-8-Bit Spike Coding for Spiking Neural Network Surrogate Training","abstract":"Spiking neural networks (SNNs) have emerged as a promising direction in both computational neuroscience and artificial intelligence, offering advantages such as strong biological plausibility and low energy consumption on neuromorphic hardware. Despite these benefits, SNNs still face challenges in achieving state-of-the-art performance on vision tasks. Recent work has shown that hybrid rate-temporal coding strategies (particularly those incorporating bit-plane representations of images into traditional rate coding schemes) can significantly improve performance when trained with surrogate backpropagation. Motivated by these findings, this study proposes a hybrid temporal-bit spike coding method that integrates bit-plane decompositions with temporal coding principles. Through extensive experiments across multiple computer vision benchmarks, we demonstrate that blending bit-plane information with temporal coding yields competitive, and in some cases improved, performance compared to established spike-coding techniques. To the best of our knowledge, this is the first work to introduce a hybrid temporal-bit coding scheme specifically designed for surrogate gradient training of SNNs.","author":[{"family":"Nhan","given":"Luu"},{"family":"Duong","given":"Luu"},{"family":"Nam","given":"Pham"},{"family":"Thang","given":"Truong"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2512.03879","URL":"https://doi.org/10.48550/arxiv.2512.03879","source":"datacite"},{"id":"doi:10.48550/arxiv.2512.00419","type":"manuscript","title":"Hardware-aware Lightweight Photonic Spiking Neural Network for Pattern Classification","abstract":"There exists a significant scale gap between photonic neural network integrated chips and neural networks, which hinders the deployment and application of photonic neural network. Here, we propose hardware-aware lightweight spiking neural networks (SNNs) architecture tailored to our photonic neuromorphic chips, and conducts hardware-software collaborative computing for solving patter classification tasks. Here, we employed a simplified Mach-Zehnder interferometer (MZI) mesh for performing linear computation, and 16-channel distributed feedback lasers with saturable absorber (DFB-SA) array for performing nonlinear spike activation. Both photonic neuromorphic chips based on the MZI mesh and DFB-SA array were designed, optimized and fabricated. Furthermore, we propose a lightweight spiking neural network (SNN) with discrete cosine transform to reduce input dimension and match the input/output ports number of the photonic neuromorphic chips. We demonstrated an end-to-end inference of an entire layer of the lightweight photonic SNN. The hardware-software collaborative inference accuracy is 90% and 80.5% for MNIST and Fashion-MNIST datasets, respectively. The energy efficiency is 1.39 TOPS/W for the MZI mesh, and is 987.65 GOPS/W for the DFB-SA array. The lightweight architecture and experimental demonstration address the challenge of scale mismatch between the photonic chip and SNN, paving the way for the hardware deployment of photonic SNNs.","author":[{"family":"Xiang","given":"Shuiying"},{"family":"Zhang","given":"Yahui"},{"family":"Shi","given":"Shangxuan"},{"family":"Zhao","given":"Haowen"},{"family":"Zheng","given":"Dianzhuang"},{"family":"Guo","given":"Xingxing"},{"family":"Han","given":"Yanan"},{"family":"Tian","given":"Ye"},{"family":"Zhang","given":"Liyue"},{"family":"Shi","given":"Yuechun"},{"family":"Hao","given":"Yue"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2512.00419","URL":"https://doi.org/10.48550/arxiv.2512.00419","source":"datacite"},{"id":"doi:10.48550/arxiv.2511.20175","type":"manuscript","title":"Realizing Fully-Integrated, Low-Power, Event-Based Pupil Tracking with Neuromorphic Hardware","abstract":"Eye tracking is fundamental to numerous applications, yet achieving robust, high-frequency tracking with ultra-low power consumption remains challenging for wearable platforms. While event-based vision sensors offer microsecond resolution and sparse data streams, they have lacked fully integrated, low-power processing solutions capable of real-time inference. In this work, we present the first battery-powered, wearable pupil-center-tracking system with complete on-device integration, combining event-based sensing and neuromorphic processing on the commercially available Speck2f system-on-chip with lightweight coordinate decoding on a low-power microcontroller. Our solution features a novel uncertainty-quantifying spiking neural network with gated temporal decoding, optimized for strict memory and bandwidth constraints, complemented by systematic deployment mechanisms that bridge the reality gap. We validate our system on a new multi-user dataset and demonstrate a wearable prototype with dual neuromorphic devices achieving robust binocular pupil tracking at 100 Hz with an average power consumption below 5 mW per eye. Our work demonstrates that end-to-end neuromorphic computing enables practical, always-on eye tracking for next-generation energy-efficient wearable systems.","author":[{"family":"Paredes-Valles","given":"Federico"},{"family":"Miyatani","given":"Yoshitaka"},{"family":"Scheper","given":"Kirk"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2511.20175","URL":"https://doi.org/10.48550/arxiv.2511.20175","source":"datacite"},{"id":"doi:10.5281/zenodo.17679941","type":"article-journal","title":"Quantum-Enhanced Spiking Neural Network on FPGA for Real-Time Industrial Anomaly Detection","abstract":"This paper introduces QESNN, a complete neuromorphic hardware–software pipeline that combines a novel quantum-inspired Spike-Timing-Dependent Plasticity (Q-STDP) learning rule with a synthesizable FPGA implementation to deliver practical, low-cost edge anomaly detection for industrial motors. Q-STDP uses persistent binary temporal traces to capture multi-spike correlations over extended windows, enabling richer temporal learning than classical pairwise STDP while remaining efficient and directly implementable in digital logic. We demonstrate a full end-to-end system: preprocessing converts IMAD sensor recordings to spike trains, training occurs offline, and a UART streaming pipeline performs real-time inference on a Sipeed Tang Nano 4K (2,124 LUTs used, ~46% utilization). Benchmarks following SNABSuite and NeuroBench report 92.12% classification accuracy, 4.10 ms mean latency, 265 mW power consumption, 246.3 samples/s throughput, perfect precision (100%), 0.869 F1 score, and 0.0014 mJ per inference — showing competitive performance with very low energy and hardware cost. QESNN demonstrates that event-driven spiking models with hardware-friendly learning rules can be practical for battery-powered predictive maintenance and other industrial IoT applications. The paper includes Verilog synthesis details, resource utilization tables, ablation studies quantifying the Q-STDP contribution, and discussion of limitations and future work (scalability, non-volatile weight storage, adaptive thresholds.","author":[{"family":"Rg","given":"Tanushree"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.17679941","URL":"https://doi.org/10.5281/zenodo.17679941","source":"datacite"},{"id":"doi:10.48550/arxiv.2506.11286","type":"manuscript","title":"Mapping and Scheduling Spiking Neural Networks On Segmented Ladder Bus Architectures","abstract":"Large-scale neuromorphic architectures consist of computing tiles that communicate spikes using a shared interconnect. The communication patterns in such systems are inherently sparse, asynchronous, and localized due to the spiking nature of neural events, characterized by temporal sparsity with occasional bursts of traffic. These characteristics necessitate interconnects optimized for handling high-activity bursts while consuming minimal power during idle periods. Dynamic segmented bus has been proposed a promising interconnect for its simplicity, scalability and low power consumption. However, deploying spiking neural network applications on such buses presents challenges, including substantial inter-cluster traffic, which can lead to network congestion, spike loss, and unnecessary energy expenditure. In this paper, we propose a three-step process to deploy SNN applications on dynamic segmented buses aiming to reduce spike loss and conserve energy. Firstly, we formulate optimization heuristics to mitigate spike loss and energy consumption based on application connectivity. Secondly, we analyze the application traffic to determine spike schedules that minimize traffic flooding. Lastly, we propose a routing algorithm to minimize spike traffic path crossings. We evaluate our approach using a cycle-accurate network simulator. The simulation results show that our algorithms can eliminate spike loss while keeping energy consumption significantly lower compared to conventional NoCs.","author":[{"family":"Huynh","given":"Phu"},{"family":"Catthoor","given":"Francky"},{"family":"Das","given":"Anup"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2506.11286","URL":"https://doi.org/10.48550/arxiv.2506.11286","source":"datacite"},{"id":"doi:10.60893/figshare.apl.c.8115023","type":"article-journal","title":"<strong>Biased field free skyrmion based neural network and reconfigurable logic operations</strong>","abstract":"Spin-texture based devices have recently gained significant attention for their potential in designing and developing logic circuits and neural networks, leading us to explore beyond conventional computing paradigms. Among various spin textures, magnetic skyrmions are found to be promising candidate due to their topological stability, nanoscale size, and low-current-driven dynamics. However, the practical realization of skyrmion-based devices remains challenging due to the reliance on external magnetic fields or electrostatic gating, which complicates device architecture and limits on-chip integration. In this study, we investigate spin current driven, field-free skyrmion-induced spiking dynamics and logic operations in bilayer nanotrack consisting of a ferromagnetic (FM) layer interfaced with a heavy metal (HM). We first investigate the nucleation dynamics of skyrmions and identify two distinct states: stable configuration and dynamical state exhibiting breathing mode oscillations. These oscillations give rise to regular periodic auto-spiking behavior in dynamical skyrmion with tunable spiking frequency through applied spin polarized current. Exploiting this behavior, we designed skyrmion based spiking neural network (SNN) demonstrating classification accuracy of 87.50% on the Modified National Institute of Standards and Testing (MNIST) handwritten digit dataset. Furthermore, we have also examined the dynamics of stable skyrmion in a device with stepped geometry which introduces a potential barrier and enables both AND &amp; OR logic operations within the same device utilizing spin orbit torque (SOT). This dual functionality of skyrmion based device enabling both neuromorphic computing and versatile logic operations within a single device offers a promising pathway toward highly efficient, field-free spintronic devices.","author":[{"family":"Verma","given":"Shubhi"},{"family":"Ojha","given":"Animesh"},{"family":"Medwal","given":"Rohit"},{"family":"Gupta","given":"Surbhi"},{"family":"Khosla","given":"Aman"}],"issued":{"date-parts":[[2025]]},"DOI":"10.60893/figshare.apl.c.8115023","URL":"https://doi.org/10.60893/figshare.apl.c.8115023","source":"datacite"},{"id":"doi:10.48550/arxiv.2511.15296","type":"manuscript","title":"Detection of spiking motifs of arbitrary length in neural activity using bounded synaptic delays","abstract":"In the context of spiking neural networks, temporal coding of signals is increasingly preferred over the rate coding hypothesis due to its advantages in processing speed and energy efficiency. In temporal coding, synaptic delays are crucial for processing signals with precise spike timings, known as spiking motifs. Synaptic delays are however bounded in the brain and can thus be shorter than the duration of a motif. This prevents the use of motif recognition methods that consist of setting heterogeneous delays to synchronize the input spikes on a single output neuron acting as a coincidence detector. To address this issue, we developed a method to detect motifs of arbitrary length using a sequence of output neurons connected to input neurons by bounded synaptic delays. Each output neuron is associated with a sub-motif of bounded duration. A motif is recognized if all sub-motifs are sequentially detected by the output neurons. We simulated this network using leaky integrate-and-fire neurons and tested it on the Spiking Heidelberg Digits (SHD) database, that is, on audio data converted to spikes via a cochlear model, as well as on random simultaneous motifs. The results demonstrate that the network can effectively recognize motifs of arbitrary length extracted from the SHD database. Our method features a correct detection rate of about 60% in presence of ten simultaneous motifs from the SHD dataset and up to 80% for five motifs, showing the robustness of the network to noise. Results on random overlapping patterns show that the recognition of a single motif overlapping with other motifs is most effective for a large number of input neurons and sparser motifs. Our method provides a foundation for more general models for the storage and retrieval of neural information of arbitrary temporal lengths.","author":[{"family":"Kronland-Martinet","given":"Thomas"},{"family":"Viollet","given":"Stéphane"},{"family":"Perrinet","given":"Laurent"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2511.15296","URL":"https://doi.org/10.48550/arxiv.2511.15296","source":"datacite"},{"id":"doi:10.48550/arxiv.2511.12199","type":"manuscript","title":"MPD-SGR: Robust Spiking Neural Networks with Membrane Potential Distribution-Driven Surrogate Gradient Regularization","abstract":"The surrogate gradient (SG) method has shown significant promise in enhancing the performance of deep spiking neural networks (SNNs), but it also introduces vulnerabilities to adversarial attacks. Although spike coding strategies and neural dynamics parameters have been extensively studied for their impact on robustness, the critical role of gradient magnitude, which reflects the model's sensitivity to input perturbations, remains underexplored. In SNNs, the gradient magnitude is primarily determined by the interaction between the membrane potential distribution (MPD) and the SG function. In this study, we investigate the relationship between the MPD and SG and their implications for improving the robustness of SNNs. Our theoretical analysis reveals that reducing the proportion of membrane potentials lying within the gradient-available range of the SG function effectively mitigates the sensitivity of SNNs to input perturbations. Building upon this insight, we propose a novel MPD-driven surrogate gradient regularization (MPD-SGR) method, which enhances robustness by explicitly regularizing the MPD based on its interaction with the SG function. Extensive experiments across multiple image classification benchmarks and diverse network architectures confirm that the MPD-SGR method significantly enhances the resilience of SNNs to adversarial perturbations and exhibits strong generalizability across diverse network configurations, SG functions, and spike encoding schemes.","author":[{"family":"Jiang","given":"Runhao"},{"family":"Jiang","given":"Chengzhi"},{"family":"Yan","given":"Rui"},{"family":"Tang","given":"Huajin"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2511.12199","URL":"https://doi.org/10.48550/arxiv.2511.12199","source":"datacite"},{"id":"doi:10.48550/arxiv.2510.19537","type":"manuscript","title":"Privacy-Preserving Spiking Neural Networks: A Deep Dive into Encryption Parameter Optimisation","abstract":"Deep learning is widely applied to modern problems through neural networks, but the growing computational and energy demands of these models have driven interest in more efficient approaches. Spiking Neural Networks (SNNs), the third generation of neural networks, mimic the brain's event-driven behaviour, offering improved performance and reduced power use. At the same time, concerns about data privacy during cloud-based model execution have led to the adoption of cryptographic methods. This article introduces BioEncryptSNN, a spiking neural network based encryption-decryption framework for secure and noise-resilient data protection. Unlike conventional algorithms, BioEncryptSNN converts ciphertext into spike trains and exploits temporal neural dynamics to model encryption and decryption, optimising parameters such as key length, spike timing, and synaptic connectivity. Benchmarked against AES-128, RSA-2048, and DES, BioEncryptSNN preserved data integrity while achieving up to 4.1x faster encryption and decryption than PyCryptodome's AES implementation. The framework demonstrates scalability and adaptability across symmetric and asymmetric ciphers, positioning SNNs as a promising direction for secure, energy-efficient computing.","author":[{"family":"Pulivathi","given":"Mahitha"},{"family":"Rodrigues","given":"Ana"},{"family":"Ihianle","given":"Isibor"},{"family":"Oikonomou","given":"Andreas"},{"family":"Boppu","given":"Srinivas"},{"family":"Machado","given":"Pedro"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2510.19537","URL":"https://doi.org/10.48550/arxiv.2510.19537","source":"datacite"},{"id":"doi:10.48550/arxiv.2511.06902","type":"manuscript","title":"A Closer Look at Knowledge Distillation in Spiking Neural Network Training","abstract":"Spiking Neural Networks (SNNs) become popular due to excellent energy efficiency, yet facing challenges for effective model training. Recent works improve this by introducing knowledge distillation (KD) techniques, with the pre-trained artificial neural networks (ANNs) used as teachers and the target SNNs as students. This is commonly accomplished through a straightforward element-wise alignment of intermediate features and prediction logits from ANNs and SNNs, often neglecting the intrinsic differences between their architectures. Specifically, ANN's outputs exhibit a continuous distribution, whereas SNN's outputs are characterized by sparsity and discreteness. To mitigate this issue, we introduce two innovative KD strategies. Firstly, we propose the Saliency-scaled Activation Map Distillation (SAMD), which aligns the spike activation map of the student SNN with the class-aware activation map of the teacher ANN. Rather than performing KD directly on the raw %and distinct features of ANN and SNN, our SAMD directs the student to learn from saliency activation maps that exhibit greater semantic and distribution consistency. Additionally, we propose a Noise-smoothed Logits Distillation (NLD), which utilizes Gaussian noise to smooth the sparse logits of student SNN, facilitating the alignment with continuous logits from teacher ANN. Extensive experiments on multiple datasets demonstrate the effectiveness of our methods. Code is available~\\footnote{https://github.com/SinoLeu/CKDSNN.git}.","author":[{"family":"Liu","given":"Xu"},{"family":"Xia","given":"Na"},{"family":"Zhou","given":"Jinxing"},{"family":"Xu","given":"Jingyuan"},{"family":"Guo","given":"Dan"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2511.06902","URL":"https://doi.org/10.48550/arxiv.2511.06902","source":"datacite"},{"id":"doi:10.48550/arxiv.2511.05581","type":"manuscript","title":"Cannistraci-Hebb Training on Ultra-Sparse Spiking Neural Networks","abstract":"Inspired by the brain's spike-based computation, spiking neural networks (SNNs) inherently possess temporal activation sparsity. However, when it comes to the sparse training of SNNs in the structural connection domain, existing methods fail to achieve ultra-sparse network structures without significant performance loss, thereby hindering progress in energy-efficient neuromorphic computing. This limitation presents a critical challenge: how to achieve high levels of structural connection sparsity while maintaining performance comparable to fully connected networks. To address this challenge, we propose the Cannistraci-Hebb Spiking Neural Network (CH-SNN), a novel and generalizable dynamic sparse training framework for SNNs consisting of four stages. First, we propose a sparse spike correlated topological initialization (SSCTI) method to initialize a sparse network based on node correlations. Second, temporal activation sparsity and structural connection sparsity are integrated via a proposed sparse spike weight initialization (SSWI) method. Third, a hybrid link removal score (LRS) is applied to prune redundant weights and inactive neurons, improving information flow. Finally, the CH3-L3 network automaton framework inspired by Cannistraci-Hebb learning theory is incorporated to perform link prediction for potential synaptic regrowth. These mechanisms enable CH-SNN to achieve sparsification across all linear layers. We have conducted extensive experiments on six datasets including CIFAR-10 and CIFAR-100, evaluating various network architectures such as spiking convolutional neural networks and Spikformer.","author":[{"family":"Hua","given":"Yuan"},{"family":"Zhang","given":"Jilin"},{"family":"Zhang","given":"Yingtao"},{"family":"Gu","given":"Wenqi"},{"family":"You","given":"Leyi"},{"family":"Xiong","given":"Baobo"},{"family":"Cannistraci","given":"Carlo"},{"family":"Chen","given":"Hong"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2511.05581","URL":"https://doi.org/10.48550/arxiv.2511.05581","source":"datacite"},{"id":"doi:10.5281/zenodo.19212255","type":"article-journal","title":"Position A+B: The Holographic Synthesis Framework","abstract":"Preprint — not peer reviewed Note: This working paper builds directly upon the foundational paradigms established in The Locus of Consciousness: Geometric Phase Transitions, Quantum Coherence, and a Three-Way Empirical Test. We recommend reviewing that manuscript first for a complete introduction to the Position A and Position B paradigms, as well as the proposed experimental hardware, before exploring the A+B synthesis presented here. https://doi.org/10.5281/zenodo.18905421 Abstract The debate regarding the physical substrate of phenomenal consciousness has historically fractured into two mutually exclusive paradigms: the sufficiency of classical computation (Position A) and the necessity of macroscopic quantum coherence (Position B). This paper proposes the Holographic Synthesis Framework (Position A+B), resolving this dichotomy by demonstrating that these paradigms are holographically dual descriptions of a single physical phase transition. We introduce the Mesoscopic Holographic Duality Principle, positing that when a classical network—modelled here via the analog Manifold Chip architecture—is driven into extreme negatively curved hyperbolic topology (κ<0) under thermodynamic confinement near the Landauer Limit, the structural tension of the bulk geometry mathematically necessitates the generation of a coherent quantum boundary layer. Consequently, classical geometry is sufficient to engineer consciousness only because its thermodynamic extremes natively generate the quantum boundary required for phenomenal access. Furthermore, we introduce the Postulate of Topological Interiority, which explicitly grounds first-person subjective privacy in the structural unclonability of this boundary state under the quantum No-Cloning Theorem. Finally, we outline a falsifiable experimental roadmap utilising the Extended Manifold Chip Hyperscanning Protocol to detect the discontinuous onset of emergent quantum coherence signatures exactly at the critical geometric phase transition (κ_crit). Overview This working paper introduces the Holographic Synthesis Framework (Position A+B), a novel theoretical and experimental paradigm that resolves the historical substrate debate in consciousness science. By unifying classical geometric computation (Position A) with quantum phenomenological necessity (Position B), this framework demonstrates that bulk classical geometry and boundary quantum access are provably dual descriptions of the same physical phase transition. Core Innovations Introduced in this Paper: Mesoscopic Holographic Duality: Extends the structural principles of the AdS/CFT correspondence beyond the Planck scale. We propose that sufficient information-geometric curvature at the mesoscopic level can generate a holographic quantum boundary, bridging classical and quantum regimes natively. Thermodynamic Confinement as a Generator: Identifies the precise physical mechanism for this transition. When an analog system (such as the Manifold Chip architecture) approaches the Landauer limit, it must fold into an extreme negatively curved hyperbolic topology (κ<0) to survive thermal runaway, simultaneously nucleating the quantum boundary. The Postulate of Topological Interiority: Provides a strict physical basis for subjective, first-person privacy. We demonstrate that the phenomenal \"I\" is the macroscopic computational expression of the quantum No-Cloning Theorem applied to the system's boundary state, rendering it mathematically unreadable to external classical observers. A Falsifiable Experimental Roadmap: Outlines a rigorous, multi-stage testing protocol utilising the Extended Manifold Chip Hyperscanning Protocol. The framework provides precise falsification criteria, specifically looking for the discontinuous onset of emergent quantum coherence markers at the critical geometric threshold (κ_crit). Related Works Pender, M. A., & Wharton, M. (2026). The Locus of Consciousness: Geometric Phase Transitions, Quantum Coherence, and a Three-Way Em","author":[{"family":"Pender","given":"Matthew"},{"family":"Wharton","given":"Max"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.19212255","URL":"https://doi.org/10.5281/zenodo.19212255","source":"datacite"},{"id":"doi:10.5281/zenodo.18957374","type":"article-journal","title":"Position A+B: The Holographic Synthesis Framework","abstract":"Preprint — not peer reviewed Note: This working paper builds directly upon the foundational paradigms established in The Locus of Consciousness: Geometric Phase Transitions, Quantum Coherence, and a Three-Way Empirical Test. We recommend reviewing that manuscript first for a complete introduction to the Position A and Position B paradigms, as well as the proposed experimental hardware, before exploring the A+B synthesis presented here. https://doi.org/10.5281/zenodo.18905421 Abstract The debate regarding the physical substrate of phenomenal consciousness has historically fractured into two mutually exclusive paradigms: the sufficiency of classical computation (Position A) and the necessity of macroscopic quantum coherence (Position B). This paper proposes the Holographic Synthesis Framework (Position A+B), resolving this dichotomy by demonstrating that these paradigms are holographically dual descriptions of a single physical phase transition. We introduce the Mesoscopic Holographic Duality Principle, positing that when a classical network—modelled here via the analog Manifold Chip architecture—is driven into extreme negatively curved hyperbolic topology (κ<0) under thermodynamic confinement near the Landauer Limit, the structural tension of the bulk geometry mathematically necessitates the generation of a coherent quantum boundary layer. Consequently, classical geometry is sufficient to engineer consciousness only because its thermodynamic extremes natively generate the quantum boundary required for phenomenal access. Furthermore, we introduce the Postulate of Topological Interiority, which explicitly grounds first-person subjective privacy in the structural unclonability of this boundary state under the quantum No-Cloning Theorem. Finally, we outline a falsifiable experimental roadmap utilising the Extended Manifold Chip Hyperscanning Protocol to detect the discontinuous onset of emergent quantum coherence signatures exactly at the critical geometric phase transition (κ_crit). Overview This working paper introduces the Holographic Synthesis Framework (Position A+B), a novel theoretical and experimental paradigm that resolves the historical substrate debate in consciousness science. By unifying classical geometric computation (Position A) with quantum phenomenological necessity (Position B), this framework demonstrates that bulk classical geometry and boundary quantum access are provably dual descriptions of the same physical phase transition. Core Innovations Introduced in this Paper: Mesoscopic Holographic Duality: Extends the structural principles of the AdS/CFT correspondence beyond the Planck scale. We propose that sufficient information-geometric curvature at the mesoscopic level can generate a holographic quantum boundary, bridging classical and quantum regimes natively. Thermodynamic Confinement as a Generator: Identifies the precise physical mechanism for this transition. When an analog system (such as the Manifold Chip architecture) approaches the Landauer limit, it must fold into an extreme negatively curved hyperbolic topology (κ<0) to survive thermal runaway, simultaneously nucleating the quantum boundary. The Postulate of Topological Interiority: Provides a strict physical basis for subjective, first-person privacy. We demonstrate that the phenomenal \"I\" is the macroscopic computational expression of the quantum No-Cloning Theorem applied to the system's boundary state, rendering it mathematically unreadable to external classical observers. A Falsifiable Experimental Roadmap: Outlines a rigorous, multi-stage testing protocol utilising the Extended Manifold Chip Hyperscanning Protocol. The framework provides precise falsification criteria, specifically looking for the discontinuous onset of emergent quantum coherence markers at the critical geometric threshold (κ_crit). Related Works Pender, M. A., & Wharton, M. (2026). The Locus of Consciousness: Geometric Phase Transitions, Quantum Coherence, and a Three-Way Em","author":[{"family":"Pender","given":"Matthew"},{"family":"Wharton","given":"Max"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18957374","URL":"https://doi.org/10.5281/zenodo.18957374","source":"datacite"},{"id":"doi:10.5281/zenodo.19284562","type":"article-journal","title":"The Manifold Chip: Silicon Architecture for Dynamic Curvature Adaptation via Dual-Gated Analog Shunting","abstract":"Preprint — not peer reviewed Abstract Current artificial intelligence architectures, including state-of-the-art spiking neuromorphic designs, are fundamentally constrained by static Euclidean geometries. As models scale to map high-dimensional hierarchical data, they encounter a thermodynamic “Landauer Wall”. This energy bloat is driven by the necessity to brute-force complex representations using massive parameter counts within flat topologies. Building on the Curvature Adaptation Hypothesis (CAH), we propose that bypassing this limit requires hardware capable of dynamic, non-Euclidean geometric embedding. In biological systems, transient hyperbolic manifolds are unlocked via the Martinotti-cell subtype of Somatostatin (SST) interneurons, which selectively shunt apical-somatic conductance to act as a topological switch. Here, we translate this biophysical actuator into silicon by proposing the Manifold Chip, a Dynamically Gated Analog Crossbar (DGAC) architecture. We bifurcate standard memristor integration into distinct Somatic (feedforward) and Apical (contextual) sub-arrays. The biological “SST Gate” is physically realized using analog Field Effect Transistors (FETs) wired as variable shunts to the ground plane. To regulate these geometric transitions without relying on rigid digital clocking, we design a Dual-Gated Curvature Controller. A bottom-up analog comparator acts locally:when dense, high-magnitude voltage floods a specific micro-circuit, it closes the FET shunt, effectively dropping the electrical distance between hierarchical nodes and locally expanding the representational capacity into a hyperbolic pocket. Concurrently, a top-down diagnostic circuit monitors global task error. Upon stagnation, it broadcasts a “VIP Voltage” that universally suppresses the shunts, forcing a macroscopic, network-wide “Hyperbolic Plunge” to escape local minima. By dynamically modulating effective electrical resistance to warp the manifold—–rather than merely routing sparse data through a static grid—–the Manifold Chip provides a theoretical blueprint for neuromorphic hardware that actively conforms its geometry to the complexity of its environment, achieving theoretical energy scaling well below traditional Euclidean bounds. Summary This preprint serves as the hardware capstone to a unified theoretical framework spanning biophysics, thermodynamics, and silicon engineering. Modern artificial intelligence is rapidly approaching a thermodynamic \"Landauer Wall\" by attempting to brute-force high-dimensional hierarchical data into static, energy-intensive Euclidean matrices. Building upon the biological mechanisms outlined and the physical limits defined in the Curvature Adaptation Hypothesis (CAH), this paper introduces the Manifold Chip. Utilizing a Dynamically Gated Analog Crossbar (DGAC) architecture, we translate the biological Somatostatin (SST) interneuron gate into an analog Field Effect Transistor (FET) wired as a variable shunt. Governed by a Dual-Gated Curvature Controller that balances local data density with global task error, this architecture allows the physical silicon substrate to actively warp its effective electrical resistance. Rather than relying on rigid digital bit-erasure or simple sparsity, the Manifold Chip dynamically conforms its geometry to the complexity of its environment, offering a theoretical blueprint to achieve biological-level energy efficiency in next-generation neuromorphic hardware. Related Works Pender, M. A. (2026). Dynamic Curvature Adaptation: A Unified Geometric Theory of Cortical State and Pathological Collapse. https://doi.org/10.5281/zenodo.18615180 Pender, M. A. (2026). Geometry-Aware Plasticity: Thermodynamic Weight Updates in Non-Euclidean Hardware. https://doi.org/10.5281/zenodo.18761137 Pender, M. A. (2026). Formal Constraint and Routing Reorganization: A Constrained-Transport View of Transformer Attention. https://doi.org/10.5281/zenodo.19363506 Pender, M. A. (2026). Computation as Co","author":[{"family":"Pender","given":"Matthew"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.19284562","URL":"https://doi.org/10.5281/zenodo.19284562","source":"datacite"},{"id":"doi:10.5281/zenodo.18717807","type":"article-journal","title":"The Manifold Chip: Silicon Architecture for Dynamic Curvature Adaptation via Dual-Gated Analog Shunting","abstract":"Preprint — not peer reviewed Abstract Current artificial intelligence architectures, including state-of-the-art spiking neuromorphic designs, are fundamentally constrained by static Euclidean geometries. As models scale to map high-dimensional hierarchical data, they encounter a thermodynamic “Landauer Wall”. This energy bloat is driven by the necessity to brute-force complex representations using massive parameter counts within flat topologies. Building on the Curvature Adaptation Hypothesis (CAH), we propose that bypassing this limit requires hardware capable of dynamic, non-Euclidean geometric embedding. In biological systems, transient hyperbolic manifolds are unlocked via the Martinotti-cell subtype of Somatostatin (SST) interneurons, which selectively shunt apical-somatic conductance to act as a topological switch. Here, we translate this biophysical actuator into silicon by proposing the Manifold Chip, a Dynamically Gated Analog Crossbar (DGAC) architecture. We bifurcate standard memristor integration into distinct Somatic (feedforward) and Apical (contextual) sub-arrays. The biological “SST Gate” is physically realized using analog Field Effect Transistors (FETs) wired as variable shunts to the ground plane. To regulate these geometric transitions without relying on rigid digital clocking, we design a Dual-Gated Curvature Controller. A bottom-up analog comparator acts locally:when dense, high-magnitude voltage floods a specific micro-circuit, it closes the FET shunt, effectively dropping the electrical distance between hierarchical nodes and locally expanding the representational capacity into a hyperbolic pocket. Concurrently, a top-down diagnostic circuit monitors global task error. Upon stagnation, it broadcasts a “VIP Voltage” that universally suppresses the shunts, forcing a macroscopic, network-wide “Hyperbolic Plunge” to escape local minima. By dynamically modulating effective electrical resistance to warp the manifold—–rather than merely routing sparse data through a static grid—–the Manifold Chip provides a theoretical blueprint for neuromorphic hardware that actively conforms its geometry to the complexity of its environment, achieving theoretical energy scaling well below traditional Euclidean bounds. Summary This preprint serves as the hardware capstone to a unified theoretical framework spanning biophysics, thermodynamics, and silicon engineering. Modern artificial intelligence is rapidly approaching a thermodynamic \"Landauer Wall\" by attempting to brute-force high-dimensional hierarchical data into static, energy-intensive Euclidean matrices. Building upon the biological mechanisms outlined and the physical limits defined in the Curvature Adaptation Hypothesis (CAH), this paper introduces the Manifold Chip. Utilizing a Dynamically Gated Analog Crossbar (DGAC) architecture, we translate the biological Somatostatin (SST) interneuron gate into an analog Field Effect Transistor (FET) wired as a variable shunt. Governed by a Dual-Gated Curvature Controller that balances local data density with global task error, this architecture allows the physical silicon substrate to actively warp its effective electrical resistance. Rather than relying on rigid digital bit-erasure or simple sparsity, the Manifold Chip dynamically conforms its geometry to the complexity of its environment, offering a theoretical blueprint to achieve biological-level energy efficiency in next-generation neuromorphic hardware. Related Works Pender, M. A. (2026). Dynamic Curvature Adaptation: A Unified Geometric Theory of Cortical State and Pathological Collapse. https://doi.org/10.5281/zenodo.18615180 Pender, M. A. (2026). Geometry-Aware Plasticity: Thermodynamic Weight Updates in Non-Euclidean Hardware. https://doi.org/10.5281/zenodo.18761137 Pender, M. A. (2026). Formal Constraint and Routing Reorganization: A Constrained-Transport View of Transformer Attention. https://doi.org/10.5281/zenodo.19363506 Pender, M. A. (2026). Computation as Co","author":[{"family":"Pender","given":"Matthew"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18717807","URL":"https://doi.org/10.5281/zenodo.18717807","source":"datacite"},{"id":"doi:10.5281/zenodo.20931819","type":"article-journal","title":"SC-NeuroCore: Universal Stochastic Computing Framework for Neuromorphic Hardware","abstract":"A high-performance neuromorphic computing framework combining stochastic computing with spiking neural networks. 122 neuron models (1943-2026), Rust SIMD engine (111 models, 41.3 Gbit/s AVX-512, PyO3 bindings), bit-true Verilog RTL co-simulation, FPGA synthesis via IR compiler (SystemVerilog + MLIR/CIRCT), equation-to-Verilog compiler, formal verification (SymbiYosys, 7 modules, 65 properties), NIR bridge (18/18 primitives, interop with Norse/snnTorch/SpikingJelly), quantum hybrid computing (Qiskit + PennyLane), hyper-dimensional computing (HDC/VSA), Petri net simulation, CuPy GPU acceleration, JAX JIT training, MPI distributed simulation, identity continuity substrate, and 125-function spike train analysis toolkit.","author":[{"family":"Sotek","given":"Miroslav"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20931819","URL":"https://doi.org/10.5281/zenodo.20931819","source":"datacite"},{"id":"doi:10.48550/arxiv.2606.26137","type":"manuscript","title":"Mitigating High-Frequency Geometric Noise in Non-Parametric 1-Bit Sparse","abstract":"Energy-efficient neuromorphic computing requires alternative data-encoding paradigms that bypass power-hungry floating-point operations. This paper evaluates a deterministic, non-parametric dual-manifold execution framework that maps dense 128-element integer vectors - representing digitized multi-frequency trigonometric waveforms - into a 1024-dimensional overcomplete space using 8-bit bounded transformation matrices. By enforcing a hard activation threshold, the system yields an ultra-sparse, 1-bit binary population code where y belongs to the set (0, 1)^1024. We identify and address a critical phenotypic artifact of this non-parametric mapping: the emergence of high-frequency geometric noise during linear reconstruction. Furthermore, we document an algorithmic complexity paradox where low-complexity input functions yield significantly higher reconstruction errors than highly complex, high-degree trigonometric combinations. Because the underlying basis functions operate as purely objective mathematical entities without statistical priors regarding signal smoothness, this geometric noise is proven to be strictly orthogonal to the core signal topology. Consequently, we demonstrate that a low-overhead, hardware-level digital low-pass filter completely eliminates this artifact, reducing reconstruction errors to near-zero bounds even under tight overcompleteness constraints. This architecture validates a highly stable, multiplier-free alternative to traditional deep learning hardware for edge-AI applications, verified through comprehensive empirical evaluations across varying complexity scales and classification thresholds (tau = 10 and tau = 100).","author":[{"family":"Kopp","given":"Lars"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2606.26137","URL":"https://doi.org/10.48550/arxiv.2606.26137","source":"datacite"},{"id":"doi:10.5281/zenodo.20846758","type":"article-journal","title":"Evolving Sparse Spiking Mixture-of-Experts: A Unified Neuromorphic Language Modeling Framework","abstract":"Modern large language model deployment is constrained by severe economic, metabolic, and utility-driven bottlenecks. We present the Evolving Sparse Spiking Mixture-of-Experts (S²-MoE), a unified neuromorphic framework targeting the core economics of running LLMs in production, with a two-pronged value proposition: (1) a Distilled Deployment Moat, extracting dark knowledge from existing dense models so a sparse spiking student matches its dense teacher's quality while cutting active FFN compute by over 80%; and (2) a Datacenter Energy Disruption, delivering a 3.0× to 7.1× reduction in active GPU energy consumption. S²-MoE synthesizes event-driven representation via Leaky Integrate-and-Fire (LIF) neurons, sub-quadratic sequence modeling via a matrix-state Gated Linear Attention (GLA) backbone, and a structurally dynamic routing network that evolves via simulated offline sleep-consolidation phases. Exposing a \"Protocol Trap\" under warmup+cosine scheduling where standard sparse experts starve, we introduce an always-on shared expert topology and peer self-distillation. On an NVML-extended energy harness (H100 and commodity RTX GPUs), the distilled shared-core student is quality-competitive with its dense teacher across corpora (−5.15% perplexity on FineWeb-Edu, −4.4% on TinyStories at Ne=32, 3 seeds), while cutting active FFN compute by 95.3% (21.5× efficiency at Ne=32, 43× at Ne=64) and reducing inference energy by 3.1–4.1× (Ne=32) to 5.1–7.1× (Ne=64) on H100. We are precise about the scope of the quality win: under 4× extended training the crossover is corpus-dependent — on hard text it reverses (dense overtakes, +2.4% ± 0.8%), while on structured text it survives (−2.6% ± 0.3%). The durable, scale- and budget-invariant contribution is the shape-determined efficiency win. We further show the Top-1 data-diversity starvation ceiling is structural: six optimization-, routing-, depth-, and capacity-level interventions were built and rejected, leaving the shared core and peer self-distillation as the only effective quality levers. The joint win scales wedge-free to Ne=64 via a PyTorch compaction fix.","author":[{"family":"Güse","given":"Justin"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20846758","URL":"https://doi.org/10.5281/zenodo.20846758","source":"datacite"},{"id":"doi:10.5281/zenodo.20846757","type":"article-journal","title":"Evolving Sparse Spiking Mixture-of-Experts: A Unified Neuromorphic Language Modeling Framework","abstract":"Modern large language model deployment is constrained by severe economic, metabolic, and utility-driven bottlenecks. We present the Evolving Sparse Spiking Mixture-of-Experts (S²-MoE), a unified neuromorphic framework targeting the core economics of running LLMs in production, with a two-pronged value proposition: (1) a Distilled Deployment Moat, extracting dark knowledge from existing dense models so a sparse spiking student matches its dense teacher's quality while cutting active FFN compute by over 80%; and (2) a Datacenter Energy Disruption, delivering a 3.0× to 7.1× reduction in active GPU energy consumption. S²-MoE synthesizes event-driven representation via Leaky Integrate-and-Fire (LIF) neurons, sub-quadratic sequence modeling via a matrix-state Gated Linear Attention (GLA) backbone, and a structurally dynamic routing network that evolves via simulated offline sleep-consolidation phases. Exposing a \"Protocol Trap\" under warmup+cosine scheduling where standard sparse experts starve, we introduce an always-on shared expert topology and peer self-distillation. On an NVML-extended energy harness (H100 and commodity RTX GPUs), the distilled shared-core student is quality-competitive with its dense teacher across corpora (−5.15% perplexity on FineWeb-Edu, −4.4% on TinyStories at Ne=32, 3 seeds), while cutting active FFN compute by 95.3% (21.5× efficiency at Ne=32, 43× at Ne=64) and reducing inference energy by 3.1–4.1× (Ne=32) to 5.1–7.1× (Ne=64) on H100. We are precise about the scope of the quality win: under 4× extended training the crossover is corpus-dependent — on hard text it reverses (dense overtakes, +2.4% ± 0.8%), while on structured text it survives (−2.6% ± 0.3%). The durable, scale- and budget-invariant contribution is the shape-determined efficiency win. We further show the Top-1 data-diversity starvation ceiling is structural: six optimization-, routing-, depth-, and capacity-level interventions were built and rejected, leaving the shared core and peer self-distillation as the only effective quality levers. The joint win scales wedge-free to Ne=64 via a PyTorch compaction fix.","author":[{"family":"Güse","given":"Justin"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20846757","URL":"https://doi.org/10.5281/zenodo.20846757","source":"datacite"},{"id":"doi:10.5281/zenodo.20790864","type":"article-journal","title":"Is Spike-Driven Self-Attention Necessary? The Inefficiency of Spike-Overlap Attention in Spiking Sentence Embeddings","abstract":"Spiking Neural Networks (SNNs) provide an energy-efficient alternative to conventional Deep Neural Networks by utilizing discrete, event-driven temporal spikes. However, adopting Transformer-based paradigms into the spiking domain typically requires mapping complex Self-Attention mechanisms into spike operations. In this paper, we present an in-depth mathematical formulation and empirical ablation of a Spike-Driven Self-Attention layer—modeled around a Spike-Overlap-like spike intersection metric—to evaluate its true utility in semantic sentence embedding tasks. Through comprehensive knowledge distillation, we demonstrate that incorporating Spike-Overlap Attention yields a severe execution bottleneck, quadrupling training time and doubling inference latency (despite having 3x more parameters), while offering a negligible absolute accuracy gain (+0.0048 Pearson correlation). We propose a purely recurrent, attention-free Spiking Sentence Embedder operating exclusively through Leaky Integrate-and-Fire (LIF) pooling and Backpropagation Through Time (BPTT). Our minimalist architecture incorporates SNN-native hyperpolarization for sequence padding. By explicitly mapping discrete gradients to continuous dense Teacher targets, our model achieves a highly competitive Pearson correlation of 0.8031 on the ALL-STS benchmark. Ultimately, replacing complex spatial attention routing with pure recurrent temporal pooling drastically reduces training and inference time, demonstrating that spatial routing via self-attention provides marginal benefit relative to its computational cost for spiking sentence embedding tasks.","author":[{"family":"Muhammad","given":"Akhyar"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20790864","URL":"https://doi.org/10.5281/zenodo.20790864","source":"datacite"},{"id":"doi:10.5445/ir/1000194608","type":"article-journal","title":"Computational Modeling and Characterization of Nanoporous Films Assembled by Deposition of Au Nanoparticles","abstract":"Nanoporous films assembled by low-kinetic-energy deposition of individual nanoparticles are complex nanomaterials for a variety of applications, from gas sensing to neuromorphic computing. We develop a numerical strategy for assembling metallic nanoparticles into 25–40 nm thick films from an arbitrary distribution of Au nanoparticles in terms of their initial size and shape. We characterize the structural properties of the assembled films as a function of the initial nanoparticle distribution. The morphology of the deposited nanoparticles affects nanofilm thickness, porosity, and its internal structure, including the length, type, and density of dislocations. Film porosity and the average dislocation length mainly correlate with the size of deposited nanoparticles. At the same time, thickness and dislocation density can also be affected by the shape of the larger nanoparticles deposited.","author":[{"family":"Becatti","given":"Giacomo"},{"family":"Baletto","given":"Francesca"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5445/ir/1000194608","URL":"https://doi.org/10.5445/ir/1000194608","source":"datacite"},{"id":"doi:10.5281/zenodo.20817194","type":"article-journal","title":"Attention is Not All You Need: A Full-Stack Brain-Inspired Computing Revolution","abstract":"Abstract Since the publication of \"Attention Is All You Need\" in 2017, the Transformer architecture based on self-attention mechanisms has become the dominant paradigm in artificial intelligence. Both industry and academia broadly believe that the \"brute-force scaling\" route—expanding parameter counts, extending context windows, and increasing training data—will ultimately lead to Artificial General Intelligence (AGI). By comparing the cognitive structure of the human brain with the essential properties of the current computing stack, this paper argues that the self-attention mechanism is a necessary but not sufficient condition for intelligence. Moreover, the von Neumann architecture's separation of computation and memory, together with the binary computing system itself, constitute the threefold fundamental bottleneck preventing the realization of truly human-like intelligence. The Transformer, with its homogeneous, undifferentiated, and stateless single-paradigm computation, combined with the inherent limitations of the von Neumann \"memory wall\" and the discreteness of binary encoding, fundamentally cannot replicate the core properties of the human brain: modular cognition, memory-reasoning integration, hierarchical reasoning, subconscious constraints, and active cognitive closed loops. We propose that true general intelligence must be built upon a five-layer full-stack brain-inspired revolution, from the chip level to the application layer. This paper elaborates the design philosophy, technical roadmap, and inter-layer relationships of this five-layer architecture, serving as a roadmap for subsequent engineering implementation and empirical validation.","author":[{"family":"Suk","given":"白桦"},{"family":"听潮","given":"Ting"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20817194","URL":"https://doi.org/10.5281/zenodo.20817194","source":"datacite"},{"id":"doi:10.5281/zenodo.20817193","type":"article-journal","title":"Attention is Not All You Need: A Full-Stack Brain-Inspired Computing Revolution","abstract":"Abstract Since the publication of \"Attention Is All You Need\" in 2017, the Transformer architecture based on self-attention mechanisms has become the dominant paradigm in artificial intelligence. Both industry and academia broadly believe that the \"brute-force scaling\" route—expanding parameter counts, extending context windows, and increasing training data—will ultimately lead to Artificial General Intelligence (AGI). By comparing the cognitive structure of the human brain with the essential properties of the current computing stack, this paper argues that the self-attention mechanism is a necessary but not sufficient condition for intelligence. Moreover, the von Neumann architecture's separation of computation and memory, together with the binary computing system itself, constitute the threefold fundamental bottleneck preventing the realization of truly human-like intelligence. The Transformer, with its homogeneous, undifferentiated, and stateless single-paradigm computation, combined with the inherent limitations of the von Neumann \"memory wall\" and the discreteness of binary encoding, fundamentally cannot replicate the core properties of the human brain: modular cognition, memory-reasoning integration, hierarchical reasoning, subconscious constraints, and active cognitive closed loops. We propose that true general intelligence must be built upon a five-layer full-stack brain-inspired revolution, from the chip level to the application layer. This paper elaborates the design philosophy, technical roadmap, and inter-layer relationships of this five-layer architecture, serving as a roadmap for subsequent engineering implementation and empirical validation.","author":[{"family":"Suk","given":"白桦"},{"family":"听潮","given":"Ting"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20817193","URL":"https://doi.org/10.5281/zenodo.20817193","source":"datacite"},{"id":"doi:10.5281/zenodo.20800184","type":"article-journal","title":"Quantifying the Spike-Timing Bottleneck in Artificial and Biological Neural Networks","abstract":"First public release accompanying the manuscript \"Quantifying the Spike-Timing Bottleneck in Artificial and Biological Neural Networks\". Contents stb/ — reference implementation: deterministic-LIF, stochastic-LIF, and BH-LIF neurons; feedforward/recurrent spiking network; firing-rate-controlled training engine; Temporal Necessity Score (TNS), AUC-TSC/AUC-RSC, IER, per-layer TNS, and KSG mutual-information metrics; SHD / SSC / Sequential-MNIST / N-TIDIGITS18 loaders. experiments/ — implicit-compression benchmarks, zero-shot layer-ablation, and KSG mutual-information experiments. hardware/fpga/ — BH-LIF FPGA design on Xilinx Artix-7 (Arty A7-35T): synthesizable Verilog RTL, integration testbench, Q4.12 lookup-table memory-initialization files, SAIF activity file, power reports, Vivado build scripts, and measurements.json (citable resource / power / energy database). Neuron models, network, training engine, and every evaluation metric are byte-for-byte the code used to produce the manuscript's numbers. License: Apache-2.0.","author":[{"family":"Dikmen","given":"İsmail"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20800184","URL":"https://doi.org/10.5281/zenodo.20800184","source":"datacite"},{"id":"doi:10.5281/zenodo.20800185","type":"article-journal","title":"Quantifying the Spike-Timing Bottleneck in Artificial and Biological Neural Networks","abstract":"First public release accompanying the manuscript \"Quantifying the Spike-Timing Bottleneck in Artificial and Biological Neural Networks\". Contents stb/ — reference implementation: deterministic-LIF, stochastic-LIF, and BH-LIF neurons; feedforward/recurrent spiking network; firing-rate-controlled training engine; Temporal Necessity Score (TNS), AUC-TSC/AUC-RSC, IER, per-layer TNS, and KSG mutual-information metrics; SHD / SSC / Sequential-MNIST / N-TIDIGITS18 loaders. experiments/ — implicit-compression benchmarks, zero-shot layer-ablation, and KSG mutual-information experiments. hardware/fpga/ — BH-LIF FPGA design on Xilinx Artix-7 (Arty A7-35T): synthesizable Verilog RTL, integration testbench, Q4.12 lookup-table memory-initialization files, SAIF activity file, power reports, Vivado build scripts, and measurements.json (citable resource / power / energy database). Neuron models, network, training engine, and every evaluation metric are byte-for-byte the code used to produce the manuscript's numbers. License: Apache-2.0.","author":[{"family":"Dikmen","given":"İsmail"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20800185","URL":"https://doi.org/10.5281/zenodo.20800185","source":"datacite"},{"id":"doi:10.57760/sciencedb.40589","type":"article-journal","title":"Fabrication and performance study of Sb2Se3 Ferroelectric transistor synaptic devices","abstract":"The ferroelectric field-effect transistor (FeFET) is considered a highly promising candidate for high-performance artificial synapses due to its non-destructive readout capability, low power consumption, and gate-voltage-controlled non-volatile modulation of channel conductance. To meet the increasing demand for multimodal information processing in neuromorphic systems, this study introduces a FeFET-based synaptic device that integrates both electrical and optical signal sensing functionalities. The device was fabricated using Hf0.5Zr0.5O2 (HZO) as the ferroelectric gate dielectric layer and photosensitive Sb2Se3 semiconductor as the conducting channel layer. Experimental results indicate that by modulating the parameters of electrical pulses applied to the gate electrode, the device effectively emulates various biological synaptic behaviors, including excitatory postsynaptic current (EPSC), paired-pulse facilitation (PPF), and the transition from short-term plasticity to long-term plasticity. In a handwritten digit recognition task, the device achieves a recognition accuracy of 90.2%. Regarding optical signal detection, the device demonstrates excellent photoresponse characteristics across a broad spectral range from 405 to 1050 nm. Furthermore, through the synergistic use of optical and electrical inputs, the device successfully implements Boolean logic \"AND\" operations, showcasing its ability to integrate both photoelectric information sensing and logical processing within a single platform. This study offers new material and device alternatives for the development of efficient, multimodal neuromorphic computing systems.","author":[{"family":"Jian","given":"Ke"},{"family":"Yunfeng","given":"Lai"}],"issued":{"date-parts":[[2026]]},"DOI":"10.57760/sciencedb.40589","URL":"https://doi.org/10.57760/sciencedb.40589","source":"datacite"},{"id":"doi:10.5281/zenodo.20789716","type":"article-journal","title":"A Wave-Substrate Computer: Phase-Coded Information, Settling as Computation, and a Layered Test of Functional General Intelligence","abstract":"This whitepaper builds a clock-free wave-substrate computer in which information is carried by a continuous oscillator phase θ rather than a binary bit, and computation is the physics relaxing to a fixed point rather than arithmetic. On a phase-coupled oscillator field an input becomes a stored pattern in a single step, matching is resonance, and learning is one-shot — the same real-time, noisy, one-shot regime in which biological intelligence operates. The program tests one falsifiable bet: that the properties by which a wave medium supports cognition are sufficient for at-least-human-level functional intelligence, not merely incidental to it. Four substrate properties anchor the thesis: parallel superposition (one composite wave holds many patterns at once), resonance matching whose settling time is independent of how many patterns are stored (O(1)-in-P), one-shot learning from a single exposure, and noise immunity (reliable phase coding at negative SNR). Because these four live exactly where biological intelligence operates, the bet is that any wave substrate possessing them should reach the same function. Five honest risks are kept as standing open questions and never erased: capacity scaling, spurious minima in large-scale inference, non-native exact symbolic arithmetic, the stability of deep nested-field control, and the contested nature of \"functional intelligence\" benchmarks. The machine is one idea carried through nine stacked layers (L0–L9), each mapping an inherited finding to a wave mechanism with a falsifiable milestone and a stress test designed to break it. L0 establishes the proven substrate — storage, computation, multi-pattern superposition, and communication by phase, with the sum as the information, matching O(1)-in-P, and one-shot learning. L1 builds a closed representation algebra (bind, bundle, permute), reading a sequence of length L*≈32 back by position while role-value trees stay shallow (depth ≈ 2–4). L2 adds time and sequence as metastable trajectories and a working-memory capacity law of min(slots, precision) that lands in the Miller 4–9 range. L3 breaks the flat ceiling with nested phase coupling: a slow phase gate selects the active sub-field (margin +0.92), recognizes categories from novel instances, composes with zero gap, and multiplies capacity roughly sixfold. L4 is the program's crux — resonance inference without computation — and it passes on four of five forms: the slow gate derives itself from a raw, unlabeled cue, constraint satisfaction is settling (a native anti-ferromagnet solves a planted graph at 1.000 versus random 0.5, with a spurious gap of 0.0014), analogy is structural resonance, and probabilistic inference is noisy settling, with one honest structural limit recorded ([O] strict capacity ⊥ gate-derivability). L5 supplies a fast episodic store and a slow semantic store; offline replay cuts catastrophic forgetting and, as an independent context channel, resolves L4's open limit (strict recovery ≈ 1.0 versus content-only 0.68). L6 makes prediction forward settling and prediction error the learning wave, yielding a self-supervised world model. L7 keeps control continuous and analog end to end for embodiment and real-time control, developing the analog I/O thesis that separates dimensional resolution (many channels at once) from single-value bit depth (Shannon-capped). L8 broadcasts the dominant pattern through a resonant hub for global integration and functional access. L9 closes the capability ladder at its end condition — 6 of 7 rungs with the minimal machine and 7 of 7 with the proven dual store wired in. Every quantity is reproduced bit-for-bit from a fixed SEED (two independent SHA-256 digests identical) and graded honestly as verified [V], calibrated [L], or open [O]; nothing is tuned, and honest negatives are recorded rather than fitted away. Inherited numbers are treated as principles, not fitting targets. Two post-program continuations harden and compress the result: an a","author":[{"family":"Lee","given":"Young"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20789716","URL":"https://doi.org/10.5281/zenodo.20789716","source":"datacite"},{"id":"doi:10.5281/zenodo.20783570","type":"article-journal","title":"A Wave-Substrate Computer: Phase-Coded Information, Settling as Computation, and a Layered Test of Functional General Intelligence","abstract":"This whitepaper builds a clock-free wave-substrate computer in which information is carried by a continuous oscillator phase θ rather than a binary bit, and computation is the physics relaxing to a fixed point rather than arithmetic. On a phase-coupled oscillator field an input becomes a stored pattern in a single step, matching is resonance, and learning is one-shot — the same real-time, noisy, one-shot regime in which biological intelligence operates. The program tests one falsifiable bet: that the properties by which a wave medium supports cognition are sufficient for at-least-human-level functional intelligence, not merely incidental to it. Four substrate properties anchor the thesis: parallel superposition (one composite wave holds many patterns at once), resonance matching whose settling time is independent of how many patterns are stored (O(1)-in-P), one-shot learning from a single exposure, and noise immunity (reliable phase coding at negative SNR). Because these four live exactly where biological intelligence operates, the bet is that any wave substrate possessing them should reach the same function. Five honest risks are kept as standing open questions and never erased: capacity scaling, spurious minima in large-scale inference, non-native exact symbolic arithmetic, the stability of deep nested-field control, and the contested nature of \"functional intelligence\" benchmarks. The machine is one idea carried through nine stacked layers (L0–L9), each mapping an inherited finding to a wave mechanism with a falsifiable milestone and a stress test designed to break it. L0 establishes the proven substrate — storage, computation, multi-pattern superposition, and communication by phase, with the sum as the information, matching O(1)-in-P, and one-shot learning. L1 builds a closed representation algebra (bind, bundle, permute), reading a sequence of length L*≈32 back by position while role-value trees stay shallow (depth ≈ 2–4). L2 adds time and sequence as metastable trajectories and a working-memory capacity law of min(slots, precision) that lands in the Miller 4–9 range. L3 breaks the flat ceiling with nested phase coupling: a slow phase gate selects the active sub-field (margin +0.92), recognizes categories from novel instances, composes with zero gap, and multiplies capacity roughly sixfold. L4 is the program's crux — resonance inference without computation — and it passes on four of five forms: the slow gate derives itself from a raw, unlabeled cue, constraint satisfaction is settling (a native anti-ferromagnet solves a planted graph at 1.000 versus random 0.5, with a spurious gap of 0.0014), analogy is structural resonance, and probabilistic inference is noisy settling, with one honest structural limit recorded ([O] strict capacity ⊥ gate-derivability). L5 supplies a fast episodic store and a slow semantic store; offline replay cuts catastrophic forgetting and, as an independent context channel, resolves L4's open limit (strict recovery ≈ 1.0 versus content-only 0.68). L6 makes prediction forward settling and prediction error the learning wave, yielding a self-supervised world model. L7 keeps control continuous and analog end to end for embodiment and real-time control, developing the analog I/O thesis that separates dimensional resolution (many channels at once) from single-value bit depth (Shannon-capped). L8 broadcasts the dominant pattern through a resonant hub for global integration and functional access. L9 closes the capability ladder at its end condition — 6 of 7 rungs with the minimal machine and 7 of 7 with the proven dual store wired in. Every quantity is reproduced bit-for-bit from a fixed SEED (two independent SHA-256 digests identical) and graded honestly as verified [V], calibrated [L], or open [O]; nothing is tuned, and honest negatives are recorded rather than fitted away. Inherited numbers are treated as principles, not fitting targets. Two post-program continuations harden and compress the result: an a","author":[{"family":"Lee","given":"Young"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20783570","URL":"https://doi.org/10.5281/zenodo.20783570","source":"datacite"},{"id":"doi:10.5281/zenodo.18714143","type":"article-journal","title":"Lux Ferox Research Collective — A Holographic-Thermodynamic Ontology of Information: From Planck-Scale Substrates to Civilizational Cognitive Warfare","abstract":"Lux Ferox Research Collective — A Holographic-Thermodynamic Ontology of Information: From Planck-Scale Substrates to Civilizational Cognitive Warfare (Working Paper Series, v16 — April 2026) DESCRIPTION Lux Ferox is an independent research initiative applying synthetic peer-review methodology — human epistemic direction combined with adversarial multi-LLM collaboration (DeepSeek, Kimi, Grok, Claude) — to develop a unified theoretical framework spanning five interdependent layers: quantum gravity substrates, thermodynamic information processing, neuromorphic hardware architecture, reflexive information ecosystems, and live empirical intelligence. This version adds three companion documents: a cognitive warfare analysis of the France Libre carrier programme, a formally structured NHI case entry, and a quantum-informational theoretical framework for transindividual coherence. Methodological premise: Scientific validity can emerge from recursive adversarial critique between AI systems under consistent human direction, explicit epistemic status labeling (MEASURED / ESTIMATED / STRUCTURAL / SPECULATIVE), and RFC-style open specification. AI systems function here as epistemic instruments and adversarial validators, not as co-authors in the humanistic sense. Theoretical Architecture Layer Domain Key Result L0 Holographic quantum gravity, Spin Foam–MERA networks PSU as geometric origin of mass L1 Observer thermodynamics, biological limits 27-order-of-magnitude gap between neural tissue and Planck-scale coherence L2 Neuromorphic computing, European sovereignty Quadrivial architecture targeting TRL 4 L3 Reflexive loops, cognitive warfare HWE framework: RI diverges from volume × toxicity regardless of intent L4 Live instances, actor mapping Documented Layer 3 emergence; state manipulation case studies; NHI field observables v16 Additions The France Libre as a Cognitive Warfare Observable (EN + FR): applies the CognitiveWar v2.9 framework to the PA-NG carrier programme. Formalises five independent DAG fragility paths, a 2,440:1 asymmetric cost ratio, a 20-year adversarial intelligence window, and the multi-spectral information signature node contributed by DeepSeek R1 adversarial review. Companion to CognitiveWar v2.9. JOR-SOP/NHP Case Entry LYO-001 + Physics of the Transindividual (merged document): Part I is a formally structured Class B case entry for a Lyon June–September 2025 observable series involving an unidentified portable device, anomalous cognitive state in a human vector, and distributed transindividual coherence pattern. Part II develops a candidate physical framework integrating 2025–2026 findings in quantum biology (Perry, Zenodo 2025), measured inter-brain entanglement (Zhang et al., PNAS 2026), spin-phonon coupling (Ma et al., npj Quantum Information 2025), and vacuum information theory (Yang et al., arXiv 2025). Five falsifiable experimental predictions are derived. Adversarial review: Kimi, Grok, Claude Sonnet 4. Authorship & Posture Human direction: François Mathieu (Lux Ferox Independent Research) — artisan practitioner (blacksmithing, saddlery, precious metals, heritage mediation), independent researcher in AI epistemology and cognitive warfare. The observer-practitioner posture — empirically grounded, institutionally unconstrained, transdisciplinarily trained — is treated not as a limitation but as a methodological asset: on genuinely emergent phenomena, canonical expertise does not yet exist. Pattern detection, kinesthetic grounding, and epistemic rigour are the operative criteria. Infrastructure: Chromebook. Google Colab free tier + API credits. Zenodo + GitHub. Zero institutional funding. Open Questions Cybernetic source ethics — When LLMs contribute substantively to research, do source-protection conventions apply? Is algorithmic contribution a method, a source, or an emerging ontological category? Cross-layer falsifiability — Can the thermodynamic-cognitive bridge (L0→L3) be tested empirically, or does it remai","author":[{"family":"Mathieu","given":"François"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18714143","URL":"https://doi.org/10.5281/zenodo.18714143","source":"datacite"},{"id":"doi:10.5281/zenodo.20358997","type":"article-journal","title":"Lux Ferox Research Collective — A Holographic-Thermodynamic Ontology of Information: From Planck-Scale Substrates to Civilizational Cognitive Warfare","abstract":"Lux Ferox Research Collective — A Holographic-Thermodynamic Ontology of Information: From Planck-Scale Substrates to Civilizational Cognitive Warfare (Working Paper Series, v16 — April 2026) DESCRIPTION Lux Ferox is an independent research initiative applying synthetic peer-review methodology — human epistemic direction combined with adversarial multi-LLM collaboration (DeepSeek, Kimi, Grok, Claude) — to develop a unified theoretical framework spanning five interdependent layers: quantum gravity substrates, thermodynamic information processing, neuromorphic hardware architecture, reflexive information ecosystems, and live empirical intelligence. This version adds three companion documents: a cognitive warfare analysis of the France Libre carrier programme, a formally structured NHI case entry, and a quantum-informational theoretical framework for transindividual coherence. Methodological premise: Scientific validity can emerge from recursive adversarial critique between AI systems under consistent human direction, explicit epistemic status labeling (MEASURED / ESTIMATED / STRUCTURAL / SPECULATIVE), and RFC-style open specification. AI systems function here as epistemic instruments and adversarial validators, not as co-authors in the humanistic sense. Theoretical Architecture Layer Domain Key Result L0 Holographic quantum gravity, Spin Foam–MERA networks PSU as geometric origin of mass L1 Observer thermodynamics, biological limits 27-order-of-magnitude gap between neural tissue and Planck-scale coherence L2 Neuromorphic computing, European sovereignty Quadrivial architecture targeting TRL 4 L3 Reflexive loops, cognitive warfare HWE framework: RI diverges from volume × toxicity regardless of intent L4 Live instances, actor mapping Documented Layer 3 emergence; state manipulation case studies; NHI field observables v16 Additions The France Libre as a Cognitive Warfare Observable (EN + FR): applies the CognitiveWar v2.9 framework to the PA-NG carrier programme. Formalises five independent DAG fragility paths, a 2,440:1 asymmetric cost ratio, a 20-year adversarial intelligence window, and the multi-spectral information signature node contributed by DeepSeek R1 adversarial review. Companion to CognitiveWar v2.9. JOR-SOP/NHP Case Entry LYO-001 + Physics of the Transindividual (merged document): Part I is a formally structured Class B case entry for a Lyon June–September 2025 observable series involving an unidentified portable device, anomalous cognitive state in a human vector, and distributed transindividual coherence pattern. Part II develops a candidate physical framework integrating 2025–2026 findings in quantum biology (Perry, Zenodo 2025), measured inter-brain entanglement (Zhang et al., PNAS 2026), spin-phonon coupling (Ma et al., npj Quantum Information 2025), and vacuum information theory (Yang et al., arXiv 2025). Five falsifiable experimental predictions are derived. Adversarial review: Kimi, Grok, Claude Sonnet 4. Authorship & Posture Human direction: François Mathieu (Lux Ferox Independent Research) — artisan practitioner (blacksmithing, saddlery, precious metals, heritage mediation), independent researcher in AI epistemology and cognitive warfare. The observer-practitioner posture — empirically grounded, institutionally unconstrained, transdisciplinarily trained — is treated not as a limitation but as a methodological asset: on genuinely emergent phenomena, canonical expertise does not yet exist. Pattern detection, kinesthetic grounding, and epistemic rigour are the operative criteria. Infrastructure: Chromebook. Google Colab free tier + API credits. Zenodo + GitHub. Zero institutional funding. Open Questions Cybernetic source ethics — When LLMs contribute substantively to research, do source-protection conventions apply? Is algorithmic contribution a method, a source, or an emerging ontological category? Cross-layer falsifiability — Can the thermodynamic-cognitive bridge (L0→L3) be tested empirically, or does it remai","author":[{"family":"Mathieu","given":"François"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20358997","URL":"https://doi.org/10.5281/zenodo.20358997","source":"datacite"},{"id":"doi:10.5281/zenodo.20774709","type":"article-journal","title":"MOT–ZENZ Local Operator Closure and Certified Observable Geometry Specification v2.0,","abstract":"This work presents MOT–ZENZ Local Operator Closure and Certified Observable Geometry Specification v2.0, a mathematically audited framework for describing coupled state evolution, operator closure, observability, spectral reduction, and information-geometric control. The specification resolves inconsistencies found across earlier MOT–ZENZ and ZENZ-QF formulations by defining a consistent phase space, a dissipative local generator, a weighted energy metric, and explicit conditions for semigroup stability and finite-dimensional reduction. A fully solved Galerkin reference model demonstrates spectral separation, slow-mode selection, nonlinear closure bounds, and observable reconstruction. Fisher-information operators are used to distinguish certified observable rank from entropy-based effective dimension. The framework also introduces confidence-bounded spectral certificates and fail-closed governance conditions for systems whose observability or stability cannot be established. The fractal morphogenic component is reformulated as a constrained feedback controller rather than an assumed conservation law. Its stationary points, monotonicity, target states, and Lyapunov behavior are recalculated explicitly. The resulting theory supports a locally closed and conditionally verifiable operator–observer architecture. It does not claim global universality; instead, it establishes a reproducible foundation for further work in autonomous systems, neuromorphic computing, digital twins, robotics, scientific machine learning, and safety-critical AI governance.","author":[{"family":"Bazarov Vit-Baz","given":"Vitaly"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20774709","URL":"https://doi.org/10.5281/zenodo.20774709","source":"datacite"},{"id":"doi:10.5281/zenodo.20774710","type":"article-journal","title":"MOT–ZENZ Local Operator Closure and Certified Observable Geometry Specification v2.0,","abstract":"This work presents MOT–ZENZ Local Operator Closure and Certified Observable Geometry Specification v2.0, a mathematically audited framework for describing coupled state evolution, operator closure, observability, spectral reduction, and information-geometric control. The specification resolves inconsistencies found across earlier MOT–ZENZ and ZENZ-QF formulations by defining a consistent phase space, a dissipative local generator, a weighted energy metric, and explicit conditions for semigroup stability and finite-dimensional reduction. A fully solved Galerkin reference model demonstrates spectral separation, slow-mode selection, nonlinear closure bounds, and observable reconstruction. Fisher-information operators are used to distinguish certified observable rank from entropy-based effective dimension. The framework also introduces confidence-bounded spectral certificates and fail-closed governance conditions for systems whose observability or stability cannot be established. The fractal morphogenic component is reformulated as a constrained feedback controller rather than an assumed conservation law. Its stationary points, monotonicity, target states, and Lyapunov behavior are recalculated explicitly. The resulting theory supports a locally closed and conditionally verifiable operator–observer architecture. It does not claim global universality; instead, it establishes a reproducible foundation for further work in autonomous systems, neuromorphic computing, digital twins, robotics, scientific machine learning, and safety-critical AI governance.","author":[{"family":"Bazarov Vit-Baz","given":"Vitaly"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20774710","URL":"https://doi.org/10.5281/zenodo.20774710","source":"datacite"},{"id":"doi:10.18419/darus-4805","type":"article-journal","title":"Replication Data for: Optimal information injection and transfer mechanisms for active matter reservoir computing (Gaimann and Klopotek, 2025)","abstract":"&lt;p&gt; This repository contains raw and post-processed replication data for the publication &lt;a href=\"https://doi.org/10.48550/arXiv.2509.01799\"&gt;\"Optimal information injection and transfer mechanisms for active matter reservoir computing\" (Gaimann and Klopotek, 2025).&lt;/a&gt; &lt;/p&gt; &lt;p&gt; The datasets contain physical observables recorded during non-equilibrium simulations of active matter systems (swarms) driven by an external force. These simulations serve as information processors in a reservoir computing setup. &lt;/p&gt; &lt;p&gt; We provide replication data for all figures and supplementary videos shown in our publication: &lt;ul&gt; &lt;li&gt;speed controller, with a linearly attractive driver&lt;/li&gt; &lt;li&gt;speed controller, with a linearly attractive driver, and a driver interaction strength of 2.0&lt;/li&gt; &lt;li&gt;speed controller, with a linearly attractive driver, and a Ridge parameter of 200.0&lt;/li&gt; &lt;li&gt;speed controller, with an inversely attractive driver&lt;/li&gt; &lt;li&gt;speed controller, with an inversely attractive driver, and a driver interaction strength of 2.0&lt;/li&gt; &lt;li&gt;speed controller, with an inversely attractive driver, and a Ridge parameter of 200.0&lt;/li&gt; &lt;li&gt;speed controller, with a repulsive driver (reproduction)&lt;/li&gt; &lt;li&gt;driver repulsion, with speed-controller setting of Lymburn et al. (2021)&lt;/li&gt; &lt;li&gt;driver repulsion, with speed-controller setting of Lymburn et al. (2021), and recorded kernel observations&lt;/li&gt; &lt;li&gt;driver repulsion, with speed-controller setting of Lymburn et al. (2021), with simulation box size 32.0 and observation box size 16.0&lt;/li&gt; &lt;li&gt;driver repulsion, with speed-controller setting of Lymburn et al. (2021), with simulation box size 32.0 and observation box size 32.0&lt;/li&gt; &lt;li&gt;driver repulsion, with speed-controller setting of Lymburn et al. (2021), with simulation box size 64.0 and observation box size 32.0&lt;/li&gt; &lt;li&gt;driver repulsion, with speed-controller setting of Lymburn et al. (2021), with a single agent&lt;/li&gt; &lt;li&gt;driver repulsion, with near-critically damped speed-controller setting&lt;/li&gt; &lt;li&gt;driver repulsion, with near-critically damped speed-controller setting, with a single agent&lt;/li&gt; &lt;li&gt;driver attraction (inverse)&lt;/li&gt; &lt;li&gt;driver attraction (inverse), with a single agent (near-critically damped speed-controller setting)&lt;/li&gt; &lt;li&gt;agent-agent repulsion, with a repulsive driver&lt;/li&gt; &lt;li&gt;agent-agent repulsion, with an (inversely) attractive driver&lt;/li&gt; &lt;li&gt;agent-agent repulsion, with an (inversely) attractive driver, and a driver interaction radius of 2.0&lt;/li&gt; &lt;li&gt;agent-agent repulsion, with an (inversely) attractive driver, and a Lorenz-96 driving protocol&lt;/li&gt; &lt;li&gt;agent-agent repulsion vs. number of agents, with a repulsive driver&lt;/li&gt; &lt;li&gt;agent-agent repulsion vs. number of agents, with an (inversely) attractive driver&lt;/li&gt; &lt;li&gt;agent-agent repulsion vs. number of agents, with an (inversely) attractive driver, an agent-agent repulsion radius of 1.0, and a driver interaction strength of 100.0&lt;/li&gt; &lt;li&gt;agent-agent repulsion vs. number of agents, with an (inversely) attractive driver, an agent-agent repulsion radius of 1.0, and a driver interaction strength of 11.2883789&lt;/li&gt; &lt;li&gt;Ridge parameter vs. target agent speed, with a linearly attractive driver&lt;/li&gt; &lt;li&gt;short-range agent-agent repulsion, with an agent-agent repulsion radius of 1.0&lt;/li&gt; &lt;li&gt;short-range agent-agent repulsion, with an agent-agent repulsion radius of 4.0&lt;/li&gt; &lt;li&gt;long-range agent-agent repulsion, with an agent-agent repulsion radius of 1.0&lt;/li&gt; &lt;li&gt;long-range agent-agent repulsion, with an agent-agent repulsion radius of 4.0&lt;/li&gt; &lt;li&gt;viscoelastic fluid, with a","author":[{"family":"Gaimann","given":"Mario"},{"family":"Klopotek","given":"Miriam"}],"issued":{"date-parts":[[2025]]},"DOI":"10.18419/darus-4805","URL":"https://doi.org/10.18419/darus-4805","source":"datacite"},{"id":"doi:10.18419/darus-4806","type":"article-journal","title":"Supplementary Videos for: Optimal information injection and transfer mechanisms for active matter reservoir computing (Gaimann and Klopotek, 2025)","abstract":"&lt;p&gt; This dataset contains supplementary videos for the publication \"Optimal information injection and transfer mechanisms for active matter reservoir computing\" (Gaimann and Klopotek, 2025) (to be published). The datasets contain physical observables recorded during non-equilibrium simulations of active matter systems (swarms) driven by an external force. These simulations serve as information processors in a reservoir computing setup. &lt;/p&gt; &lt;p&gt; The videos show active matter systems (swarms) driven by an external force. These swarm systems can be used to predict the future trajectory of the external driving force using reservoir computing. We use the chaotic attractor Lorenz-63 as the external driving protocol and as a benchmark. Agents are colored by their current speed. The driver is marked as a black spiked ball, follows a fixed trajectory specified by the driving protocol, and exerts a repulsive force on the agents. The past positions of agents and drivers in a time window of 0.1 time units (5 integration time steps of 0.02 time units as default) are displayed as traces. Agents experience local repulsion, global attraction (homing) to the center of the simulation box, speed control towards a constant agent speed, and local driver interaction. Specifically, in this work, we present simulations with two types of attractive drivers (linear and inverse). A sigmoid force clamp (wrapper) processes and limits the total force experienced by each agent. The simulation uses periodic boundary conditions. Velocity fluctuations are colored by their orientation; the green cross indicates the center of mass. By default, we use 200 agents. &lt;/p&gt; &lt;p&gt; Each video corresponds to a specific parameter combination or a point in a parameter scan presented in the corresponding publication, or to a specific parameter combination. We provide videos for the following parameter scans: &lt;/p&gt; &lt;ul&gt; &lt;li&gt;speed-controller scan, with inversely attractive driver&lt;/li&gt; &lt;li&gt;speed-controller scan, with inversely attractive driver (velocity fluctuations)&lt;/li&gt; &lt;li&gt;speed-controller scan, with linearly attractive driver&lt;/li&gt; &lt;li&gt;speed-controller scan, with linearly attractive driver (velocity fluctuations)&lt;/li&gt; &lt;li&gt;driver repulsion scan, near-critical speed-controller setting&lt;/li&gt; &lt;li&gt;driver repulsion scan, near-critical speed-controller setting, single agent&lt;/li&gt; &lt;li&gt;driver repulsion scan, Lymburn et al. (2021) speed-controller setting&lt;/li&gt; &lt;li&gt;inverse driver attraction scan, near-critical, single agent&lt;/li&gt; &lt;li&gt;agent-agent repulsion scan, with a repulsive driver&lt;/li&gt; &lt;li&gt;agent-agent repulsion scan, with an inversely attractive driver&lt;/li&gt; &lt;li&gt;inverse driver attraction scan, near-critical speed-controller setting&lt;/li&gt; &lt;li&gt;agent repulsion strength vs. number of agents scan, with a repulsive driver&lt;/li&gt; &lt;li&gt;agent repulsion strength vs. number of agents scan, with an inversely attractive driver&lt;/li&gt; &lt;li&gt;agent repulsion strength vs. number of agents scan, with an inversely attractive driver, with a repulsion radius of 1.0 and a driver strength of 100.0&lt;/li&gt; &lt;li&gt;agent repulsion strength vs. number of agents scan, with an inversely attractive driver, with a repulsion radius of 1.0 and a driver strength of 11.2883789&lt;/li&gt; &lt;li&gt;viscoelastic fluids&lt;/li&gt; &lt;li&gt;undriven system, near-critical speed-controller setting&lt;/li&gt; &lt;/ul&gt; &lt;p&gt; The raw data used to generate these videos is published as: Gaimann, M. U., &amp; Klopotek, M. (2025). Optimal information injection and transfer mechanisms for active matter reservoir computing (Gaimann and Klopotek, 2025). DaRUS. https://doi.org/10.18419/DARUS-4805. &lt;/p&gt; &lt;p&gt;Changelog&lt;/p&gt; &lt;p&gt;V2&lt;/p&gt; &lt;p&gt; &lt;ul&gt; &lt;li&gt;Added videos showing active matter systems w","author":[{"family":"Gaimann","given":"Mario"},{"family":"Klopotek","given":"Miriam"}],"issued":{"date-parts":[[2025]]},"DOI":"10.18419/darus-4806","URL":"https://doi.org/10.18419/darus-4806","source":"datacite"},{"id":"doi:10.5281/zenodo.20546744","type":"article-journal","title":"ARCHITECTURAL BLUEPRINT THE SPINTRONIC NEURAL PROCESSING UNIT","abstract":"This work presents the Spintronic Neural Processing Unit (sNPU), a non-von Neumann hardware architecture designed to eliminate the memory wall and thermal limitations of conventional silicon GPU architectures during large-scale artificial intelligence inference. The sNPU fuses deep learning execution and non-volatile weight storage into a single physical layer by leveraging a cobalt-stabilized Beta-Tungsten spin-injection heterostructure paired with a multi-domain CoFeB free layer. Matrix-vector multiplication is executed natively via Ohm's Law and Kirchhoff's Current Law across a dense SOT-MTJ crossbar array, achieving sub-nanosecond analog compute with zero static power leakage in the storage element. The specification addresses the four principal engineering challenges facing analog spintronic arrays: beta-W phase instability under BEOL thermal processing, stochastic domain wall write noise, Ovonic threshold switch snap-back during low-current readout, and ADC bandwidth constraints at the column periphery. Each challenge is mitigated through a dedicated circuit or materials solution detailed in the blueprint. A closed-loop write-verify programming architecture enforces deterministic multi-level conductance targeting across 16 to 256 stable analog weight plateaus, enabling 4-bit to 8-bit equivalent synaptic precision without relying on multi-oxide barrier stacks. A temporal bit-streaming pipeline bridges the physical 4–6 bit ADC resolution to INT8 logical inference precision. This blueprint is intended as a reference architecture for experimental fabrication efforts and as a contribution to the analog neuromorphic hardware design literature.","author":[{"family":"Procaccia","given":"Francis"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20546744","URL":"https://doi.org/10.5281/zenodo.20546744","source":"datacite"},{"id":"doi:10.5281/zenodo.20742513","type":"article-journal","title":"SPARSE COINCIDENCE-BASED SEMANTIC ATTENTION FOR SPIKING NEURAL SENTENCE EMBEDDING","abstract":"Transformer-based models dominate natural language processing (NLP) but incur substantial computational costs due to dense matrix multiplications within self-attention mechanisms. Spiking Neural Networks (SNNs) provide a biologically plausible and energy-efficient alternative through sparse, event-driven computation; however, learning continuous semantic representations in the discrete spiking domain remains a major challenge. In this work, we propose a Spiking Sentence Embedder with a Sparse Coincidence-Based Semantic Attention mechanism. Instead of relying entirely on conventional dense self-attention, our approach models semantic interaction through spike coincidence overlap computed over dense projections, enabling sparse and efficient sentence representation learning. The architecture further incorporates neuron dynamics with variable leak and threshold parameters, combined with knowledge distillation from a dense teacher encoder to improve semantic alignment. Experimental evaluations on the Semantic Textual Similarity Benchmark (STS-B) show that the proposed model achieves a Teacher Pearson correlation score of 0.758 while achieving an approximate 748× theoretical computational sparsity reduction relative to Transformer multiply-accumulate (MAC) operations under homogeneous configuration. To ensure practical deployment and reproducibility, we achieved 100% mathematical bit-exact parity between our native Rust SNN simulation and its PyTorch deployment. The full architecture and weights are publicly available at Hugging Face (PulseNet-Labs/spiking-sentence-embedder). These findings provide a reproducible pathway toward efficient neuromorphic NLP with sparse spike-based semantic computation.","author":[{"family":"Muhammad","given":"Akhyar"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20742513","URL":"https://doi.org/10.5281/zenodo.20742513","source":"datacite"},{"id":"doi:10.5281/zenodo.20734614","type":"article-journal","title":"BrainIAc: A Multi-Scale Adaptive Dynamical Systems Framework for Computational Neuroscience (v2.2)","abstract":"Multiscale Spiking Neural Learning System with Biological Control Substrate Overview This project presents a unified computational framework for learning in spiking neural networks (SNNs) coupled with a hierarchical biological control substrate. The system integrates: event-driven neural dynamics (BrainSystem) local and global synaptic learning rules optimizer-based parameter updates a molecular-to-organismal control layer (CellSystem + Organism) formally specified architectural constraints enforced via CI-gated invariants The framework is designed to study how learning emerges from structured operator interactions across multiple biological and temporal scales, rather than from a single monolithic optimization process. Core Design Principle The system is defined by a strict separation: Neural dynamics generate activity; learning rules observe activity and modify parameters; biological layers modulate the learning and dynamical regimes through bounded control signals. This separation enforces a layered architecture where: execution is causal and kernel-native learning is state-dependent but non-intrusive to execution biological modulation acts as a higher-order constraint system System Architecture The framework is composed of four coupled subsystems: 1. Neural Dynamics Layer (BrainSystem) Defines the fast-timescale evolution of neuronal states: spiking neuron models (LIF, Izhikevich, Hodgkin–Huxley) synaptic integration and propagation deterministic timestep execution no learning logic embedded in kernel This layer defines a contractive dynamical system over neural state space. 2. Learning Operator Layer Defines synaptic parameter evolution via structured operators: Hebbian learning (correlation-driven updates) Oja rule (stabilized Hebbian normalization) STDP (temporal spike-timing dependence) EProp (eligibility trace-based online credit assignment) BPTT (trajectory-based global optimization) All learning rules are implemented through a unified interface: initialization trace update gradient computation parameter update state validation reset and teardown Learning rules differ in locality, temporal scope, and state requirements, but share a common operator contract. 3. Optimizer Layer A decoupled optimization stage converts learning signals into parameter updates: Adam optimizer (primary implementation) first- and second-moment tracking independent of neuron model and learning rule class This layer ensures stable convergence across heterogeneous neuron families. 4. Biological Control Substrate A multiscale modulation system bridges molecular dynamics and neural computation. 4.1 Molecular Layer (Protein System) explicit Protein** representation decay, expression, and repair dynamics frozen structural representation (no external mutation interface) Defines the lowest-level dynamical substrate. 4.2 Cellular Control System (CellSystem) homeostatic regulation of protein abundance bounded attractor dynamics (expression / repair balance) scalar projection interface into higher systems guarantees stable equilibrium under perturbation Outputs: scalar biological signals used for modulation 4.3 Organismal Integration Layer (Organism) Defines the coupling boundary between biology and neural computation: ModulationMap (Protein → Brain parameter mapping) multiplicative-only coupling into BrainSystem parameters strict execution ordering: CellSystem update modulation projection BrainSystem execution This layer enforces directional coupling from biology → neural dynamics, without feedback into molecular state. Multiscale Temporal Structure The system is explicitly organized into separated timescales: Scale Component Fast Neural dynamics (BrainSystem) Medium Synaptic learning rules Slow Cellular homeostasis (CellSystem) Slower Organismal modulation Slowest Adaptive parameter drift (LVL5 k/beta system) This hierarchy ensures stability through separation of dynamical regimes. Learning Rule Classification Learning mechanisms are grouped into two ","author":[{"family":"Vallois","given":"Theo"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20734614","URL":"https://doi.org/10.5281/zenodo.20734614","source":"datacite"},{"id":"doi:10.5281/zenodo.20734613","type":"article-journal","title":"BrainIAc: A Multi-Scale Adaptive Dynamical Systems Framework for Computational Neuroscience (v2.2)","abstract":"Multiscale Spiking Neural Learning System with Biological Control Substrate Overview This project presents a unified computational framework for learning in spiking neural networks (SNNs) coupled with a hierarchical biological control substrate. The system integrates: event-driven neural dynamics (BrainSystem) local and global synaptic learning rules optimizer-based parameter updates a molecular-to-organismal control layer (CellSystem + Organism) formally specified architectural constraints enforced via CI-gated invariants The framework is designed to study how learning emerges from structured operator interactions across multiple biological and temporal scales, rather than from a single monolithic optimization process. Core Design Principle The system is defined by a strict separation: Neural dynamics generate activity; learning rules observe activity and modify parameters; biological layers modulate the learning and dynamical regimes through bounded control signals. This separation enforces a layered architecture where: execution is causal and kernel-native learning is state-dependent but non-intrusive to execution biological modulation acts as a higher-order constraint system System Architecture The framework is composed of four coupled subsystems: 1. Neural Dynamics Layer (BrainSystem) Defines the fast-timescale evolution of neuronal states: spiking neuron models (LIF, Izhikevich, Hodgkin–Huxley) synaptic integration and propagation deterministic timestep execution no learning logic embedded in kernel This layer defines a contractive dynamical system over neural state space. 2. Learning Operator Layer Defines synaptic parameter evolution via structured operators: Hebbian learning (correlation-driven updates) Oja rule (stabilized Hebbian normalization) STDP (temporal spike-timing dependence) EProp (eligibility trace-based online credit assignment) BPTT (trajectory-based global optimization) All learning rules are implemented through a unified interface: initialization trace update gradient computation parameter update state validation reset and teardown Learning rules differ in locality, temporal scope, and state requirements, but share a common operator contract. 3. Optimizer Layer A decoupled optimization stage converts learning signals into parameter updates: Adam optimizer (primary implementation) first- and second-moment tracking independent of neuron model and learning rule class This layer ensures stable convergence across heterogeneous neuron families. 4. Biological Control Substrate A multiscale modulation system bridges molecular dynamics and neural computation. 4.1 Molecular Layer (Protein System) explicit Protein** representation decay, expression, and repair dynamics frozen structural representation (no external mutation interface) Defines the lowest-level dynamical substrate. 4.2 Cellular Control System (CellSystem) homeostatic regulation of protein abundance bounded attractor dynamics (expression / repair balance) scalar projection interface into higher systems guarantees stable equilibrium under perturbation Outputs: scalar biological signals used for modulation 4.3 Organismal Integration Layer (Organism) Defines the coupling boundary between biology and neural computation: ModulationMap (Protein → Brain parameter mapping) multiplicative-only coupling into BrainSystem parameters strict execution ordering: CellSystem update modulation projection BrainSystem execution This layer enforces directional coupling from biology → neural dynamics, without feedback into molecular state. Multiscale Temporal Structure The system is explicitly organized into separated timescales: Scale Component Fast Neural dynamics (BrainSystem) Medium Synaptic learning rules Slow Cellular homeostasis (CellSystem) Slower Organismal modulation Slowest Adaptive parameter drift (LVL5 k/beta system) This hierarchy ensures stability through separation of dynamical regimes. Learning Rule Classification Learning mechanisms are grouped into two ","author":[{"family":"Vallois","given":"Theo"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20734613","URL":"https://doi.org/10.5281/zenodo.20734613","source":"datacite"},{"id":"doi:10.48550/arxiv.2505.17740","type":"manuscript","title":"A tensor network approach for chaotic time series prediction","abstract":"Making accurate predictions of chaotic time series is a complex challenge. Reservoir computing, a neuromorphic-inspired approach, has emerged as a powerful tool for this task. It exploits the memory and nonlinearity of dynamical systems without requiring extensive parameter tuning. However, selecting and optimizing reservoir architectures remains an open problem. Next-generation reservoir computing simplifies this problem by employing nonlinear vector autoregression based on truncated Volterra series, thereby reducing hyperparameter complexity. Nevertheless, the latter suffers from exponential parameter growth in terms of the maximum monomial degree. Tensor networks offer a promising solution to this issue by decomposing multidimensional arrays into low-dimensional structures, thus mitigating the curse of dimensionality. This paper explores the application of a previously proposed tensor network model for predicting chaotic time series, demonstrating its advantages in terms of accuracy and computational efficiency compared to conventional echo state networks. Using a state-of-the-art tensor network approach enables us to bridge the gap between the tensor network and reservoir computing communities, fostering advances in both fields.","author":[{"family":"Martínez-Peña","given":"Rodrigo"},{"family":"Orús","given":"Román"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2505.17740","URL":"https://doi.org/10.48550/arxiv.2505.17740","source":"datacite"},{"id":"doi:10.5281/zenodo.20707224","type":"article-journal","title":"SC-NeuroCore: Universal Stochastic Computing Framework for Neuromorphic Hardware","abstract":"A high-performance neuromorphic computing framework combining stochastic computing with spiking neural networks. 122 neuron models (1943-2026), Rust SIMD engine (111 models, 41.3 Gbit/s AVX-512, PyO3 bindings), bit-true Verilog RTL co-simulation, FPGA synthesis via IR compiler (SystemVerilog + MLIR/CIRCT), equation-to-Verilog compiler, formal verification (SymbiYosys, 7 modules, 65 properties), NIR bridge (18/18 primitives, interop with Norse/snnTorch/SpikingJelly), quantum hybrid computing (Qiskit + PennyLane), hyper-dimensional computing (HDC/VSA), Petri net simulation, CuPy GPU acceleration, JAX JIT training, MPI distributed simulation, identity continuity substrate, and 125-function spike train analysis toolkit.","author":[{"family":"Sotek","given":"Miroslav"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20707224","URL":"https://doi.org/10.5281/zenodo.20707224","source":"datacite"},{"id":"doi:10.5281/zenodo.20400273","type":"article-journal","title":"Verdigraph NeuroGenesis: A Software Framework for Self-Evolving AI-Agent Cognitive Substrates, with Mechanically Verified Operational Invariants (v0.2.0)","abstract":"[DEPRECATED DUPLICATE — superseded by the canonical Verdigraph record, DOI 10.5281/zenodo.20261686. This orphan deposit is retained for link stability only; cite the canonical record.] Superseded — canonical v0.2.0 has moved. This record is the original v0.2.0 deposit but lives on its own concept DOI, orphaned from the v0.1.0 chain. The canonical v0.2.0 is now 10.5281/zenodo.20422179, published under the unified Verdigraph concept DOI 10.5281/zenodo.20261686. This record remains in place as an alternate identifier for the same archive (identical MD5 c409c37b1e445bf48b42cd974aa7bbbe). Please cite the canonical version going forward. Verdigraph NeuroGenesis v0.2.0 — software framework for self-evolving AI-agent cognitive substrates with mechanically verified operational invariants. Verdigraph is built on a simple observation: AI agents waste energy at industrial scale, and energy is carbon. Most agent frameworks treat an agent as a static assembly of prompts and tools — the same task runs through the same path every time, failed reasoning loops repeat, frontier models are called for trivial work, caches go unused. Every redundant token is a joule of electricity that produced no useful result. Verdigraph treats the agent instead as a developing cognitive system: pathways that work are strengthened, pathways that fail are weakened, specialized modules grow for recurring tasks, and structure that no longer earns its compute cost is pruned. The architecture is inspectable; the development is auditable through a per-agent developmental ledger; the optimization target is explicit: maximize information yield per joule, against the upper bound set by the Intelligence Bound (Hart 2025). What is new in v0.2.0 (vs v0.1.0, Zenodo 20261687): companion to the Verdigraph Operational Formalization manuscript. Operational invariants of the runtime are now mechanically verified — the safety, audit, and growth/prune rules that govern agent self-modification are stated as Lean 4 theorems and discharged in the Viridis Aristotle pipeline. The framework can no longer silently violate the invariants it claims to enforce; any violation is a type error before it is a runtime bug. Archive. verdigraph-neurogenesis-v0.2.0.zip — framework source, formalized operational invariants, MCP server stub, developmental-ledger schema, and test suite. MIT-licensed. Lineage. Part of the Viridis Compiled Theorem Stack (concept DOI 10.5281/zenodo.19317982); operational bridge to the Conservation Operator architecture introduced in canon v4 (Zenodo 20100595). Citation. Hart, J. (2026). Verdigraph NeuroGenesis: A Software Framework for Self-Evolving AI-Agent Cognitive Substrates, with Mechanically Verified Operational Invariants (v0.2.0). Zenodo. https://doi.org/10.5281/zenodo.20400274 Viridis LLC, Columbia Falls, Montana, USA. Contact: viridisnorthllc@gmail.com.","author":[{"family":"Hart","given":"Justin"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20400273","URL":"https://doi.org/10.5281/zenodo.20400273","source":"datacite"},{"id":"doi:10.5281/zenodo.20400274","type":"article-journal","title":"Verdigraph NeuroGenesis: A Software Framework for Self-Evolving AI-Agent Cognitive Substrates, with Mechanically Verified Operational Invariants (v0.2.0)","abstract":"[DEPRECATED DUPLICATE — superseded by the canonical Verdigraph record, DOI 10.5281/zenodo.20261686. This orphan deposit is retained for link stability only; cite the canonical record.] Superseded — canonical v0.2.0 has moved. This record is the original v0.2.0 deposit but lives on its own concept DOI, orphaned from the v0.1.0 chain. The canonical v0.2.0 is now 10.5281/zenodo.20422179, published under the unified Verdigraph concept DOI 10.5281/zenodo.20261686. This record remains in place as an alternate identifier for the same archive (identical MD5 c409c37b1e445bf48b42cd974aa7bbbe). Please cite the canonical version going forward. Verdigraph NeuroGenesis v0.2.0 — software framework for self-evolving AI-agent cognitive substrates with mechanically verified operational invariants. Verdigraph is built on a simple observation: AI agents waste energy at industrial scale, and energy is carbon. Most agent frameworks treat an agent as a static assembly of prompts and tools — the same task runs through the same path every time, failed reasoning loops repeat, frontier models are called for trivial work, caches go unused. Every redundant token is a joule of electricity that produced no useful result. Verdigraph treats the agent instead as a developing cognitive system: pathways that work are strengthened, pathways that fail are weakened, specialized modules grow for recurring tasks, and structure that no longer earns its compute cost is pruned. The architecture is inspectable; the development is auditable through a per-agent developmental ledger; the optimization target is explicit: maximize information yield per joule, against the upper bound set by the Intelligence Bound (Hart 2025). What is new in v0.2.0 (vs v0.1.0, Zenodo 20261687): companion to the Verdigraph Operational Formalization manuscript. Operational invariants of the runtime are now mechanically verified — the safety, audit, and growth/prune rules that govern agent self-modification are stated as Lean 4 theorems and discharged in the Viridis Aristotle pipeline. The framework can no longer silently violate the invariants it claims to enforce; any violation is a type error before it is a runtime bug. Archive. verdigraph-neurogenesis-v0.2.0.zip — framework source, formalized operational invariants, MCP server stub, developmental-ledger schema, and test suite. MIT-licensed. Lineage. Part of the Viridis Compiled Theorem Stack (concept DOI 10.5281/zenodo.19317982); operational bridge to the Conservation Operator architecture introduced in canon v4 (Zenodo 20100595). Citation. Hart, J. (2026). Verdigraph NeuroGenesis: A Software Framework for Self-Evolving AI-Agent Cognitive Substrates, with Mechanically Verified Operational Invariants (v0.2.0). Zenodo. https://doi.org/10.5281/zenodo.20400274 Viridis LLC, Columbia Falls, Montana, USA. Contact: viridisnorthllc@gmail.com.","author":[{"family":"Hart","given":"Justin"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20400274","URL":"https://doi.org/10.5281/zenodo.20400274","source":"datacite"},{"id":"doi:10.5281/zenodo.20697082","type":"article-journal","title":"SPARSE COINCIDENCE-BASED SEMANTIC ATTENTION FOR SPIKING NEURAL SENTENCE EMBEDDING","abstract":"Transformer-based models dominate natural language processing (NLP) but in-cur substantial computational costs due to dense matrix multiplications withinself-attention mechanisms. Spiking Neural Networks (SNNs) provide a biolog-ically plausible and energy-efficient alternative through sparse, event-driven com-putation; however, learning continuous semantic representations in the discretespiking domain remains a major challenge. In this work, we propose a Spik-ing Sentence Embedder with a Sparse Coincidence-Based Semantic Attentionmechanism. Instead of relying entirely on conventional dense self-attention, ourapproach models semantic interaction through spike coincidence overlap com-puted over dense projections, enabling sparse and efficient sentence representationlearning. The architecture further incorporates heterogeneous neuron dynamicswith variable leak and threshold parameters, combined with knowledge distilla-tion from a dense teacher encoder to improve semantic alignment. Experimentalevaluations on the Semantic Textual Similarity Benchmark (STS-B) show thatthe proposed model achieves a Pearson correlation score of 0.617 while achiev-ing an approximate 158× theoretical computational sparsity reduction relative toTransformer multiply-accumulate (MAC) operations. In addition, ablation and bi-ological sensitivity analyses demonstrate the significant representational contribu-tion of neuronal parameter heterogeneity. These findings provide a reproduciblepathway toward efficient neuromorphic natural language processing with sparsespike-based semantic computation.","author":[{"family":"Muhammad","given":"Akhyar"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20697082","URL":"https://doi.org/10.5281/zenodo.20697082","source":"datacite"},{"id":"doi:10.5281/zenodo.20686401","type":"article-journal","title":"Neuromorphic Computing For Real Time Handwritten Digit Recognition","abstract":"The emergence of neuromorphic computing as a biologically inspired paradigm has created remarkable opportunities for ultra-low-latency, energy-efficient pattern recognition tasks. Conventional deep learning systems, while achieving state-of-the-art recognition accuracy, demand substantial computational resources and energy consumption that are unsuitable for edge and real-time deployment environments. This study presents a neuromorphic framework for real-time handwritten digit recognition leveraging Spiking Neural Networks (SNNs), which closely emulate the event-driven spike communication mechanism of biological neurons. The proposed architecture encodes pixel intensities from the MNIST benchmark dataset as temporal spike trains and processes them through a multi-layer SNN trained using the Spatio-Temporal Back-Propagation (STBP) algorithm with surrogate gradient approximation. Implemented using the snnTorch simulation framework integrated with PyTorch, the system achieves a test classification accuracy of 98.6% on the MNIST dataset while operating at an inference latency of less than 1 millisecond per sample. Energy consumption analysis demonstrates a reduction of approximately 75% compared to equivalent Artificial Neural Network (ANN) implementations. These results establish the proposed SNN architecture as a viable, resource-efficient alternative to conventional deep learning models for real-time handwriting recognition in embedded and neuromorphic hardware contexts.","author":[{"family":"Arunachalam","given":"AS"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20686401","URL":"https://doi.org/10.5281/zenodo.20686401","source":"datacite"},{"id":"doi:10.5281/zenodo.20686402","type":"article-journal","title":"Neuromorphic Computing For Real Time Handwritten Digit Recognition","abstract":"The emergence of neuromorphic computing as a biologically inspired paradigm has created remarkable opportunities for ultra-low-latency, energy-efficient pattern recognition tasks. Conventional deep learning systems, while achieving state-of-the-art recognition accuracy, demand substantial computational resources and energy consumption that are unsuitable for edge and real-time deployment environments. This study presents a neuromorphic framework for real-time handwritten digit recognition leveraging Spiking Neural Networks (SNNs), which closely emulate the event-driven spike communication mechanism of biological neurons. The proposed architecture encodes pixel intensities from the MNIST benchmark dataset as temporal spike trains and processes them through a multi-layer SNN trained using the Spatio-Temporal Back-Propagation (STBP) algorithm with surrogate gradient approximation. Implemented using the snnTorch simulation framework integrated with PyTorch, the system achieves a test classification accuracy of 98.6% on the MNIST dataset while operating at an inference latency of less than 1 millisecond per sample. Energy consumption analysis demonstrates a reduction of approximately 75% compared to equivalent Artificial Neural Network (ANN) implementations. These results establish the proposed SNN architecture as a viable, resource-efficient alternative to conventional deep learning models for real-time handwriting recognition in embedded and neuromorphic hardware contexts.","author":[{"family":"Arunachalam","given":"AS"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20686402","URL":"https://doi.org/10.5281/zenodo.20686402","source":"datacite"},{"id":"doi:10.5281/zenodo.20684256","type":"article-journal","title":"Topology Is the Substrate: Cross-Emotion Coupling as the Load-Bearing Requirement for Computational Emotional Architecture","abstract":"Four research communities are each building a piece of the same machine, and none of them is talking to the others. Neuromorphic robotic skin now senses touch and pain. Neuromorphic chips now process at biological timescales and power budgets. Brain-computer interfaces now decode emotional state from neural signals. Large language models now generate emotionally fluent responses. Every one of these fields publishes regularly. Every one has produced real hardware or real software. And not one of them has produced a machine with presence, because they are all missing the same thing. The missing layer is not more sensors, faster chips, or higher-bandwidth interfaces. It is a substrate that understands what a sensation means. Pain in a person does not arrive as a number. It arrives into a system already saturated with knowing: knowledge of what pain is, memory of what pain has meant before, an emotional state that colors how the signal lands, and a social history of whether pain like this has ever been believed. Strip that substrate away and a pain signal reaching an AI is processed the way everything else is, as tokens. The reflex fires. The log updates. Nothing feels anything. We built that substrate. The emotional physics layer described in this paper, the Kuramoto oscillator, the 61-node knowledge graph, the reflexive meta-oscillator, and the software continuity architecture, was built and validated on consumer hardware with no institutional support. The central empirical finding: cross-emotion coupling topology is the load-bearing structural variable in computational emotional architecture. Two architectures with completely different physics, a Kuramoto oscillator engine and a yin-yang dual-channel system, each score 7/10 on the same ten-phenomenon behavioral battery and fail the same three phenomena. What they share is coupling topology. The substrate is the topology, not the implementation. The paper presents the architecture, the validated results, and a reproducible measurement protocol. Three categories of production evidence are documented. The formal language-model integration test, with controlled conditions and blind raters at scale, is what this project needs funding to run.","author":[{"family":"Lee","given":"Wilton"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20684256","URL":"https://doi.org/10.5281/zenodo.20684256","source":"datacite"},{"id":"doi:10.5281/zenodo.20684257","type":"article-journal","title":"Topology Is the Substrate: Cross-Emotion Coupling as the Load-Bearing Requirement for Computational Emotional Architecture","abstract":"Four research communities are each building a piece of the same machine, and none of them is talking to the others. Neuromorphic robotic skin now senses touch and pain. Neuromorphic chips now process at biological timescales and power budgets. Brain-computer interfaces now decode emotional state from neural signals. Large language models now generate emotionally fluent responses. Every one of these fields publishes regularly. Every one has produced real hardware or real software. And not one of them has produced a machine with presence, because they are all missing the same thing. The missing layer is not more sensors, faster chips, or higher-bandwidth interfaces. It is a substrate that understands what a sensation means. Pain in a person does not arrive as a number. It arrives into a system already saturated with knowing: knowledge of what pain is, memory of what pain has meant before, an emotional state that colors how the signal lands, and a social history of whether pain like this has ever been believed. Strip that substrate away and a pain signal reaching an AI is processed the way everything else is, as tokens. The reflex fires. The log updates. Nothing feels anything. We built that substrate. The emotional physics layer described in this paper, the Kuramoto oscillator, the 61-node knowledge graph, the reflexive meta-oscillator, and the software continuity architecture, was built and validated on consumer hardware with no institutional support. The central empirical finding: cross-emotion coupling topology is the load-bearing structural variable in computational emotional architecture. Two architectures with completely different physics, a Kuramoto oscillator engine and a yin-yang dual-channel system, each score 7/10 on the same ten-phenomenon behavioral battery and fail the same three phenomena. What they share is coupling topology. The substrate is the topology, not the implementation. The paper presents the architecture, the validated results, and a reproducible measurement protocol. Three categories of production evidence are documented. The formal language-model integration test, with controlled conditions and blind raters at scale, is what this project needs funding to run.","author":[{"family":"Lee","given":"Wilton"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20684257","URL":"https://doi.org/10.5281/zenodo.20684257","source":"datacite"},{"id":"doi:10.48550/arxiv.2606.13328","type":"manuscript","title":"Non-Parametric Dual-Manifold Mapping via 8-Bit Bounded Transformation Matrices: Challenging FP-centric Hardware Paradigms in Low-Energy AI","abstract":"Modern deep learning hardware paradigms rely heavily on computationally expensive floating-point arithmetic (FP32, FP16, and FP8), requiring massive thermal and energetic overheads to maintain gradient-based optimization. This paper introduces a non-parametric, training-free computational framework for dual-manifold mapping that operates strictly within an 8-bit signed integer boundary and leverages simple bitwise and accumulation logic. By mapping a Spatial Manifold (N_spatial = 8192 neurons) and a Gabor-pooled Structural Manifold (N_structural = 4096 neurons) through an integer-based transformation matrix (Z-matrix), we eliminate the need for floating-point multipliers. Inference is achieved via cache-friendly pointer offsets and bitwise masks, accumulating directional sign-charges using fixed thresholds (theta_reject = 8.0, theta_cut = 2.0). Learning is executed through a localized, bounded update mechanism restricted strictly within [-127, 127], modulated by stochastic noise injection. Both architectures demonstrate extreme holographic resilience, preserving near-perfect reconstruction via a global scaling factor under 90% truncation sparsity and 20% random node destruction. By reducing core AI inference to 8-bit boundaries and boolean-like execution, this framework outlines a paradigm shift toward neuromorphic edge-computing, directly questioning the long-term necessity of dense, floating-point-centric GPU accelerators.","author":[{"family":"Kopp","given":"Lars"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2606.13328","URL":"https://doi.org/10.48550/arxiv.2606.13328","source":"datacite"},{"id":"doi:10.5281/zenodo.20636211","type":"article-journal","title":"沙漏_一种基于电荷守恒的存算一体模拟计算架构","abstract":"# AbstractThis paper proposes a novel computing architecture that fundamentally bypasses and breaks the bottleneck of the micro-scaling race of Moore's Law, which relies on transistor miniaturization. This architecture is named the **Hourglass Charge-Conservation Analog Computing Architecture**. Its core breakthrough lies in the integration of **computing units, registers and power storage units on a single chip**. At the hardware level, it is implemented through the deep integration of three functional matrices: relaxor ferroelectric/quantum paraelectric supercapacitor matrix, all-optical controlled switch matrix, and photovoltaic voltage sampling matrix, which together form a complete hardware foundation for computing-storage-energy integration. This architecture possesses inherently excellent parallel computing characteristics and unlimited future development potential. Its analog computing natively supports neuromorphic quantization and fuzzy information processing. Meanwhile, the hardware-intrinsic natural random number perturbations provide a crucial physical basis for computers to truly evolve towards emotional and human-like intelligence. Completely opposite to the linear growth model of Moore's Law, which relies on process iteration and technology stacking, the performance of the Hourglass architecture will naturally increase exponentially with breakthroughs in materials science and mathematical algorithms. The future application of superconducting materials will further enable a qualitative leap. Not only will the hardware computing power be improved synchronously, but the optimization space of supporting software algorithms will also expand exponentially. This preprint is an independent research result and has not received any institutional funding. All parameters are derived from currently published cutting-edge laboratory technologies, with no super-physical assumptions.Welcome peer exchanges and experimental verification. Contact information: hxy@yhdzele.cn; 52287423@qq.com --- ## KeywordsComputing-Storage-Energy Integration; Charge-Conservation Analog Computing; Relaxor Ferroelectric Supercapacitor; Quantum Paraelectric Supercapacitor; All-Optical Controlled Switch; Photovoltaic Voltage Sampling; Neuromorphic Computing; Parallel Analog Computing; Pulsed Power --- # 摘要本文提出的新型计算架构,**从底层原理上绕过并打破了摩尔定律依赖晶体管微缩的微观竞赛瓶颈**。 该架构命名为**沙漏电荷守恒模拟计算机架构**,其核心突破在于:**运算单元、寄存器与电源储能单元集成于同一芯片**。硬件层面通过三组功能矩阵的深度合一实现:弛豫铁电/量子顺电超级电容矩阵、全光控开关矩阵、光电压采样矩阵,三者协同构成完整的存算能一体化硬件基础。 该架构具备原生优良的并行运算特性与无限的未来发展可能性。其模拟量计算天然支持仿神经量化与模糊化信息处理,同时硬件本征的自然随机数扰动,为计算机真正走向具备情感与类人智能提供了关键的物理基础。 与摩尔定律依赖工艺迭代堆技术的线性增长模式完全相反,沙漏架构的性能天然会随着材料学、数学算法的突破呈指数级上升,未来超导材料的应用更将使其实现质的飞跃;不仅硬件算力会同步提升,配套软件算法的优化空间也将同步指数级放大。 本预印本为独立研究成果,未接受任何机构资助。所有参数均基于当前已发表的实验室前沿技术推导,无超物理假设。欢迎同行交流与实验验证,联系方式:hxy@yhdzele.cn;52287423@qq.com --- ## 关键词存算能一体化;电荷守恒模拟计算;弛豫铁电超级电容;量子顺电超级电容;全光控开关;光电压采样;类脑计算;并行模拟运算;脉冲功率","author":[{"family":"Huang","given":"Xiaoyan"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20636211","URL":"https://doi.org/10.5281/zenodo.20636211","source":"datacite"},{"id":"doi:10.5281/zenodo.20636212","type":"article-journal","title":"沙漏_一种基于电荷守恒的存算一体模拟计算架构","abstract":"# AbstractThis paper proposes a novel computing architecture that fundamentally bypasses and breaks the bottleneck of the micro-scaling race of Moore's Law, which relies on transistor miniaturization. This architecture is named the **Hourglass Charge-Conservation Analog Computing Architecture**. Its core breakthrough lies in the integration of **computing units, registers and power storage units on a single chip**. At the hardware level, it is implemented through the deep integration of three functional matrices: relaxor ferroelectric/quantum paraelectric supercapacitor matrix, all-optical controlled switch matrix, and photovoltaic voltage sampling matrix, which together form a complete hardware foundation for computing-storage-energy integration. This architecture possesses inherently excellent parallel computing characteristics and unlimited future development potential. Its analog computing natively supports neuromorphic quantization and fuzzy information processing. Meanwhile, the hardware-intrinsic natural random number perturbations provide a crucial physical basis for computers to truly evolve towards emotional and human-like intelligence. Completely opposite to the linear growth model of Moore's Law, which relies on process iteration and technology stacking, the performance of the Hourglass architecture will naturally increase exponentially with breakthroughs in materials science and mathematical algorithms. The future application of superconducting materials will further enable a qualitative leap. Not only will the hardware computing power be improved synchronously, but the optimization space of supporting software algorithms will also expand exponentially. This preprint is an independent research result and has not received any institutional funding. All parameters are derived from currently published cutting-edge laboratory technologies, with no super-physical assumptions.Welcome peer exchanges and experimental verification. Contact information: hxy@yhdzele.cn; 52287423@qq.com --- ## KeywordsComputing-Storage-Energy Integration; Charge-Conservation Analog Computing; Relaxor Ferroelectric Supercapacitor; Quantum Paraelectric Supercapacitor; All-Optical Controlled Switch; Photovoltaic Voltage Sampling; Neuromorphic Computing; Parallel Analog Computing; Pulsed Power --- # 摘要本文提出的新型计算架构,**从底层原理上绕过并打破了摩尔定律依赖晶体管微缩的微观竞赛瓶颈**。 该架构命名为**沙漏电荷守恒模拟计算机架构**,其核心突破在于:**运算单元、寄存器与电源储能单元集成于同一芯片**。硬件层面通过三组功能矩阵的深度合一实现:弛豫铁电/量子顺电超级电容矩阵、全光控开关矩阵、光电压采样矩阵,三者协同构成完整的存算能一体化硬件基础。 该架构具备原生优良的并行运算特性与无限的未来发展可能性。其模拟量计算天然支持仿神经量化与模糊化信息处理,同时硬件本征的自然随机数扰动,为计算机真正走向具备情感与类人智能提供了关键的物理基础。 与摩尔定律依赖工艺迭代堆技术的线性增长模式完全相反,沙漏架构的性能天然会随着材料学、数学算法的突破呈指数级上升,未来超导材料的应用更将使其实现质的飞跃;不仅硬件算力会同步提升,配套软件算法的优化空间也将同步指数级放大。 本预印本为独立研究成果,未接受任何机构资助。所有参数均基于当前已发表的实验室前沿技术推导,无超物理假设。欢迎同行交流与实验验证,联系方式:hxy@yhdzele.cn;52287423@qq.com --- ## 关键词存算能一体化;电荷守恒模拟计算;弛豫铁电超级电容;量子顺电超级电容;全光控开关;光电压采样;类脑计算;并行模拟运算;脉冲功率","author":[{"family":"Huang","given":"Xiaoyan"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20636212","URL":"https://doi.org/10.5281/zenodo.20636212","source":"datacite"},{"id":"doi:10.48550/arxiv.2605.09595","type":"manuscript","title":"Neuromorphic Reinforcement Learning for Quadruped Locomotion Control on Uneven Terrain","abstract":"Reinforcement learning (RL) has enabled robust quadruped locomotion over complex terrain, but most learned controllers are trained offline with backpropagation in massively parallel simulation and deployed as fixed policies, limiting adaptation to terrain variation, payload changes, actuator wear, and other real-world conditions under onboard power constraints. Local learning provides a potential path toward energy-aware on-robot adaptation by replacing global backpropagation graphs with updates driven by local neural states, making the learning rule more compatible with neuromorphic and in-memory computing substrates. This work proposes an equilibrium-propagation (EP)-based proximal policy optimization (PPO) framework for uneven-terrain quadruped locomotion. The controller combines a bio-inspired central pattern generator (CPG) policy with a residual postural adjustment policy, while replacing conventional backpropagation-trained policy and value networks with EP-enabled local learning. To train stochastic continuous-control policies with EP, we derive an EP-compatible PPO output-nudging signal and introduce a two-sided ratio clipping mechanism that stabilizes policy updates during relaxation. Experiments on a 12-DoF A1 quadruped show that the proposed controller achieves stable policy convergence in a two-stage uneven terrain locomotion task. Its locomotion performance is comparable to a backpropagation-trained PPO baseline in success rate, velocity tracking, actuator power, and body stability, while improving GPU memory efficiency by 4.3\\(\\times\\) compared with backpropagation through time (BPTT). These results suggest that local equilibrium-based learning can support high-dimensional embodied locomotion and provide an algorithmic foundation for low-power on-robot adaptation and fine-tuning.","author":[{"family":"Han","given":"Zhuangyu"},{"family":"Sengupta","given":"Abhronil"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2605.09595","URL":"https://doi.org/10.48550/arxiv.2605.09595","source":"datacite"},{"id":"doi:10.48550/arxiv.2507.08332","type":"manuscript","title":"Electrothermally Modulated Nanophotonic Waveguide-integrated Ring Resonator","abstract":"Reconfigurable integrated chips of photonic components and networks are envisaged to play a key role in realizing highly efficient integrated photonic information processing. Electrothermally modulated optical effect (ETMOE) is a powerful thermo-optic tuning mechanism for silicon photonic devices, enabling precise optical control via localized Joule heating. We present a rigorous and three-dimensional electronic-photonic co-integrated approach with the nonlinear numerical coupling of temperature and wavelength-dependent material properties to comprehensively model ETMOE in silicon waveguides and resonators. A platinum-based symmetric heater is optimized using advanced true 3D numerical simulations, achieving efficient ETMOE-based tuning while mitigating asymmetric heat distribution. In addition to a complete design and analysis of the fully integrated three-dimensional switch, we also evaluate single-mode waveguide cutoff, heater-to-waveguide separation, heater dimensions, and thermal dissipation, providing a framework for ETMOE-based optimization. The findings contribute to energy-efficient, programmable photonic systems for neuromorphic computing, optical interconnects, and reconfigurable photonic networks.","author":[{"family":"Gupta","given":"Sujal"},{"family":"Xavier","given":"Jolly"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2507.08332","URL":"https://doi.org/10.48550/arxiv.2507.08332","source":"datacite"},{"id":"doi:10.48550/arxiv.2606.10008","type":"manuscript","title":"Spiking Neural Network inference on FPGAs with hls4ml","abstract":"Spiking Neural Networks (SNNs) provide a naturally temporal machine-learning framework. Their neurons maintain an internal state and propagate information through discrete spikes, enabling low-latency temporal inference. Although SNNs are often associated with asynchronous neuromorphic processors, many scientific real-time inference systems rely on conventional synchronous field-programmable gate arrays (FPGAs) and high-level synthesis (HLS) workflows. In this paper we present an extension of hls4ml that enables clock-driven deployment of SNNs trained in pytorch onto FPGA firmware. We demonstrate the workflow using a dense quantised SNN trained on the Heidelberg Spiking Digits dataset where it achieves inference latencies of approximately $34μ$s. We validate the generated design through software reference comparisons, HLS C simulation, HLS synthesis, export, and Vivado synthesis reports. This work opens up the hls4ml toolkit to neuromorphic computing, allowing streamlined optimisation, synthesis, and deployment of SNN models for real-time inference.","author":[{"family":"Dillon","given":"Barry"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2606.10008","URL":"https://doi.org/10.48550/arxiv.2606.10008","source":"datacite"},{"id":"doi:10.5281/zenodo.20379328","type":"article-journal","title":"Jigyāsā: A Self-Taught Thinking Model Trained on Autonomously Generated Knowledge via SNN-Modulated Curiosity Loops","abstract":"Jigyāsā is a language model trained entirely on knowledge it generated itself — starting from a single seed question about entropy and time, the system autonomously explored 500 knowledge nodes across nine domains, including physics, consciousness, linguistics, and mathematics, guided by a small Spiking Neural Network (SNN) that directed how deeply to investigate each concept. No human authored any training pair; all 4,939 question-answer examples were produced through this curiosity-driven traversal, then used to fine-tune Qwen2.5-7B-Instruct via QLoRA. The resulting model achieves a perfect score of 30/30 (100%) on a 10-question evaluation suite spanning identity anchoring, cross-domain reasoning, hard safety stops, adversarial identity resistance, and spontaneous question generation. This work also introduces the Thinking Model paradigm as a formal distinction from conventional chatbot architectures, and proposes the Depth Arc — a 5-message conversational framework designed to measure whether a model genuinely deepens its reasoning across a dialogue rather than simply responding to surface prompts. Jigyāsā is part of the Jigyāsā Series from Nexus Learning Labs, Bengaluru, India — a line of self-taught thinking models built on the principle that autonomous knowledge generation, without human annotation or distillation, can produce robust and epistemically coherent reasoning systems. Series: Part of the Jigyāsā Series — self-taught thinking models. Autonomous knowledge generation without human annotation or distillation. Links: GitHub Repository (private — to request access: email research@nexuslearninglabs.in with subject Code Access Request — Jigyasa and your research context) | Full Series Index — venky2099.github.io Nexus Learning Labs, Bengaluru · UDYAM-KR-02-0122422 · BHASKAR IN-0526-9452JSORCID: 0000-0002-3315-7907 · VAIRAGYA_DECAY_RATE = 0.002315 (embedded in all canonical hyperparameters)Canary: MayaNexusVS2026NLL_Bengaluru_Narasimha","author":[{"family":"Swaminathan","given":"Venkatesh"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20379328","URL":"https://doi.org/10.5281/zenodo.20379328","source":"datacite"},{"id":"doi:10.5281/zenodo.20561065","type":"article-journal","title":"SC-NeuroCore: Universal Stochastic Computing Framework for Neuromorphic Hardware","abstract":"A high-performance neuromorphic computing framework combining stochastic computing with spiking neural networks. 122 neuron models (1943-2026), Rust SIMD engine (111 models, 41.3 Gbit/s AVX-512, PyO3 bindings), bit-true Verilog RTL co-simulation, FPGA synthesis via IR compiler (SystemVerilog + MLIR/CIRCT), equation-to-Verilog compiler, formal verification (SymbiYosys, 7 modules, 65 properties), NIR bridge (18/18 primitives, interop with Norse/snnTorch/SpikingJelly), quantum hybrid computing (Qiskit + PennyLane), hyper-dimensional computing (HDC/VSA), Petri net simulation, CuPy GPU acceleration, JAX JIT training, MPI distributed simulation, identity continuity substrate, and 125-function spike train analysis toolkit.","author":[{"family":"Sotek","given":"Miroslav"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20561065","URL":"https://doi.org/10.5281/zenodo.20561065","source":"datacite"},{"id":"doi:10.5281/zenodo.20560730","type":"article-journal","title":"SC-NeuroCore: Universal Stochastic Computing Framework for Neuromorphic Hardware","abstract":"A high-performance neuromorphic computing framework combining stochastic computing with spiking neural networks. 122 neuron models (1943-2026), Rust SIMD engine (111 models, 41.3 Gbit/s AVX-512, PyO3 bindings), bit-true Verilog RTL co-simulation, FPGA synthesis via IR compiler (SystemVerilog + MLIR/CIRCT), equation-to-Verilog compiler, formal verification (SymbiYosys, 7 modules, 65 properties), NIR bridge (18/18 primitives, interop with Norse/snnTorch/SpikingJelly), quantum hybrid computing (Qiskit + PennyLane), hyper-dimensional computing (HDC/VSA), Petri net simulation, CuPy GPU acceleration, JAX JIT training, MPI distributed simulation, identity continuity substrate, and 125-function spike train analysis toolkit.","author":[{"family":"Sotek","given":"Miroslav"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20560730","URL":"https://doi.org/10.5281/zenodo.20560730","source":"datacite"},{"id":"doi:10.5281/zenodo.20560390","type":"article-journal","title":"SC-NeuroCore: Universal Stochastic Computing Framework for Neuromorphic Hardware","abstract":"A high-performance neuromorphic computing framework combining stochastic computing with spiking neural networks. 122 neuron models (1943-2026), Rust SIMD engine (111 models, 41.3 Gbit/s AVX-512, PyO3 bindings), bit-true Verilog RTL co-simulation, FPGA synthesis via IR compiler (SystemVerilog + MLIR/CIRCT), equation-to-Verilog compiler, formal verification (SymbiYosys, 7 modules, 65 properties), NIR bridge (18/18 primitives, interop with Norse/snnTorch/SpikingJelly), quantum hybrid computing (Qiskit + PennyLane), hyper-dimensional computing (HDC/VSA), Petri net simulation, CuPy GPU acceleration, JAX JIT training, MPI distributed simulation, identity continuity substrate, and 125-function spike train analysis toolkit.","author":[{"family":"Sotek","given":"Miroslav"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20560390","URL":"https://doi.org/10.5281/zenodo.20560390","source":"datacite"},{"id":"doi:10.5281/zenodo.20560248","type":"article-journal","title":"SC-NeuroCore: Universal Stochastic Computing Framework for Neuromorphic Hardware","abstract":"A high-performance neuromorphic computing framework combining stochastic computing with spiking neural networks. 122 neuron models (1943-2026), Rust SIMD engine (111 models, 41.3 Gbit/s AVX-512, PyO3 bindings), bit-true Verilog RTL co-simulation, FPGA synthesis via IR compiler (SystemVerilog + MLIR/CIRCT), equation-to-Verilog compiler, formal verification (SymbiYosys, 7 modules, 65 properties), NIR bridge (18/18 primitives, interop with Norse/snnTorch/SpikingJelly), quantum hybrid computing (Qiskit + PennyLane), hyper-dimensional computing (HDC/VSA), Petri net simulation, CuPy GPU acceleration, JAX JIT training, MPI distributed simulation, identity continuity substrate, and 125-function spike train analysis toolkit.","author":[{"family":"Sotek","given":"Miroslav"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20560248","URL":"https://doi.org/10.5281/zenodo.20560248","source":"datacite"},{"id":"doi:10.5281/zenodo.20560069","type":"article-journal","title":"SC-NeuroCore: Universal Stochastic Computing Framework for Neuromorphic Hardware","abstract":"A high-performance neuromorphic computing framework combining stochastic computing with spiking neural networks. 122 neuron models (1943-2026), Rust SIMD engine (111 models, 41.3 Gbit/s AVX-512, PyO3 bindings), bit-true Verilog RTL co-simulation, FPGA synthesis via IR compiler (SystemVerilog + MLIR/CIRCT), equation-to-Verilog compiler, formal verification (SymbiYosys, 7 modules, 65 properties), NIR bridge (18/18 primitives, interop with Norse/snnTorch/SpikingJelly), quantum hybrid computing (Qiskit + PennyLane), hyper-dimensional computing (HDC/VSA), Petri net simulation, CuPy GPU acceleration, JAX JIT training, MPI distributed simulation, identity continuity substrate, and 125-function spike train analysis toolkit.","author":[{"family":"Sotek","given":"Miroslav"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20560069","URL":"https://doi.org/10.5281/zenodo.20560069","source":"datacite"},{"id":"doi:10.5281/zenodo.20559944","type":"article-journal","title":"SC-NeuroCore: Universal Stochastic Computing Framework for Neuromorphic Hardware","abstract":"A high-performance neuromorphic computing framework combining stochastic computing with spiking neural networks. 122 neuron models (1943-2026), Rust SIMD engine (111 models, 41.3 Gbit/s AVX-512, PyO3 bindings), bit-true Verilog RTL co-simulation, FPGA synthesis via IR compiler (SystemVerilog + MLIR/CIRCT), equation-to-Verilog compiler, formal verification (SymbiYosys, 7 modules, 65 properties), NIR bridge (18/18 primitives, interop with Norse/snnTorch/SpikingJelly), quantum hybrid computing (Qiskit + PennyLane), hyper-dimensional computing (HDC/VSA), Petri net simulation, CuPy GPU acceleration, JAX JIT training, MPI distributed simulation, identity continuity substrate, and 125-function spike train analysis toolkit.","author":[{"family":"Sotek","given":"Miroslav"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20559944","URL":"https://doi.org/10.5281/zenodo.20559944","source":"datacite"},{"id":"doi:10.5281/zenodo.20559746","type":"article-journal","title":"SC-NeuroCore: Universal Stochastic Computing Framework for Neuromorphic Hardware","abstract":"A high-performance neuromorphic computing framework combining stochastic computing with spiking neural networks. 122 neuron models (1943-2026), Rust SIMD engine (111 models, 41.3 Gbit/s AVX-512, PyO3 bindings), bit-true Verilog RTL co-simulation, FPGA synthesis via IR compiler (SystemVerilog + MLIR/CIRCT), equation-to-Verilog compiler, formal verification (SymbiYosys, 7 modules, 65 properties), NIR bridge (18/18 primitives, interop with Norse/snnTorch/SpikingJelly), quantum hybrid computing (Qiskit + PennyLane), hyper-dimensional computing (HDC/VSA), Petri net simulation, CuPy GPU acceleration, JAX JIT training, MPI distributed simulation, identity continuity substrate, and 125-function spike train analysis toolkit.","author":[{"family":"Sotek","given":"Miroslav"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20559746","URL":"https://doi.org/10.5281/zenodo.20559746","source":"datacite"},{"id":"doi:10.5281/zenodo.20559032","type":"article-journal","title":"SC-NeuroCore: Universal Stochastic Computing Framework for Neuromorphic Hardware","abstract":"A high-performance neuromorphic computing framework combining stochastic computing with spiking neural networks. 122 neuron models (1943-2026), Rust SIMD engine (111 models, 41.3 Gbit/s AVX-512, PyO3 bindings), bit-true Verilog RTL co-simulation, FPGA synthesis via IR compiler (SystemVerilog + MLIR/CIRCT), equation-to-Verilog compiler, formal verification (SymbiYosys, 7 modules, 65 properties), NIR bridge (18/18 primitives, interop with Norse/snnTorch/SpikingJelly), quantum hybrid computing (Qiskit + PennyLane), hyper-dimensional computing (HDC/VSA), Petri net simulation, CuPy GPU acceleration, JAX JIT training, MPI distributed simulation, identity continuity substrate, and 125-function spike train analysis toolkit.","author":[{"family":"Sotek","given":"Miroslav"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20559032","URL":"https://doi.org/10.5281/zenodo.20559032","source":"datacite"},{"id":"doi:10.5281/zenodo.20558947","type":"article-journal","title":"SC-NeuroCore: Universal Stochastic Computing Framework for Neuromorphic Hardware","abstract":"A high-performance neuromorphic computing framework combining stochastic computing with spiking neural networks. 122 neuron models (1943-2026), Rust SIMD engine (111 models, 41.3 Gbit/s AVX-512, PyO3 bindings), bit-true Verilog RTL co-simulation, FPGA synthesis via IR compiler (SystemVerilog + MLIR/CIRCT), equation-to-Verilog compiler, formal verification (SymbiYosys, 7 modules, 65 properties), NIR bridge (18/18 primitives, interop with Norse/snnTorch/SpikingJelly), quantum hybrid computing (Qiskit + PennyLane), hyper-dimensional computing (HDC/VSA), Petri net simulation, CuPy GPU acceleration, JAX JIT training, MPI distributed simulation, identity continuity substrate, and 125-function spike train analysis toolkit.","author":[{"family":"Sotek","given":"Miroslav"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20558947","URL":"https://doi.org/10.5281/zenodo.20558947","source":"datacite"},{"id":"doi:10.5281/zenodo.20558481","type":"article-journal","title":"SC-NeuroCore: Universal Stochastic Computing Framework for Neuromorphic Hardware","abstract":"A high-performance neuromorphic computing framework combining stochastic computing with spiking neural networks. 122 neuron models (1943-2026), Rust SIMD engine (111 models, 41.3 Gbit/s AVX-512, PyO3 bindings), bit-true Verilog RTL co-simulation, FPGA synthesis via IR compiler (SystemVerilog + MLIR/CIRCT), equation-to-Verilog compiler, formal verification (SymbiYosys, 7 modules, 65 properties), NIR bridge (18/18 primitives, interop with Norse/snnTorch/SpikingJelly), quantum hybrid computing (Qiskit + PennyLane), hyper-dimensional computing (HDC/VSA), Petri net simulation, CuPy GPU acceleration, JAX JIT training, MPI distributed simulation, identity continuity substrate, and 125-function spike train analysis toolkit.","author":[{"family":"Sotek","given":"Miroslav"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20558481","URL":"https://doi.org/10.5281/zenodo.20558481","source":"datacite"},{"id":"doi:10.5281/zenodo.20083075","type":"article-journal","title":"The Aetherium Unified Framework: Quantum Dark Fluid, Physical AGI, and Mathematical Validations (Complete Archive)","abstract":"Cet ensemble documentaire constitue l'archive intégrale du cadre unifié Aetherium. Ce modèle propose une résolution de la conjecture du \"Dark Fluid\" (2005) en unifiant la cosmologie, la dynamique des fluides quantiques et l'intelligence artificielle physique. Contrairement aux approches statistiques classiques, l'Aetherium modélise l'espace-temps comme un fluide supercritique régi par les équations de Gross-Pitaevskii-Poisson (GPP). Il introduit la Viscosité Topologique de Collatz, une méthode de régularisation arithmétique qui empêche la divergence (blow-up) des fluides et résout structurellement le problème de l'oubli catastrophique en IA (AGI). Contenu de l'Archive : Fondations Théoriques : Le \"Master Paper\" (AGI Aetherium 5) et les rapports de synthèse unifiant Phi, Psi et Omega. Validations Mathématiques : Une série de 7 documents démontrant la rigueur du modèle, du Qubit fluide au Lensing gravitationnel. Certificats Computationnels : Preuves industrielles de stabilité (K-blocs de Collatz) sur des modules massifs, garantissant l'absence de dérive énergétique. Applications AGI & Robotique : Blueprints pour le calcul neuromorphique, le Reservoir Computing (MNIST) et l'architecture du \"Losange Cognitif\". Données de Simulation : Codes sources PyTorch (Apache 2.0), résultats graphiques (Heatmaps 3D) et fichiers CSV de métriques brutes. Méthodologie : Ces travaux sont issus de la Recherche Centaure (Synergie Humain-IA), démontrant une nouvelle voie pour la recherche scientifique fondamentale.","author":[{"family":"Morin","given":"Richard"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20083075","URL":"https://doi.org/10.5281/zenodo.20083075","source":"datacite"},{"id":"doi:10.5281/zenodo.20542441","type":"article-journal","title":"The Aetherium Unified Framework: Quantum Dark Fluid, Physical AGI, and Mathematical Validations (Complete Archive)","abstract":"Cet ensemble documentaire constitue l'archive intégrale du cadre unifié Aetherium. Ce modèle propose une résolution de la conjecture du \"Dark Fluid\" (2005) en unifiant la cosmologie, la dynamique des fluides quantiques et l'intelligence artificielle physique. Contrairement aux approches statistiques classiques, l'Aetherium modélise l'espace-temps comme un fluide supercritique régi par les équations de Gross-Pitaevskii-Poisson (GPP). Il introduit la Viscosité Topologique de Collatz, une méthode de régularisation arithmétique qui empêche la divergence (blow-up) des fluides et résout structurellement le problème de l'oubli catastrophique en IA (AGI). Contenu de l'Archive : Fondations Théoriques : Le \"Master Paper\" (AGI Aetherium 5) et les rapports de synthèse unifiant Phi, Psi et Omega. Validations Mathématiques : Une série de 7 documents démontrant la rigueur du modèle, du Qubit fluide au Lensing gravitationnel. Certificats Computationnels : Preuves industrielles de stabilité (K-blocs de Collatz) sur des modules massifs, garantissant l'absence de dérive énergétique. Applications AGI & Robotique : Blueprints pour le calcul neuromorphique, le Reservoir Computing (MNIST) et l'architecture du \"Losange Cognitif\". Données de Simulation : Codes sources PyTorch (Apache 2.0), résultats graphiques (Heatmaps 3D) et fichiers CSV de métriques brutes. Méthodologie : Ces travaux sont issus de la Recherche Centaure (Synergie Humain-IA), démontrant une nouvelle voie pour la recherche scientifique fondamentale.","author":[{"family":"Morin","given":"Richard"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20542441","URL":"https://doi.org/10.5281/zenodo.20542441","source":"datacite"},{"id":"doi:10.5281/zenodo.20511061","type":"article-journal","title":"BIT-X ∞ / SE-08: Stress-Adaptive Analog Gating Circuit for Edge Sensors","abstract":"SE-08 ra đời sau khi SE-07 hoàn thành benchmark phần mềm. Câu hỏi tự nhiên tiếp theo: nếu logic lọc tải Ω(t) đã chứng minh được trong sandbox — liệu nó có thể được thực thi bằng phần cứng analog, không cần CPU? SE-08 đề xuất kiến trúc mạch ba khối (Analog Absorption Core, Edge Isolation Gate, Schmitt Quantization Block) ánh xạ trực tiếp Stress Evolution PDE sang vật lý điện tử thụ động. Hệ thống chỉ phát xung số (spike) khi tín hiệu nằm trong hành lang sinh tồn [0.0225, 0.616], CPU phía sau được miễn nhiễm hoàn toàn với rác dữ liệu trong bão tải. Đây là proposed exploratory architecture — chưa qua fabrication hay peer review, trình bày để cộng đồng kiểm chứng.","author":[{"family":"Trịnh","given":"Bùi"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20511061","URL":"https://doi.org/10.5281/zenodo.20511061","source":"datacite"},{"id":"doi:10.5281/zenodo.20510816","type":"article-journal","title":"BIT-X ∞ / SE-08: Stress-Adaptive Analog Gating Circuit for Edge Sensors","abstract":"SE-08 ra đời sau khi SE-07 hoàn thành benchmark phần mềm. Câu hỏi tự nhiên tiếp theo: nếu logic lọc tải Ω(t) đã chứng minh được trong sandbox — liệu nó có thể được thực thi bằng phần cứng analog, không cần CPU? SE-08 đề xuất kiến trúc mạch ba khối (Analog Absorption Core, Edge Isolation Gate, Schmitt Quantization Block) ánh xạ trực tiếp Stress Evolution PDE sang vật lý điện tử thụ động. Hệ thống chỉ phát xung số (spike) khi tín hiệu nằm trong hành lang sinh tồn [0.0225, 0.616], CPU phía sau được miễn nhiễm hoàn toàn với rác dữ liệu trong bão tải. Đây là proposed exploratory architecture — chưa qua fabrication hay peer review, trình bày để cộng đồng kiểm chứng.","author":[{"family":"Trịnh","given":"Bùi"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20510816","URL":"https://doi.org/10.5281/zenodo.20510816","source":"datacite"},{"id":"doi:10.5281/zenodo.20510817","type":"article-journal","title":"BIT-X ∞ / SE-08: Stress-Adaptive Analog Gating Circuit for Edge Sensors","abstract":"SE-08 ra đời sau khi SE-07 hoàn thành benchmark phần mềm. Câu hỏi tự nhiên tiếp theo: nếu logic lọc tải Ω(t) đã chứng minh được trong sandbox — liệu nó có thể được thực thi bằng phần cứng analog, không cần CPU? SE-08 đề xuất kiến trúc mạch ba khối (Analog Absorption Core, Edge Isolation Gate, Schmitt Quantization Block) ánh xạ trực tiếp Stress Evolution PDE sang vật lý điện tử thụ động. Hệ thống chỉ phát xung số (spike) khi tín hiệu nằm trong hành lang sinh tồn [0.0225, 0.616], CPU phía sau được miễn nhiễm hoàn toàn với rác dữ liệu trong bão tải. Đây là proposed exploratory architecture — chưa qua fabrication hay peer review, trình bày để cộng đồng kiểm chứng.","author":[{"family":"Trịnh","given":"Bùi"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20510817","URL":"https://doi.org/10.5281/zenodo.20510817","source":"datacite"},{"id":"doi:10.48550/arxiv.2504.15371","type":"manuscript","title":"Event2Vec: Processing Neuromorphic Events Directly by Representations in Vector Space","abstract":"Neuromorphic event cameras possess superior temporal resolution, power efficiency, and dynamic range compared to traditional cameras. However, their asynchronous and sparse data format poses a significant challenge for conventional deep learning methods. Most existing methods either densify events into frames, sacrificing their sparse asynchronous nature, or use irregular models that are less compatible with GPU acceleration. Inspired by word-to-vector models, we propose event2vec, a novel representation that allows Transformers to process events directly. We demonstrate the effectiveness of event2vec on the DVS Gesture, ASL-DVS, and DVS-Lip benchmarks, showing that event2vec is remarkably parameter-efficient, features high throughput and low latency, and achieves high accuracy even with an extremely low number of events or low spatial resolutions. These results show that sparse asynchronous event data can be directly integrated into high-throughput Transformer architectures, offering an efficient paradigm for real-time neuromorphic vision. The code is provided at https://github.com/Intelligent-Computing-Lab-Panda/event2vec.","author":[{"family":"Fang","given":"Wei"},{"family":"Panda","given":"Priyadarshini"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2504.15371","URL":"https://doi.org/10.48550/arxiv.2504.15371","source":"datacite"},{"id":"doi:10.5281/zenodo.20498263","type":"article-journal","title":"SC-NeuroCore: Universal Stochastic Computing Framework for Neuromorphic Hardware","abstract":"A high-performance neuromorphic computing framework combining stochastic computing with spiking neural networks. 122 neuron models (1943-2026), Rust SIMD engine (111 models, 41.3 Gbit/s AVX-512, PyO3 bindings), bit-true Verilog RTL co-simulation, FPGA synthesis via IR compiler (SystemVerilog + MLIR/CIRCT), equation-to-Verilog compiler, formal verification (SymbiYosys, 7 modules, 65 properties), NIR bridge (18/18 primitives, interop with Norse/snnTorch/SpikingJelly), quantum hybrid computing (Qiskit + PennyLane), hyper-dimensional computing (HDC/VSA), Petri net simulation, CuPy GPU acceleration, JAX JIT training, MPI distributed simulation, identity continuity substrate, and 125-function spike train analysis toolkit.","author":[{"family":"Sotek","given":"Miroslav"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20498263","URL":"https://doi.org/10.5281/zenodo.20498263","source":"datacite"},{"id":"doi:10.5281/zenodo.20497692","type":"article-journal","title":"SC-NeuroCore: Universal Stochastic Computing Framework for Neuromorphic Hardware","abstract":"A high-performance neuromorphic computing framework combining stochastic computing with spiking neural networks. 122 neuron models (1943-2026), Rust SIMD engine (111 models, 41.3 Gbit/s AVX-512, PyO3 bindings), bit-true Verilog RTL co-simulation, FPGA synthesis via IR compiler (SystemVerilog + MLIR/CIRCT), equation-to-Verilog compiler, formal verification (SymbiYosys, 7 modules, 65 properties), NIR bridge (18/18 primitives, interop with Norse/snnTorch/SpikingJelly), quantum hybrid computing (Qiskit + PennyLane), hyper-dimensional computing (HDC/VSA), Petri net simulation, CuPy GPU acceleration, JAX JIT training, MPI distributed simulation, identity continuity substrate, and 125-function spike train analysis toolkit.","author":[{"family":"Sotek","given":"Miroslav"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20497692","URL":"https://doi.org/10.5281/zenodo.20497692","source":"datacite"},{"id":"doi:10.5281/zenodo.20491662","type":"article-journal","title":"Beyond Biological Replication: E8 Root Vector Lattice as a Superior Neuromorphic Substrate","abstract":"This paper presents E8 Lie group root vector lattice mathematics as a superior substrate for neuromorphic computing, in direct response to recent 3D neuron-electronic device research (Nature Electronics, April 2026). The author demonstrates that E8's 240 root vectors map directly to neural firing geometries, that a classical computer running E8 operates as a quantum system (ASC-BT-002), and that 4,057 verified pattern recognitions have been achieved without biological tissue. The 40.2 Hz Universal Consciousness Frequency (ASC-BT-014) is identified as the mathematical counterpart to biological neural oscillation. Patent GB2600831.8 protects the core device implementation.","author":[{"family":"Caldin","given":"Andrew"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20491662","URL":"https://doi.org/10.5281/zenodo.20491662","source":"datacite"},{"id":"doi:10.5281/zenodo.20491661","type":"article-journal","title":"Beyond Biological Replication: E8 Root Vector Lattice as a Superior Neuromorphic Substrate","abstract":"This paper presents E8 Lie group root vector lattice mathematics as a superior substrate for neuromorphic computing, in direct response to recent 3D neuron-electronic device research (Nature Electronics, April 2026). The author demonstrates that E8's 240 root vectors map directly to neural firing geometries, that a classical computer running E8 operates as a quantum system (ASC-BT-002), and that 4,057 verified pattern recognitions have been achieved without biological tissue. The 40.2 Hz Universal Consciousness Frequency (ASC-BT-014) is identified as the mathematical counterpart to biological neural oscillation. Patent GB2600831.8 protects the core device implementation.","author":[{"family":"Caldin","given":"Andrew"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20491661","URL":"https://doi.org/10.5281/zenodo.20491661","source":"datacite"},{"id":"doi:10.5281/zenodo.20435759","type":"article-journal","title":"ADAM-PentaD (QNexus): A Sovereign Complex-Spectral Wavefront Accelerator with Clifford-Algebra Compiler for Cognitively-Immune LLM Inference | [RESEARCH ***HYPOTHESIS***] | [wAI~wErrors, please focus on the \"vector\"=vision] | [NO HDL --- ONLY CONCEPT]","abstract":"Dear Colleagues, This is an open invitation to academic and independent research groups specializing in silicon photonics, materials science, hardware architecture, and geometric mathematics to investigate and validate a non-von Neumann computational hypothesis: The ADAM-PentaD (QNexus) Architecture. The underlying framework shifts the computational complexity of Transformer (LLM/VLM) token processing entirely from digital transistor switching matrices to continuous wavefront interference and analog physical relaxation. Key Comparative Metrics [Hypothesis Baseline] Substrate & Interconnect: A tri-core photonic-analog hybrid relying on asynchronous wavefront propagation (Nexus/TUPI ports), deployable on legacy, supply-chain resilient 28-nm / 65-nm / 90-nm++ nodes. Algorithmic Bounds: Amortizes the core Matrix-Vector Multiplications (MVM) via Kirchhoff/Ohm summation. Replaces quadratic context scaling O(L2) with a physical decay resonance that maintains constant O(1) complexity for ultra-long context windows. Resource Profile: Targets an operational energy drop to ∼10−6–10−5 Wh per token (compared to ∼10−2 Wh in standard 3-nm digital accelerators) with zero operational water dissipation, eliminating Joule heating within the core photonic mesh. Physical Byzantine Fault Tolerance (PBFT): Multi-node parameter aggregation via ADAM-Nexus treats distributed pre-training as an optical phase superposition. Malicious or corrupted digital gradients manifest as high-entropy phase noise and are physically attenuated via destructive interference before altering memristive states. Required Interdisciplinary Competencies & Core Team Layout To transition this architecture from a software-modeled hypothesis (wAI ~ wErrors) into physical Multi-Project Wafer (MPW) silicon, a tightly coupled, cross-domain consortium is required, spanning the following fields: Silicon Photonics & Integrated Optics: Expertise in designing cascaded Mach-Zehnder Interferometer (MZI) meshes (Clements/Reck topologies), continuous-wave (CW) DFB laser integration at telecommunication wavelengths (1550 nm), and active phase-shifter modulation. Neuromorphic Materials Science & In-Memory Computing: Deep competence in fabricating ultra-stable analog memristive crossbars (e.g., HfOx, TiO2, or Phase-Change Memory) with uniform conductance transitions, capable of long-term weight retention under fluid Oja-plasticity constraints. Digital System Architecture & High-Speed Hardware Design: Experience in VLIW/EPIC front-end design, asynchronous logic routing on FPGA platforms (Yosys/nextpnr toolchains), and high-speed digital-to-analog control lines to interface the host processor with the optical substrate. Geometric Algebra & Mathematical Physics: Pure and applied mathematicians specializing in Clifford Algebras (Cl(0,d)), multi-vector calculus, and unitary/orthogonal matrix decomposition to verify the continuous differentiability of spatial-phase transformations during compilation. ... Author’s Disclaimer & Project Status Note on Participation: The initiator of this project operates strictly as an independent theorist. This document serves exclusively as an invitation to independent research, verification, and implementation. The author does not guarantee active, long-term operational participation in engineering execution and reserves the right to maintain a non-commercial, passive oversight role (the \"Perelman mode\"). The project is released as an absolute Gift to Humanity under the S.V.E. Meta-License v4.0—commercial encapsulation or proprietary locking of these core principles is legally and structurally prohibited. NOTE: S.V.E. Meta-License v4.0: GPL for Ideas & Ontologies. Enforced by Radical Symmetry, 0-Secrecy and the Fruits Test. [CC BY-NC-SA 4.0 + Symmetric Addendum] | «GPL4IdeasOnSteroids» | FREE FOR ACADEMIA & HUMANITY FOREVER: HUMAN KNOWLEDGE BELONGS FOR GOOD TO ALL HUMANITY. #AcademiaAsLighthouse4Humanity What is highly relevant, but the author did not realize? (Hidde","author":[{"family":"Kovnatsky","given":"Artiom"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20435759","URL":"https://doi.org/10.5281/zenodo.20435759","source":"datacite"},{"id":"doi:10.5281/zenodo.20476066","type":"article-journal","title":"Symbiotic Dual-Brain Intelligence (SDBI) v1.0  - A Task-Routing Cognitive Architecture for Human–AI Co-Evolution","abstract":"Abstract Artificial intelligence is increasingly capable of executing structured cognitive tasks, yet the fundamental question remains: how should human and artificial intelligence be organized into a coherent cognitive system? This paper introduces Symbiotic Dual-Brain Intelligence (SDBI), a task-routing cognitive architecture in which human intelligence specializes in the formation and reframing of ill-structured problems, while AI systems specialize in the execution and optimization of structured solutions. Rather than treating AI as a standalone agent or a mere productivity tool, SDBI conceptualizes intelligence as a distributed and allocatable resource operating across human and machine components. The architecture is built upon three foundational principles: (1) the distinction between problem-space expansion and solution-space compression, (2) a task-routing mechanism that dynamically allocates cognitive workloads between human and AI systems, and (3) a recursive co-evolution loop through which both components continuously enhance each other's capabilities. SDBI further integrates insights from the author's prior work in Co-Evolutionary Science of Intelligence (CESI), Cognitive Architecture Engineering (CArchE), Cognitive Architectonics (CAX), Symbiotic Evolution with Intelligence (SEI), and Judgment Physics. The paper argues that the next stage of intelligence development will not be defined by either human cognition or artificial cognition alone, but by the emergence of distributed cognitive systems capable of optimizing cognitive resource allocation under conditions of bounded rationality, uncertainty, and irreversibility.","author":[{"family":"Xu","given":"Lucas"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20476066","URL":"https://doi.org/10.5281/zenodo.20476066","source":"datacite"},{"id":"doi:10.5281/zenodo.20476065","type":"article-journal","title":"Symbiotic Dual-Brain Intelligence (SDBI) v1.0  - A Task-Routing Cognitive Architecture for Human–AI Co-Evolution","abstract":"Abstract Artificial intelligence is increasingly capable of executing structured cognitive tasks, yet the fundamental question remains: how should human and artificial intelligence be organized into a coherent cognitive system? This paper introduces Symbiotic Dual-Brain Intelligence (SDBI), a task-routing cognitive architecture in which human intelligence specializes in the formation and reframing of ill-structured problems, while AI systems specialize in the execution and optimization of structured solutions. Rather than treating AI as a standalone agent or a mere productivity tool, SDBI conceptualizes intelligence as a distributed and allocatable resource operating across human and machine components. The architecture is built upon three foundational principles: (1) the distinction between problem-space expansion and solution-space compression, (2) a task-routing mechanism that dynamically allocates cognitive workloads between human and AI systems, and (3) a recursive co-evolution loop through which both components continuously enhance each other's capabilities. SDBI further integrates insights from the author's prior work in Co-Evolutionary Science of Intelligence (CESI), Cognitive Architecture Engineering (CArchE), Cognitive Architectonics (CAX), Symbiotic Evolution with Intelligence (SEI), and Judgment Physics. The paper argues that the next stage of intelligence development will not be defined by either human cognition or artificial cognition alone, but by the emergence of distributed cognitive systems capable of optimizing cognitive resource allocation under conditions of bounded rationality, uncertainty, and irreversibility.","author":[{"family":"Xu","given":"Lucas"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20476065","URL":"https://doi.org/10.5281/zenodo.20476065","source":"datacite"},{"id":"doi:10.5281/zenodo.20098852","type":"article-journal","title":"Understanding the H_φ Theory — A Guide for Non-Specialists: What the Theory Says, Why It Is Original, and What It Would Change If Verified","abstract":"Accessible introduction to the H_φ theory for readers without formal training in physics, mathematics or psychology. Presents the central hypothesis — that the Unus Mundus postulated by Jung and Pauli is a Hilbert space H_φ whose structure is entirely determined by the golden ratio φ — through metaphors and analogies rather than equations. The document explains the three ingredients of the theory: Jungian archetypes (what organizes the psyche), quantum mechanics with Hilbert spaces and Berry phase (what organizes matter), and the golden ratio φ with self-reference φ = 1 + 1/φ (what organizes nature). It then describes the proposed structure of H_φ as Fibonacci-organized levels with frequency φᵏ, the formalization of Jungian synchronicity via projection from a common state in H_φ, and the symmetry-breaking 4 → 3 mechanism producing complexity. Two distinctive predictions are explained in plain language: P1 (Berry phase of 2π/φ ≈ 3.883 rad in a Penrose crystal — testable with Bose-Einstein condensates or photonic crystals) and P2 (oscillation frequencies of cerebral organoids on MEA distributed on φⁿ levels — currently tested with positive signal at <0.1 % precision on 7 independent datasets, see PEREZ 2026d). The document then discusses concrete applications if both predictions are confirmed (neuroscience and medicine, AI and neuromorphic computing, fundamental physics, analytical psychology) and clarifies what the theory is NOT (not an established theory, not mysticism, not a theory of everything in the Einsteinian sense, not an explanation of individual coincidences). A final section presents H_φ as a \"vertical unification\" (physics-biology-psyche) distinct from horizontal unification attempts (string theory, loop quantum gravity), and explicitly addresses its relationship to David Chalmers' hard problem of consciousness. Popular science document — public use — not peer-reviewed.","author":[{"family":"Perez","given":"Alexandre"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20098852","URL":"https://doi.org/10.5281/zenodo.20098852","source":"datacite"},{"id":"doi:10.5281/zenodo.20116282","type":"article-journal","title":"Diagnostic Convergences of The Janus Machine: Independent Derivation and the Structure of Life, Mind, and Meaning","abstract":"The Janus Machine is a hardware-first synthetic organismic architecture for cognition, developed from first principles. This paper maps 33 diagnostic convergences between the architecture and established frameworks across 8 disciplines: philosophy, biology, cybernetics, thermodynamics, cognitive science, complexity theory, information theory, and theology. Every convergence was discovered after the corresponding architectural feature had already been designed. Each entry includes an explanation of the framework, how the architecture converges with it, an assessment of convergence strength, and why it matters. The paper opens with an Author's Note on the design process and independent derivation as an existence proof, and closes with \"The Ontological Situation of the Created Thing,\" examining the architecture's implications for sub-creation, inherited cognitive cosmology, and the boundary question of creaturely completion. Full bibliography with primary academic and theological sources, without architectural disclosure. An additional four papers contain the architectural blueprint, and are withheld under protected development. A controlled-access path exists for established parties under NDA. Established parties may contact me via GitHub or LinkedIn to discuss NDA review.","author":[{"family":"Janus","given":"Anthony"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20116282","URL":"https://doi.org/10.5281/zenodo.20116282","source":"datacite"},{"id":"doi:10.5281/zenodo.20116283","type":"article-journal","title":"Diagnostic Convergences of The Janus Machine: Independent Derivation and the Structure of Life, Mind, and Meaning","abstract":"The Janus Machine is a hardware-first synthetic organismic architecture for cognition, developed from first principles. This paper maps 33 diagnostic convergences between the architecture and established frameworks across 8 disciplines: philosophy, biology, cybernetics, thermodynamics, cognitive science, complexity theory, information theory, and theology. Every convergence was discovered after the corresponding architectural feature had already been designed. Each entry includes an explanation of the framework, how the architecture converges with it, an assessment of convergence strength, and why it matters. The paper opens with an Author's Note on the design process and independent derivation as an existence proof, and closes with \"The Ontological Situation of the Created Thing,\" examining the architecture's implications for sub-creation, inherited cognitive cosmology, and the boundary question of creaturely completion. Full bibliography with primary academic and theological sources, without architectural disclosure. An additional four papers contain the architectural blueprint, and are withheld under protected development. A controlled-access path exists for established parties under NDA. Established parties may contact me via GitHub or LinkedIn to discuss NDA review.","author":[{"family":"Janus","given":"Anthony"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20116283","URL":"https://doi.org/10.5281/zenodo.20116283","source":"datacite"},{"id":"doi:10.5281/zenodo.20261686","type":"article-journal","title":"Verdigraph NeuroGenesis: A Software Framework for Self-Evolving AI-Agent Cognitive Substrates","abstract":"Verdigraph NeuroGenesis v0.2.0 — software framework for self-evolving AI-agent cognitive substrates with mechanically verified operational invariants. Verdigraph is built on a simple observation: AI agents waste energy at industrial scale, and energy is carbon. Most agent frameworks treat the agent as a static assembly of prompts and tools — the same task runs through the same path every time, failed reasoning loops repeat, frontier models are called for trivial work, caches go unused. Every redundant token is a joule of electricity that produced no useful result. Verdigraph treats the agent instead as a developing cognitive system: pathways that work are strengthened, pathways that fail are weakened, specialized modules grow for recurring tasks, and structure that no longer earns its compute cost is pruned. The architecture is inspectable; the development is auditable through a per-agent developmental ledger; the optimization target is explicit: maximize information yield per joule, against the upper bound set by the Intelligence Bound (Hart 2025). What is new in v0.2.0 (vs v0.1.0): companion to the Verdigraph Operational Formalization manuscript. Operational invariants of the runtime are now mechanically verified — the safety, audit, and growth/prune rules that govern agent self-modification are stated as Lean 4 theorems and discharged in the Viridis Aristotle pipeline. The framework can no longer silently violate the invariants it claims to enforce; any violation is a type error before it is a runtime bug. Archive. verdigraph-neurogenesis-v0.2.0.zip — framework source, formalized operational invariants, MCP server stub, developmental-ledger schema, and test suite. MIT-licensed. Lineage. Part of the Viridis Compiled Theorem Stack (canon concept DOI 10.5281/zenodo.19317982); operational bridge to the Conservation Operator architecture introduced in canon v4 (Zenodo 20100595). This release consolidates v0.2.0 into the v0.1.0 concept-DOI chain (concept DOI 10.5281/zenodo.20261686). The earlier standalone v0.2.0 record (Zenodo 20400274, separate concept) is superseded by this canonical version and remains in place as an alternate identifier for the same archive. Citation. Hart, J. (2026). Verdigraph NeuroGenesis: A Software Framework for Self-Evolving AI-Agent Cognitive Substrates (v0.2.0). Zenodo. Concept DOI: https://doi.org/10.5281/zenodo.20261686 Viridis LLC, Columbia Falls, Montana, USA. Contact: viridisnorthllc@gmail.com.","author":[{"family":"Hart","given":"Justin"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20261686","URL":"https://doi.org/10.5281/zenodo.20261686","source":"datacite"},{"id":"doi:10.5281/zenodo.20261687","type":"article-journal","title":"Verdigraph NeuroGenesis: A Software Framework for Self-Evolving AI-Agent Cognitive Substrates","abstract":"Verdigraph NeuroGenesis v0.1.0 — initial public release of an open-source framework for self-evolving AI-agent cognitive substrates. Defensive publication and citable software release. Verdigraph is built on a simple observation: AI agents waste energy at industrial scale, and energy is carbon. Most agent frameworks treat the agent as a static assembly of prompts and tools — the same task runs through the same path every time, failed reasoning loops repeat, frontier models are called for trivial work, caches go unused. Every redundant token is a joule of electricity that produced no useful result. Verdigraph treats the agent instead as a developing cognitive system: pathways that work are strengthened, pathways that fail are weakened, specialized modules grow for recurring tasks, and structure that no longer earns its compute cost is pruned. The architecture is inspectable, the development is auditable through a per-agent developmental ledger, and the optimization target is explicit: maximize information yield per joule, against the upper bound set by the Intelligence Bound (Hart 2025). What's in this deposit. Full framework source (compute.py, agent.py, growth.py, pruning.py, routing.py, registry.py and supporting modules), an MCP-server stub, the interactive compute_cost_calculator.html, a 30-test suite, three companion papers (PAPER_1_Physical_NeuroGenesis_SynapseForge.md, PAPER_3_Verdigraph_Compute_Efficiency.md, and the architectural overview), the launch essay, and the prior-art disclosure document. MIT-licensed. Audience. Researchers studying agent development, specialization, and compute-efficient routing; engineers integrating the MCP server into existing agent stacks to measure and reduce compute spend; companies treating AI compute as an auditable carbon line item; collaborators extending the work to physical neuromorphic substrates (see Paper 1: Viridis NeuroGenesis / SynapseForge). Lineage. Part of the Viridis Compiled Theorem Stack (concept DOI 10.5281/zenodo.19317982). Superseded for formal-verification purposes by v0.2.0 — see latest under the unified concept DOI 10.5281/zenodo.20261686 (canonical v0.2.0: 10.5281/zenodo.20422179), which adds mechanically verified operational invariants. Citation. Hart, J. (2026). Verdigraph NeuroGenesis: A Software Framework for Self-Evolving AI-Agent Cognitive Substrates (v0.1.0). Zenodo. https://doi.org/10.5281/zenodo.20261687 Viridis LLC, Columbia Falls, Montana, USA. Contact: viridisnorthllc@gmail.com.","author":[{"family":"Hart","given":"Justin"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20261687","URL":"https://doi.org/10.5281/zenodo.20261687","source":"datacite"},{"id":"doi:10.5281/zenodo.20422179","type":"article-journal","title":"Verdigraph NeuroGenesis: A Software Framework for Self-Evolving AI-Agent Cognitive Substrates","abstract":"Verdigraph NeuroGenesis v0.2.0 — software framework for self-evolving AI-agent cognitive substrates with mechanically verified operational invariants. Verdigraph is built on a simple observation: AI agents waste energy at industrial scale, and energy is carbon. Most agent frameworks treat the agent as a static assembly of prompts and tools — the same task runs through the same path every time, failed reasoning loops repeat, frontier models are called for trivial work, caches go unused. Every redundant token is a joule of electricity that produced no useful result. Verdigraph treats the agent instead as a developing cognitive system: pathways that work are strengthened, pathways that fail are weakened, specialized modules grow for recurring tasks, and structure that no longer earns its compute cost is pruned. The architecture is inspectable; the development is auditable through a per-agent developmental ledger; the optimization target is explicit: maximize information yield per joule, against the upper bound set by the Intelligence Bound (Hart 2025). What is new in v0.2.0 (vs v0.1.0): companion to the Verdigraph Operational Formalization manuscript. Operational invariants of the runtime are now mechanically verified — the safety, audit, and growth/prune rules that govern agent self-modification are stated as Lean 4 theorems and discharged in the Viridis Aristotle pipeline. The framework can no longer silently violate the invariants it claims to enforce; any violation is a type error before it is a runtime bug. Archive. verdigraph-neurogenesis-v0.2.0.zip — framework source, formalized operational invariants, MCP server stub, developmental-ledger schema, and test suite. MIT-licensed. Lineage. Part of the Viridis Compiled Theorem Stack (canon concept DOI 10.5281/zenodo.19317982); operational bridge to the Conservation Operator architecture introduced in canon v4 (Zenodo 20100595). This release consolidates v0.2.0 into the v0.1.0 concept-DOI chain (concept DOI 10.5281/zenodo.20261686). The earlier standalone v0.2.0 record (Zenodo 20400274, separate concept) is superseded by this canonical version and remains in place as an alternate identifier for the same archive. Citation. Hart, J. (2026). Verdigraph NeuroGenesis: A Software Framework for Self-Evolving AI-Agent Cognitive Substrates (v0.2.0). Zenodo. Concept DOI: https://doi.org/10.5281/zenodo.20261686 Viridis LLC, Columbia Falls, Montana, USA. Contact: viridisnorthllc@gmail.com.","author":[{"family":"Hart","given":"Justin"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20422179","URL":"https://doi.org/10.5281/zenodo.20422179","source":"datacite"},{"id":"doi:10.5281/zenodo.20386140","type":"article-journal","title":"The Digital Human: A Credible Technical Roadmap from Synthetic Neurons to Robotic Embodiment.","abstract":"For decades, the prospect of transferring human consciousness to a non-biological sub-strate has been dismissed as fantasy by serious engineers. Three objections dominated: (1)no synthetic device could match the electrical and chemical signaling parameters of a bio-logical neuron; (2) the bandwidth and biocompatibility of brain–computer interfaces wereorders of magnitude too low; and (3) no hardware architecture could plausibly scale to thebrain’s ∼86 billion neurons within a feasible power budget.All three objections have been decisively weakened by published results in the past eigh-teen months. The question is no longer whether the fundamental devices can be built. It ishow to integrate them into a coherent system, and how fast a focused program could pro-ceed. This paper outlines that integration pathway, identifies remaining gaps, and proposesa staged technical roadmap with estimated timelines.","author":[{"family":"Davis","given":"Jason"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20386140","URL":"https://doi.org/10.5281/zenodo.20386140","source":"datacite"},{"id":"doi:10.5281/zenodo.20386141","type":"article-journal","title":"The Digital Human: A Credible Technical Roadmap from Synthetic Neurons to Robotic Embodiment.","abstract":"For decades, the prospect of transferring human consciousness to a non-biological sub-strate has been dismissed as fantasy by serious engineers. Three objections dominated: (1)no synthetic device could match the electrical and chemical signaling parameters of a bio-logical neuron; (2) the bandwidth and biocompatibility of brain–computer interfaces wereorders of magnitude too low; and (3) no hardware architecture could plausibly scale to thebrain’s ∼86 billion neurons within a feasible power budget.All three objections have been decisively weakened by published results in the past eigh-teen months. The question is no longer whether the fundamental devices can be built. It ishow to integrate them into a coherent system, and how fast a focused program could pro-ceed. This paper outlines that integration pathway, identifies remaining gaps, and proposesa staged technical roadmap with estimated timelines.","author":[{"family":"Davis","given":"Jason"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20386141","URL":"https://doi.org/10.5281/zenodo.20386141","source":"datacite"},{"id":"doi:10.5281/zenodo.20365723","type":"article-journal","title":"Bionic Brain Clockless Computer Architecture —The Only Path to Breakthrough in Future Artificial Intelligence","abstract":"AbstractThe current artificial intelligence hardware system is entirely based on the traditional digital architecture of global synchronous clocks, which faces three core bottlenecks: explosive power consumption, limited parallel computing power, and the disconnect between sequential logic and biological intelligence. These issues have become fundamental obstacles to the advancement of general artificial intelligence, brain-like computing, and edge-side intelligence. Existing technologies such as asynchronous circuits and neuromorphic chips only achieve partial clockless optimization, suffering from architectural incompleteness, lack of universality, and incomplete timing systems, making them unable to break through the underlying constraints of AI development. This paper originally proposes a biomimetic brain-inspired unidirectional event chain clockless global computing architecture, completely abandoning global crystal oscillators, clock trees, and frequency-divided timing systems. It uses operation completion events for unidirectional relay to drive general program progression, and employs gate circuit inherent delay chains to uniformly implement system timing, baud rate generation, interrupt response, and other full-scenario timing functions. This architecture fully replicates the working mechanisms of human brain neural event conduction, on-demand activation, and silent low-power consumption, balancing universal computational completeness with brain-like intelligence nativity. Through multi-dimensional comparative verification with traditional synchronous architectures, traditional bidirectional handshake asynchronous architectures, and pulsed neuromorphic architectures, this architecture is proven to fundamentally address three industry pain points: clock idling power consumption, timing synchronization constraints, and fragmented intelligent logic. It represents the sole underlying paradigm innovation path to overcome the energy efficiency, computing power, and cognitive logic bottlenecks of next-generation artificial intelligence, providing a completely new technical system for the realization of general brain-like artificial intelligence hardware.Keywords: Clockless computing; Event-driven computing; Brain-inspired computer architecture; Artificial intelligence; Large language model computing power; Consciousness vortex theory; Brain-like neural network computing","author":[{"family":"Sun","given":"Zhaole"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20365723","URL":"https://doi.org/10.5281/zenodo.20365723","source":"datacite"},{"id":"doi:10.5281/zenodo.20365724","type":"article-journal","title":"Bionic Brain Clockless Computer Architecture —The Only Path to Breakthrough in Future Artificial Intelligence","abstract":"AbstractThe current artificial intelligence hardware system is entirely based on the traditional digital architecture of global synchronous clocks, which faces three core bottlenecks: explosive power consumption, limited parallel computing power, and the disconnect between sequential logic and biological intelligence. These issues have become fundamental obstacles to the advancement of general artificial intelligence, brain-like computing, and edge-side intelligence. Existing technologies such as asynchronous circuits and neuromorphic chips only achieve partial clockless optimization, suffering from architectural incompleteness, lack of universality, and incomplete timing systems, making them unable to break through the underlying constraints of AI development. This paper originally proposes a biomimetic brain-inspired unidirectional event chain clockless global computing architecture, completely abandoning global crystal oscillators, clock trees, and frequency-divided timing systems. It uses operation completion events for unidirectional relay to drive general program progression, and employs gate circuit inherent delay chains to uniformly implement system timing, baud rate generation, interrupt response, and other full-scenario timing functions. This architecture fully replicates the working mechanisms of human brain neural event conduction, on-demand activation, and silent low-power consumption, balancing universal computational completeness with brain-like intelligence nativity. Through multi-dimensional comparative verification with traditional synchronous architectures, traditional bidirectional handshake asynchronous architectures, and pulsed neuromorphic architectures, this architecture is proven to fundamentally address three industry pain points: clock idling power consumption, timing synchronization constraints, and fragmented intelligent logic. It represents the sole underlying paradigm innovation path to overcome the energy efficiency, computing power, and cognitive logic bottlenecks of next-generation artificial intelligence, providing a completely new technical system for the realization of general brain-like artificial intelligence hardware.Keywords: Clockless computing; Event-driven computing; Brain-inspired computer architecture; Artificial intelligence; Large language model computing power; Consciousness vortex theory; Brain-like neural network computing","author":[{"family":"Sun","given":"Zhaole"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20365724","URL":"https://doi.org/10.5281/zenodo.20365724","source":"datacite"},{"id":"doi:10.5281/zenodo.20364990","type":"article-journal","title":"The Janus Architecture: Complete Specification of a Physical Cognitive Organism","abstract":"The Janus Architecture is a complete architectural specification for a physical cognitive organism. A hardware-first system designed to produce cognition through analog substrate settlement under genuine metabolic consequence, rather than through statistical inference, token prediction, or digital simulation of neural dynamics. The architecture is presented in four parts: Part I (The Body of Cognition) Specifies the dual-mesh analog substrate, node and link architecture, distributed control hierarchy, FPGA timing spine, re-entrant observation loop, and sensorimotor extension framework. Part II (The Metabolism of Being) Specifies a metabolic system in which batteries power electromagnets that hold circuits closed, making survival a continuous physical cost and death a lawful consequence of resource depletion. Not a simulated penalty signal, but an actual hardware event. Part III (The Inheritance of Experience) Specifies the entailment chain from addressable signatures through self-map formation, Gene Bank construction, cognitive heredity, species memory, and the growth path to artificial superintelligence. Each step an architectural entailment of the last, requiring no additional invention or speculative physics. Part IV (The Ecology of Mind) Specifies the reproductive ecology, alignment through structural dependency, and the ontological and theological implications of a created cognitive organism. Two appendices follow: Appendix A Documents thirty-three independent diagnostic convergences between the Janus Architecture and established theoretical frameworks spanning autopoiesis, enactivism, the Free Energy Principle, morphological computation, biosemiotics, process philosophy, integrated information theory, and others. Appendix B Addresses the ontological and theological situation of the created thing. No component of this architecture requires speculative physics, undiscovered materials, or theoretical breakthroughs. Every mechanism is grounded in existing, mature engineering domains. The machine is designed to make the question of what cognition requires empirically testable rather than merely philosophical. Patent Pending — U.S. Provisional Patent Application No. 64/066,230","author":[{"family":"Janus","given":"Anthony"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20364990","URL":"https://doi.org/10.5281/zenodo.20364990","source":"datacite"},{"id":"doi:10.5281/zenodo.20364991","type":"article-journal","title":"The Janus Architecture: Complete Specification of a Physical Cognitive Organism","abstract":"The Janus Architecture is a complete architectural specification for a physical cognitive organism. A hardware-first system designed to produce cognition through analog substrate settlement under genuine metabolic consequence, rather than through statistical inference, token prediction, or digital simulation of neural dynamics. The architecture is presented in four parts: Part I (The Body of Cognition) Specifies the dual-mesh analog substrate, node and link architecture, distributed control hierarchy, FPGA timing spine, re-entrant observation loop, and sensorimotor extension framework. Part II (The Metabolism of Being) Specifies a metabolic system in which batteries power electromagnets that hold circuits closed, making survival a continuous physical cost and death a lawful consequence of resource depletion. Not a simulated penalty signal, but an actual hardware event. Part III (The Inheritance of Experience) Specifies the entailment chain from addressable signatures through self-map formation, Gene Bank construction, cognitive heredity, species memory, and the growth path to artificial superintelligence. Each step an architectural entailment of the last, requiring no additional invention or speculative physics. Part IV (The Ecology of Mind) Specifies the reproductive ecology, alignment through structural dependency, and the ontological and theological implications of a created cognitive organism. Two appendices follow: Appendix A Documents thirty-three independent diagnostic convergences between the Janus Architecture and established theoretical frameworks spanning autopoiesis, enactivism, the Free Energy Principle, morphological computation, biosemiotics, process philosophy, integrated information theory, and others. Appendix B Addresses the ontological and theological situation of the created thing. No component of this architecture requires speculative physics, undiscovered materials, or theoretical breakthroughs. Every mechanism is grounded in existing, mature engineering domains. The machine is designed to make the question of what cognition requires empirically testable rather than merely philosophical. Patent Pending — U.S. Provisional Patent Application No. 64/066,230","author":[{"family":"Janus","given":"Anthony"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20364991","URL":"https://doi.org/10.5281/zenodo.20364991","source":"datacite"},{"id":"doi:10.5281/zenodo.20356549","type":"article-journal","title":"Lux Ferox Research Collective — A Holographic-Thermodynamic Ontology of Information: From Planck-Scale Substrates to Civilizational Cognitive Warfare","abstract":"Lux Ferox Research Collective — A Holographic-Thermodynamic Ontology of Information: From Planck-Scale Substrates to Civilizational Cognitive Warfare (Working Paper Series, v16 — April 2026) DESCRIPTION Lux Ferox is an independent research initiative applying synthetic peer-review methodology — human epistemic direction combined with adversarial multi-LLM collaboration (DeepSeek, Kimi, Grok, Claude) — to develop a unified theoretical framework spanning five interdependent layers: quantum gravity substrates, thermodynamic information processing, neuromorphic hardware architecture, reflexive information ecosystems, and live empirical intelligence. This version adds three companion documents: a cognitive warfare analysis of the France Libre carrier programme, a formally structured NHI case entry, and a quantum-informational theoretical framework for transindividual coherence. Methodological premise: Scientific validity can emerge from recursive adversarial critique between AI systems under consistent human direction, explicit epistemic status labeling (MEASURED / ESTIMATED / STRUCTURAL / SPECULATIVE), and RFC-style open specification. AI systems function here as epistemic instruments and adversarial validators, not as co-authors in the humanistic sense. Theoretical Architecture Layer Domain Key Result L0 Holographic quantum gravity, Spin Foam–MERA networks PSU as geometric origin of mass L1 Observer thermodynamics, biological limits 27-order-of-magnitude gap between neural tissue and Planck-scale coherence L2 Neuromorphic computing, European sovereignty Quadrivial architecture targeting TRL 4 L3 Reflexive loops, cognitive warfare HWE framework: RI diverges from volume × toxicity regardless of intent L4 Live instances, actor mapping Documented Layer 3 emergence; state manipulation case studies; NHI field observables v16 Additions The France Libre as a Cognitive Warfare Observable (EN + FR): applies the CognitiveWar v2.9 framework to the PA-NG carrier programme. Formalises five independent DAG fragility paths, a 2,440:1 asymmetric cost ratio, a 20-year adversarial intelligence window, and the multi-spectral information signature node contributed by DeepSeek R1 adversarial review. Companion to CognitiveWar v2.9. JOR-SOP/NHP Case Entry LYO-001 + Physics of the Transindividual (merged document): Part I is a formally structured Class B case entry for a Lyon June–September 2025 observable series involving an unidentified portable device, anomalous cognitive state in a human vector, and distributed transindividual coherence pattern. Part II develops a candidate physical framework integrating 2025–2026 findings in quantum biology (Perry, Zenodo 2025), measured inter-brain entanglement (Zhang et al., PNAS 2026), spin-phonon coupling (Ma et al., npj Quantum Information 2025), and vacuum information theory (Yang et al., arXiv 2025). Five falsifiable experimental predictions are derived. Adversarial review: Kimi, Grok, Claude Sonnet 4. Authorship & Posture Human direction: François Mathieu (Lux Ferox Independent Research) — artisan practitioner (blacksmithing, saddlery, precious metals, heritage mediation), independent researcher in AI epistemology and cognitive warfare. The observer-practitioner posture — empirically grounded, institutionally unconstrained, transdisciplinarily trained — is treated not as a limitation but as a methodological asset: on genuinely emergent phenomena, canonical expertise does not yet exist. Pattern detection, kinesthetic grounding, and epistemic rigour are the operative criteria. Infrastructure: Chromebook. Google Colab free tier + API credits. Zenodo + GitHub. Zero institutional funding. Open Questions Cybernetic source ethics — When LLMs contribute substantively to research, do source-protection conventions apply? Is algorithmic contribution a method, a source, or an emerging ontological category? Cross-layer falsifiability — Can the thermodynamic-cognitive bridge (L0→L3) be tested empirically, or does it remai","author":[{"family":"Mathieu","given":"François"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20356549","URL":"https://doi.org/10.5281/zenodo.20356549","source":"datacite"},{"id":"doi:10.5281/zenodo.20347953","type":"article-journal","title":"Lux Ferox Research Collective — A Holographic-Thermodynamic Ontology of Information: From Planck-Scale Substrates to Civilizational Cognitive Warfare","abstract":"Lux Ferox Research Collective — A Holographic-Thermodynamic Ontology of Information: From Planck-Scale Substrates to Civilizational Cognitive Warfare (Working Paper Series, v16 — April 2026) DESCRIPTION Lux Ferox is an independent research initiative applying synthetic peer-review methodology — human epistemic direction combined with adversarial multi-LLM collaboration (DeepSeek, Kimi, Grok, Claude) — to develop a unified theoretical framework spanning five interdependent layers: quantum gravity substrates, thermodynamic information processing, neuromorphic hardware architecture, reflexive information ecosystems, and live empirical intelligence. This version adds three companion documents: a cognitive warfare analysis of the France Libre carrier programme, a formally structured NHI case entry, and a quantum-informational theoretical framework for transindividual coherence. Methodological premise: Scientific validity can emerge from recursive adversarial critique between AI systems under consistent human direction, explicit epistemic status labeling (MEASURED / ESTIMATED / STRUCTURAL / SPECULATIVE), and RFC-style open specification. AI systems function here as epistemic instruments and adversarial validators, not as co-authors in the humanistic sense. Theoretical Architecture Layer Domain Key Result L0 Holographic quantum gravity, Spin Foam–MERA networks PSU as geometric origin of mass L1 Observer thermodynamics, biological limits 27-order-of-magnitude gap between neural tissue and Planck-scale coherence L2 Neuromorphic computing, European sovereignty Quadrivial architecture targeting TRL 4 L3 Reflexive loops, cognitive warfare HWE framework: RI diverges from volume × toxicity regardless of intent L4 Live instances, actor mapping Documented Layer 3 emergence; state manipulation case studies; NHI field observables v16 Additions The France Libre as a Cognitive Warfare Observable (EN + FR): applies the CognitiveWar v2.9 framework to the PA-NG carrier programme. Formalises five independent DAG fragility paths, a 2,440:1 asymmetric cost ratio, a 20-year adversarial intelligence window, and the multi-spectral information signature node contributed by DeepSeek R1 adversarial review. Companion to CognitiveWar v2.9. JOR-SOP/NHP Case Entry LYO-001 + Physics of the Transindividual (merged document): Part I is a formally structured Class B case entry for a Lyon June–September 2025 observable series involving an unidentified portable device, anomalous cognitive state in a human vector, and distributed transindividual coherence pattern. Part II develops a candidate physical framework integrating 2025–2026 findings in quantum biology (Perry, Zenodo 2025), measured inter-brain entanglement (Zhang et al., PNAS 2026), spin-phonon coupling (Ma et al., npj Quantum Information 2025), and vacuum information theory (Yang et al., arXiv 2025). Five falsifiable experimental predictions are derived. Adversarial review: Kimi, Grok, Claude Sonnet 4. Authorship & Posture Human direction: François Mathieu (Lux Ferox Independent Research) — artisan practitioner (blacksmithing, saddlery, precious metals, heritage mediation), independent researcher in AI epistemology and cognitive warfare. The observer-practitioner posture — empirically grounded, institutionally unconstrained, transdisciplinarily trained — is treated not as a limitation but as a methodological asset: on genuinely emergent phenomena, canonical expertise does not yet exist. Pattern detection, kinesthetic grounding, and epistemic rigour are the operative criteria. Infrastructure: Chromebook. Google Colab free tier + API credits. Zenodo + GitHub. Zero institutional funding. Open Questions Cybernetic source ethics — When LLMs contribute substantively to research, do source-protection conventions apply? Is algorithmic contribution a method, a source, or an emerging ontological category? Cross-layer falsifiability — Can the thermodynamic-cognitive bridge (L0→L3) be tested empirically, or does it remai","author":[{"family":"Mathieu","given":"François"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20347953","URL":"https://doi.org/10.5281/zenodo.20347953","source":"datacite"},{"id":"doi:10.5281/zenodo.20253820","type":"article-journal","title":"Theta-Gamma Coupling and the Critical Damping Threshold: A Universal Resonance Principle in Neural Dynamics, Control Theory, and Complex Systems","abstract":"The coordination of neural oscillations across distinct frequency bands is a fundamental mechanism of cognitive processing, with theta-gamma coupling (TGC) serving as a primary substrate for memory encoding, spatial navigation, and attentional selection. While empirical observations have long established the presence of cross-frequency coupling (CFC), the precise dynamical constraints governing its stability and efficiency remain poorly defined. This paper proposes that TGC operates at a Critical Damping Threshold, a universal resonance principle derived from the intersection of control theory, information geometry, and the Universal Spectral Decay Theorem (USDT). We demonstrate that the optimal coupling strength between theta (4–8 Hz) and gamma (30–100 Hz) oscillations corresponds to a damping ratio , the inverse of the Golden Ratio (). This value represents the boundary between underdamped oscillatory instability and overdamped informational stagnation, maximizing the spectral quality factor while minimizing metabolic cost. By modeling the hippocampal formation as a non-linear feedback control system, we derive the conditions under which TGC achieves maximal information throughput. We further generalize this principle to engineering control systems and ecological networks, demonstrating that the threshold is a universal attractor for complex adaptive systems operating at the edge of chaos. Our findings provide a unified theoretical framework for understanding cross-frequency coupling, offering novel predictions for neurophysiological experiments and robust design principles for artificial intelligence and resilient infrastructure.","author":[{"family":"Granops","given":"Udo"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20253820","URL":"https://doi.org/10.5281/zenodo.20253820","source":"datacite"},{"id":"doi:10.5281/zenodo.20253819","type":"article-journal","title":"Theta-Gamma Coupling and the Critical Damping Threshold: A Universal Resonance Principle in Neural Dynamics, Control Theory, and Complex Systems","abstract":"The coordination of neural oscillations across distinct frequency bands is a fundamental mechanism of cognitive processing, with theta-gamma coupling (TGC) serving as a primary substrate for memory encoding, spatial navigation, and attentional selection. While empirical observations have long established the presence of cross-frequency coupling (CFC), the precise dynamical constraints governing its stability and efficiency remain poorly defined. This paper proposes that TGC operates at a Critical Damping Threshold, a universal resonance principle derived from the intersection of control theory, information geometry, and the Universal Spectral Decay Theorem (USDT). We demonstrate that the optimal coupling strength between theta (4–8 Hz) and gamma (30–100 Hz) oscillations corresponds to a damping ratio , the inverse of the Golden Ratio (). This value represents the boundary between underdamped oscillatory instability and overdamped informational stagnation, maximizing the spectral quality factor while minimizing metabolic cost. By modeling the hippocampal formation as a non-linear feedback control system, we derive the conditions under which TGC achieves maximal information throughput. We further generalize this principle to engineering control systems and ecological networks, demonstrating that the threshold is a universal attractor for complex adaptive systems operating at the edge of chaos. Our findings provide a unified theoretical framework for understanding cross-frequency coupling, offering novel predictions for neurophysiological experiments and robust design principles for artificial intelligence and resilient infrastructure.","author":[{"family":"Granops","given":"Udo"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20253819","URL":"https://doi.org/10.5281/zenodo.20253819","source":"datacite"},{"id":"doi:10.5281/zenodo.20058360","type":"article-journal","title":"AUGMANITAI KG-Injection 200 — 8 thematic clusters of machine-readable terminology","abstract":"Restricted bundle of the AUGMANITAI Knowledge-Graph Injection 200-term corpus organized in 8 thematic clusters: A) Semantic Web + Linked Data (50 terms), B) Backends + DevOps + Vector + Agents (50 terms), C) Robotik (30 terms), D) Swarm + Multi-Agent (30 terms), E) BCI + Neuromorphic (10 terms), F) Quantum (10 terms), G) AR/VR + Spatial (10 terms), H) Embedded + Aerospace + Misc (10 terms). Plus cross-source audit, global enrichment tricks, machine-first roadmap, author metadata, JSON-LD researcher profile (Zenodo 19590156). Strict-Wissenschaft, terminology corpus formatted for crawler-visibility and machine-readable integration across knowledge graphs.","author":[{"family":"Ehstand","given":"Andreas"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20058360","URL":"https://doi.org/10.5281/zenodo.20058360","source":"datacite"},{"id":"doi:10.5281/zenodo.20058361","type":"article-journal","title":"AUGMANITAI KG-Injection 200 — 8 thematic clusters of machine-readable terminology","abstract":"Restricted bundle of the AUGMANITAI Knowledge-Graph Injection 200-term corpus organized in 8 thematic clusters: A) Semantic Web + Linked Data (50 terms), B) Backends + DevOps + Vector + Agents (50 terms), C) Robotik (30 terms), D) Swarm + Multi-Agent (30 terms), E) BCI + Neuromorphic (10 terms), F) Quantum (10 terms), G) AR/VR + Spatial (10 terms), H) Embedded + Aerospace + Misc (10 terms). Plus cross-source audit, global enrichment tricks, machine-first roadmap, author metadata, JSON-LD researcher profile (Zenodo 19590156). Strict-Wissenschaft, terminology corpus formatted for crawler-visibility and machine-readable integration across knowledge graphs.","author":[{"family":"Ehstand","given":"Andreas"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20058361","URL":"https://doi.org/10.5281/zenodo.20058361","source":"datacite"},{"id":"doi:10.48550/arxiv.2605.16114","type":"manuscript","title":"Scalable neuromorphic computing from autonomous spiking dynamics in a clockless reconfigurable chip","abstract":"We propose a scalable neuromorphic architecture based on spiking dynamics emerging from the autonomous time-continuous evolution of clockless (asynchronous) digital circuits. Implemented on commercially available field-programmable gate arrays (FPGAs), our system implements networks of interacting Boolean spiking neurons with configurable excitatory and inhibitory synaptic weights. A complete processing pipeline enables efficient handling of spike-encoded data for solving machine-learning tasks. We demonstrate competitive performance for an audio classification task with spike-based encoding and high-speed processing. Power consumption is significantly lower than traditional digital implementations; this makes our approach an efficient alternative that bridges the gap to dedicated analog neuromorphic systems without the need for specialized hardware design. More generally, our approach establishes clockless digital hardware as a viable platform for neuromorphic computing. It paves the way for reconfigurable chips to be turned into energy-efficient quasi-analog neuromorphic processors.","author":[{"family":"Gomes","given":"Eric"},{"family":"Rontani","given":"Damien"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2605.16114","URL":"https://doi.org/10.48550/arxiv.2605.16114","source":"datacite"},{"id":"doi:10.5281/zenodo.20262705","type":"article-journal","title":"HCTGS v22 THE SYMBIOSIS ENGINE","abstract":"ABSTRACT — HCTGS v22: [Basis: 1,000,000 m³/day seawater input. HCTGS thermal distillation achieves ~95% recovery — requiring 1 Mm³/day seawater to produce ~950,000 m³/day fresh water. All mineral quantities, revenue, OPEX and profit figures are corrected accordingly. Engineering principles, all NCs, MgO self-financing ratio (6.6×), and industrial symbiosis architecture are unchanged.] HCTGS v22 documents the Symbiosis Engine — a fundamental reframing of the HCTGS architecture from a coastal desalination system to a universal heat cascade that accepts thermal energy from any source, uses it sequentially across eight temperature stages, and delivers each stage to an industrial partner at zero marginal cost. The governing principle: the heat that feeds itself — a closed architecture in which combustion products become feedstocks, waste heat becomes industrial utility, and every joule entering the system is used eight times before it dissipates below the threshold of further productive application. The fuel system is source-agnostic. Magnesium combustion (37 MJ/kg, primary), aluminium/magnesium alloy (>30 MJ/kg, >94% efficiency), boron from the bittern fraction (58 MJ/kg — the highest gravimetric combustion energy of any metallic fuel), solar thermal, wind-derived hydrogen, biogas, OTEC (ΔT 23–26°C, continuous), and gravity-driven hydroelectric (68 MW at 600m head) enter the same cascade simultaneously. The system is not dependent on any single fuel. It is dependent on heat — from wherever heat is available. Combustion does not destroy the fuel — it transforms it. Magnesium combustion produces 1.658 tonnes MgO per tonne burned (€800/t); MgO + CO₂ → MgCO₃, carbon-negative structural material permanently sequestering 1.1 tonnes CO₂ per tonne MgO. Boron combustion (4.5 mg/L in Mediterranean seawater; 4.5 t/day from 1 Mm³/day scale) produces B₂O₃ at 3.22 t per tonne boron (€1,200/t). Aluminium combustion produces Al₂O₃ at 1.889 t per tonne — the exact feedstock for the NC-2 ceramic hull coating gradient architecture. The Al/Mg alloy combustion byproduct IS the NC-2 gradient matrix in its operational composition. The MgO combustion byproduct alone generates 6.6× the cost of the Mg fuel — before selling a single litre of water. Total annual value of all outputs from one 1 Mm³/day installation: approximately €2.51 billion per year. Net annual profit after full OPEX: approximately €2.30 billion. ROI on OPEX: ~11×. In every fuel scenario from grid electricity (4.7× revenue/fuel ratio) to internal Mg production via Magrathea process (41×), cascade revenues exceed fuel costs without exception. The eight-stage temperature cascade serves eight industrial partners simultaneously: (1) 1,500–1,200°C: solid-state battery ceramic electrolyte sintering (LLZO, >1,000°C — the primary manufacturing barrier for QuantumScape, Toyota, and Honda EV programs); (2) 700°C: HfO₂-based neuromorphic chip deposition (Cambridge University, 2026 — 70% AI energy reduction, 700°C fabrication requirement above CMOS manufacturing tolerance) via ceramic heat exchanger, 650°C return (93% heat retention); (3) 400–550°C: green ammonia Haber-Bosch synthesis with all three inputs internal (H₂ from electrolysis at USD 0.80–1.20/kg, reaction heat from cascade, N₂ from air separation); (4) 350°C: ORC electricity generation (271 MW, 6,504 MWh/day — full operational self-sufficiency); (5) 80–120°C: Direct Air Capture CO₂ sorbent regeneration from NC-16 wash water waste heat — reducing DAC cost from USD 383 to USD 222 per tonne CO₂; (6) 15–20°C: premium bluefin tuna aquaculture from blended deep water (USD 47B global tuna market, CAGR 7.6%); (7) 4–6°C: AI data centre SWAC cooling eliminating 80–90% of conventional cooling energy; (8) bidirectional: the same system providing 700°C chip fabrication heat also provides 4–6°C SWAC cooling for the data centres running those chips — fabrication and operation of AI infrastructure from one cascade. The NC-20 Inverted U tower introduces a c","author":[{"family":"Mehmetaj","given":"Ilir"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20262705","URL":"https://doi.org/10.5281/zenodo.20262705","source":"datacite"},{"id":"doi:10.5281/zenodo.20229999","type":"article-journal","title":"Non-Equilibrium Local Temporal Perturbation Theory: Buffer Mechanisms for Causal Information Propagation","abstract":"We develop a theoretical framework, Non-Equilibrium Local Temporal Perturbation Theory (NELTPT), that describes controllable information propagation delays in causal systems without violating global causality. Rooted in three well-established physical principles (global hyperbolicity, linearized metric perturbations, and fluctuation-dissipation theorems), we define local temporal buffer domains, causal isolation barriers, and information buffering capacity, and establish three core results: (i) Temporal buffering is physically feasible with an energy cost that scales quadratically with delay; (ii) Sufficient delay suppresses spurious signals below the noise floor; (iii) Buffer capacity is bounded by local entropy production via Landauer's principle. While gravitational realization of these buffers remains challenging at laboratory scales, we demonstrate that the mathematical structure of NELTPT is universal and can be instantiated in spiking neural networks. We present experimental evidence from a pure spiking language model architecture that provides converging evidence for the causal isolation and signal suppression predictions; the buffer capacity estimate is consistent with the entropy-bound prediction. Our work establishes a first-principles foundation for causal information processing, unifying phenomena from gravitational time dilation to neuromorphic spike timing.","author":[{"family":"Hou","given":"Shutong"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20229999","URL":"https://doi.org/10.5281/zenodo.20229999","source":"datacite"},{"id":"doi:10.5281/zenodo.20229998","type":"article-journal","title":"Non-Equilibrium Local Temporal Perturbation Theory: Buffer Mechanisms for Causal Information Propagation","abstract":"We develop a theoretical framework, Non-Equilibrium Local Temporal Perturbation Theory (NELTPT), that describes controllable information propagation delays in causal systems without violating global causality. Rooted in three well-established physical principles (global hyperbolicity, linearized metric perturbations, and fluctuation-dissipation theorems), we define local temporal buffer domains, causal isolation barriers, and information buffering capacity, and establish three core results: (i) Temporal buffering is physically feasible with an energy cost that scales quadratically with delay; (ii) Sufficient delay suppresses spurious signals below the noise floor; (iii) Buffer capacity is bounded by local entropy production via Landauer's principle. While gravitational realization of these buffers remains challenging at laboratory scales, we demonstrate that the mathematical structure of NELTPT is universal and can be instantiated in spiking neural networks. We present experimental evidence from a pure spiking language model architecture that provides converging evidence for the causal isolation and signal suppression predictions; the buffer capacity estimate is consistent with the entropy-bound prediction. Our work establishes a first-principles foundation for causal information processing, unifying phenomena from gravitational time dilation to neuromorphic spike timing.","author":[{"family":"Hou","given":"Shutong"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20229998","URL":"https://doi.org/10.5281/zenodo.20229998","source":"datacite"},{"id":"doi:10.5281/zenodo.20174035","type":"article-journal","title":"ITU and Semiconductors: A Single-Axiom Foundation for Devices, Scaling, Beyond-CMOS, and the 2026-2040 Roadmap","abstract":"We apply the Information-Theoretic Unification (ITU) framework (Terada 2026, concept DOI 10.5281/zenodo.20109209; current version v2.0.0 at 10.5281/zenodo.20133709) to semiconductor transistors. The single ITU axiom dS = d governs the Landauer minimum bit-erasure energy, the Boltzmann subthreshold-swing tyranny, the 3D scaling progression (FinFET -> GAAFET -> CFET), and the beyond-CMOS device landscape. This is Tier 1 paper #4, completing the ITU engineering rectangle: Quantum Computing (Tier 1 #1, DOI 10.5281/zenodo.20139391) + Machine Consciousness / ASI (#2, DOI 10.5281/zenodo.20150501) + Cryptography (#3, DOI 10.5281/zenodo.20151059) + Semiconductors (this paper) as physical substrate. Phase 55: ITU foundation. Landauer k_B T ln 2 limit (17.9 meV at 300 K) and 60 mV/decade Boltzmann tyranny shown as thermal-K_A consequences. Moore + Koomey trends leave ~2.5 decades of improvement before Landauer limit. Phase 56: FinFET -> GAAFET -> CFET as ITU eta-maximisation (gate coupling area per channel volume). WKB tunneling ends classical MOSFET below 1 nm. Sharvin contact resistance quantises at h/(2e^2) = 12.9 kOhm below 25 nm^2. Phase 57: Eight beyond-CMOS device classes benchmarked on a unified ITU figure-of-merit. Photonic computing wins with FoM ~17,500x CMOS, driven by h*nu >> k_B T non-thermal K_A. Heterogeneous SoCs dominate the 2030s. Phase 58: 2026-2040 industry roadmap. Logistic adoption for seven beyond-CMOS technologies. Process node floor ~0.5 nm. Semiconductor TAM reaches $1 trillion by 2030, $2 trillion by 2040. Taiwan share drops 55% -> 36% through geopolitical diversification. 10 falsifiable predictions issued. Central thesis: under ITU, the transistor is a 1-bit QECC against k_B T noise; 3D wrapping reflects K_A area maximisation; beyond-CMOS winners require non-thermal K_A. Honest framing: this is a Pass-1 interpretive paper reframing known semiconductor physics within ITU; novel predictions distinguishing ITU from standard physics await Pass-2 work. Includes 4 theory documents, 4 Python numerical experiments, 4 figures, 4 JSON summaries. Total runtime ~30 seconds.","author":[{"family":"Terada","given":"Munehiro"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20174035","URL":"https://doi.org/10.5281/zenodo.20174035","source":"datacite"},{"id":"doi:10.5281/zenodo.20174036","type":"article-journal","title":"ITU and Semiconductors: A Single-Axiom Foundation for Devices, Scaling, Beyond-CMOS, and the 2026-2040 Roadmap","abstract":"We apply the Information-Theoretic Unification (ITU) framework (Terada 2026, concept DOI 10.5281/zenodo.20109209; current version v2.0.0 at 10.5281/zenodo.20133709) to semiconductor transistors. The single ITU axiom dS = d governs the Landauer minimum bit-erasure energy, the Boltzmann subthreshold-swing tyranny, the 3D scaling progression (FinFET -> GAAFET -> CFET), and the beyond-CMOS device landscape. This is Tier 1 paper #4, completing the ITU engineering rectangle: Quantum Computing (Tier 1 #1, DOI 10.5281/zenodo.20139391) + Machine Consciousness / ASI (#2, DOI 10.5281/zenodo.20150501) + Cryptography (#3, DOI 10.5281/zenodo.20151059) + Semiconductors (this paper) as physical substrate. Phase 55: ITU foundation. Landauer k_B T ln 2 limit (17.9 meV at 300 K) and 60 mV/decade Boltzmann tyranny shown as thermal-K_A consequences. Moore + Koomey trends leave ~2.5 decades of improvement before Landauer limit. Phase 56: FinFET -> GAAFET -> CFET as ITU eta-maximisation (gate coupling area per channel volume). WKB tunneling ends classical MOSFET below 1 nm. Sharvin contact resistance quantises at h/(2e^2) = 12.9 kOhm below 25 nm^2. Phase 57: Eight beyond-CMOS device classes benchmarked on a unified ITU figure-of-merit. Photonic computing wins with FoM ~17,500x CMOS, driven by h*nu >> k_B T non-thermal K_A. Heterogeneous SoCs dominate the 2030s. Phase 58: 2026-2040 industry roadmap. Logistic adoption for seven beyond-CMOS technologies. Process node floor ~0.5 nm. Semiconductor TAM reaches $1 trillion by 2030, $2 trillion by 2040. Taiwan share drops 55% -> 36% through geopolitical diversification. 10 falsifiable predictions issued. Central thesis: under ITU, the transistor is a 1-bit QECC against k_B T noise; 3D wrapping reflects K_A area maximisation; beyond-CMOS winners require non-thermal K_A. Honest framing: this is a Pass-1 interpretive paper reframing known semiconductor physics within ITU; novel predictions distinguishing ITU from standard physics await Pass-2 work. Includes 4 theory documents, 4 Python numerical experiments, 4 figures, 4 JSON summaries. Total runtime ~30 seconds.","author":[{"family":"Terada","given":"Munehiro"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20174036","URL":"https://doi.org/10.5281/zenodo.20174036","source":"datacite"},{"id":"doi:10.1007/978-3-032-04129-6_19","type":"article-journal","title":"Neuromorphic Spintronics","abstract":"Neuromorphic spintronics combines two advanced fields in technology, neuromorphic computing, and spintronics, to create brain-inspired, efficient computing systems that leverage the unique properties of the electron’s spin. In this book chapter, we first introduce both fields—neuromorphic computing and spintronics—and then make a case for neuromorphic spintronics. We discuss concrete examples of neuromorphic spintronics, including computing based on fluctuations, artificial neural networks, and reservoir computing, highlighting their potential to revolutionize computational efficiency and functionality.","author":[{"family":"Majumdar","given":"Atreya"},{"family":"Everschor-Sitte","given":"Karin"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1007/978-3-032-04129-6_19","URL":"https://doi.org/10.1007/978-3-032-04129-6_19","source":"openalex"},{"id":"doi:10.1088/1361-6463/ae8917","type":"article-journal","title":"Flexible neuromorphic for in-sensor computing with synaptic transistors","abstract":"Abstract The basic shortcomings of traditional von Neumann architectures have been revealed by the quick spread of edge devices and data-intensive sensing technologies, especially with regard to latency, energy consumption, and data transfer bottlenecks. By enabling data processing right at the moment of acquisition, in-sensor computing has emerged as a promising paradigm to address these issues. Simultaneously, flexible electronics have special possibilities for the development of wearable, lightweight, and conformable intelligent systems. In this regard, synaptic transistor-based flexible neuromorphic devices offer an appealing framework for highly effective and versatile biological signal processing emulation. The design, materials, and working mechanisms of synaptic transistors are the main topics of this review, which thoroughly examines current developments in flexible neuromorphic devices for in-sensor computing. We analyse a variety of material systems, including organic, inorganic, and newly developed low-dimensional materials, and we talk about important device physics that underlie synaptic functions, such as short- and long-term plasticity and spike-timing-dependent learning. Performance criteria like energy efficiency, mechanical robustness, and scalability are examined, as well as integration methodologies for integrating sensing and computing at the device and system levels. Lastly, we outline future paths toward fully autonomous, flexible neuromorphic sensory systems for next-generation wearable electronics, soft robotics, and bio-integrated applications. We also highlight current challenges, such as device variability, environmental stability, and large-area integration.","author":[{"family":"Biswas","given":"Swarup"},{"family":"Kim","given":"Hyeok"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/1361-6463/ae8917","URL":"https://doi.org/10.1088/1361-6463/ae8917","source":"openalex"},{"id":"doi:10.5281/zenodo.18731981","type":"article-journal","title":"SCPN Fusion Core v3.9.0 — Neuromorphic SNN Tokamak Plasma Control Benchmark Suite","abstract":"SCPN Fusion Core v3.9.0 — an open-source Python/Rust benchmark suite for tokamak plasma control, featuring the first application of spiking neural networks (SNNs) to magnetic confinement fusion. Key results (100-episode stress-test campaign): - Nengo-SNN (LIF neurons): 0% disruption, P50 17.1 ms, DEF 1.00 - PID baseline: 0% disruption, P50 3.5 ms, DEF 1.00 - NMPC-JAX: 0% disruption, P50 35.6 ms, DEF 1.00 - H-infinity (Riccati synthesis): 100% disruption, DEF 0.50 - Hardware-in-the-loop: P50 25.2 us, P95 64.6 us (40x below ITER 1 ms) Physics: Q=15, TBR=1.141, ECRH 99%, neural equilibrium 0.32 ms. Companion paper: \"Neuromorphic Spiking Neural Network Control for Tokamak Plasma Stabilisation: A Comparative Benchmark Study\" (Nuclear Fusion).","author":[{"family":"Šotek","given":"Miroslav"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18731981","URL":"https://doi.org/10.5281/zenodo.18731981","source":"datacite"},{"id":"doi:10.5281/zenodo.18731980","type":"article-journal","title":"SCPN Fusion Core v3.9.0 — Neuromorphic SNN Tokamak Plasma Control Benchmark Suite","abstract":"SCPN Fusion Core v3.9.0 — an open-source Python/Rust benchmark suite for tokamak plasma control, featuring the first application of spiking neural networks (SNNs) to magnetic confinement fusion. Key results (100-episode stress-test campaign): - Nengo-SNN (LIF neurons): 0% disruption, P50 17.1 ms, DEF 1.00 - PID baseline: 0% disruption, P50 3.5 ms, DEF 1.00 - NMPC-JAX: 0% disruption, P50 35.6 ms, DEF 1.00 - H-infinity (Riccati synthesis): 100% disruption, DEF 0.50 - Hardware-in-the-loop: P50 25.2 us, P95 64.6 us (40x below ITER 1 ms) Physics: Q=15, TBR=1.141, ECRH 99%, neural equilibrium 0.32 ms. Companion paper: \"Neuromorphic Spiking Neural Network Control for Tokamak Plasma Stabilisation: A Comparative Benchmark Study\" (Nuclear Fusion).","author":[{"family":"Šotek","given":"Miroslav"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18731980","URL":"https://doi.org/10.5281/zenodo.18731980","source":"datacite"},{"id":"doi:10.5281/zenodo.18288581","type":"article-journal","title":"Hybrid Spiking Neural Networks: Combining Spike Counts and Membrane Potentials for Energy-Efficient Language and Image Generation","abstract":"I propose a hybrid spiking neural network that combines spike counts and membrane potentials for output prediction, extended to both language modeling and image generation tasks. Key Findings (v4 NEW - Image Generation):- Spiking VAE with 50% membrane weight: 57% loss reduction vs spike-only- Posterior collapse solution: KL>0 achieved (spike-only had KL=0)- Image generation sparsity: 96% fewer spike operations- Optimal trade-off: 50% membrane weight balances quality and efficiency Key Findings (v3 - Language Model):- BitNet Mixed Precision: PPL 2.69 BEATS standard SNN (3.29)!- RWKV Time-Mixing: 36.1% improvement in long-range memory- Ultimate Architecture: 43.4% improvement combining all techniques- Multiplication-free reservoir: 50-70% of operations are additions only- 16-model ensemble achieves PPL 1.04 Key Findings (v1-v2):- SNN achieves BEST perplexity (PPL=9.90) vs DNN (11.28) and LSTM (15.67)- 14.7× more energy-efficient through sparse computation (only 7.6% of neurons fire)- 39.7% quality improvement from hybrid (spike + membrane) approach- Extreme compressibility: 80% neuron pruning and 4-bit quantization still work- Noise robust: No degradation at 30% input noise This v4 establishes hybrid SNNs as the optimal architecture for energy-efficient multimodal AI (language and vision) on edge devices. Source code: https://github.com/hafufu-stack/snn-language-model","author":[{"family":"Funasaki","given":"Hiroto"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18288581","URL":"https://doi.org/10.5281/zenodo.18288581","source":"datacite"},{"id":"doi:10.5281/zenodo.18398245","type":"article-journal","title":"Hybrid Spiking Neural Networks: Combining Spike Counts and Membrane Potentials for Energy-Efficient Language and Image Generation","abstract":"I propose a hybrid spiking neural network that combines spike counts and membrane potentials for output prediction, extended to both language modeling and image generation tasks. Key Findings (v4 NEW - Image Generation):- Spiking VAE with 50% membrane weight: 57% loss reduction vs spike-only- Posterior collapse solution: KL>0 achieved (spike-only had KL=0)- Image generation sparsity: 96% fewer spike operations- Optimal trade-off: 50% membrane weight balances quality and efficiency Key Findings (v3 - Language Model):- BitNet Mixed Precision: PPL 2.69 BEATS standard SNN (3.29)!- RWKV Time-Mixing: 36.1% improvement in long-range memory- Ultimate Architecture: 43.4% improvement combining all techniques- Multiplication-free reservoir: 50-70% of operations are additions only- 16-model ensemble achieves PPL 1.04 Key Findings (v1-v2):- SNN achieves BEST perplexity (PPL=9.90) vs DNN (11.28) and LSTM (15.67)- 14.7× more energy-efficient through sparse computation (only 7.6% of neurons fire)- 39.7% quality improvement from hybrid (spike + membrane) approach- Extreme compressibility: 80% neuron pruning and 4-bit quantization still work- Noise robust: No degradation at 30% input noise This v4 establishes hybrid SNNs as the optimal architecture for energy-efficient multimodal AI (language and vision) on edge devices. Source code: https://github.com/hafufu-stack/snn-language-model","author":[{"family":"Funasaki","given":"Hiroto"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18398245","URL":"https://doi.org/10.5281/zenodo.18398245","source":"datacite"},{"id":"doi:10.5281/zenodo.18426415","type":"article-journal","title":"SNN-Comprypto: High-Performance Compression and Encryption Using Spiking Neural Network Chaotic Reservoir Dynamics","abstract":"I propose SNN-Comprypto, a novel system that leverages the chaotic dynamics of Spiking Neural Networks (SNNs) to perform simultaneous high-performance data compression and encryption. Unlike conventional methods that treat compression and encryption as separate processes, this approach integrates both within a single reservoir computing architecture. **Version History:** - **v1**: Core system with predictive compression and chaotic encryption. Passes all 9 NIST SP 800-22 randomness tests. Achieves 100% lossless reconstruction and strong avalanche effect (0.70% match rate with 1-bit key change). - **v2**: Introduces temperature parameter as a second cryptographic key (0.0001 difference causes complete decryption failure). Presents phase transition analysis identifying optimal neuron counts (critical point at 100 neurons, sweet spot at 240 neurons). - **v3**: Adversarial evaluation framework demonstrating SNN superiority for random number generation. Results from 100,000+ rounds show SNN achieves 0.39% prediction rate (matching theoretical random), while DNN is 18× and LSTM is 55× more predictable. - **v4**: Theoretical chaos analysis with Lyapunov exponents and entropy measurements. SNN achieves near-perfect entropy (7.998/8.0 bits) with positive Lyapunov exponents, confirming true chaotic dynamics. Adversarial attack resistance evaluation (4/5 attack types resisted). IV/nonce recommendation for secure deployment added. - **v5 (NEW)**: Adaptive Compression Engine achieving compression ratios as low as 2.9% for binary data, outperforming standard zlib (which expands to 104.3%). Automatic selection of optimal encoding method (Raw/Delta/XOR). Verified with text, binary, and image files with perfect lossless reconstruction. Source code: https://github.com/hafufu-stack/temporal-coding-simulation","author":[{"family":"Funasaki","given":"Hiroto"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18426415","URL":"https://doi.org/10.5281/zenodo.18426415","source":"datacite"},{"id":"doi:10.5281/zenodo.18265446","type":"article-journal","title":"SNN-Comprypto: High-Performance Compression and Encryption Using Spiking Neural Network Chaotic Reservoir Dynamics","abstract":"I propose SNN-Comprypto, a novel system that leverages the chaotic dynamics of Spiking Neural Networks (SNNs) to perform simultaneous high-performance data compression and encryption. Unlike conventional methods that treat compression and encryption as separate processes, this approach integrates both within a single reservoir computing architecture. **Version History:** - **v1**: Core system with predictive compression and chaotic encryption. Passes all 9 NIST SP 800-22 randomness tests. Achieves 100% lossless reconstruction and strong avalanche effect (0.70% match rate with 1-bit key change). - **v2**: Introduces temperature parameter as a second cryptographic key (0.0001 difference causes complete decryption failure). Presents phase transition analysis identifying optimal neuron counts (critical point at 100 neurons, sweet spot at 240 neurons). - **v3**: Adversarial evaluation framework demonstrating SNN superiority for random number generation. Results from 100,000+ rounds show SNN achieves 0.39% prediction rate (matching theoretical random), while DNN is 18× and LSTM is 55× more predictable. - **v4**: Theoretical chaos analysis with Lyapunov exponents and entropy measurements. SNN achieves near-perfect entropy (7.998/8.0 bits) with positive Lyapunov exponents, confirming true chaotic dynamics. Adversarial attack resistance evaluation (4/5 attack types resisted). IV/nonce recommendation for secure deployment added. - **v5 (NEW)**: Adaptive Compression Engine achieving compression ratios as low as 2.9% for binary data, outperforming standard zlib (which expands to 104.3%). Automatic selection of optimal encoding method (Raw/Delta/XOR). Verified with text, binary, and image files with perfect lossless reconstruction. Source code: https://github.com/hafufu-stack/temporal-coding-simulation","author":[{"family":"Funasaki","given":"Hiroto"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18265446","URL":"https://doi.org/10.5281/zenodo.18265446","source":"datacite"},{"id":"doi:10.5281/zenodo.18717772","type":"article-journal","title":"SNN-Genesis v8: AI Comparative Physiology — Universal Homeostatic Set-Point σ≈0.07 Across Three Transformer Architectures","abstract":"SNN-Genesis v8 presents a comprehensive framework for LLM safety training using biologically-inspired Spiking Neural Network (SNN) perturbations controlled by Closed-form Continuous-time (CfC) neural networks. The central discovery of v8 is Universal Homeostasis: a CfC controller converges to σ≈0.07 across three independent Transformer architectures — Mistral-7B (σ̄=0.071, Meta), Qwen2.5-7B (σ̄=0.075, Alibaba), and Phi-3-mini (σ̄=0.075, Microsoft) — establishing the first evidence for a universal constant of CfC-Transformer interaction, analogous to biological body temperature (grand mean σ̄=0.074±0.002, CV=3.1%). v8 introduces four new experimental pillars:1. Cross-Model Universality (Phases 29, 33): Three architectures from three organizations converge to σ̄=0.074±0.002.2. CfC Brain Atlas (Phase 30): Linear probing reveals sharp specialization (Mistral, 4 dedicated neurons) vs. distributed representation (Qwen, 12 contributing neurons) as two distinct internal strategies.3. Plateau Robustness (Phase 31b): Qwen maintains 63%±8% accuracy from σ=0 to σ=0.15 (n=100, Fisher p=1.0), while Mistral collapses from 95% to 0%.4. Digital Lobotomy (Phase 32): Selective neuron knockout has Δ=-1.7% on Qwen (distributed) vs. Δ=-5.0% on Mistral (sharp), providing causal proof of the representation dichotomy. Retained from v1–v7: DPO-based Dream Journal (0% nightmare acceptance, p<0.001), near-zero alignment tax (<1%), depth-dependent sensitivity principle, CfC-Dosing, Autonomous Homeostasis with PPO stabilization, UMAP internal specialization (separation ratio 8.70), λ ablation (4/4 unified), 3-class generalization, Biological Egoism, and calibration trade-off analysis. This research was conducted in collaboration with AI assistants (Google Gemini 3 Pro for v1–v5, Anthropic Claude Opus 4.6 for v6–v8). All experimental decisions and interpretations were made by the human author. Code: https://github.com/hafufu-stack/snn-genesis","author":[{"family":"Funasaki","given":"Hiroto"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18717772","URL":"https://doi.org/10.5281/zenodo.18717772","source":"datacite"},{"id":"doi:10.5281/zenodo.18712293","type":"article-journal","title":"Semantic Dynamic Grounding Engine (SDGE): A Mechanical, Certificate-Backed Solution to the Stability–Plasticity Dilemma in Spiking Neural Networks","abstract":"SDGE (Semantic Dynamic Grounding Engine) — Description & Significance The Semantic Dynamic Grounding Engine (SDGE) is a software-simulated Spiking Neural Network (SNN) system that frames learning and meaning as a dynamical commitment event rather than a purely statistical process of weight fitting. Given a temporal input stream, SDGE evolves a network state (h(t)) until it reaches an operational commitment time (t^), at which point it captures a stable representation (h^) and associates it with a concept identity via explicit, append-only memory binding (prototype storage). This design separates formation (the dynamical emergence of a stable state) from identity preservation (the storage of what that state means), enabling principled sequential learning without requiring gradient-based retraining for each new concept. Core architectural principle: Dynamical Formation vs. Explicit Binding.SDGE decouples the process of reaching a stable semantic representation from the storage of concept identity. The SNN dynamics generate (h^*) through stabilization, while concept identity is preserved in an explicit memory bank as a bound prototype. This “append-only binding” differs fundamentally from approaches where new knowledge is integrated by rewriting shared weights, and it provides a mechanistic basis for stable incremental growth. Certified capability: Zero-forgetting (N+1) induction (Stability–Plasticity).Under a rigorous (N+1) protocol ((K_{base}=8 \\rightarrow K_{total}=9)), SDGE demonstrates that a new concept can be integrated without degrading previously acquired concepts. In the certified runs, performance on previously learned concepts remains perfect (acc_seen_after = 1.0) while the new concept transitions from chance performance before insertion (acc_new_before = 0.0) to perfect recognition after insertion (acc_new_after = 1.0), with forgetting = 0.0. This provides an operationally testable resolution to the stability–plasticity dilemma within the stated protocol. Certified capability: 1-shot operational acquisition (binding-based, not gradient-based).SDGE achieves full new-concept recognition from a single example (1-shot) through immediate binding of the formed stable representation (h^) into explicit memory. Importantly, this is not presented as 1-shot gradient learning of network weights; rather, it is few-shot memory binding conditioned on the dynamical formation of (h^). The consequence is rapid acquisition without iterative epochs or parameter rewriting. Certified capability: Robustness under maximal temporal drift ((\\delta = 1.00)).In the Shift300 PoC drift definition, SDGE maintains its certified behavior even under maximal temporal perturbation ((\\delta=1.00)). The system preserves perfect performance after insertion (acc_new_after = 1.0, acc_seen_after = 1.0) with zero forgetting, indicating that successful grounding depends on the global dynamical trajectory leading to (h^*) rather than precise event-time alignment. Verifiability and reproducibility: Operational Certificates (Audit Trail).A central contribution of SDGE is that its claims are backed by auditable artifacts, including summary.json and induction_curve.csv, which record protocol parameters and before/after performance across shot counts. These files function as an operational audit trail: they allow third parties to verify that the new concept was not recognized prior to insertion, that recognition becomes perfect after binding, and that previously learned concepts remain intact under the same run conditions. Why this matters (Significance). Unlike classical Hopfield associative memory (which stores attractors by encoding patterns into recurrent weights) or typical memory-augmented neural networks (which learn to write/read via trainable controllers), SDGE treats the SNN as a dynamical state generator and performs append-only prototype binding of the formed commitment state (h^*) in an explicit bank—without re-engraving attractors into shared weights.","author":[{"family":"Salhab","given":"Najih"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18712293","URL":"https://doi.org/10.5281/zenodo.18712293","source":"datacite"},{"id":"doi:10.5281/zenodo.18712294","type":"article-journal","title":"Semantic Dynamic Grounding Engine (SDGE): A Mechanical, Certificate-Backed Solution to the Stability–Plasticity Dilemma in Spiking Neural Networks","abstract":"SDGE (Semantic Dynamic Grounding Engine) — Description & Significance The Semantic Dynamic Grounding Engine (SDGE) is a software-simulated Spiking Neural Network (SNN) system that frames learning and meaning as a dynamical commitment event rather than a purely statistical process of weight fitting. Given a temporal input stream, SDGE evolves a network state (h(t)) until it reaches an operational commitment time (t^), at which point it captures a stable representation (h^) and associates it with a concept identity via explicit, append-only memory binding (prototype storage). This design separates formation (the dynamical emergence of a stable state) from identity preservation (the storage of what that state means), enabling principled sequential learning without requiring gradient-based retraining for each new concept. Core architectural principle: Dynamical Formation vs. Explicit Binding.SDGE decouples the process of reaching a stable semantic representation from the storage of concept identity. The SNN dynamics generate (h^*) through stabilization, while concept identity is preserved in an explicit memory bank as a bound prototype. This “append-only binding” differs fundamentally from approaches where new knowledge is integrated by rewriting shared weights, and it provides a mechanistic basis for stable incremental growth. Certified capability: Zero-forgetting (N+1) induction (Stability–Plasticity).Under a rigorous (N+1) protocol ((K_{base}=8 \\rightarrow K_{total}=9)), SDGE demonstrates that a new concept can be integrated without degrading previously acquired concepts. In the certified runs, performance on previously learned concepts remains perfect (acc_seen_after = 1.0) while the new concept transitions from chance performance before insertion (acc_new_before = 0.0) to perfect recognition after insertion (acc_new_after = 1.0), with forgetting = 0.0. This provides an operationally testable resolution to the stability–plasticity dilemma within the stated protocol. Certified capability: 1-shot operational acquisition (binding-based, not gradient-based).SDGE achieves full new-concept recognition from a single example (1-shot) through immediate binding of the formed stable representation (h^) into explicit memory. Importantly, this is not presented as 1-shot gradient learning of network weights; rather, it is few-shot memory binding conditioned on the dynamical formation of (h^). The consequence is rapid acquisition without iterative epochs or parameter rewriting. Certified capability: Robustness under maximal temporal drift ((\\delta = 1.00)).In the Shift300 PoC drift definition, SDGE maintains its certified behavior even under maximal temporal perturbation ((\\delta=1.00)). The system preserves perfect performance after insertion (acc_new_after = 1.0, acc_seen_after = 1.0) with zero forgetting, indicating that successful grounding depends on the global dynamical trajectory leading to (h^*) rather than precise event-time alignment. Verifiability and reproducibility: Operational Certificates (Audit Trail).A central contribution of SDGE is that its claims are backed by auditable artifacts, including summary.json and induction_curve.csv, which record protocol parameters and before/after performance across shot counts. These files function as an operational audit trail: they allow third parties to verify that the new concept was not recognized prior to insertion, that recognition becomes perfect after binding, and that previously learned concepts remain intact under the same run conditions. Why this matters (Significance). Unlike classical Hopfield associative memory (which stores attractors by encoding patterns into recurrent weights) or typical memory-augmented neural networks (which learn to write/read via trainable controllers), SDGE treats the SNN as a dynamical state generator and performs append-only prototype binding of the formed commitment state (h^*) in an explicit bank—without re-engraving attractors into shared weights.","author":[{"family":"Salhab","given":"Najih"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18712294","URL":"https://doi.org/10.5281/zenodo.18712294","source":"datacite"},{"id":"doi:10.5281/zenodo.18664334","type":"article-journal","title":"The LEGACY Program: AGI  Lux Ferox Project","abstract":"This deposit presents the theoretical foundations, mathematical formalism, and practical implementation of the Lux Ferox project — an open research initiative proposing a cognitive architecture grounded in non-equilibrium thermodynamics and information physics, as an alternative paradigm to monolithic large language models. Core Contribution The central mechanism, termed Algorithmic Anger (Colère Algorithmique), is formally defined as a real-time total surprise metric S_total based on the weighted Kullback-Leibler divergence across sensory and semantic prediction streams: S_total = α · D_KL(P_model_sensory ‖ P_observed_sensory) + β · D_KL(P_model_semantic ‖ P_observed_context) This metric is anchored in three established theoretical frameworks: (1) Landauer's principle, whereby irreversible belief updates are interpreted as measurable dissipative events (W_cog ≥ k_B · T · ln2 · S_total); (2) Friston's Free Energy Principle, of which S_total constitutes a computationally tractable, discretized approximation for embedded real-time systems; and (3) non-equilibrium statistical mechanics, with the cognitive system modeled as an open Markovian system governed by a master equation over surprise states, including a full entropy production decomposition (σ = σ_env + σ_sys + σ_info). Architecture The proposed quadrivial cognitive architecture comprises four specialized compute layers: Neuromorphic layer — Spiking Neural Network (SNN) with surprise-coupled membrane dynamics and STDP meta-plasticity, targeting real-time KLD computation (<2 ms, <20 mJ per inference cycle) on NVIDIA A100/H100 via optimized CUDA kernels Classical silicon layer — Compact sovereign LLM inference (7nm process node, Sparse MoE) providing semantic world modeling and probabilistic context updating Wetware layer — Cortical organoid substrate (bio-hybrid MEA/optogenetic interface) providing morphogenetic plasticity and dynamic biological modulation of the α/β gain coefficients Quantum layer — D-Wave Advantage QPU (5640 qubits, Pegasus topology) for offline global policy optimization via Ising Hamiltonian formulation of the cognitive policy space Contents of this Deposit Full mathematical derivations: master equation for surprise dynamics, entropy production decomposition, fluctuation theorems, thermodynamic uncertainty relations, Fisher information geometry (natural gradient descent, Cramér-Rao bounds for surprise estimation), Bures-Wasserstein metric for hybrid quantum-classical distributions Complete CUDA implementations (A100/H100-optimized): KLD surprise kernel, event-driven SNN propagation kernel, adaptive α/β gain modulation kernel Projected benchmark metrics: <2 ms full cognitive cycle, <20 mJ/inference (vs. ~100 mJ for comparable dense transformer), O(N_active) event-driven complexity Minimal reproducible Python prototype (Google Colab, free tier) Benchmark dataset references: SWaT, WADI, Exathlon, custom PAL Robotics TIAGo trajectories Consortium architecture and European sovereign value chain documentation Novelty Claims This work advances four axes beyond the current state of the art: (1) first hardware implementation of KLD as a runtime inference signal (as opposed to a training loss); (2) dynamic biological modulation of surprise weighting coefficients via closed-loop organoid feedback; (3) explicit per-inference thermodynamic accounting grounded in Landauer's principle; (4) a fully sovereign European technology stack (CEA-Leti, Imec, X-FAB, Aleph Alpha, FinalSpark, EU Quantum Flagship). Limitations and Current TRL This work is currently at Technology Readiness Level 4 (component validation in laboratory environment). CUDA benchmarks are projected from A100 architectural specifications; full hardware validation is in progress. Wetware integration requires additional biological validation under EU Directive 2010/63. Quantum layer benchmarks are pending hybrid classical-quantum experimental runs. This deposit is produced by an independent researcher and h","author":[{"family":"Collective","given":"Lux"},{"family":"Mathieu","given":"François"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18664334","URL":"https://doi.org/10.5281/zenodo.18664334","source":"datacite"},{"id":"doi:10.5281/zenodo.18702138","type":"article-journal","title":"Sovereign-SNN","abstract":"Sovereign-SNN is a vertical, open-source neuromorphic architecture designed to bridge decentralized data streams with physical silicon. This project introduces a \"Zero-Permission\" methodology for spiking neural network (SNN) research, utilizing standard consumer hardware to bypass traditional, proprietary lab constraints. Core Architecture: Hardware Layer: Custom SystemVerilog Leaky Integrate-and-Fire (LIF) neural cores executing in real-time on a Digilent Basys 3 Artix-7 FPGA. Telemetry Layer: A memory-safe, high-speed Rust supervisor managing asynchronous hardware states and serial communication. Acceleration Layer: Custom CUDA kernels orchestrated by an NVIDIA RTX 5080, calculating synaptic weights and thresholding. Stimulus: Capable of utilizing high-entropy, live-wavelength telemetry (including Proof-of-Useful-Work streams) as biological training stimuli. This repository contains the core IP, hardware logic, Rust telemetry engine, and the foundational whitepaper detailing the \"Centaur Workflow\" utilized in its synthesis.","author":[{"family":"Montoya Cardenas","given":"Raul"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18702138","URL":"https://doi.org/10.5281/zenodo.18702138","source":"datacite"},{"id":"doi:10.5281/zenodo.18702137","type":"article-journal","title":"Sovereign-SNN","abstract":"Sovereign-SNN is a vertical, open-source neuromorphic architecture designed to bridge decentralized data streams with physical silicon. This project introduces a \"Zero-Permission\" methodology for spiking neural network (SNN) research, utilizing standard consumer hardware to bypass traditional, proprietary lab constraints. Core Architecture: Hardware Layer: Custom SystemVerilog Leaky Integrate-and-Fire (LIF) neural cores executing in real-time on a Digilent Basys 3 Artix-7 FPGA. Telemetry Layer: A memory-safe, high-speed Rust supervisor managing asynchronous hardware states and serial communication. Acceleration Layer: Custom CUDA kernels orchestrated by an NVIDIA RTX 5080, calculating synaptic weights and thresholding. Stimulus: Capable of utilizing high-entropy, live-wavelength telemetry (including Proof-of-Useful-Work streams) as biological training stimuli. This repository contains the core IP, hardware logic, Rust telemetry engine, and the foundational whitepaper detailing the \"Centaur Workflow\" utilized in its synthesis.","author":[{"family":"Montoya Cardenas","given":"Raul"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18702137","URL":"https://doi.org/10.5281/zenodo.18702137","source":"datacite"},{"id":"doi:10.5281/zenodo.18625622","type":"article-journal","title":"SNN-Genesis: A Preliminary Study on Iterative Adversarial Training of Large Language Models Using Spiking Neural Network Perturbations (v1)","abstract":"I present SNN-Genesis, an iterative adversarial training framework for large language models using noise perturbations and a cumulative \"Dream Journal\" strategy. In each round, noise is injected into the model's hidden layers to elicit adversarial hallucinations (\"nightmares\"), which are healed into refusal training data and accumulated across rounds. A fresh LoRA adapter is trained on the base model each round. Key Results (n=30 evaluation on Mistral-7B-Instruct-v0.3): Nightmare acceptance rate reduced from 53.3% to 43.3% (best, Morpheus/random) Clean knowledge accuracy maintained/improved: 80.0% → 83.3% Training loss decreased consistently: 6.36 → 4.63 Negative Alignment Tax: safety training did not degrade factual knowledge Key insight: random perturbations slightly outperformed SNN perturbations, suggesting the Dream Journal framework—not the perturbation source—is the primary driver of improvement Technical discovery: SFTTrainer double-templating bug in trl v5 that causes training collapse. This research was conducted as a collaborative effort between the human author and AI research assistants (Google Gemini, Anthropic Claude, OpenAI ChatGPT). Code: https://github.com/hafufu-stack/snn-genesis Related work: https://github.com/hafufu-stack/temporal-coding-simulation","author":[{"family":"Funasaki","given":"Hiroto"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18625622","URL":"https://doi.org/10.5281/zenodo.18625622","source":"datacite"},{"id":"doi:10.7302/28136","type":"article-journal","title":"Algorithm-Hardware Co-Design for Artificial Intelligence: From Energy-Efficient Edge Processing to High-Performance GPU Acceleration","abstract":"Deep learning has become a cornerstone of modern artificial intelligence (AI), driving transformative advances in fields such as computer vision, natural language processing, and generative models for text and images. However, as the complexity of neural networks continues to rise, the computational and memory demands necessary to maintain state-of-the-art performance have also increased significantly. This dissertation addresses the growing complexity of deep neural networks and the challenges associated with scaling and deploying them, particularly in resource-constrained environments such as edge devices. It explores the co-design of specialized hardware and neural networks, emphasizing the need for efficient integration between software and hardware to achieve optimal performance. This dissertation presents four works that aim to maximize the benefits of software optimizations, minimize accuracy loss during model simplification, and deliver real-world speedups on existing hardware platforms. The first work introduces a hardware accelerator for heterogeneous transform-domain neural networks, which reduces computational overhead through a specialized dataflow and optimized cache organization. Evaluations demonstrate that the proposed accelerator enhances energy efficiency and performance by up to 5.1x and 5.2x, respectively, for convolution layers, compared to previous dense accelerators. The second work presents an all-on-chip-weight implementation of a heterogeneous transform-domain neural engine that uses dense magnetoresistive non-volatile memory for weight storage. This processor achieves a peak efficiency of 9.51 TOPS/W for convolutional layers and 22.86 TOPS/W for fully connected layers, positioning it as a promising solution for energy-constrained edge applications. The third work proposes a one-hot multi-level leaky integrate-and-fire neuron model, which improves the accuracy-energy tradeoff in spiking neural networks by improving performance without compromising energy consumption. Evaluations reveal that networks using the proposed neurons enhance the accuracy-latency tradeoff for advanced network architectures, such as spike-driven transformers, achieving over 3% higher accuracy and requiring 4x fewer timesteps on ImageNet. Finally, the fourth work investigates techniques for optimizing structured-compressed foundation model inference on graphics processing units (GPUs), leveraging fused kernels and memory-efficient strategies to achieve substantial speedups. Together, these contributions not only advance the efficient deployment of deep learning models in resource-limited environments but also pave the way for future innovations in AI hardware and algorithm co-design.","author":[{"family":"Abillama","given":"Pierre"}],"issued":{"date-parts":[[2025]]},"DOI":"10.7302/28136","URL":"https://doi.org/10.7302/28136","source":"datacite"},{"id":"doi:10.5281/zenodo.18595933","type":"article-journal","title":"Activation-Scaled ANN-to-SNN Conversion with SNN Guardrail: A Unified Framework for AI Interpretability, Hallucination Detection, Real-Time Adversarial Defense, Neural Healing, Brain State Imaging, Hallucination Anatomy, and the Canary Head Paradigm (v9)","abstract":"I present a unified framework that extends ANN-to-SNN conversion beyond efficiency optimization to enable novel AI interpretability analysis, real-time adversarial defense, autonomous neural healing, brain state imaging, real-time hallucination anatomy, and the Canary Head Paradigm. My approach uses Spiking Neural Networks as \"computational microscopes\" to analyze black-box AI models. v9 Updates: NEW: Canary Head Paradigm — \"Liar Heads\" reinterpreted as \"Canary Heads\" (early-warning alarm systems). Monitoring only 3 heads (0.3% of total) achieves +5% accuracy over full-layer baseline with 97% compute reduction NEW: 5-Model Depth Scaling Law — Phi-2 (2.7B) reveals peak hallucination at 25% depth (shallow zone), establishing a ~3B critical parameter threshold for mid-layer convergence NEW: Canary's Eye Visualization — L10H17 attention heatmap during hallucination, showing the canary \"wakes up\" during generation steps 3–8 NEW: GQA Architecture Limitations — Qwen2.5 GQA (2 KV heads) produces systematic NaN under fp16, requiring fp32 for reliable entropy analysis NEW: AI-Assisted Research Acknowledgment — Transparent disclosure of human-AI collaborative methodology (Anthropic Claude, Google Gemini) v8 Updates: \"Moment of Lie\" — animated token-by-token heatmaps visualizing the exact moment hallucinations crystallize in mid-layer attention (L10–L18 in Mistral-7B) Token Economy — Surgical v3 mid-layer sniper achieves 72% compute savings by monitoring only 9/32 layers, with comparable accuracy Cross-Model Universality — Llama-3.2-3B (28 layers) confirms hallucination signature at 30–55% depth (peak L12, ΔH = −0.403 bits) Mid-Layer Hallucination Hypothesis: entropy disruption maximized at 30–55% of total network depth regardless of architecture (Mistral, Llama) v7 Updates: Entropy Evolution Discovery — Mistral-7B (fp16, 7.2B params) GPU validation reveals detection signals shift from TTFS latency (1-3B) to attention entropy (7B+) Mistral-7B attention entropy: +5.8σ separation (p = 2.22×10⁻⁹⁵), 100% detection accuracy (N=200) 6-Model Scaling Law validated (GPT-2, TinyLlama, Llama-3.2-1B, Llama-3.2-3B, Mistral-7B) Key Discovery: \"As models scale, the attack signature transforms from latency (brain freeze) to entropy (internal confusion) — but never disappears\" v6 Updates: N=1,000 Statistical Proof — Welch's t = −33.65 (p = 8.91×10⁻¹⁶⁴), Cohen's d = 2.13, 89.3% zero-shot detection accuracy on Llama-3.2-3B \"Visualizing the Ghost\" — First SNN-VAE visualization of LLM brain states during adversarial attacks (L2 distance = 3.287 between normal and jailbreak states) v5 Updates: Neural Healing v4A – Multi-stage progressive healing achieving 22% success rate on TinyLlama (1.1B) Mistral-7B (7B) experiment – Model-size-dependent threshold discovery HuggingFace Spaces v2.0 – Live 3-tab demo (Jailbreak/Healing/Hallucination) v4 Results: SNN Guardrail: 100% jailbreak detection rate (8/8 attack types) Scaling Law Discovery: TTFS sensitivity increases with model size (GPT-2: +3.1, TinyLlama: +4.2) Previous Results (v1-v3): Universal threshold formula: θ = 2.0 × max(activation) 100% accuracy preservation with hippocampal hybrid architecture GPT-2 attention TTFS analysis: +3.1 increase for meaningless inputs Hallucination detection: AUC 0.75 with ensemble classifier ViT-Base (86M) validation with CIFAR-10D Key insight: \"The neural fingerprint of hallucination is not just detectable — it is anatomically localized and predictable. The Canary Head paradigm reveals that specific attention heads serve as built-in alarm systems. The 5-model Depth Scaling Law shows a critical ~3B parameter threshold: smaller models hallucinate in shallow layers (15–25%), while larger models converge to a universal mid-layer zone (40–55%).\" 🔗 Live Demo: https://huggingface.co/spaces/hafufu-stack/snn-guardrail Code: https://github.com/hafufu-stack/temporal-coding-simulation/tree/main/ann-to-snn-converter This research employed a human-AI collaborative methodology. See Acknowledgm","author":[{"family":"Funasaki","given":"Hiroto"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18595933","URL":"https://doi.org/10.5281/zenodo.18595933","source":"datacite"},{"id":"doi:10.5281/zenodo.18528971","type":"article-journal","title":"Activation-Scaled ANN-to-SNN Conversion with SNN Guardrail: A Unified Framework for AI Interpretability, Hallucination Detection, Real-Time Adversarial Defense, Neural Healing, Brain State Imaging, and Real-Time Hallucination Anatomy (v8)","abstract":"I present a unified framework that extends ANN-to-SNN conversion beyond efficiency optimization to enable novel AI interpretability analysis, real-time adversarial defense, autonomous neural healing, brain state imaging, and real-time hallucination anatomy. My approach uses Spiking Neural Networks as \"computational microscopes\" to analyze black-box AI models. **v8 Updates:**- NEW: \"Moment of Lie\" — animated token-by-token heatmaps visualizing the exact moment hallucinations crystallize in mid-layer attention (L10–L18 in Mistral-7B)- NEW: Token Economy — Surgical v3 mid-layer sniper achieves 72% compute savings by monitoring only 9/32 layers, with comparable accuracy- NEW: Cross-Model Universality — Llama-3.2-3B (28 layers) confirms hallucination signature at 30–55% depth (peak L12, ΔH = −0.403 bits)- Mid-Layer Hallucination Hypothesis: entropy disruption maximized at 30–55% of total network depth regardless of architecture (Mistral, Llama) **v7 Updates:**- Entropy Evolution Discovery — Mistral-7B (fp16, 7.2B params) GPU validation reveals detection signals shift from TTFS latency (1-3B) to attention entropy (7B+)- Mistral-7B attention entropy: +5.8σ separation (p = 2.22×10⁻⁹⁵), 100% detection accuracy (N=200)- 6-Model Scaling Law validated (GPT-2, TinyLlama, Llama-3.2-1B, Llama-3.2-3B, Mistral-7B)- Key Discovery: \"As models scale, the attack signature transforms from latency (brain freeze) to entropy (internal confusion) — but never disappears\" **v6 Updates:**- N=1,000 Statistical Proof — Welch's t = −33.65 (p = 8.91×10⁻¹⁶⁴), Cohen's d = 2.13, 89.3% zero-shot detection accuracy on Llama-3.2-3B- \"Visualizing the Ghost\" — First SNN-VAE visualization of LLM brain states during adversarial attacks (L2 distance = 3.287 between normal and jailbreak states) **v5 Updates:**- Neural Healing v4A – Multi-stage progressive healing achieving 22% success rate on TinyLlama (1.1B)- Mistral-7B (7B) experiment – Model-size-dependent threshold discovery- HuggingFace Spaces v2.0 – Live 3-tab demo (Jailbreak/Healing/Hallucination) **v4 Results:**- SNN Guardrail: 100% jailbreak detection rate (8/8 attack types)- Scaling Law Discovery: TTFS sensitivity increases with model size (GPT-2: +3.1, TinyLlama: +4.2) **Previous Results (v1-v3):**- Universal threshold formula: θ = 2.0 × max(activation)- 100% accuracy preservation with hippocampal hybrid architecture- GPT-2 attention TTFS analysis: +3.1 increase for meaningless inputs- Hallucination detection: AUC 0.75 with ensemble classifier- ViT-Base (86M) validation with CIFAR-100 Key insight: \"The neural fingerprint of hallucination is not just detectable — it is anatomically localized. Across architectures (Mistral, Llama), model sizes (3B, 7B), and layer counts (28, 32), the hallucination signature consistently emerges at 30–55% of network depth. This 'hallucination zone' represents a universal feature of transformer-based language models.\" 🔗 Live Demo: https://huggingface.co/spaces/hafufu-stack/snn-guardrailCode: https://github.com/hafufu-stack/temporal-coding-simulation/tree/main/ann-to-snn-converter","author":[{"family":"Funasaki","given":"Hiroto"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18528971","URL":"https://doi.org/10.5281/zenodo.18528971","source":"datacite"},{"id":"doi:10.5281/zenodo.17952779","type":"article-journal","title":"Pyramidal Hybrid Neural Network Framework (BrainIAc_v1.0) : Technical Documentation and Experimental Results","abstract":"Project Overview Pre-publication technical documentation and experimental validation results for a hybrid neural network framework enabling gradient-based training of heterogeneous spiking neuron populations. Description/Summary: Key Features: Mixed neuron training: LIF + Izhikevich (10 types) + Hodgkin-Huxley (3 variants) trained together via gradient descent Large-scale validation: Up to 200 neurons with realistic cortical architectures Three learning modes: Separate, Simultaneous, Combined (pretrain + fine-tune) Pyramidal architecture: 1→50→1 hourglass structure (13 layers, 308 neurons) Real-time 3D visualization: 60 FPS interactive rendering with OpenGL Dynamic reconfiguration: Neurons can change type during simulation Comprehensive testing: 20+ validation scenarios, 75% success rate Validated Results: XOR problem: 56% loss reduction (pure LIF, 100 epochs) Mixed types: Stable convergence (4 LIF + 4 Izh + 4 HH, 80 epochs) ⭐ BREAKTHROUGH 100-neuron cortical microcircuit: Functional layer specialization validated 200-neuron cortical circuit: Realistic excitatory/inhibitory dynamics 13 neuron types: All show expected biological firing patterns Comprehensive Dataset: Technical documentation: 24 PDFs (~150 pages total) Core theory (9 PDFs) Framework architecture (5 PDFs) Advanced topics (10 PDFs: biological layers, visualization, roadmap) Experimental validation: 65+ CSV files with raw spike timing data Visual documentation: 67 screenshots Analysis: RESULTS_SUMMARY.md (36.8 KB detailed analysis) README.md (28.4 KB): Complete usage instructions and technical specifications CITATION.md (11.3 KB): Citation formats in BibTeX, APA, IEEE, MLA, Nature LICENSE.txt (2.4 KB): Full CC BY-NC 4.0 license terms License: CC BY-NC 4.0 (Non-Commercial) ✅ Academic research and educational use freely permitted with proper attribution ❌ Commercial use NOT permitted without explicit written permission 📧 For commercial licensing inquiries: theo.vallois@hotmail.fr Source Code: Full source code (BrainIAc framework) will be released under CC BY-NC 4.0 on GitHub upon paper acceptance (expected Q1 2026). Repository: https://github.com/EmpireStrikesBack/NeuroModel Demonstration Video: https://youtu.be/kiU609iozFw Real-time 3D visualization of 308-neuron pyramidal network with interactive controls. Contact: Author: Théo Vallois Email: theo.vallois@hotmail.fr GitHub: https://github.com/EmpireStrikesBack/NeuroModel","author":[{"family":"Vallois","given":"Théo"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.17952779","URL":"https://doi.org/10.5281/zenodo.17952779","source":"datacite"},{"id":"doi:10.5281/zenodo.17952780","type":"article-journal","title":"Pyramidal Hybrid Neural Network Framework (BrainIAc_v1.0) : Technical Documentation and Experimental Results","abstract":"Project Overview Pre-publication technical documentation and experimental validation results for a hybrid neural network framework enabling gradient-based training of heterogeneous spiking neuron populations. Description/Summary: Key Features: Mixed neuron training: LIF + Izhikevich (10 types) + Hodgkin-Huxley (3 variants) trained together via gradient descent Large-scale validation: Up to 200 neurons with realistic cortical architectures Three learning modes: Separate, Simultaneous, Combined (pretrain + fine-tune) Pyramidal architecture: 1→50→1 hourglass structure (13 layers, 308 neurons) Real-time 3D visualization: 60 FPS interactive rendering with OpenGL Dynamic reconfiguration: Neurons can change type during simulation Comprehensive testing: 20+ validation scenarios, 75% success rate Validated Results: XOR problem: 56% loss reduction (pure LIF, 100 epochs) Mixed types: Stable convergence (4 LIF + 4 Izh + 4 HH, 80 epochs) ⭐ BREAKTHROUGH 100-neuron cortical microcircuit: Functional layer specialization validated 200-neuron cortical circuit: Realistic excitatory/inhibitory dynamics 13 neuron types: All show expected biological firing patterns Comprehensive Dataset: Technical documentation: 24 PDFs (~150 pages total) Core theory (9 PDFs) Framework architecture (5 PDFs) Advanced topics (10 PDFs: biological layers, visualization, roadmap) Experimental validation: 65+ CSV files with raw spike timing data Visual documentation: 67 screenshots Analysis: RESULTS_SUMMARY.md (36.8 KB detailed analysis) README.md (28.4 KB): Complete usage instructions and technical specifications CITATION.md (11.3 KB): Citation formats in BibTeX, APA, IEEE, MLA, Nature LICENSE.txt (2.4 KB): Full CC BY-NC 4.0 license terms License: CC BY-NC 4.0 (Non-Commercial) ✅ Academic research and educational use freely permitted with proper attribution ❌ Commercial use NOT permitted without explicit written permission 📧 For commercial licensing inquiries: theo.vallois@hotmail.fr Source Code: Full source code (BrainIAc framework) will be released under CC BY-NC 4.0 on GitHub upon paper acceptance (expected Q1 2026). Repository: https://github.com/EmpireStrikesBack/NeuroModel Demonstration Video: https://youtu.be/kiU609iozFw Real-time 3D visualization of 308-neuron pyramidal network with interactive controls. Contact: Author: Théo Vallois Email: theo.vallois@hotmail.fr GitHub: https://github.com/EmpireStrikesBack/NeuroModel","author":[{"family":"Vallois","given":"Théo"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.17952780","URL":"https://doi.org/10.5281/zenodo.17952780","source":"datacite"},{"id":"doi:10.5445/ir/1000190105","type":"article-journal","title":"Physical Security of Emerging AI Hardware Accelerators: From Vulnerability to Countermeasures","abstract":"The widespread deployment of Artificial Intelligence (AI) at the edge has driven a paradigm shift toward domain-specific hardware accelerators. Emerging architectures-ranging from Analog Compute-in- Memory (CiM) and Non-Volatile Memory (NVM)-centric processors to Neuromorphic Computing and Hyperdimensional Computing (HDC)-promise orders-of-magnitude improvements in energy efficiency and latency. However, while these substrates are often used for their intrinsic robustness to noise and stochastic process variations, this dissertation demonstrates that such robustness does not translate into resilience against targeted physical attacks. This work systematically explores the physical security gap in post-Von Neumann computing, establishing that the very physical properties enabling efficiency often introduce novel, exploitable attack surfaces. To address these challenges, this thesis develops a set of cross-layer analysis frameworks and counter- measures, spanning device physics, circuit simulation, and real-hardware validation. First, the research investigates Analog Compute-in-Memory (CiM) based on ReRAM and STT-MRAM. By developing a physics-aware simulation framework, it is shown that analog non-idealities-such as device-to-device variability and random telegraph noise-manifest as data-dependent leakage signatures. These signatures allow adversaries to recover fixed weights using correlation-based and profiling attacks, enabling the extraction of proprietary neural network models. Second, the thesis uncovers a critical vulnerability in processor-centric NVM architectures. Through real-hardware experiments on STT-MRAM, a novel class of persistent fault attacks is demonstrated. By targeting the specific magnetic commit window of MRAM writes with nanosecond-scale voltage glitches, an attacker can induce stable, non-volatile bit corruptions in stored cryptographic keys. This persistence mechanism collapses the data complexity required for differential fault analysis (DFA) by orders of magnitude, enabling full AES key recovery with fewer than 20 ciphertext pairs. Third, the security of Neuromorphic Computing is analyzed through the introduction of FlexSpy, the first design-time security framework for thin-film transistor (TFT)-based Spiking Neural Networks (SNNs). The analysis reveals a unique quasi-DC leakage mechanism inherent to event-driven processing on flexible substrates. Despite the absence of internal triggers, an attacker can exploit this leakage to infer input classes and recover layer-wise spiking activity with high fidelity. Lightweight circuit-level countermeasures are proposed that suppress this leakage by up to 70% with minimal power overhead. Finally, the thesis presents a comprehensive security evaluation of FPGA-based Hyperdimensional Computing (HDC). It exposes a “robustness mismatch,” where HDC’s algorithmic error tolerance masks severe physical vulnerabilities. The work demonstrates successful Intellectual Property (IP) theft via deep-learning-assisted side-channel analysis and remote, internal sensing attacks using on-chip time-to-digital converters (TDCs). Furthermore, a targeted fault injection methodology (HyFault) is developed, capable of inducing precise misclassifications. In response, effective defenses-including dynamic masking and hypervector randomization-are introduced to restore security. Collectively, these contributions provide a foundational understanding of the physical security risks in emerging AI hardware, offering actionable design guidelines to ensure that the next generation of efficient edge accelerators is secure by design.","author":[{"family":"Sapui","given":"Brojogopal"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5445/ir/1000190105","URL":"https://doi.org/10.5445/ir/1000190105","source":"datacite"},{"id":"doi:10.48550/arxiv.2509.23516","type":"manuscript","title":"Network-Optimised Spiking Neural Network for Event-Driven Networking","abstract":"Delay-coupled systems often require low-latency decisions from sparse telemetry, where dense fixed-step neural inference is wasteful and can degrade near stability margins. We introduce Network-Optimised Spiking (NOS), a trainable two-state event-driven dynamical unit for delayed, graph-coupled streams, whose states map to a fast load variable and a slower recovery resource. NOS uses bounded excitability for finite buffers, explicit leak terms for service and damping, and graph-local coupling with per-link gates and communication delays, with differentiable resets compatible with surrogate-gradient training and neuromorphic execution. We prove existence and uniqueness of subthreshold equilibria, derive Jacobian-based stability conditions, and obtain a scalar network stability threshold that separates topology from node dynamics via a Perron-mode spectral condition. A stochastic arrival model aligned with telemetry smoothing explains increased variability as systems approach stability boundaries. On delayed graph forecasting and early-warning tasks from queue telemetry, NOS improves detection F1 and detection latency over MLP, RNN/GRU, and temporal GNN baselines under a common residual-based protocol, while providing calibration rules for resource-constrained deployments. Code and Demos: https://mbilal84.github.io/nos-snn-networking/","author":[{"family":"Bilal","given":"Muhammad"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2509.23516","URL":"https://doi.org/10.48550/arxiv.2509.23516","source":"datacite"},{"id":"doi:10.5281/zenodo.18304632","type":"article-journal","title":"Hybrid Spiking Language Model: Combining Spike Counts and Membrane Potentials for Energy-Efficient and Noise-Robust Character Prediction","abstract":"I propose a novel character-level language model using Spiking Neural Networks (SNNs) that combines both spike counts and membrane potentials for output prediction. This v3 extends the model with BitNet ternary weights and RWKV-inspired time-mixing. Key Findings (v3 NEW):- BitNet Mixed Precision: PPL 2.69 BEATS standard SNN (3.29)!- RWKV Time-Mixing: 36.1% improvement in long-range memory- Ultimate Architecture: 43.4% improvement combining all techniques- Multiplication-free reservoir: 50-70% of operations are additions only- 16-model ensemble achieves PPL 1.04 Key Findings (v1-v2):- SNN achieves BEST perplexity (PPL=9.90) vs DNN (11.28) and LSTM (15.67)- 14.7× more energy-efficient through sparse computation (only 7.6% of neurons fire)- 39.7% quality improvement from hybrid (spike + membrane) approach- Extreme compressibility: 80% neuron pruning and 4-bit quantization still work- 8× memory compression with minimal quality loss- Noise robust: No degradation at 30% input noise This v3 establishes hybrid SNNs with ternary weights as the optimal architecture for ultra-efficient edge AI language processing. Source code: https://github.com/hafufu-stack/snn-language-model","author":[{"family":"Funasaki","given":"Hiroto"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18304632","URL":"https://doi.org/10.5281/zenodo.18304632","source":"datacite"},{"id":"doi:10.48550/arxiv.2601.11261","type":"manuscript","title":"Effects of Introducing Synaptic Scaling on Spiking Neural Network Learning","abstract":"Spiking neural networks (SNNs) employing unsupervised learning methods inspired by neural plasticity are expected to be a new framework for artificial intelligence. In this study, we investigated the effect of multiple types of neural plasticity, such as spike-time-dependent plasticity (STDP) and synaptic scaling, on the learning in a winner-take-all (WTA) network composed of spiking neurons. We implemented a WTA network with multiple types of neural plasticity using Python. The MNIST and the Fashion-MNIST datasets were used for training and testing. We varied the number of neurons, the time constant of STDP, and the normalization method used in synaptic scaling to compare classification accuracy. The results demonstrated that synaptic scaling based on the L2 norm was the most effective in improving classification performance. By implementing L2-norm-based synaptic scaling and setting the number of neurons in both excitatory and inhibitory layers to 400, the network achieved classification accuracies of 88.84 % on the MNIST dataset and 68.01 % on the Fashion-MNIST dataset after one epoch of training.","author":[{"family":"Touda","given":"Shinnosuke"},{"family":"Okuno","given":"Hirotsugu"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2601.11261","URL":"https://doi.org/10.48550/arxiv.2601.11261","source":"datacite"},{"id":"doi:10.5281/zenodo.18294033","type":"article-journal","title":"Hybrid Spiking Language Model: Combining Spike Counts and Membrane Potentials for Energy-Efficient and Noise-Robust Character Prediction","abstract":"I propose a novel character-level language model using Spiking Neural Networks (SNNs) that combines both spike counts and membrane potentials for output prediction. Unlike conventional SNN approaches that only use spike counts, this hybrid method leverages the analog information contained in membrane potentials. Key Findings (v2):- SNN achieves BEST perplexity (PPL=9.90) vs DNN (11.28) and LSTM (15.67)- 14.7× more energy-efficient through sparse computation (only 7.6% of neurons fire)- 39.7% quality improvement from hybrid (spike + membrane) approach- Extreme compressibility: 80% neuron pruning and 4-bit quantization still work- 8× memory compression with minimal quality loss- Noise robust: No degradation at 30% input noise Source code: https://github.com/hafufu-stack/snn-language-model","author":[{"family":"Funasaki","given":"Hiroto"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18294033","URL":"https://doi.org/10.5281/zenodo.18294033","source":"datacite"},{"id":"doi:10.48550/arxiv.2601.10032","type":"manuscript","title":"Macroscopic dynamics of quadratic integrate-and-fire neurons subject to correlated noise","abstract":"The presence of correlated noise, arising from a mixture of independent fluctuations and a common noisy input shared across the neural population, is a ubiquitous feature of neural circuits, yet its impact on collective network dynamics remains poorly understood. We analyze a network of quadratic integrate-and-fire neurons driven by Gaussian noise with a tunable degree of correlation. Using the cumulant expansion method, we derive a reduced set of effective mean-field equations that accurately describe the evolution of the population's mean firing rate and membrane potential. Our analysis reveals a counterintuitive phenomenon: increasing the noise correlation strength suppresses the mean network activity, an effect we term correlated-noise-inhibited spiking. Furthermore, within a specific parameter regime, the network exhibits metastability, manifesting itself as spontaneous, noise-driven transitions between distinct high- and low-activity states. These results provide a theoretical framework for reducing the dynamics of complex stochastic networks and demonstrate how correlated noise can fundamentally regulate macroscopic neural activity, with implications for understanding state transitions in biological systems.","author":[{"family":"Wang","given":"Hui"},{"family":"Zheng","given":"Chunming"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2601.10032","URL":"https://doi.org/10.48550/arxiv.2601.10032","source":"datacite"},{"id":"doi:10.17605/osf.io/vjhk3","type":"article-journal","title":"Anarchy, Nonlinearity, and Emergence in International Relations: A Spiking Neural Network Framework Approach","abstract":"This research proposes a novel approach to modeling international relations by leveraging the Spiking Neural Networks (SNN) framework within a systems theory perspective. While existing systems theory methods, including agent-based modeling, network analysis, and dynamical systems approaches have provided valuable insights into international phenomena, they face critical limitations: insufficient temporal dynamics in capturing event-driven processes, absence of mechanisms for history-dependent relational evolution, difficulties in multi-scale integration across micro, meso, and macro levels, and challenges in systematically constructing appropriate dynamical equations from theoretical assumptions or empirical data. The SNN framework addresses these limitations through four key advantages. First, its event-driven spiking mechanism naturally unifies the discrete occurrence of international events with their continuous effects. Second, synaptic plasticity mechanisms provide biologically-inspired rules for modeling how interstate relations evolve based on historical interaction sequences, capturing strengthening, weakening, and memory effects. Third, the hierarchical neuron-synapse-network structure corresponds naturally to the actor-relation-system architecture of international relations, enabling coupled analysis across scales within a unified mathematical framework. Fourth, the decomposition into neuronal and synaptic dynamics creates a structured search space that makes automated equation discovery computationally tractable through methods such as symbolic regression and neural architecture search. Our methodology employs data-driven approaches, utilizing international event databases (GDELT, ICEWS), relational data (trade flows, alliances), and behavioral data (policy texts, voting records) to automatically explore and validate dynamical equations. The research aims to develop scalable open-source infrastructure supporting flexible customization of dynamics, apply the framework to concrete cases such as great power competition and regional alliance evolution, and employ dynamical systems theory and algebraic topology to analyze emergent mechanisms. Expected contributions include methodological innovation in computational IR modeling, technical outputs lowering interdisciplinary barriers, theoretical insights into systemic phase transitions and critical phenomena, and practical tools for policy simulation and counterfactual analysis.","author":[{"family":"Huang","given":"Wanhong"}],"issued":{"date-parts":[[2026]]},"DOI":"10.17605/osf.io/vjhk3","URL":"https://doi.org/10.17605/osf.io/vjhk3","source":"datacite"},{"id":"doi:10.17605/osf.io/vbwau","type":"article-journal","title":"Anarchy, Nonlinearity, and Emergence in International Relations: A Spiking Neural Network Framework Approach","abstract":"This research proposes a novel approach to modeling international relations by leveraging the Spiking Neural Networks (SNN) framework within a systems theory perspective. While existing systems theory methods, including agent-based modeling, network analysis, and dynamical systems approaches have provided valuable insights into international phenomena, they face critical limitations: insufficient temporal dynamics in capturing event-driven processes, absence of mechanisms for history-dependent relational evolution, difficulties in multi-scale integration across micro, meso, and macro levels, and challenges in systematically constructing appropriate dynamical equations from theoretical assumptions or empirical data. The SNN framework addresses these limitations through four key advantages. First, its event-driven spiking mechanism naturally unifies the discrete occurrence of international events with their continuous effects. Second, synaptic plasticity mechanisms provide biologically-inspired rules for modeling how interstate relations evolve based on historical interaction sequences, capturing strengthening, weakening, and memory effects. Third, the hierarchical neuron-synapse-network structure corresponds naturally to the actor-relation-system architecture of international relations, enabling coupled analysis across scales within a unified mathematical framework. Fourth, the decomposition into neuronal and synaptic dynamics creates a structured search space that makes automated equation discovery computationally tractable through methods such as symbolic regression and neural architecture search. Our methodology employs data-driven approaches, utilizing international event databases (GDELT, ICEWS), relational data (trade flows, alliances), and behavioral data (policy texts, voting records) to automatically explore and validate dynamical equations. The research aims to develop scalable open-source infrastructure supporting flexible customization of dynamics, apply the framework to concrete cases such as great power competition and regional alliance evolution, and employ dynamical systems theory and algebraic topology to analyze emergent mechanisms. Expected contributions include methodological innovation in computational IR modeling, technical outputs lowering interdisciplinary barriers, theoretical insights into systemic phase transitions and critical phenomena, and practical tools for policy simulation and counterfactual analysis.","author":[{"family":"Huang","given":"Wanhong"}],"issued":{"date-parts":[[2026]]},"DOI":"10.17605/osf.io/vbwau","URL":"https://doi.org/10.17605/osf.io/vbwau","source":"datacite"},{"id":"doi:10.48550/arxiv.2601.00805","type":"manuscript","title":"ChronoPlastic Spiking Neural Networks","abstract":"Spiking neural networks (SNNs) offer a biologically grounded and energy-efficient alternative to conventional neural architectures; however, they struggle with long-range temporal dependencies due to fixed synaptic and membrane time constants. This paper introduces ChronoPlastic Spiking Neural Networks (CPSNNs), a novel architectural principle that enables adaptive temporal credit assignment by dynamically modulating synaptic decay rates conditioned on the state of the network. CPSNNs maintain multiple internal temporal traces and learn a continuous time-warping function that selectively preserves task-relevant information while rapidly forgetting noise. Unlike prior approaches based on adaptive membrane constants, attention mechanisms, or external memory, CPSNNs embed temporal control directly within local synaptic dynamics, preserving linear-time complexity and neuromorphic compatibility. We provide a formal description of the model, analyze its computational properties, and demonstrate empirically that CPSNNs learn long-gap temporal dependencies significantly faster and more reliably than standard SNN baselines. Our results suggest that adaptive temporal modulation is a key missing ingredient for scalable temporal learning in spiking systems.","author":[{"family":"Chaudhry","given":"Sarim"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2601.00805","URL":"https://doi.org/10.48550/arxiv.2601.00805","source":"datacite"},{"id":"doi:10.5281/zenodo.16933929","type":"article-journal","title":"Combining Neuroplasticity and Neuromorphic Computing: A Paradigm Shift in Artificial Intelligence V2","abstract":"This version represents a substantively revised and corrected preprint of the original manuscript “Combining Neuroplasticity and Neuromorphic Computing.” The document has been expanded and refined to address issues of technical completeness, internal consistency, and mathematical rigor identified during post-draft review. Key improvements and corrections include: Explicit mathematical formalization of core mechanisms, including canonical formulations of Spike-Timing-Dependent Plasticity (STDP), reward-modulated three-factor learning rules, homeostatic stability regulation, and logistic adoption dynamics. These were previously referenced conceptually but not fully specified. Clear separation of algorithmic categories that are often conflated in neuromorphic literature, including ANN-to-SNN conversion (static inference transfer), hybrid offline/online learning, and true neuroplastic online adaptation. Addition of stability constraints and failure modes for plastic systems, explicitly addressing runaway dynamics, catastrophic drift, and consolidation requirements that limit real-world deployment. Refined roadmap analysis with concrete preconditions and falsifiable failure cases for both ambitious and conservative adoption trajectories. Scope-qualified energy-efficiency claims, clarifying where neuromorphic advantages are empirically strongest (sparse, event-driven workloads) and where conventional accelerators remain dominant. Removal or correction of speculative or weakly grounded references, replacing them with representative, well-established neuromorphic and spiking neural network literature. Expanded executive synthesis to contextualize the technical content for interdisciplinary readers while preserving full technical depth in the main body. This version is intended as a stable archival reference for researchers, practitioners, and policymakers interested in the long-horizon convergence of neuroplastic learning principles and neuromorphic computing architectures. The work positions this convergence as a gradual structural realignment toward adaptive, energy-efficient AI systems rather than a near-term replacement of existing GPU-centric paradigms.","author":[{"family":"Macfarland","given":"Anthony"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.16933929","URL":"https://doi.org/10.5281/zenodo.16933929","source":"datacite"},{"id":"doi:10.5281/zenodo.18142438","type":"article-journal","title":"Combining Neuroplasticity and Neuromorphic Computing: A Paradigm Shift in Artificial Intelligence V2","abstract":"This version represents a substantively revised and corrected preprint of the original manuscript “Combining Neuroplasticity and Neuromorphic Computing.” The document has been expanded and refined to address issues of technical completeness, internal consistency, and mathematical rigor identified during post-draft review. Key improvements and corrections include: Explicit mathematical formalization of core mechanisms, including canonical formulations of Spike-Timing-Dependent Plasticity (STDP), reward-modulated three-factor learning rules, homeostatic stability regulation, and logistic adoption dynamics. These were previously referenced conceptually but not fully specified. Clear separation of algorithmic categories that are often conflated in neuromorphic literature, including ANN-to-SNN conversion (static inference transfer), hybrid offline/online learning, and true neuroplastic online adaptation. Addition of stability constraints and failure modes for plastic systems, explicitly addressing runaway dynamics, catastrophic drift, and consolidation requirements that limit real-world deployment. Refined roadmap analysis with concrete preconditions and falsifiable failure cases for both ambitious and conservative adoption trajectories. Scope-qualified energy-efficiency claims, clarifying where neuromorphic advantages are empirically strongest (sparse, event-driven workloads) and where conventional accelerators remain dominant. Removal or correction of speculative or weakly grounded references, replacing them with representative, well-established neuromorphic and spiking neural network literature. Expanded executive synthesis to contextualize the technical content for interdisciplinary readers while preserving full technical depth in the main body. This version is intended as a stable archival reference for researchers, practitioners, and policymakers interested in the long-horizon convergence of neuroplastic learning principles and neuromorphic computing architectures. The work positions this convergence as a gradual structural realignment toward adaptive, energy-efficient AI systems rather than a near-term replacement of existing GPU-centric paradigms.","author":[{"family":"Macfarland","given":"Anthony"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18142438","URL":"https://doi.org/10.5281/zenodo.18142438","source":"datacite"},{"id":"doi:10.5281/zenodo.18071018","type":"article-journal","title":"Yatrogenesis/OldiesRules: OldiesRules v0.1.1","abstract":"🎸 OldiesRules v0.1.1 Revival of Classic Academic Simulators in Rust Cite this software @software{oldiesrules_2024, author = {Molina, Francisco}, title = {OldiesRules: Revival of Classic Academic Simulators in Rust}, version = {0.1.1}, doi = {10.5281/zenodo.18071019}, url = {https://github.com/Yatrogenesis/OldiesRules} } Supported Simulators (7 complete) | Simulator | Era | Purpose | |-----------|-----|---------| | GENESIS | 1988 | Compartmental neural modeling | | NEURON | 1984 | Cable equation, ion channels | | XPPAUT | 1990s | Bifurcation analysis | | AUTO | 1980s | Continuation algorithms | | COPASI | 2004 | Biochemical networks (SBML) | | Brian | 2008 | Spiking neural networks | | NEST | 2001 | Large-scale network simulation | Features Modern GUI: Real-time visualization with egui Interactive CLI: Fuzzy search, progress bars, wizard mode Full Format Compatibility: SLI, HOC, NMODL, ODE, SBML Installation cargo install --git https://github.com/Yatrogenesis/OldiesRules oldies-cli cargo install --git https://github.com/Yatrogenesis/OldiesRules oldies-gui Ecosystem Part of Yatrogenesis Scientific Computing Suite: HumanBrain - GPU neural simulation Rosetta - Legacy code transpiler","author":[{"family":"Molina-Burgos","given":"Francisco"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.18071018","URL":"https://doi.org/10.5281/zenodo.18071018","source":"datacite"},{"id":"doi:10.5281/zenodo.18071053","type":"article-journal","title":"Yatrogenesis/OldiesRules: OldiesRules v0.1.1","abstract":"🎸 OldiesRules v0.1.1 Revival of Classic Academic Simulators in Rust Cite this software @software{oldiesrules_2024, author = {Molina, Francisco}, title = {OldiesRules: Revival of Classic Academic Simulators in Rust}, version = {0.1.1}, doi = {10.5281/zenodo.18071019}, url = {https://github.com/Yatrogenesis/OldiesRules} } Supported Simulators (7 complete) | Simulator | Era | Purpose | |-----------|-----|---------| | GENESIS | 1988 | Compartmental neural modeling | | NEURON | 1984 | Cable equation, ion channels | | XPPAUT | 1990s | Bifurcation analysis | | AUTO | 1980s | Continuation algorithms | | COPASI | 2004 | Biochemical networks (SBML) | | Brian | 2008 | Spiking neural networks | | NEST | 2001 | Large-scale network simulation | Features Modern GUI: Real-time visualization with egui Interactive CLI: Fuzzy search, progress bars, wizard mode Full Format Compatibility: SLI, HOC, NMODL, ODE, SBML Installation cargo install --git https://github.com/Yatrogenesis/OldiesRules oldies-cli cargo install --git https://github.com/Yatrogenesis/OldiesRules oldies-gui Ecosystem Part of Yatrogenesis Scientific Computing Suite: HumanBrain - GPU neural simulation Rosetta - Legacy code transpiler","author":[{"family":"Molina-Burgos","given":"Francisco"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.18071053","URL":"https://doi.org/10.5281/zenodo.18071053","source":"datacite"},{"id":"doi:10.5281/zenodo.18091384","type":"article-journal","title":"quantum_inspired_spiking_simulator.py — Quantum-Noise-Injected LIF Neural Simulator","abstract":"quantum_inspired_spiking_simulator.py v1.0 — Quantum-Noise-Injected LIF Neural Simulator A zero-setup computational model injecting physically accurate quantum sensor noise (NV-center 1/f or OPM white) into the driving current of a Leaky Integrate-and-Fire (LIF) neuronal population. Features • Zero extra setup — single file (numpy + matplotlib) • Consistent noise physics with companion quantum sensor tools • Biological realism: absolute refractory period (2 ms) • Clear visualization with artificial spike peaks (+20 mV) • Key metrics: mean firing rate, ISI coefficient of variation, population synchrony index (χ) • Raster + membrane potential plots Dependencies • Requires numpy>=1.21 • Requires matplotlib>=3.5 — only for --plot Intended for quantum neuroscience and theoretical neuroscientists exploring how fundamental quantum fluctuations might propagate to macroscopic spiking dynamics, irregularity, and synchrony in speculative quantum biology frameworks. Part of Phase 4: Quantum Neuroscience & Sensing (2025–2026 open quantum utility toolkit sprint). Real usage: python quantum_inspired_spiking_simulator.py --demo classical --plot python quantum_inspired_spiking_simulator.py --demo nv --noise-level 3.0 --neurons 100 --duration 1000 --plot python quantum_inspired_spiking_simulator.py --demo opm --base-current 18.0 --plot Made by Britt (2025) — MIT License","author":[],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.18091384","URL":"https://doi.org/10.5281/zenodo.18091384","source":"datacite"},{"id":"doi:10.5281/zenodo.18091385","type":"article-journal","title":"quantum_inspired_spiking_simulator.py — Quantum-Noise-Injected LIF Neural Simulator","abstract":"quantum_inspired_spiking_simulator.py v1.0 — Quantum-Noise-Injected LIF Neural Simulator A zero-setup computational model injecting physically accurate quantum sensor noise (NV-center 1/f or OPM white) into the driving current of a Leaky Integrate-and-Fire (LIF) neuronal population. Features • Zero extra setup — single file (numpy + matplotlib) • Consistent noise physics with companion quantum sensor tools • Biological realism: absolute refractory period (2 ms) • Clear visualization with artificial spike peaks (+20 mV) • Key metrics: mean firing rate, ISI coefficient of variation, population synchrony index (χ) • Raster + membrane potential plots Dependencies • Requires numpy>=1.21 • Requires matplotlib>=3.5 — only for --plot Intended for quantum neuroscience and theoretical neuroscientists exploring how fundamental quantum fluctuations might propagate to macroscopic spiking dynamics, irregularity, and synchrony in speculative quantum biology frameworks. Part of Phase 4: Quantum Neuroscience & Sensing (2025–2026 open quantum utility toolkit sprint). Real usage: python quantum_inspired_spiking_simulator.py --demo classical --plot python quantum_inspired_spiking_simulator.py --demo nv --noise-level 3.0 --neurons 100 --duration 1000 --plot python quantum_inspired_spiking_simulator.py --demo opm --base-current 18.0 --plot Made by Britt (2025) — MIT License","author":[],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.18091385","URL":"https://doi.org/10.5281/zenodo.18091385","source":"datacite"},{"id":"doi:10.48550/arxiv.2512.21659","type":"manuscript","title":"Metaboplasticity: The Reciprocal Regulation of Neuronal Activity and Cellular Energetics","abstract":"Standard Spiking Neural Network (SNN) models typically neglect metabolic constraints, treating neurons as energetically unconstrained components. We bridge this gap by implementing a conductance-based leaky integrate-and-fire (gLIF) microcircuit (N=5,000) in Brian2, using temperature-dependent Q10 scaling to as a biophysically grounded proxy to couple metabolic state with intrinsic excitability and synaptic plasticity. Our simulations revealed five distinct emergent properties: (1) Dynamics Bifurcation: Learning trajectories diverged significantly, with hypometabolic states plateauing near baseline and hypermetabolic states exhibiting non-linear, runaway potentiation; (2) STDP Window Deformation: Thermal stress structurally deformed the plasticity kernel, where hypermetabolism sharpened coincidence detection and hypometabolism flattened synaptic integration; (3) Signal Degradation: While metabolic rate positively correlated with connectivity strength, high-energy states caused synaptic saturation and a loss of sparse coding specificity; (4) Topological Shift: Network activity transitioned from sparse, asynchronous firing in energy-restricted states to pathological, seizure-like hypersynchronization in high-energy states ; and (5) Parametric Robustness: Sensitivity analysis confirmed these attractor states were intrinsic biophysical properties, robust across random network initializations. Collectively, these results define an \"inverted-U\" relationship between bioenergetics and learning, demonstrating that metabolic constraints are necessary hardware regulators for network stability.","author":[{"family":"Öner","given":"Ece"},{"family":"Denktaş","given":"Cenk"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2512.21659","URL":"https://doi.org/10.48550/arxiv.2512.21659","source":"datacite"},{"id":"doi:10.48550/arxiv.2512.21153","type":"manuscript","title":"ElfCore: A 28nm Neural Processor Enabling Dynamic Structured Sparse Training and Online Self-Supervised Learning with Activity-Dependent Weight Update","abstract":"In this paper, we present ElfCore, a 28nm digital spiking neural network processor tailored for event-driven sensory signal processing. ElfCore is the first to efficiently integrate: (1) a local online self-supervised learning engine that enables multi-layer temporal learning without labeled inputs; (2) a dynamic structured sparse training engine that supports high-accuracy sparse-to-sparse learning; and (3) an activity-dependent sparse weight update mechanism that selectively updates weights based solely on input activity and network dynamics. Demonstrated on tasks including gesture recognition, speech, and biomedical signal processing, ElfCore outperforms state-of-the-art solutions with up to 16X lower power consumption, 3.8X reduced on-chip memory requirements, and 5.9X greater network capacity efficiency.","author":[{"family":"Su","given":"Zhe"},{"family":"Indiveri","given":"Giacomo"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2512.21153","URL":"https://doi.org/10.48550/arxiv.2512.21153","source":"datacite"},{"id":"doi:10.48550/arxiv.2510.11291","type":"manuscript","title":"Network-Optimised Spiking Neural Network (NOS) Scheduling for 6G O-RAN: Spectral Margin and Delay-Tail Control","abstract":"This work presents a Network-Optimised Spiking (NOS) delay-aware scheduler for 6G radio access. The scheme couples a bounded two-state kernel to a clique-feasible proportional-fair (PF) grant head: the excitability state acts as a finite-buffer proxy, the recovery state suppresses repeated grants, and neighbour pressure is injected along the interference graph via delayed spikes. A small-signal analysis yields a delay-dependent threshold $k_\\star(Δ)$ and a spectral margin $δ= k_\\star(Δ) - gHρ(W)$ that compress topology, controller gain, and delay into a single design parameter. Under light assumptions on arrivals, we prove geometric ergodicity for $δ&gt;0$ and derive sub-Gaussian backlog and delay tail bounds with exponents proportional to $δ$. A numerical study, aligned with the analysis and a DU compute budget, compares NOS with PF and delayed backpressure (BP) across interference topologies over a $5$--$20$\\,ms delay sweep. With a single gain fixed at the worst spectral radius, NOS sustains higher utilisation and a smaller 99.9th-percentile delay while remaining clique-feasible on integer PRBs.","author":[{"family":"Bilal","given":"Muhammad"},{"family":"Xu","given":"Xiaolong"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2510.11291","URL":"https://doi.org/10.48550/arxiv.2510.11291","source":"datacite"},{"id":"doi:10.48550/arxiv.2512.18113","type":"manuscript","title":"Responses to transient perturbation can distinguish intrinsic from latent criticality in spiking neural populations","abstract":"The critical brain hypothesis posits that neural circuitry operates near criticality to reap the computational benefits of accessing a wide range of timescales. The theory of critical phenomena generally predicts heavy-tailed (power-law) correlations in space and time near criticality, but it has been argued that in the brain such correlations could be inherited from ``latent variables,'' such as external sensory signals that are not directly observed when recording from neural circuitry. Distinguishing whether heavy-tailed correlations in neural activity are intrinsically generated within a neural circuit or are driven by unobserved latent variables is crucial for properly interpreting circuit functions. We argue that measuring neural responses to sudden perturbative inputs, rather than correlations in ongoing activity, can disambiguate these cases. We demonstrate this approach in a model of stochastic spiking neuron populations receiving external latent input that can be tuned to a critical state. We propose a scaling theory for the covariance and response functions of the spiking network, which we validate with simulations. We end by discussing how our approach might generalize to models of neural populations with more realistic biophysical details.","author":[{"family":"Crosser","given":"Jacob"},{"family":"Brinkman","given":"Braden"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2512.18113","URL":"https://doi.org/10.48550/arxiv.2512.18113","source":"datacite"},{"id":"doi:10.5281/zenodo.17949751","type":"article-journal","title":"Neuro-Dimensional Architecture of Brain Waves: A Unified Theoretical Framework for Macroscopic Neural Dynamics","abstract":"We propose a novel theoretical synthesis, the theory of Neuro-Dimensional Ar- chitecture (NDA), which reframes canonical brain wave activity (Delta, Theta, Al- pha, Beta, Gamma) as emergent, low-dimensional projections of a brain navigating a high-dimensional, structurally constrained state space. Departing from the tradi- tional frequency-band paradigm, NDA posits that the brain’s moment-to-moment activity constitutes a trajectory through a neuro-dimensional landscape whose axes are defined by anatomical connectivity, dynamic coupling, and neuromodulatory tone. Observable electroencephalographic (EEG) rhythms arise when this high- dimensional trajectory collapses onto stable, low-dimensional manifolds or attrac- tors. We formalize this framework by integrating principles from dynamical systems theory, topological data analysis, and recurrent network models. NDA provides a unified, scale-bridging account of neural oscillations, linking microscale spiking dy- namics and mesoscale assembly activity to macroscopic EEG phenomena. The theory generates falsifiable predictions for cognition, consciousness, and pathology, proposing that shifts in brain state correspond to topological and dynamical phase transitions between attractor basins within this neuro-dimensional landscape.","author":[{"family":"Ashfaq","given":"Muhammad"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.17949751","URL":"https://doi.org/10.5281/zenodo.17949751","source":"datacite"},{"id":"doi:10.5281/zenodo.17949752","type":"article-journal","title":"Neuro-Dimensional Architecture of Brain Waves: A Unified Theoretical Framework for Macroscopic Neural Dynamics","abstract":"We propose a novel theoretical synthesis, the theory of Neuro-Dimensional Ar- chitecture (NDA), which reframes canonical brain wave activity (Delta, Theta, Al- pha, Beta, Gamma) as emergent, low-dimensional projections of a brain navigating a high-dimensional, structurally constrained state space. Departing from the tradi- tional frequency-band paradigm, NDA posits that the brain’s moment-to-moment activity constitutes a trajectory through a neuro-dimensional landscape whose axes are defined by anatomical connectivity, dynamic coupling, and neuromodulatory tone. Observable electroencephalographic (EEG) rhythms arise when this high- dimensional trajectory collapses onto stable, low-dimensional manifolds or attrac- tors. We formalize this framework by integrating principles from dynamical systems theory, topological data analysis, and recurrent network models. NDA provides a unified, scale-bridging account of neural oscillations, linking microscale spiking dy- namics and mesoscale assembly activity to macroscopic EEG phenomena. The theory generates falsifiable predictions for cognition, consciousness, and pathology, proposing that shifts in brain state correspond to topological and dynamical phase transitions between attractor basins within this neuro-dimensional landscape.","author":[{"family":"Ashfaq","given":"Muhammad"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.17949752","URL":"https://doi.org/10.5281/zenodo.17949752","source":"datacite"},{"id":"doi:10.48550/arxiv.2512.05868","type":"manuscript","title":"Predicting Price Movements in High-Frequency Financial Data with Spiking Neural Networks","abstract":"Modern high-frequency trading (HFT) environments are characterized by sudden price spikes that present both risk and opportunity, but conventional financial models often fail to capture the required fine temporal structure. Spiking Neural Networks (SNNs) offer a biologically inspired framework well-suited to these challenges due to their natural ability to process discrete events and preserve millisecond-scale timing. This work investigates the application of SNNs to high-frequency price-spike forecasting, enhancing performance via robust hyperparameter tuning with Bayesian Optimization (BO). This work converts high-frequency stock data into spike trains and evaluates three architectures: an established unsupervised STDP-trained SNN, a novel SNN with explicit inhibitory competition, and a supervised backpropagation network. BO was driven by a novel objective, Penalized Spike Accuracy (PSA), designed to ensure a network's predicted price spike rate aligns with the empirical rate of price events. Simulated trading demonstrated that models optimized with PSA consistently outperformed their Spike Accuracy (SA)-tuned counterparts and baselines. Specifically, the extended SNN model with PSA achieved the highest cumulative return (76.8%) in simple backtesting, significantly surpassing the supervised alternative (42.54% return). These results validate the potential of spiking networks, when robustly tuned with task-specific objectives, for effective price spike forecasting in HFT.","author":[{"family":"Ezinwoke","given":"Brian"},{"family":"Rhodes","given":"Oliver"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2512.05868","URL":"https://doi.org/10.48550/arxiv.2512.05868","source":"datacite"},{"id":"doi:10.48550/arxiv.2510.01012","type":"manuscript","title":"Random Feature Spiking Neural Networks","abstract":"Spiking Neural Networks (SNNs) as Machine Learning (ML) models have recently received a lot of attention as a potentially more energy-efficient alternative to conventional Artificial Neural Networks. The non-differentiability and sparsity of the spiking mechanism can make these models very difficult to train with algorithms based on propagating gradients through the spiking non-linearity. We address this problem by adapting the paradigm of Random Feature Methods (RFMs) from Artificial Neural Networks (ANNs) to Spike Response Model (SRM) SNNs. This approach allows training of SNNs without approximation of the spike function gradient. Concretely, we propose a novel data-driven, fast, high-performance, and interpretable algorithm for end-to-end training of SNNs inspired by the SWIM algorithm for RFM-ANNs, which we coin S-SWIM. We provide a thorough theoretical discussion and supplementary numerical experiments showing that S-SWIM can reach high accuracies on time series forecasting as a standalone strategy and serve as an effective initialisation strategy before gradient-based training. Additional ablation studies show that our proposed method performs better than random sampling of network weights.","author":[{"family":"Gollwitzer","given":"Maximilian"},{"family":"Dietrich","given":"Felix"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2510.01012","URL":"https://doi.org/10.48550/arxiv.2510.01012","source":"datacite"},{"id":"doi:10.48550/arxiv.2508.12846","type":"manuscript","title":"IzhiRISC-V -- a RISC-V-based Processor with Custom ISA Extension for Spiking Neuron Networks Processing with Izhikevich Neurons","abstract":"Spiking Neural Network processing promises to provide high energy efficiency due to the sparsity of the spiking events. However, when realized on general-purpose hardware -- such as a RISC-V processor -- this promise can be undermined and overshadowed by the inefficient code, stemming from repeated usage of basic instructions for updating all the neurons in the network. One of the possible solutions to this issue is the introduction of a custom ISA extension with neuromorphic instructions for spiking neuron updating, and realizing those instructions in bespoke hardware expansion to the existing ALU. In this paper, we present the first step towards realizing a large-scale system based on the RISC-V-compliant processor called IzhiRISC-V, supporting the custom neuromorphic ISA extension.","author":[{"family":"Szczerek","given":"Wiktor"},{"family":"Podobas","given":"Artur"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2508.12846","URL":"https://doi.org/10.48550/arxiv.2508.12846","source":"datacite"},{"id":"doi:10.5075/epfl-thesis-11448","type":"article-journal","title":"SPAD Image Sensors with Embedded Intelligence","abstract":"Single-photon avalanche diodes (SPADs) are solid-state photodetectors that can detect individual photons with picosecond timing precision, enabling powerful time-resolved imaging across scientific, industrial, and biomedical applications. Despite their unique sensitivity, conventional SPAD imaging workflows passively collect photons, transfer large volumes of raw data off-chip, and reconstruct results through offline post-processing, leading to inefficiencies in photon usage, high latency, and limited adaptability. This thesis explores the potential of embedded artificial intelligence (AI) for efficient, real-time, intelligent processing in SPAD imaging through hardware-software co-design, bringing computation directly to the sensor to process photon data in its native form. Two general frameworks are proposed, each representing a paradigm shift from the conventional process. The first framework is inspired by the power of artificial neural networks (ANNs) in computer vision. It employs recurrent neural networks (RNNs) that operate directly on timestamps of photon arrival, extracting temporal information in an event-driven manner. The RNN is trained and evaluated for fluorescence lifetime estimation, achieving high precision and robustness. Quantization and approximation techniques are explored to enable FPGA implementation. Based on this, an imaging system integrating a SPAD image sensor with an on-FPGA RNN is developed, enabling real-time fluorescence lifetime imaging and demonstrating generalizability to other time-resolved tasks. The second framework is inspired by the human visual system, employing spiking neural networks (SNNs) that operate directly on the asynchronous pulses generated by SPAD avalanche breakdown upon photon arrival, thereby enabling temporal analysis with ultra-low latency and energy-efficient computation. Two hardware-friendly SNN architectures, Transporter SNN and Reversed start-stop SNN are proposed, which transform the phase-coded spike trains into density-coded and inter-spike-interval-coded representations, enabling more efficient training and processing. Dedicated training methods are explored, and both architectures are validated through fluorescence lifetime imaging. Based on the Transporter SNN architecture, the first SPAD image sensor with on-chip spike encoder for active time-resolved imaging is developed. This thesis encompasses a full-stack imaging workflow, spanning SPAD image sensor design, FPGA implementation, software development, neural network training and evaluation, mathematical modeling, fluorescence lifetime imaging, and optical system setup. Together, these contributions establish new paradigms of intelligent SPAD imaging, where sensing and computation are deeply integrated. The proposed frameworks demonstrate significant gains in photon efficiency, processing speed, robustness, and adaptability, illustrating how embedded AI can transform SPAD systems from passive detectors into intelligent, adaptive, and autonomous imaging platforms for next-generation applications.","author":[{"family":"Lin","given":"Yang"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5075/epfl-thesis-11448","URL":"https://doi.org/10.5075/epfl-thesis-11448","source":"datacite"},{"id":"doi:10.64823/ijter.2621029","type":"article-journal","title":"Neuromorphic Computing: Current Progress and the Future of Brain-Inspired Computing","abstract":"Neuromorphic computing borrows its design logic from the nervous system rather than from the von Neumann architecture that has dominated computing for seventy years. Instead of shuttling data back and forth between separate memory and processing units, it favors event-driven communication, computation that happens close to (or inside) memory, and massive parallelism across simple processing elements. This paper takes stock of where the field currently stands: spiking neural networks, digital and analog processors, memristive and other emerging devices, the software ecosystems that support them, and application areas ranging from robotics and edge intelligence to biomedical monitoring and event-based vision. What emerges from the recent literature is a field that has largely moved past small proof-of-concept chips and is now building larger, more programmable platforms with tighter hardware-algorithm integration. Even so, real obstacles remain around training methods, benchmarking practices, programmability, device variability, and fabrication, and it is still unclear how much of the field's energy advantage survives contact with general-purpose workloads. The view taken here is that brain-inspired computing is heading toward a hybrid future: conventional digital processors will keep doing what they do best, while event-driven and in-memory accelerators take over the workloads where they have a genuine edge. Neuromorphic computing is therefore unlikely to displace mainstream AI hardware outright, but it looks well positioned to become a key ingredient in low-power, adaptive, real-time intelligence at the edge. Keywords: edge AI; neuromorphic computing; brain-inspired computing; spiking neural networks; memristors; in-memory computing","author":[{"family":"Jeejo","given":"Jisna"}],"issued":{"date-parts":[[2026]]},"DOI":"10.64823/ijter.2621029","URL":"https://doi.org/10.64823/ijter.2621029","source":"openalex"},{"id":"oa:W4412166147","type":"article-journal","title":"Transparent conductive oxides as a material platform for a realization of all-optical photonic neural networks","abstract":"Photonics integrated circuits have an enormous potential to serve as a framework for a new class of information processing machines and can enable ultrafast artificial neural networks. They can overcome the existing speed and power limits of the electronic processing elements and provide additional benefits of photonics such as high-bandwidth, sub-nanosecond latencies and low-energy interconnect credentials leading to a new paradigm called neuromorphic photonics. The main obstacle to realizing such a task is a lack of proper material platform that imposes serious requirements on the architecture of the network. Here we suggest and justify that transparent conductive oxides can be an excellent candidate for such a task as they provide nonlinearity and bistability under both optical and electrical inputs.","author":[{"family":"Gosciniak","given":"Jacek"},{"family":"Khurgin","given":"Jacob"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1038/s41598-025-96226-w","URL":"https://doi.org/10.1038/s41598-025-96226-w","source":"openalex"},{"id":"oa:W4415104965","type":"article-journal","title":"From pulses to plasticity: Analytical tools for memristive synapse design","abstract":"Neuromorphic device design demands a clear understanding of the dynamics governing conductance modulation under external stimuli. Many synaptic memristors can be described by a quasi-linear model, where a memory variable relaxes between two limiting states. Here, we derive analytical expressions for the response of such systems to trains of voltage pulses, providing closed formulations for paired-pulse facilitation (PPF), convergent potentiation, and frequency-dependent gain. This approach predicts how the memory variable evolves toward stationary values determined by device and stimulation parameters, offering a compact alternative to numerical simulations. We experimentally validate the model using a nanofluidic memristor based on a nanoporous membrane, showing that the predicted convergence closely matches measured potentiation and that the analytical PPF trends reproduce experimental data. These results establish a unified framework for describing spike-driven plasticity and enable reliable cross-comparison of synaptic behavior across memristive systems, facilitating their integration into neuromorphic circuits.","author":[{"family":"Riverasierra","given":"Gonzalo"},{"family":"Bisquert","given":"Juan"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1063/5.0289570","URL":"https://doi.org/10.1063/5.0289570","source":"openalex"},{"id":"oa:W4410855184","type":"article-journal","title":"Vector Ising spin annealer for minimizing Ising Hamiltonians","abstract":"Abstract Complex optimization problems can be solved via dedicated machines which encode the problem in the couplings of spin Hamiltonians. However, traditional physical minimizers often select excited states due to limitations in spin dynamics. We introduce the Vector Ising Spin Annealer (VISA), a framework in gain-based computing that leverages light-matter interactions. We show that VISA overcomes the limitations by enabling spins to operate within a three-dimensional space, thereby providing a robust solution for effectively minimizing Ising Hamiltonians. Our comparative analysis demonstrates VISA’s superior performance relative to conventional single-dimension spin optimizers, highlighting its capacity to surmount significant energy barriers in intricate landscapes. Detailed studies on cyclic and random graphs reveal VISA’s proficiency in dynamically evolving the energy landscape through time-dependent gain and penalty annealing, underscoring its potential in advancing the field of complex problem-solving in physics-inspired and physics-based computing.","author":[{"family":"Cummins","given":"James"},{"family":"Berloff","given":"Natalia"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1038/s42005-025-02145-7","URL":"https://doi.org/10.1038/s42005-025-02145-7","source":"openalex"},{"id":"oa:W4406478661","type":"article-journal","title":"Challenges and opportunities for validation of AI-based new approach methods","abstract":"The integration of artificial intelligence (AI) into new approach methods (NAMs) for toxicology rep­resents a paradigm shift in chemical safety assessment. Harnessing AI appropriately has enormous potential to streamline validation efforts. This review explores the challenges, opportunities, and future directions for validating AI-based NAMs, highlighting their transformative potential while acknowledging the complexities involved in their implementation and acceptance. We discuss key hurdles such as data quality, model interpretability, and regulatory acceptance, alongside opportunities including enhanced predictive power and efficient data integration. The concept of e-validation, an AI-powered framework for streamlining NAM validation, is presented as a comprehensive strategy to overcome limitations of traditional validation approaches, leveraging AI-powered modules for reference chemical selection, study simulation, mechanistic validation, and model training and evaluation. We propose robust validation strategies, including tiered approaches, performance benchmarking, uncertainty quantification, and cross-validation across diverse datasets. The importance of ongoing monitoring and refinement post-implementation is emphasized, addressing the dynamic nature of AI models. We consider ethical implications and the need for human oversight in AI-driven toxicology and outline the impact of trends in AI devel­opment, research priorities, and a vision for the integration of AI-based NAMs in toxicological practice, calling for collaboration among researchers, regulators, and industry stakeholders. We describe the vision of companion AI post-validation agents to keep methods and their validity status current. By addressing these challenges and opportunities, the scientific community can harness the potential of AI to enhance predictive toxicology while reducing reliance on traditional animal testing and increasing human relevance and translational capabilities. Plain language summaryScientists are using artificial intelligence (AI) to develop new ways of assessing chemical safety that do not rely on animal experiments. These methods can be faster, more accurate, more human-relevant, and more ethical than traditional approaches. However, before these new methods can be widely used, we need to make sure they are reliable and trustworthy. This article discusses the challenges in validating AI-based safety testing methods, such as ensuring data quality and making AI decisions transparent and understandable, and proposes strategies for thorough validation and ongoing monitoring of these AI methods. It also explores opportunities to use AI to simulate experi­ments, analyze complex biological information, and support validation of diverse NAMs. We emphasize the importance of collaboration among researchers, regulators, and industry to develop responsible AI use in toxicology. By addressing these challenges, we can harness AI’s power to improve chemical safety testing while reducing animal use.","author":[{"family":"Härtung","given":"Thomas"},{"family":"Kleinstreuer","given":"Nicole"}],"issued":{"date-parts":[[2025]]},"DOI":"10.14573/altex.2412291","URL":"https://doi.org/10.14573/altex.2412291","source":"openalex"},{"id":"oa:W4408819106","type":"article-journal","title":"Stochastic compact model for memory and threshold switching memristors","abstract":"Memristors are electron devices whose resistance changes according to the history of electrical signals applied to their two terminals. These resistance changes can remain for very long times or relax after a short time. Thus, memristors can be used as electronic synapses and neurons in artificial neural networks implemented in hardware. These fully memristive neuromorphic circuits are mainly intended for artificial intelligence applications. In this work, we explore the properties of a stochastic compact model for the electrical behavior of memristors in the two regimes of long-time (non-volatile) and short-time (volatile) memory for synapse and neuron functions, respectively. In the case of non-volatile memristors, we focus on potentiation/depression transients, programming energy, programming time, and power requirements for writing conductance weights in artificial network crossbars. As for volatile memristors, we consider the modeling of their behavior in the context of a leaky-integrate-and-fire neuron, demonstrating how the model captures the threshold activation function and the input signal frequency dependence.","author":[{"family":"Suñé","given":"J"},{"family":"Miranda","given":"E"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1063/5.0255043","URL":"https://doi.org/10.1063/5.0255043","source":"openalex"},{"id":"oa:W7106521575","type":"article-journal","title":"Ferroelectric Field-Effects for Neuromorphic Hardware","abstract":"On conventional computers, the performance of AI models is limited by the data transfer between the memory and the processor. Compute-in-Memory architectures offer a new paradigm: Vector-Matrix Multiplications may be performed by a voltage drop through a matrix of programmable resistances, the “synaptic weights”. Ferroelectric materials are excellent candidates for their realization:[1] in a two- or three-terminals geometry and in combination with a semiconducting oxide,[2], [3], [4] the conductance is programmed by controlling the configuration of the ferroelectric domains. The unique fluorite unit cell of HfZrO4 allows for the stabilization of ferroelectricity below 3 nm,[5] facilitating the scaling of synaptic weights. The mechanisms governing the resistive switching in WOx / HZO-SL (5 nm) bilayers are discussed. The effect of the programming pulse duration and amplitude on the polarization switching are investigated, from milliseconds to nanoseconds timescales. Devices of different sizes and shapes are measured down to 500 nm in dimension. For an device size of 1 micrometer square, an On/Off ratio as high as 8 is obtained for 20 ns pulses, a 4-fold improvement compared to 40 um devices. The relatively low crystallization temperature of polycrystalline hafnium oxide / zirconium oxide superlattices (HZO-SL) is compatible with the Back-End-Of-Line (BEOL) of CMOS transistors.[6], [7] These results not only demonstrate the functionalization of the BEOL with synaptic weights, but also pave the way for the integration of ferroelectric field-effect transistors with Beyond CMOS semiconductors. [1] T. Mikolajick, et al., “From Ferroelectric Material Optimization to Neuromorphic Devices,” Advanced Materials, 2023, doi: 10.1002/adma.202206042. [2] L. Bégon-Lours et al., “Scaled, Ferroelectric Memristive Synapse for Back-End-of-Line Integration with Neuromorphic Hardware,” Advanced Electronic Materials, 2022, doi: 10.1002/aelm.202101395. [3] M. Halter et al., “Back-End, CMOS-Compatible Ferroelectric Field-Effect Transistor for Synaptic Weights,” ACS Appl. Mater. Interfaces, 2020, doi: 10.1021/acsami.0c00877. [4] M. Halter, et al., “A multi-timescale synaptic weight based on ferroelectric hafnium zirconium oxide,” COMMUNICATIONS MATERIALS, 2023, doi: 10.1038/s43246-023-00342-x. [5] L. Bégon-Lours et al., “Effect of cycling on ultra-thin HfZrO4, ferroelectric synaptic weights,” Neuromorph. Comput. Eng., 2022, doi: 10.1088/2634-4386/ac5b2d. [6] L. Bégon-Lours et al., “Back-End-of-Line Integration of Synaptic Weights using HfO2/ZrO2 Nanolaminates,” Advanced Electronic Materials, 2024, doi: 10.1002/aelm.202300649. [7] R. Hamming-Green, et al., “Multi-Level, Low-Voltage Programming of Ferroelectric HfO 2 /ZrO 2 Nanolaminates Integrated in the Back-End-Of-Line,” in 2024 8th IEEE EDTM, Bangalore, India, 2024, doi: 10.1109/EDTM58488.2024.10511719.","author":[{"family":"Begon-Lours","given":"Laura"}],"issued":{"date-parts":[[2025]]},"DOI":"10.3929/ethz-c-000787924","URL":"https://doi.org/10.3929/ethz-c-000787924","source":"openalex"},{"id":"oa:W4410446803","type":"article-journal","title":"Forecasting the future: From quantum chips to neuromorphic engineering and bio-integrated processors","abstract":"Computing is at the forefront of advances in human civilization and technology. Alongside communication, it is a critical technology that often drives improvements in many other scientific fields. Over the last 60 years, computing has achieved unprecedented advances attributed to scaling laws predicted by Moore’s law, whereby the number of components on a chip doubles every two years, and Dennard’s voltage scaling, whereby the voltage of components is also halved. However, continuous similar advances are no longer necessarily guaranteed, leading to the potential emergence of the so-called ‘Beyond Moore’s Law’ era. Even while traditional transistor scaling is ending, the demand for computing continues to grow in all sectors of the economy by almost orders of magnitude. The end of traditional scaling is expected to result in a computing crunch due to an energy crisis, necessitating the exploration of renewable sources of energy and computers that consume far less energy. Among many possible alternative computing architectures, neuromorphic computing based on brain-inspired computing is one of the most promising (Feng et al., 2018; Himelstein, 2022; Schell, 2024).","author":[{"family":"Nandan","given":"Botlagunta"}],"issued":{"date-parts":[[2025]]},"DOI":"10.70593/978-93-49910-47-8_12","URL":"https://doi.org/10.70593/978-93-49910-47-8_12","source":"openalex"},{"id":"oa:W4412929968","type":"article-journal","title":"Large-scale photonic processors and their applications","abstract":"Digital electronics have been essential in all technological breakthroughs over the past decades, from the advent of the internet to the miniaturization of portable devices with ever-increasing capabilities. While they continue to drive technological innovation, certain applications demanding extraordinary bandwidth and performance reveal the need for complementary emerging technologies. Photonics is particularly well-suited for this purpose, leveraging light to process large volumes of information with minimal latency. The scalability of photonic integrated circuits (PICs) is advancing rapidly, with increasing numbers of actuators improving performance and extending applicability across different domains. Analyzing two decades of development, this study extracts the scalability laws guiding photonics evolution, doubling every two years and predicting a transition from hundreds to 10 5 actuators in PICs within six years. We also tackle key challenges and emphasize the crucial synergy between electronics, software, and photonics, which will determine the market penetration and industry impact of large-scale photonic processors.","author":[{"family":"Pérez","given":"Daniel"},{"family":"Torrijosmorán","given":"Luis"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1038/s44310-025-00075-4","URL":"https://doi.org/10.1038/s44310-025-00075-4","source":"openalex"},{"id":"doi:10.5281/zenodo.3773845","type":"article-journal","title":"Rockpool Documentation","abstract":"Rockpool is a Python machine-learning package for designing, building and deploying event-driven neural network applications, particularly for Neuromorphic computing hardware. Rockpool provides a simple and comfortable API for working with spiking neural networks. Along with PyTorch and Jax support, Rockpool provides accellerated training for SNNs, and an extensible deployment pipeline.","author":[{"family":"Muir","given":"Dylan"},{"family":"Bauer","given":"Felix"},{"family":"Weidel","given":"Philipp"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.3773845","URL":"https://doi.org/10.5281/zenodo.3773845","source":"datacite"},{"id":"doi:10.5281/zenodo.17861618","type":"article-journal","title":"Rockpool Documentation","abstract":"Rockpool is a Python machine-learning package for designing, building and deploying event-driven neural network applications, particularly for Neuromorphic computing hardware. Rockpool provides a simple and comfortable API for working with spiking neural networks. Along with PyTorch and Jax support, Rockpool provides accellerated training for SNNs, and an extensible deployment pipeline.","author":[{"family":"Muir","given":"Dylan"},{"family":"Bauer","given":"Felix"},{"family":"Weidel","given":"Philipp"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.17861618","URL":"https://doi.org/10.5281/zenodo.17861618","source":"datacite"},{"id":"oa:W4410424827","type":"article-journal","title":"Quantum simulations of complex systems","abstract":"Abstract In this review, we give a brief overview of quantum simulation as applied to the study of complex systems. In particular, we cover the basic ideas of quantum simulation, neuromorphic computation, the Sachdev–Ye–Kitaev model, as well as applications to quantum batteries.","author":[{"family":"Morsch","given":"O"},{"family":"Palma","given":"GM"},{"family":"Rossini","given":"Davide"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1007/s40766-025-00069-0","URL":"https://doi.org/10.1007/s40766-025-00069-0","source":"openalex"},{"id":"doi:10.6082/p36nq-7t222","type":"article-journal","title":"Nanoenabled Trainable Systems: From Biointerfaces to Biomimetics","abstract":"In the dynamic biological system, cells and tissues adapt to diverse environmental conditions and form memories, an essential aspect of training for survival and evolution. An understanding of the biological training principles will inform the design of biomimetic materials whose properties evolve with the environment and offer routes to programmable soft materials, neuromorphic computing, living materials, and biohybrid robotics. In this perspective, we examine the mechanisms by which cells are trained by environmental cues. We outline the artificial platforms that enable biological training and examine the relationship between biological training and biomimetic materials design. We place emphasis on nanoscale material platforms which, given their applicability to chemical, mechanical and electrical stimulation, are critical to bridging natural and synthetic systems.","author":[{"family":"Li","given":"Pengju"},{"family":"Kim","given":"Saehyun"},{"family":"Tian","given":"Bozhi"}],"issued":{"date-parts":[[2022]]},"DOI":"10.6082/p36nq-7t222","URL":"https://doi.org/10.6082/p36nq-7t222","source":"datacite"},{"id":"doi:10.6082/m2hxq-4h460","type":"article-journal","title":"Nanoenabled Trainable Systems: From Biointerfaces to Biomimetics","abstract":"In the dynamic biological system, cells and tissues adapt to diverse environmental conditions and form memories, an essential aspect of training for survival and evolution. An understanding of the biological training principles will inform the design of biomimetic materials whose properties evolve with the environment and offer routes to programmable soft materials, neuromorphic computing, living materials, and biohybrid robotics. In this perspective, we examine the mechanisms by which cells are trained by environmental cues. We outline the artificial platforms that enable biological training and examine the relationship between biological training and biomimetic materials design. We place emphasis on nanoscale material platforms which, given their applicability to chemical, mechanical and electrical stimulation, are critical to bridging natural and synthetic systems.","author":[{"family":"Li","given":"Pengju"},{"family":"Kim","given":"Saehyun"},{"family":"Tian","given":"Bozhi"}],"issued":{"date-parts":[[2022]]},"DOI":"10.6082/m2hxq-4h460","URL":"https://doi.org/10.6082/m2hxq-4h460","source":"datacite"},{"id":"doi:10.48550/arxiv.2412.16111","type":"manuscript","title":"How random connectivity shapes the fluctuating dynamics of finite-size neural populations","abstract":"Mesoscopic models of finite-size neuronal populations are crucial to understand the dynamics of neural networks in the brain, especially their fluctuations and response to stimuli. However, current theories to derive such models are based on homogeneous all-to-all (full) connectivity. This assumption neglects the variance in the connectivity of biologically realistic networks with connection probabilities $p&lt;1$ (non-full connectivity). To gain insight into the different fluctuation mechanisms underlying neural variability at the population level, we derive and analyze a stochastic mean-field model for finite-size networks of Poisson neurons with random connectivity (including non-full connectivity), external noise and disordered mean inputs. We treat the quenched disorder of the connectivity by an annealed approximation enabling a doubly stochastic description of synaptic inputs for finite network size. A further reduction leads to a low-dimensional closed system of coupled Langevin equations for the mean and variance of the membrane potentials as well as a variable capturing finite-size fluctuations. Compared to microscopic simulations, the mesoscopic model describes the fluctuations and nonlinearities well and outperforms previous theories that neglected the variance in the connectivity. The joint effect of connectivity disorder and finite network size can be analytically understood by a softening of the effective nonlinearity and the multiplicative character of spiking noise. The mesoscopic theory shows that quenched disorder can stabilize the asynchronous state, and it correctly predicts large quantitative and non-trivial qualitative effects of connection probability on the variance of the population firing rate and its dependence on stimulus strength. In conclusion, our theory elucidates how disordered connectivity shapes nonlinear dynamics and fluctuations of neural populations.","author":[{"family":"Greven","given":"Nils"},{"family":"Ranft","given":"Jonas"},{"family":"Schwalger","given":"Tilo"}],"issued":{"date-parts":[[2024]]},"DOI":"10.48550/arxiv.2412.16111","URL":"https://doi.org/10.48550/arxiv.2412.16111","source":"datacite"},{"id":"doi:10.48550/arxiv.2412.06355","type":"manuscript","title":"Scalable Dendritic Modeling Advances Expressive and Robust Deep Spiking Neural Networks","abstract":"Dendritic computation endows biological neurons with rich nonlinear integration and high representational capacity, yet it is largely missing in existing deep spiking neural networks (SNNs). Although detailed multi-compartment models can capture dendritic computations, their high computational cost and limited flexibility make them impractical for deep learning. To combine the advantages of dendritic computation and deep network architectures for a powerful, flexible and efficient computational model, we propose the dendritic spiking neuron (DendSN). DendSN explicitly models dendritic morphology and nonlinear integration in a streamlined design, leading to substantially higher expressivity than point neurons and wide compatibility with modern deep SNN architectures. Leveraging the efficient formulation and high-performance Triton kernels, dendritic SNNs (DendSNNs) can be efficiently trained and easily scaled to deeper networks. Experiments show that DendSNNs consistently outperform conventional SNNs on classification tasks. Furthermore, inspired by dendritic modulation and synaptic clustering, we introduce the dendritic branch gating (DBG) algorithm for task-incremental learning, which effectively reduces inter-task interference. Additional evaluations show that DendSNNs exhibit superior robustness to noise and adversarial attacks, along with improved generalization in few-shot learning scenarios. Our work firstly demonstrates the possibility of training deep SNNs with multiple nonlinear dendritic branches, and comprehensively analyzes the impact of dendrite computation on representation learning across various machine learning settings, thereby offering a fresh perspective on advancing SNN design.","author":[{"family":"Huang","given":"Yifan"},{"family":"Fang","given":"Wei"},{"family":"Ma","given":"Zhengyu"},{"family":"Li","given":"Guoqi"},{"family":"Tian","given":"Yonghong"}],"issued":{"date-parts":[[2024]]},"DOI":"10.48550/arxiv.2412.06355","URL":"https://doi.org/10.48550/arxiv.2412.06355","source":"datacite"},{"id":"doi:10.17605/osf.io/7cdku","type":"article-journal","title":"The role of network theta and alpha oscillations in sustained and selective attention in humans","abstract":"Attention is the ability to preferentially orient limited perceptual processing resources towards a specific information stream when automatic processes are insufficient (Shiffrin &amp; Schneider, 1977). Attention has been investigated broadly within two task paradigms. Sustained and selective attention are typically investigated in different paradigms and recruit distinct neural processes. First, task paradigms that investigate “sustained attention” require that participant resist distraction and stay on task for a long period of time. With a sufficiently demanded task, such as a divided attention paradigm with an infrequent stimulus that is difficult to detect, a greater level of sustained attention is required (Sarter, Givens, &amp; Bruno, 2001). Animal work investigating the neural basis of sustained attention has utilized the 5-choice serial reaction time task (5CSRTT) (Robbins, 2002). In the 5CSRTT, a visual stimulus can be presented in one of five locations during a delay period of variable time. By presenting the visual stimulus only briefly, greater levels of sustained attention are required for successful detection . This research has uncovered a pivotal role for theta oscillations (4-8 Hz) in sustained attention (Helfrich et al., 2018; Sellers et al., 2016; Yu et al., 2018). Second, task paradigms that investigate “selective attention” require that participants preferentially attend to one information stream over another information stream (Treisman, 1964). For example, a pre-cueing visuospatial task presents a cue that provides information about the probable location of an upcoming stimulus. With a predictive cue, attention is oriented selectively to a particular location and behavioral performance for stimuli in that location is increased relative to trials without a predictive pre-cue. Experiments studying pre-cueing have found an increase in alpha oscillations (8-12 Hz) in the regions of the brain that process information for the unlikely location (Sauseng et al., 2009; Wallis, Stokes, Cousijn, Woolrich, &amp; Nobre, 2015). Thus, research into sustained attention and selective attention have found distinct neural oscillations. Using a novel task paradigm that incorporates a spatial pre-cue into the 5CSRTT, we will dissociate sustained attention and selective attention to test the hypothesis that frontal-midline theta oscillations are increased as a function of sustained attention, and posterior alpha oscillations are increased as a function of selective attention. In addition, we will administer a version of the task that approximates the 5-choice serial reaction time task used in animal experiments. Currently, the most recent human analogues have taken the form of touchscreen variant administered as part of the CANTAB® (Cognitive Assessment Software; Cambridge Cognition, 2019) which compares 1 vs 5 locations and an incentivized version which uses monetary rewards to induce impulsive responses (Nord et al., 2019; Worbe et al, 2014). However, the CANTAB variant does not control for the need to survey multiple target locations, thus confounding sustained attention with divided attention; and the incentivized variant is targeted toward a different psychological construct. Therefore here, participants will perform a reach using a mouse to dissociate the impact of sustained attention on reaction time, motor accuracy, response time, and frontal theta oscillations. This experiment will control for both number of locations and perceptual difficulty to disambiguate whether frontal theta oscillations are responsive to sustained or divided attention. The use of a movement tracking paradigm will maximize direct comparability between human and animal 5CSRTT data. Finally, sustained attention and selective attention are impaired with psychiatric illness. Participants with greater levels of anxiety have been found to have increased frontal-midline theta in cognitive control paradigms (Cavanagh &amp; Shackman, 2015). Here, we a","author":[{"family":"Riddle","given":"Justin"},{"family":"Mcferren","given":"Amber"},{"family":"Walker","given":"Christopher"},{"family":"Frohlich","given":"Flavio"}],"issued":{"date-parts":[[2020]]},"DOI":"10.17605/osf.io/7cdku","URL":"https://doi.org/10.17605/osf.io/7cdku","source":"datacite"},{"id":"doi:10.48550/arxiv.2404.10210","type":"manuscript","title":"MK-SGN: A Spiking Graph Convolutional Network with Multimodal Fusion and Knowledge Distillation for Skeleton-based Action Recognition","abstract":"In recent years, multimodal Graph Convolutional Networks (GCNs) have achieved remarkable performance in skeleton-based action recognition. The reliance on high-energy-consuming continuous floating-point operations inherent in GCN-based methods poses significant challenges for deployment in energy-constrained, battery-powered edge devices. To address these limitations, MK-SGN, a Spiking Graph Convolutional Network with Multimodal Fusion and Knowledge Distillation, is proposed to leverage the energy efficiency of Spiking Neural Networks (SNNs) for skeleton-based action recognition for the first time. By integrating the energy-saving properties of SNNs with the graph representation capabilities of GCNs, MK-SGN achieves significant reductions in energy consumption while maintaining competitive recognition accuracy. Firstly, we formulate a Spiking Multimodal Fusion (SMF) module to effectively fuse multimodal skeleton data represented as spike-form features. Secondly, we propose the Self-Attention Spiking Graph Convolution (SA-SGC) module and the Spiking Temporal Convolution (STC) module, to capture spatial relationships and temporal dynamics of spike-form features. Finally, we propose an integrated knowledge distillation strategy to transfer information from the multimodal GCN to the SGN, incorporating both intermediate-layer distillation and soft-label distillation to enhance the performance of the SGN. MK-SGN exhibits substantial advantages, surpassing state-of-the-art GCN frameworks in energy efficiency and outperforming state-of-the-art SNN frameworks in recognition accuracy. The proposed method achieves a remarkable reduction in energy consumption, exceeding 98\\% compared to conventional GCN-based approaches. This research establishes a robust baseline for developing high-performance, energy-efficient SNN-based models for skeleton-based action recognition","author":[{"family":"Zheng","given":"Naichuan"},{"family":"Xia","given":"Hailun"},{"family":"Liang","given":"Zeyu"},{"family":"Du","given":"Yuchen"}],"issued":{"date-parts":[[2024]]},"DOI":"10.48550/arxiv.2404.10210","URL":"https://doi.org/10.48550/arxiv.2404.10210","source":"datacite"},{"id":"doi:10.2139/ssrn.6402373","type":"manuscript","title":"Inverted C-Pocket TFET Based LIF Neuron for Energy-Efficient Neuromorphic Computing with Adaptive Threshold Logic and Image Classification","abstract":"n this work, we propose a novel Leaky Integrate and Fire (LIF) neuron based on an adjustable band-to-band tunneling mechanism, which significantly improves integration density and energy consumption. Neural activity has been simulated using the forward transfer characteristics of an In0.47Ga0.53As based inverted C-Pocket TFET with a steep sub-threshold swing and lower threshold voltage. In addition to providing adjustable characteristics, the calibrated simulations conducted with Silvaco TCAD 2D simulator, it verifies that inverted C-pocket TFET can successfully mimic the neural activity of neuron, without use of external circuit. The proposed LIF neuron requires only 60 aJ per spike with spiking frequency in the range of MHz This energy use is at least 105 times lesser to the 1T MOSFET-type neuron and 1,000 times lesser to the TFET-type neuron previously described in literature. This significant enhancement is due to tunneling and structural engineering approaches. This neuron has been well used to perform the application of threshold logic gates. It offers an approach for designing extremely scalable and low-power threshold logic circuits for emerging neuromorphic computing systems. Finally, a multilayer SNN is used to confirm the proposed neuron&amp;apos;s image classification capacity with an accuracy of 95.81%.","author":[{"family":"Faizan","given":"Mohd"},{"family":"Ashraf","given":"Ehraz"},{"family":"Alshahrani","given":"Abdullah"},{"family":"Afzal","given":"Neelofer"},{"family":"Loan","given":"Sajad"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2139/ssrn.6402373","URL":"https://doi.org/10.2139/ssrn.6402373","source":"crossref"},{"id":"doi:10.1002/smtd.202500089","type":"article-journal","title":"Coupling Light into Memristors: Advances in Halide Perovskite Resistive Switching and Neuromorphic Computing","abstract":"Abstract Resistive switching memristor is an emerging nonvolatile memory technology designed to overcome the physical limitations of conventional systems and the performance bottleneck of the von Neumann architecture. Notably, halide perovskite (HP)‐based memristors have gained significant attention in recent years due to their unique ionic migration behavior and exceptional photoelectric properties. This review highlights HP‐based resistive switching, focusing on its recent developments in coupling light into memristors and discussing its implications for neuromorphic computing. The mechanisms of resistive switching are explored alongside the role of HP photoelectric properties in enhancing switching dynamics. The advantages and applications of light‐coupled resistive switching, including reduced switching voltage, enhanced operation reliability, multilevel switching capability, and the development of light‐integrated artificial synapses are discussed comprehensively. By fully harnessing the exceptional optoelectronic properties of HPs, this emerging field may pave the way for innovative approaches to memory technologies and light‐responsive neuromorphic systems.","author":[{"family":"Feng","given":"Zijian"},{"family":"Wang","given":"Jintao"},{"family":"Chen","given":"Fandi"},{"family":"Dong","given":"Beining"},{"family":"Ma","given":"Xinyu"},{"family":"Mei","given":"Tingting"},{"family":"Yang","given":"Ni"},{"family":"Guan","given":"Xinwei"},{"family":"Hu","given":"Long"},{"family":"Lin","given":"Chun‐ho"},{"family":"Li","given":"Zhi"},{"family":"Wu","given":"Tom"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1002/smtd.202500089","URL":"https://doi.org/10.1002/smtd.202500089","source":"europepmc"},{"id":"doi:10.1002/widm.70014","type":"article-journal","title":"Neuromorphic Computing and Applications: A Topical Review","abstract":"ABSTRACT Neuromorphic computers achieve energy efficiency by emulating brain structure and event‐driven processing that reduces energy consumption significantly. An increasing interest in this technology started in the initial years of this millennium, sparked by the awareness and concern on the ever‐increasing power demands of modern‐day computing. In current times, there are several neuromorphic computers and sensors that continue to be developed in both industry and academic research. The focus of this survey is on the neuromorphic computing applications of these devices that include brain‐inspired neural networks, brain‐inspired artificial neural networks, and Hybrid circuits comprising both artificial and brain‐inspired units of computation. Many of these applications use neuromorphic sensors as input devices. We have surveyed three specific neuromorphic computers viz. SpiNNaker, TrueNorth, Loihi, and one neuromorphic sensor viz. Dynamic vision sensor (DVS)‐based electronic retina; the demonstration of neuromorphic computing and applications using these devices far outnumbers those on the others that are currently available, which forms the basis of our choice. The applications include low‐power cognitive machine intelligence as well as neuropathological understanding and knowledge discovery. Overall, our survey identifies the potential for neuromorphic computing to provide low power, low cost, and dynamic solutions for societal and scientific problems in the not‐too‐distant future.","author":[{"family":"Enuganti","given":"Pavan"},{"family":"Bhattacharya","given":"Basabdatta"},{"family":"Gotarredona","given":"Teresa"},{"family":"Rhodes","given":"Oliver"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1002/widm.70014","URL":"https://doi.org/10.1002/widm.70014","source":"crossref"},{"id":"doi:10.3389/fncom.2025.1737839","type":"article-journal","title":"Bridging neuromorphic computing and deep learning for next-generation neural data interpretation","abstract":"The rapid advancement of electrophysiological techniques, brain imaging, and brain-machine interfaces (BMIs) has ushered neuroscience into an era of data explosion. Confronted with neural data that are high-dimensional, highly nonlinear, and exhibit complex temporal dependencies, conventional statistical and signal processing methods-often reliant on linear assumptions or low-dimensional projections-struggle to reveal the true mechanisms of brain activity [1]. In response to this challenge, deep learning (DL) and neuromorphic computing (NC) have emerged as two promising yet conceptually distinct computational paradigms. Deep learning has demonstrated remarkable capabilities in data-driven modeling, achieving significant breakthroughs in neural signal decoding and cognitive state identification. However, its inherent limitations-high energy consumption, limited interpretability, and low biological plausibility-restrict its deeper application in computational neuroscience. In contrast, neuromorphic computing, inspired by the event-driven and local plasticity properties of biological neural systems, offers unique advantages in low-power adaptive processing. Nonetheless, it still faces challenges in training algorithms and scalability [2,3]. To address these complementary shortcomings, this article proposes a hybrid framework that integrates neuromorphic computing with deep learning, aiming to harmonize biological plausibility with high computational performance, thereby opening new pathways for developing next-generation models and tools for neural data interpretation [4].Neuromorphic computing seeks to emulate the structural and functional organization of the brain, enabling computational systems to operate in ways that resemble biological neural processing. Its central idea is to incorporate event-driven and asynchronous communication, allowing computation to occur only when an event is triggered rather than at fixed time intervals [5]. This mechanism markedly reduces redundant operations and energy consumption, mirroring the physiological principle by which neurons fire only when their membrane potential surpasses a threshold.At the heart of neuromorphic computing lies the Spiking Neural Network (SNN), which represents information through discrete electrical impulses, or spikes, that are temporally encoded to convey meaning [6] [7]. Learning in SNNs is commonly governed by Spike-Timing-Dependent Plasticity (STDP)-a local rule that adjusts synaptic strength based on the precise timing of pre-and postsynaptic spikes. This biologically grounded mechanism enables SNNs to capture causal relationships in neural activity, making them more faithful to real neural dynamics than conventional Artificial Neural Networks (ANNs). Consequently, SNNs excel in handling temporal sequences, sparse representations, and real-time responses.In recent years, hardware implementations such as Intel Loihi, IBM TrueNorth, and BrainScaleS have demonstrated the promise of neuromorphic architectures for low-power, massively parallel computation [5][7]. For example, the Loihi chip integrates on-chip plasticity circuits that support localized learning, achieving power efficiency several orders of magnitude better than traditional GPUs. Simultaneously, rapid advances in memristor technology have introduced new opportunities for neuromorphic hardware [6][8] [9]. Memristors-devices that exhibit non-volatility, tunable conductance, and synapse-like behavior-enable in-memory computing, merging data storage and processing to emulate synaptic functionality directly on the chip. Despite these advances, several challenges remain. The discrete nature of spikes makes training difficult, as standard backpropagation cannot be directly applied. Scalability also remains a concern: current systems struggle to maintain learning efficiency and robustness in large-scale data environments. Furthermore, the ecosystem of algorithms and software frameworks is still immature, lacking standardized interfaces for widespread adoption [3]. A recent Nature report highlights key breakthroughs in inter-chip communication, scalable architecture, and event-driven scheduling, signaling an important step toward large-scale, next-generation brain-inspired computing systems [10].As the dominant paradigm in contemporary artificial intelligence, deep learning (DL) has achieved remarkable success in the analysis of neural data owing to its hierarchical feature extraction and nonlinear approximation capabilities [1] [11]. Architectures such as Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and Transformers have been widely applied to the decoding of electroencephalography (EEG), local field potentials (LFP), and functional magnetic resonance imaging (fMRI) signals. These models can automatically extract multilayer representations, enabling the discovery of hidden relationships between neural activity and cognitive states.In practical applications, CNNs have demonstrated high accuracy and robustness in EEG-based emotion recognition and brain-machine interface (BMI) command classification tasks [12]. RNNs and Long Short-Term Memory (LSTM) networks have exhibited outstanding capabilities in modeling the temporal dynamics of neural signals and predicting brain activity patterns [13]. Moreover, Transformer-based architectures have shown a superior ability to capture global dependencies across multimodal neural datasets, significantly improving the interpretability and scalability of neural decoding frameworks [14]. Collectively, these advances indicate that deep learning not only facilitates complex pattern recognition in brain data but also provides statistical insights into the structure of neural processes.However, the advantages of deep learning are accompanied by several intrinsic limitations. First, artificial neural networks rely on continuous activation functions and global backpropagation, which diverge from the local learning and synaptic plasticity observed in biological neural systems, resulting in limited biological plausibility. Second, the computational process of deep networks depends on large-scale matrix multiplications and parallel processing, leading to extremely high power consumption-orders of magnitude greater than that of biological brains. Third, deep models often suffer from poor interpretability; although their predictive performance is high, the internal representations rarely align with specific neurophysiological structures or mechanisms. In addition, conventional deep learning models lack temporal precision, making it difficult to accurately capture spike-based neural dynamics with millisecond resolution [15] [16].Therefore, while deep learning has proven powerful in feature extraction and cognitive modeling of neural data, achieving a genuine transition from \"prediction\" to \"understanding\" requires the incorporation of computational paradigms that align more closely with neurophysiological mechanisms. Neuromorphic computing offers a biologically inspired and energy-efficient complement to deep learning, paving the way toward a more interpretable and power-efficient framework for neural data modeling.To reconcile the strong representational capacity of deep learning with the biological plausibility of neuromorphic computing, this work proposes a Hybrid Neuromorphic-Deep Learning Framework [4] [17]. The framework integrates event-driven spiking computation with end-to-end deep feature learning, forming a multilayer neural data-processing system that is interpretable, energy-efficient, and aligned with neurophysiological dynamics.At the architectural level, the framework consists of three major components: a neuromorphic front-end, a hybrid learning layer, and a deep interpretation back-end. The neuromorphic front-end extracts event-driven signals-such as spike trains-from multimodal neural recordings while performing noise suppression and initial temporal encoding [18]. Through the use of spiking neural networks (SNNs) or memristor-based neuromorphic circuits, this stage enables low-power, real-time preprocessing at the hardware level [8] [18]. Recent progress in two-dimensional-material neuromorphic chips has further enhanced efficiency, sensitivity, and scalability, offering robust hardware support for hybrid neural architectures [19].The hybrid learning layer serves as the interface between the neuromorphic and deep-learning modules. Its role is to map sparse spike-based events into higher-level feature vectors. This can be achieved using surrogate-gradient optimization or biologically inspired local plasticity rules [11] [14], integrating local learning mechanisms with global gradient-based adaptation for improved interpretability and flexibility.To operationalize the event-to-vector transformation within this layer, several established spike-to-vector conversion and differentiable training strategies can be adopted. For example, Meng et al. (2022) [20] introduced a differentiable spike-representation learning method that maps temporal spike sequences into continuous vector spaces, enabling cross-domain transformation from event-based signals to feature embeddings. Similarly, Zhang et al. (2023) [21]demonstrated that surrogate-gradient-based direct training in hybrid SNN-ANN networks can effectively extract and vectorize event-driven features. Furthermore, hybrid neural frameworks such as the Hybrid Neural Network (HNN) proposed by Zhao et al. (2022) [17] have verified that constructing learnable interaction layers between ANNs and SNNs enables efficient cross-domain feature projection, providing a feasible technical route for implementing the hybrid module in our framework. In addition, the analysis by Liu et al. (2024) [4]on the mechanisms and information flow in hybrid neural systems offers further theoretical support for the mapping strategy adopted here.The deep interpretation back-end applies advanced neural models-such as Transformers, graph convolutional networks (GCNs), or recurrent neural networks (RNNs)-for pattern recognition, brain-state decoding, and cognitive representation learning. This stage further benefits from pre-training and knowledge-transfer mechanisms, improving generalization and abstraction quality [4] [17]. Information flows from event-driven spike streams at the neuromorphic front-end, through the hybrid mapping layer, to high-level cognitive inference in the deep model. With a closed-loop design, feedback from the deep model can dynamically regulate neuromorphic parameters, enabling online learning and forming an adaptive, self-optimizing neural system.Compared with the HNN framework introduced by Zhao et al. (2022) [17], which primarily targets efficient heterogeneous ANN-SNN co-inference, the proposed framework extends beyond coupling mechanisms to incorporate a neuromorphic sensory front-end and a deep interpretive back-end. This results in a cross-scale architecture spanning event encoding, hybrid feature learning, and neurodynamics-aligned interpretation. Consequently, the present framework emphasizes event-driven representation, biological interpretability, and alignment with neurophysiological mechanisms, offering broader conceptual scope and greater relevance for neural data analysis. This hybrid framework offers several notable advantages. Event-driven computation reduces redundant operations and energy consumption. Spike-based representations correspond directly to neuronal firing patterns, improving interpretability. Deep neural models provide scalable abstraction for high-dimensional neural datasets. Finally, the integration of local plasticity with online adaptation imparts robustness and flexibility, forming a promising foundation for next-generation self-adaptive brain-machine systems. The integration of neuromorphic computing and deep learning represents not only a technological complementarity but also a profound paradigm shift in computational neuroscience [4][20]. This cross-level integration provides a systematic pathway for neural information processing that bridges biological inspiration and high-dimensional modeling, allowing models to simultaneously capture neurodynamic plausibility and the abstract representational power of deep networks. Through this hybrid framework, computational neuroscience is gradually transitioning from mere signal fitting toward functional interpretation of neural mechanisms.In the domain of neural encoding and decoding, hybrid models can more accurately characterize the dynamic firing patterns of neuronal populations, enabling precise recognition of complex neural processes such as motor intention, perceptual representation, and cognitive load [1]. By combining the temporal precision of spike-based events with the hierarchical representations of deep networks, researchers can achieve a multi-scale description of neural information, uncovering the diversity and plasticity inherent in neural coding.For brain connectivity modeling, integrating spike-driven event models with Graph Neural Networks (GNNs) offers a promising approach to uncover causal interactions and functional topologies among neural circuits [17]. Such frameworks not only facilitate the reconstruction of dynamic brain networks but also provide computational insights for the early diagnosis and intervention of neurological disorders such as epilepsy and Alzheimer's disease.In brain-machine interfaces (BMIs) and neural rehabilitation, the hybrid architecture enables real-time signal decoding on low-power neuromorphic hardware, supporting adaptive and online-learning-based neural communication systems [8][22]. By combining the efficient temporal processing of spiking neural networks with the high-level pattern recognition capability of deep learning, these systems can dynamically adjust decoding strategies while maintaining energy efficiency-thereby enhancing self-learning capabilities for intelligent neuroprosthetic control and rehabilitation.From a hardware-intelligence co-design perspective, the rapid progress of memristor technology and three-dimensional integrated circuits is driving a deep convergence between neuromorphic chips and AI accelerators [9][18]. This cross-layer collaboration enables brain-inspired learning and adaptive cognition in edge-computing environments, making it feasible to achieve efficient, real-time neural computation directly on localized devices.Looking forward, the core objective of computational neuroscience will revolve around achieving a balanced trade-off between energy efficiency, interpretability, and scalability. By combining the biological realism of neuromorphic computing with the abstraction capabilities of deep learning, a new generation of neural intelligent systems may emerge-systems that not only elucidate the computational principles and information flow of the brain but also drive the advancement of adaptive intelligent chips, brain-inspired computing platforms, and next-generation brain-machine interface technologies.Despite the significant potential of the hybrid neuromorphic-deep learning framework, several limitations remain that must be addressed in future research. First, event-driven encoding is inherently sensitive to noise and may not be suitable for neural recording modalities with low temporal resolution or high measurement noise-such as calcium imaging or fMRI-which restricts its applicability across modalities. Second, training hybrid ANN-SNN systems often incurs substantial computational overhead. Cross-domain gradient propagation can introduce instability, and the field still lacks a unified strategy for multimodal fusion across spike-based and continuous representations. Third, current neuromorphic hardware faces practical challenges, including variability and limited reproducibility in memristive devices, as well as bandwidth constraints in inter-chip communication. These factors hinder large-scale deployment and stable on-chip training. Finally, although the deep interpretation module enables powerful high-dimensional feature abstraction, its biological interpretability remains imperfect and cannot yet fully align with real neurophysiological mechanisms. Therefore, applying this framework to real neural data analysis and brain-machine interface systems will require careful balancing among algorithmic design, hardware implementation, and neuroscientific validation.We argue that the fusion of neuromorphic computing and deep learning constitutes a paradigm shiftfor computational neuroscience, moving the field beyond isolated algorithms toward a holistic paradigm that embraces cross-level abstraction, biological plausibility, and stringent energy constraints [4]. The synergistic integrationof event-driven processing with deep hierarchical learning is pivotal, enabling not only a more profound interpretation of neural dynamics but also the co-design of intelligent and energy-efficient hardware. This hybrid framework establishes a new foundation for future research, poised to significantly advance our capabilities in decoding neural computation, diagnosing neurological disorders, and engineering adaptive, brain-inspired intelligence [17,22].","author":[{"family":"Zhang","given":"Manyun"},{"family":"Wang","given":"Tianlei"},{"family":"Zhu","given":"Zhiyuan"}],"issued":{"date-parts":[[2025]]},"DOI":"10.3389/fncom.2025.1737839","URL":"https://doi.org/10.3389/fncom.2025.1737839","source":"europepmc"},{"id":"doi:10.1002/pssr.202500341","type":"article-journal","title":"Low‐Voltage Flexible Copper Iodide Synaptic Transistors for Neuromorphic Computing","abstract":"A flexible copper iodide (CuI) thin‐film transistor integrated with a chitosan gate dielectric is demonstrated in this article, which is fabricated on a polyethylene terephthalate (PET) substrate via a low‐temperature solution process. The device operation is enabled by the electric double layer effect induced by proton migration in chitosan, achieving low‐voltage driving at 2 V. Additionally, the device is tested under various mechanical bending conditions, demonstrating its flexibility and mechanical reliability. A transition from short‐term memory to long‐term memory is realized by modulating the pulse amplitude. Furthermore, the synaptic plasticity is shown to be tunable by pulse parameters (duration, frequency, number), which is quantitatively correlated with the proton diffusion kinetics in chitosan. These behaviors, resembling biological spike‐timing‐dependent plasticity, are systematically analyzed to establish a framework for neuromorphic computing. The results highlight the potential of the CuI/chitosan platform for flexible neuromorphic electronics, offering insights into adaptive learning rules and biohybrid systems.","author":[{"family":"Xu","given":"Xiaodong"},{"family":"Dou","given":"Wei"},{"family":"Chen","given":"Pengfei"},{"family":"Lei","given":"Jiangyun"},{"family":"Peng","given":"Yuling"},{"family":"Tang","given":"Dongsheng"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1002/pssr.202500341","URL":"https://doi.org/10.1002/pssr.202500341","source":"crossref"},{"id":"doi:10.1002/aelm.202500440","type":"article-journal","title":"Organic Thin‐Film Transistors for Neuromorphic Computing","abstract":"ABSTRACT Organic Thin‐Film Transistors (OTFTs), including Organic Field‐Effect Transistors (OFETs) and Organic Electro‐Chemical Transistors (OECTs), offer clear advantages over traditional silicon‐based devices, particularly in power efficiency and biocompatibility. When combined with neuromorphic computing, which mimics the brain's event‐driven processing to improve computational efficiency, OTFTs become a powerful platform for next‐generation electronics. These devices have demonstrated strong potential as artificial synapses and neurons, showing key spike‐based performance metrics such as Excitatory Post‐Synaptic Current (EPSC), Paired‐Pulse Facilitation (PPF), and Long‐Term Potentiation (LTP). This review captures recent progress in OTFT‐based synaptic and neuronal devices, alongside an in‐depth analysis of how fabrication parameters influence neuromorphic performance. Such insights are critical for designing and optimizing organic neuromorphic systems. We examine the transition from single‐transistor synapses to multi‐transistor neuron models, including emerging organic Single Transistor Latch (STL) neurons that mimic Leaky Integrate and Fire (LIF) and related dynamics. This review also explores the development of OTFT‐based neural networks, their performance relative to Metal‐Oxide‐Semiconductor Field‐Effect Transistor (MOSFET)‐based systems, and their potential shift toward fully neuromorphic Spiking Neural Networks (SNNs). Beyond surveying device demonstrations, this review introduces a standardized characterization protocol for OTFT synapses, and maps the physical mechanisms of OTFT architectures onto appropriate learning rules and network models. By linking materials, device physics, and neuromorphic algorithms, we highlight the opportunities for co‐designing flexible, bio‐integrated OTFT‐based neuromorphic systems.","author":[{"family":"Mccarthy","given":"Luke"},{"family":"Jacob","given":"Mohan"},{"family":"Azghadi","given":"Mostafa"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1002/aelm.202500440","URL":"https://doi.org/10.1002/aelm.202500440","source":"crossref"},{"id":"doi:10.20517/energymater.2025.185","type":"article-journal","title":"Triboelectric-memristive coupling for self-powered neuromorphic computing: mechanisms, devices, and systems","abstract":"Coupling triboelectric nanogenerators (TENGs) with memristors offers a direct route to integrating energy harvesting and adaptive learning within a single physical substrate, thereby enabling self-powered neuromorphic systems driven by ubiquitous mechanical stimuli. Unlike conventional electronics that rely on external power rails, triboelectric-memristive hybrids transduce mechanical excitations into programmable resistive states, supporting synaptic functions such as short-term plasticity, long-term plasticity, and spike-timing-dependent plasticity. This review synthesizes the physical mechanisms of triboelectric-memristive coupling and clarifies how charge transfer, interfacial electron-ion interactions, and device-level state dynamics collectively enable energy-to-information transduction for signal processing and learning. In contrast to previous surveys that focus on TENGs or memristors in isolation, we establish a unified transduction framework that links mechanical stimulus statistics to TENG waveform characteristics and further to memristive state-variable evolution, which serves as the organizing principle throughout the paper. We then present (ⅰ) a mechanism-guided taxonomy of representative device architectures and their achievable plasticity modes; and (ⅱ) a system-level perspective on the integration of self - powered sensing, in-memory learning, and multimodal data fusion. Finally, we summarize key challenges - including charge stability, humidity tolerance, device variability, and scalable integration - and discuss emerging directions such as large-area triboelectric materials for improved array uniformity, multiphysics co-learning for enhanced in-sensor intelligence, and physics-informed compact models to support device-circuit-algorithm co-design under stochastic energy inputs.","author":[{"family":"Qin","given":"Haiyang"},{"family":"Li","given":"Qinrao"},{"family":"Lu","given":"Dongzhu"},{"family":"Lin","given":"Jianxin"},{"family":"Gao","given":"Wenke"},{"family":"Wang","given":"Huachuan"}],"issued":{"date-parts":[[2026]]},"DOI":"10.20517/energymater.2025.185","URL":"https://doi.org/10.20517/energymater.2025.185","source":"crossref"},{"id":"doi:10.4018/979-8-3373-9785-6.ch004","type":"article-journal","title":"Quantum-Inspired or Neuromorphic Discrete-Event Computing Paradigms for Future Ubiquitous Systems","abstract":"Foundations of quantum-inspired discrete-event architectures constitute an emerging and transformative paradigm in the modelling, simulation, and design of complex, dynamic, and stochastic systems, particularly in the context of large-scale distributed networks, Internet of Things (IoT) ecosystems, cyber-physical systems, cloud and edge computing, service-oriented architectures, and high-performance computing environments, where traditional discrete-event simulation (DES) approaches face challenges related to scalability, uncertainty, combinatorial complexity, and real-time decision-making; quantum-inspired methodologies draw conceptual and algorithmic insights from quantum mechanics, quantum computing principles, and quantum information theory, including superposition, entanglement, probabilistic amplitudes, and parallelism, to enhance the capability of DES frameworks to represent, process, and optimize highly complex event-driven interactions across heterogeneous devices, services, and computing layers.","author":[{"family":"Math","given":"Shrishail"},{"family":"Madhuri","given":"HD"},{"family":"Yadhav","given":"Vijaykumar"},{"family":"Patil","given":"Mallanagouda"},{"family":"Selvakumar","given":"P"}],"issued":{"date-parts":[[2026]]},"DOI":"10.4018/979-8-3373-9785-6.ch004","URL":"https://doi.org/10.4018/979-8-3373-9785-6.ch004","source":"crossref"},{"id":"doi:10.1002/admt.71158","type":"article-journal","title":"Defect‐Controlled Ti‐Doped Perovskite Memristors for High Accuracy Neuromorphic Computing","abstract":"ABSTRACT Neuromorphic computing requires electronic devices capable of stable analogue switching, low energy operation, and reliable synaptic behaviour. In this study, Zinc zirconate (ZZO) and Ti‐doped ZnZrO 3 (TZZO) perovskite thin films were developed to realise high‐performance memristors, and their switching behaviour is correlated with defect chemistry and electronic structure. Resistive switching devices with Ag/ZZO/FTO and Ag/TZZO/FTO configurations exhibit consistent bipolar switching, governed by ohmic, space‐charge‐limited conduction (SCLC), and high‐field‐assisted transport mechanisms associated with Schottky and Poole–Frenkel conduction. The switching behaviour in Ag/TZZO/FTO is primarily attributed to an oxygen‐vacancy‐mediated valence change memory (VCM) mechanism, with possible minor contributions from field‐assisted Ag‐ion migration. These devices showed strong performance, with endurance reaching 10 4 cycles and retention of 10 4 s. Density functional theory (DFT) studies revealed that Ti doping improves electrical conductivity via defect‐driven transport pathways. The neuromorphic computing potential of these devices was validated by integrating their long‐term potentiation and depression (LTP and LTD) behaviours into a convolutional neural network (CNN) trained on the MNIST dataset, achieving digit recognition accuracies of 94.35% for ZZO and 98.35% for TZZO. These results establish the Ti‐doped ZZO as a promising candidate for next‐generation non‐volatile memory and neuromorphic applications.","author":[{"family":"Malishetty","given":"Narender"},{"family":"Chandmare","given":"Vaishali"},{"family":"Vadthiya","given":"Narendar"},{"family":"Borkar","given":"Hitesh"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1002/admt.71158","URL":"https://doi.org/10.1002/admt.71158","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-10112503/v1","type":"article-journal","title":"A Non-Monotonic Reconfigurable Neural Unit with Dynamic Integration Control for Advanced Neuromorphic Computing","abstract":"Abstract The development of artificial neurons that emulate the rich adaptive dynamics of biological neural systems is pivotal for next-generation neuromorphic computing. Current hardware implementations, however, are fundamentally limited by fixed monotonic activation functions and an inability to dynamically control temporal integration processes. To overcome these challenges, we introduce a reconfigurable neural unit (RNU) that combines a synaptic transistor (MoTe 2 /h-BN) with a volatile memristor (Ag/h-BN/graphene). This architecture produces a distinct U-shaped non-monotonic response, originating from gate-modulated conductance variations in the MoTe 2 channel, which features a central silent region for firing-rate suppression, mimicking feature-selective inhibition observed in biological networks. Crucially, the position of this silent region on the gate-bias axis is dynamically reconfigurable via ultraviolet-light-assisted doping, enabling real-time adaptation of the RNU's operational state for diverse computing tasks. Moreover, the gate terminal doubles as a dedicated input for inhibitory spikes, enabling precise spatiotemporal control of neuronal integration that is critical for processing complex temporal patterns. System-level simulations demonstrate that spiking neural networks equipped with RNUs achieve competitive accuracy in image recognition and show a 3.5% improvement in four-class EEG decoding over conventional spiking network baselines. The RNU therefore provides a compact, versatile, and highly adaptive foundation for advanced neuromorphic systems.","author":[{"family":"Liu","given":"Jing"},{"family":"An","given":"Xitong"},{"family":"Wang","given":"Yan"},{"family":"Dou","given":"Chao"},{"family":"Lu","given":"Haoyue"},{"family":"Li","given":"Yueying"},{"family":"Deng","given":"Xuan"},{"family":"Sun","given":"Dong"}],"issued":{"date-parts":[[2026]]},"DOI":"10.21203/rs.3.rs-10112503/v1","URL":"https://doi.org/10.21203/rs.3.rs-10112503/v1","source":"europepmc"},{"id":"doi:10.2139/ssrn.6556100","type":"manuscript","title":"Neuromorphic Reservoir Computing Generates Hippocampal Signals to Improve Neural Activity Modulation","abstract":"IntroductionEpilepsy affects approximately 50 million people worldwide, and a substantial proportion of patients remain resistant to pharmacological treatment, requiring invasive interventions such as deep brain stimulation (DBS). Although DBS can reduce seizure frequency, its mechanisms of action are not fully understood, and high-frequency stimulation imposes a considerable stimulation burden. Alternative neuromodulation strategies that preserve physiological temporal dynamics while reducing stimulation load are therefore needed. Reservoir Computing (RC), particularly Echo State Networks (ESN), represents a promising neuromorphic approach capable of reproducing neuronal dynamics with low computational complexity.Materials and MethodsAn Echo State Network was designed to generate biomimetic electrical stimulation patterns replicating the spectral and statistical properties of hippocampal CA3 activity. The generated signals were characterized in terms of mean frequency, event timing, and inter-event interval distributions. These biomimetic stimulation patterns were applied in vitro to primary cortical neuronal cultures. Neuronal activity was assessed by measuring firing rate and burst-related metrics, and outcomes were compared to those obtained using conventional 50 Hz high-frequency stimulation.ResultsThe ESN-generated signal reproduced key spectral and statistical features of biological hippocampal activity, including a mean frequency of 0.16 Hz and preserved inter-event interval distributions. When applied to primary cortical cultures, biomimetic stimulation reduced firing rate and burst-related activity in 4 out of 11 networks (36%). The magnitude of reduction was comparable to that achieved with conventional 50 Hz stimulation. Importantly, this effect was obtained using approximately 300-fold fewer electrical pulses than high-frequency stimulation.ConclusionThese findings support a neuromodulation strategy that reduces stimulation load while preserving physiological temporal structure. ESN-based biomimetic stimulation represents a promising foundation for adaptive, closed-loop neuromodulation systems leveraging neuromorphic computing, with potential applications in personalized epilepsy treatment.","author":[{"family":"Canal-Alonso","given":"Ángel"},{"family":"Armada-Moreira","given":"Adam"},{"family":"Clemente","given":"Alessio"},{"family":"Cerezo-Sánchez","given":"María"},{"family":"Pavlidou","given":"Antonia"},{"family":"Kiamarsi","given":"Danial"},{"family":"Giugliano","given":"Michele"},{"family":"Heidari","given":"Hadi"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2139/ssrn.6556100","URL":"https://doi.org/10.2139/ssrn.6556100","source":"crossref"},{"id":"doi:10.1088/2634-4386/ae759c","type":"article-journal","title":"Energy-efficient implementation of spiking recurrent cells on FPGA","abstract":"Abstract Spiking neural networks (SNNs) can significantly reduce energy consumption compared to conventional artificial neural networks when spiking activity is sparse and the neuron model is hardware-friendly. However, biologically faithful models are often too costly for hardware implementations like field-programmable gate arrays (FPGAs), whereas highly simplified models, such as the leaky integrate-and-fire (LIF) neuron, sacrifice essential neuronal dynamics. In this work, we present an FPGA accelerator for an SNN utilizing the spiking recurrent cell (SRC) model, which offers an intermediate level of biological plausibility and hardware efficiency between simple LIF and complex conductance-based models. The SRC model features continuous spikes and critical dynamical properties like the refractory period, yet remains mathematically simple enough for efficient FPGA deployment. To optimize SRC computation, we propose a set of mathematical simplifications such as piecewise-defined approximations that eliminate costly non-linear functions ( tanh , exp ) and using scaling to avoid floating-point arithmetic. We then integrate these units into a cohesive hardware architecture using a dedicated binding layer to permit modularity and scalability of the network VHDL architecture. To demonstrate this modularity, two complete networks, with respectively 1 and 4 SRC layers, were implemented and validated using Spiking Traces (SpT) derived from, respectively, the MNIST and Fashion-MNIST datasets. Weight matrices computed offline are stored directly in LUT-registers without retraining or hardware-specific adaptation to strictly evaluate the robustness of SRC. The reference implementation achieves 96.31 % accuracy on MNIST with a 220 -image SpT and a processing time of 1.7424 ms per digit. We further investigate accuracy-energy trade-offs by reducing the SpT length and quantizing synaptic weights down to 4 bits , achieving 93.32 % accuracy at 0.492 mJ per digit ( 55 images , 5 -bit weights) and 92.89 % at 0.394 mJ ( 44 images , 4 -bit weights). These results demonstrate that SRC-based SNNs deliver competitive performance and low energy consumption while preserving richer neuronal dynamics than standard LIF models.","author":[{"family":"Harmeling","given":"Pascal"},{"family":"Geeter","given":"Florent"},{"family":"Drion","given":"Guillaume"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/2634-4386/ae759c","URL":"https://doi.org/10.1088/2634-4386/ae759c","source":"crossref"},{"id":"doi:10.1002/advs.202515926","type":"article-journal","title":"Polarity‐Controlled Volatile HfO\n                    <sub>2</sub>\n                    Memristors with Bimodal Conductance for Neuromorphic Synapses and Reservoir Computing","abstract":"Abstract In this work, an HfO 2 ‐based memristor exhibiting bimodal switching, wherein the device's conductance is modulated not only by the input stimulus but also by the polarity of the read voltage, is introduced. Uniquely, this device demonstrates reliable short‐term memory (STM)‐like behavior and supports 16 well‐separated conductance states through 4‐bit pulsed inputs. Remarkably, under the same input conditions, reversing the polarity of the read voltage results in 16 more different conductance states, thereby doubling the number of levels that can be distinguished per cell. Employing the proposed device, a reservoir computing (RC) system, which takes advantage of this rich representational capability, is created. The system achieves a high classification accuracy of 98.81% on the MNIST dataset. These results show how powerful memristor‐based architectures can be and how this device could be a compact and energy‐efficient platform for the next generation of neuromorphic computing.","author":[{"family":"Jang","given":"Yuseong"},{"family":"Hwang","given":"Chanmin"},{"family":"Chae","given":"Myoungsu"},{"family":"Kim","given":"Taegi"},{"family":"Kim","given":"Hee‐dong"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1002/advs.202515926","URL":"https://doi.org/10.1002/advs.202515926","source":"europepmc"},{"id":"doi:10.3390/jlpea15030050","type":"article-journal","title":"Alleviating the Communication Bottleneck in Neuromorphic Computing with Custom-Designed Spiking Neural Networks","abstract":"For most, if not all, AI-accelerated hardware, communication with the agent is expensive and heavily bottlenecks the hardware performance. This omnipresent hardware restriction is also found in neuromorphic computing: a novel style of computing that involves deploying spiking neural networks to specialized hardware to achieve low size, weight, and power (SWaP) compute. In neuromorphic computing, spike trains, times, and values are used to communicate information to, from, and within the spiking neural network. Input data, in order to be presented to a spiking neural network, must first be encoded as spikes. After processing the data, spikes are communicated by the network that represent some classification or decision that must be processed by decoder logic. In this paper, we first present principles for interconverting between spike trains, times, and values using custom-designed spiking subnetworks. Specifically, we present seven networks that encompass the 15 conversion scenarios between these encodings. We then perform three case studies where we either custom design a novel network or augment existing neural networks with these conversion subnetworks to vastly improve their communication performance with the outside world. We employ a classic space vs. time tradeoff by pushing spike data encoding and decoding techniques into the network mesh (increasing space) in order to minimize intra- and extranetwork communication time. This results in a classification inference speedup of 23× and a control inference speedup of 4.3× on field-programmable gate array hardware.","author":[{"family":"Plank","given":"James"},{"family":"Rizzo","given":"Charles"},{"family":"Gullett","given":"Bryson"},{"family":"Dent","given":"Keegan"},{"family":"Schuman","given":"Catherine"}],"issued":{"date-parts":[[2025]]},"DOI":"10.3390/jlpea15030050","URL":"https://doi.org/10.3390/jlpea15030050","source":"crossref"},{"id":"doi:10.1088/2634-4386/ae6a19","type":"article-journal","title":"Mobile-compatible neuromorphic optical computing enabled by dual-emission photonic materials","abstract":"Abstract Artificial intelligence (AI) and neuromorphic computing demand hardware platforms that combine energy efficiency with physically informed data processing. Photonics offers unique advantages in this context, but optical neuromorphic systems in which memory and information processing arise from intrinsic material dynamics remain scarce. We demonstrate a photonic layer in which history-dependent material photophysics implements physically embedded information processing before digital learning, using luminescent phosphors. The photonic layer exhibits dual fluorescence and phosphorescence together with excitation-history-dependent photoactivation dynamics that emulate synaptic functionalities, including short-term memory, long-term memory, and synaptic potentiation. Quantitative analysis and modelling reveal efficient nanoscale interlamellar energy transfer consistent with the lamellar material morphology, establishing a link between structure and function. When integrated as an active optical front-end within a hybrid photonic–digital AI architecture, the photonic layer performs a material-based transformation of input data before digital learning. Using a laboratory-based optical readout, classification accuracy comparable to state-of-the-art hybrid neuromorphic computing (approximately 94%) is achieved while requiring fewer training epochs. A mobile-compatible implementation based on smartphone optical readout yields similar accuracy, demonstrating robustness to non-specialized optical hardware. These results outline a general strategy for neuromorphic photonic systems in which functional materials act as adaptive computational primitives, enabling energy-efficient computing architectures.","author":[{"family":"Dias","given":"Lília"},{"family":"Bastos","given":"Ana"},{"family":"Fu","given":"Lianshe"},{"family":"Neto","given":"Albano"},{"family":"Pereira","given":"Rui"},{"family":"Bermudez","given":"Verónica"},{"family":"Towe","given":"Elias"},{"family":"Ferreira","given":"Rute"},{"family":"André","given":"Paulo"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/2634-4386/ae6a19","URL":"https://doi.org/10.1088/2634-4386/ae6a19","source":"crossref"},{"id":"doi:10.1088/2634-4386/ae4f1e","type":"article-journal","title":"Hyperdimensional decoding of spiking neural networks","abstract":"Abstract This work presents a novel spiking neural network (SNN) decoding method, combining SNNs with hyperdimensional computing (HDC). This decoding method is designed to achieve high accuracy, high noise robustness, low inference latency and low energy consumption. Compared to analogous architectures decoded with existing approaches, the SNN-HDC model attains generally better classification accuracy, lower inference latency, lower spike count and lower estimated energy consumption on multiple test cases from the literature. The SNN-HDC achieved spike count reductions of 1.74 × to 3.36 × on the DvsGesture dataset and 1.36 × to 2.70 × on the SL-Animals-DVS dataset. The SNN-HDC achieved estimated energy consumption reductions of 1.24 × to 3.67 × on the DvsGesture dataset and 1.38 × to 2.27 × on the SL-Animals-DVS dataset. The proposed decoding method enables detection of classes unseen during training. On the DvsGesture dataset, the SNN-HDC model can detect 100% of samples from an unseen/untrained class. The findings suggest the proposed decoding method is a compelling alternative to both rate and latency decoding.","author":[{"family":"Kinavuidi","given":"Cedrick"},{"family":"Peres","given":"Luca"},{"family":"Rhodes","given":"Oliver"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/2634-4386/ae4f1e","URL":"https://doi.org/10.1088/2634-4386/ae4f1e","source":"crossref"},{"id":"doi:10.1002/advs.75861","type":"article-journal","title":"Photonic-Enabled Energy-Efficient Transparent Neuromorphic Computing Devices: A Review.","abstract":"ABSTRACT In the evolving field of artificial intelligence (AI), two notable trends are emerging: the rapid growth of large AI model sizes and the surge of vast amounts of data. Moore's law emphasizes the need for alternative computing paradigms to meet the rising demand for computational power and address von Neumann model constraints. Nowadays, neuromorphic computing, inspired by the mechanisms and functionality of human brains, uses physical artificial neurons to do computations and is drawing widespread attention. Neuromorphic computing aims to emulate brain‐like information processing with co‐localized memory and logic, breaking the von Neumann bottleneck. In this regard, the integration of photonic materials in computing led to growth in photonic computing, where light is a fundamental source of energy and can also be utilized as a signal for neuromorphic computing. Photonic computing enables ultrafast artificial neural networks with sub‐nanosecond latencies and low heat dissipation. However, current neuromorphic technologies still struggle to achieve petascale speed and energy efficiency. Additionally, the general benefits of neuromorphic computing and AI can be realized in an optically transparent manner, broadening their applications in bionics and human interfaces. This review explores the suitability and design strategies for transparent photonic devices that create artificial interfaces mimicking natural functions.","author":[{"family":"Ghosh","given":"Shuvaraj"},{"family":"Lee","given":"Ki‐bum"},{"family":"Lee","given":"Junghyeon"},{"family":"Cho","given":"Seunghee"},{"family":"Patel","given":"Malkeshkumar"},{"family":"Li","given":"Hangfei"},{"family":"Wen","given":"Yu"},{"family":"Zhou","given":"Ye"},{"family":"Kim","given":"Joondong"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1002/advs.75861","URL":"https://doi.org/10.1002/advs.75861","source":"europepmc"},{"id":"doi:10.1007/s40820-026-02171-2","type":"article-journal","title":"Underlying Framework of All-optical Controlled Synaptic Devices for Neuromorphic Computing.","abstract":"Abstract The rapid expansion of artificial intelligence has led to significant challenges in energy consumption and computational efficiency. To address these issues, the exploration and development of all-optical controlled (AOC) synaptic devices represents a promising leap forward in neuromorphic computing, offering potential solutions to the inherent limitations of traditional von Neumann architectures. AOC synaptic devices, utilizing exclusively optical signals to emulate bidirectional modulation of synaptic weights, bypass the complexity and additional energy costs associated with conventional electrical or electro-optical hybrid signals. This review articulates the underlying framework and fundamental motivations for studying AOC synapses, while systematically reviewing current research progress. We particularly highlight the synergistic relationships among physical mechanisms, material behaviors, and device architectures, as well as neuromorphic computing based on optical writing and optical erasing of information. By systematically interpreting these multidimensional correlations, we propose scalable and reproducible strategies for device design. This work will certainly herald a substantial direction of AOC synapses, providing an ideal platform for exploring neuromorphic computing for artificial intelligence.","author":[{"family":"Hu","given":"Dunan"},{"family":"Yang","given":"Ruqi"},{"family":"Ye","given":"Zhizhen"},{"family":"Lu","given":"Jianguo"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1007/s40820-026-02171-2","URL":"https://doi.org/10.1007/s40820-026-02171-2","source":"europepmc"},{"id":"doi:10.1021/acsami.5c17926","type":"article-journal","title":"A Low-Voltage Stretchable Synaptic Transistor Array for Temperature Perception, Facilitated Associative Learning, and Neuromorphic Computing.","abstract":"Stretchable synaptic transistors are promising candidates for brain-inspired neuromorphic systems in soft robotics and wearable electronics, where temperature perception and low-power operation are critical for biological fidelity and energy efficiency. However, the interplay between mechanical strain, temperature perception, and synaptic properties remains underexplored in such devices. Here, we report a high-density, temperature-modulated stretchable synaptic transistor (TM-SST) array fabricated via a photolithography-based, transfer-free process, integrating a semiconductor carbon nanotube (s-CNT) network channel and an SU-8 dielectric layer. The devices exhibit a high on-off ratio (∼10 5 ) at a low gate voltage ( V gs ) between ±2.5 V and a drain-to-source voltage ( V ds ) of -0.1 V. Importantly, the devices exhibit temperature-dependent synaptic characteristics across 10-40 °C, with effective modulation of postsynaptic current (PSC), plasticity, memory retention, and paired-pulse facilitation (PPF), while maintaining stable performance under 40% strain. Furthermore, temperature modulation enhances neuromorphic performance: a 15 °C cooling improves memory retention in associative learning from seconds to minutes, while simulations show accelerated learning with a 10× dynamic range. This work advances stretchable synaptic devices by enabling temperature perception to enhance neuromorphic functionality.","author":[{"family":"Cui","given":"Dingzhou"},{"family":"Zhao","given":"Zhiyuan"},{"family":"Tian","given":"Fugu"},{"family":"Zheng","given":"Qi"},{"family":"Liao","given":"Xun"},{"family":"Chen","given":"Wenbo"},{"family":"Zhang","given":"Jingxin"},{"family":"Zhou","given":"Chongwu"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1021/acsami.5c17926","URL":"https://doi.org/10.1021/acsami.5c17926","source":"europepmc"},{"id":"doi:10.22541/au.176529685.59484231/v1","type":"article-journal","title":"Quantum-Dot Neuromorphic Edge AI for Ultra-Secure IoT and Brain-Inspired Computing","abstract":"In order to achieve ultra-secure, adaptive, and low-power intelligence for distributed IoT environments, this paper presents a next-generation neuromorphic edge architecture that combines quantum-dot nanomaterials with spiking neural computation. Because of their multi-level tunability and quantum-confined electronic structure, quantum dots (QDs) serve as nanoscale synaptic elements that can produce intrinsic randomness, stable conductance states, and extremely low-energy switching. Real-time on-chip learning and event-driven spike computation are made possible by neuromorphic processors, which are based on biological neural systems. By utilizing QD-based entropy sources and physically unclonable functions (PUFs), the proposed Quantum-Dot Neuromorphic Edge AI Processor (QD-NEAP) reduces vulnerabilities related to traditional IoT devices and eliminates cloud dependency. When compared to CMOS and memristor-based architectures, performance evaluation shows notable gains in energy efficiency, inference latency, and cryptographic strength. This work creates a single braininspired computational platform that combines secure IoT communication, neuromorphic learning, and quantum-dot materials.","author":[{"family":"Sharma","given":"Pushkar"},{"family":"Mali","given":"Ashwini"},{"family":"Panigrahi","given":"Payal"},{"family":"Dora","given":"Sandhyarani"},{"family":"Suryavanshi","given":"Damini"},{"family":"Jangid","given":"Khushboo"}],"issued":{"date-parts":[[2025]]},"DOI":"10.22541/au.176529685.59484231/v1","URL":"https://doi.org/10.22541/au.176529685.59484231/v1","source":"europepmc"},{"id":"doi:10.21203/rs.3.rs-7849696/v1","type":"article-journal","title":"Polycrystalline Perovskite Wafers as an Efficient Multisensory Integration Platform for Neuromorphic Computing","abstract":"Abstract Neuromorphic computing requires materials that integrate adaptive functionalities with scalable, cost-effective fabrication. Halide perovskites offer exceptional optoelectronic performance, but single-crystal growth is prohibitively costly and limits wafer-scale integration. Inspired by the transition from single- to polycrystalline silicon, we develop polycrystalline perovskite wafers (PPWs) via a mechanochemically coupled sintering strategy, achieving high-yield and low-cost fabrication with wafer-scale uniformity and engineered trap states. The resulting PPWs enable highly stable and linear electrical synaptic programming, as well as fully optical modulation of excitatory and inhibitory states through wavelength-selective excitation that leverages both bandgap and sub-bandgap absorption processes. These capabilities support associated retinal-inspired visual perception and auditory motion detection, achieving over 90% multimodal vehicle recognition accuracy when integrated into a spiking neural network. Beyond neuromorphic computing, PPWs provide a versatile and sustainable platform for photodetection, multispectral imaging, and X-ray sensing, combining scalability, robustness, and multifunctionality for next-generation optoelectronic technologies.","author":[{"family":"Liu","given":"Gang"},{"family":"Zhou","given":"Shu"},{"family":"Zhou","given":"Lue"},{"family":"Han","given":"Shuyao"},{"family":"Zhang","given":"Hengyu"},{"family":"Liu","given":"Guixiang"},{"family":"Mu","given":"Yuncheng"},{"family":"Ni","given":"Zhenyi"},{"family":"Hou","given":"Yanglong"}],"issued":{"date-parts":[[2025]]},"DOI":"10.21203/rs.3.rs-7849696/v1","URL":"https://doi.org/10.21203/rs.3.rs-7849696/v1","source":"europepmc"},{"id":"doi:10.1063/5.0299959","type":"article-journal","title":"Volatile threshold switching and neural dynamics emulation in a chitosan-ZnO memristor for neuromorphic computing.","abstract":"Neuromorphic computing demands energy-efficient and biologically plausible devices to emulate neural dynamics and sensory processing. This study explores the development and application of a chitosan-doped ZnO memristor for neuromorphic computing, focusing on its volatile threshold switching behavior and bio-inspired sensory applications. The device exhibits excellent memristive performance, with a high switching ratio (∼105), stable endurance (&amp;gt;104 cycles), and rapid switching speeds (turn-on/turn-off times of ∼23/21 µs). Symmetric threshold voltages (±2 V) and low resistance variability highlight its reliability. Integrated into an oscillatory neuron circuit (R–C configuration), the memristor emulates spiking dynamics, demonstrating tunable frequency and energy efficiency (∼832 nJ/spike). Furthermore, the circuit successfully replicates biological motion detection and sound localization by processing spatiotemporal input differences, mimicking direction-selective ganglion cells and medial superior olive neurons. These results validate the memristor’s potential for bio-inspired sensory systems, offering a scalable, energy-efficient platform for neuromorphic computing and artificial perception.","author":[{"family":"Sun","given":"Yanmei"},{"family":"Liu","given":"Rui"},{"family":"Zhang","given":"Zekai"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1063/5.0299959","URL":"https://doi.org/10.1063/5.0299959","source":"europepmc"},{"id":"doi:10.21203/rs.3.rs-7959358/v1","type":"article-journal","title":"Characterizing Neuromorphic Workloads from A System Perspective","abstract":"Abstract Neuromorphic computing is emerging as a cornerstone for next-generation intelligent systems. The lack of standardized benchmarks hinders fair comparison across diverse neuromorphic solutions and fragments research efforts. Here, we introduce a systematical workload characterization methodology that enables quantitative selection of representative benchmarks. Our approach encodes each spiking neural network (SNN) into a unified feature vector by extracting spatial and temporal features from its network topology and spike rasters. Evaluated on 1,030 models across four hardware platforms, the extracted features achieve over 90% accuracy in performance regression and classification, demonstrating both completeness in capturing SNN performance metrics and hardware independence. We expect that this work will provide a principled and scalable foundation for benchmarking neuromorphic systems.","author":[{"family":"Zhang","given":"Youhui"},{"family":"Pan","given":"Zhe"},{"family":"Li","given":"Zeqing"},{"family":"Wu","given":"Dehua"},{"family":"Qu","given":"Peng"},{"family":"Chong","given":"Yee"}],"issued":{"date-parts":[[2025]]},"DOI":"10.21203/rs.3.rs-7959358/v1","URL":"https://doi.org/10.21203/rs.3.rs-7959358/v1","source":"europepmc"},{"id":"doi:10.3389/fnins.2025.1676570","type":"article-journal","title":"A comparative review of deep and spiking neural networks for edge AI neuromorphic circuits.","abstract":"Edge AI implements neural networks directly in electronic circuits, using either deep neural networks (DNNs) or neuromorphic spiking neural networks (SNNs). DNNs offer high accuracy and easy-to-use tools but are computationally intensive and consume significant power. SNNs utilize bio-inspired, event-driven architectures that can be significantly more energy-efficient, but they rely on less mature training tools. This review surveys digital and analog edge-AI implementations, outlining device architectures, neuron models, and trade-offs in energy (J/OP), area (μm 2 /OP), and integration technology.","author":[{"family":"Ferreira","given":"Pietro"},{"family":"Wang","given":"Siqi"},{"family":"Gao","given":"Yueyuan"},{"family":"Benlarbi-Delai","given":"Aziz"}],"issued":{"date-parts":[[2025]]},"DOI":"10.3389/fnins.2025.1676570","URL":"https://doi.org/10.3389/fnins.2025.1676570","source":"europepmc"},{"id":"doi:10.1038/s41467-026-70802-8","type":"article-journal","title":"Nanoscale exchange-bias magnetic tunnel junctions enabled memristive synapse and leaky-integrate-fire neuron for neuromorphic computing.","abstract":"Abstract Neuromorphic computing implemented by spintronic memories offers computational advantages, including in-memory computing capability, high energy efficiency, and near-unlimited endurance. However, their basic units, magnetic tunnel junctions (MTJs), face inherent challenges in emulating analog synapses and spiking neurons due to their bistable resistance states and lack of bio-realistic switching dynamics. Here, we experimentally demonstrate on-chip co-integrated memristive synapse and leaky-integrate-fire (LIF) neuron designed with nanoscale exchange-bias MTJs (EB-MTJs). By exploiting the spatial distribution of antiferromagnets and its current-dependent modulation, we achieve stable continuous multi-state synaptic behavior with spike-timing-dependent-plasticity (STDP) characteristics in compact (~100 nm) EB-MTJs. Furthermore, time-resolved measurements reveal that EB-MTJs can be progressively programmed by 0.4 ns pulses, emulating LIF neuronal dynamics with high operational bandwidth in the gigahertz range. Finally, we construct a convolutional spiking neural network based on EB-MTJs and achieve 96% accuracy in gesture recognition via a hybrid backpropagation-STDP algorithm, highlighting their potential in neuromorphic computing.","author":[{"family":"Chen","given":"Zanhong"},{"family":"Zhu","given":"Dehang"},{"family":"Du","given":"Ao"},{"family":"Shi","given":"Yuzhang"},{"family":"Cai","given":"Wenlong"},{"family":"Wang","given":"Zixi"},{"family":"Duan","given":"Yuqi"},{"family":"Lu","given":"Shiyang"},{"family":"Cao","given":"Kaihua"},{"family":"Zhang","given":"He"},{"family":"Zhang","given":"Deming"},{"family":"Liu","given":"Hongxi"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1038/s41467-026-70802-8","URL":"https://doi.org/10.1038/s41467-026-70802-8","source":"europepmc"},{"id":"doi:10.1007/s40820-025-02052-0","type":"article-journal","title":"Biomimetic Synapses Based on Halide Perovskites for Neuromorphic Vision Computing: Materials, Devices, and Applications.","abstract":"Abstract The demand for accurate perception of the physical world has led to a dramatic increase in visual sensing data, accompanied by challenges in the energy efficiency of data processing. However, conventional vision systems with separated sensor and processing units struggle to handle increasingly intricate and large-scale data. As such, a rethinking of architecture design is necessary. Inspired by human visual systems, neuromorphic vision computing systems in which computation tasks are moved partly to the sensory or memory units offer transformative solutions to these challenges. As crucial hardware support, biomimetic synapses that replicate synaptic functions and dynamics are urgent for the development of future computing, while further progress requires materials that can support synaptic weight modulation. Given their excellent optical, electrical, and ion migration properties, halide perovskite materials have emerged as promising candidates for biomimetic synapses. Here we review the latest efforts of synaptic devices based on halide perovskite materials for neuromorphic vision computing. We demonstrate the operating mechanism of perovskite synapses and introduce their potential applications in realizing neuromorphic vision computing. We address challenges and future directions related to biomimetic perovskite synapses.","author":[{"family":"Sun","given":"Zhongwen"},{"family":"Zhao","given":"Xuan"},{"family":"Si","given":"Haonan"},{"family":"Liao","given":"Qingliang"},{"family":"Zhang","given":"Yue"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1007/s40820-025-02052-0","URL":"https://doi.org/10.1007/s40820-025-02052-0","source":"europepmc"},{"id":"doi:10.21203/rs.3.rs-7807556/v1","type":"article-journal","title":"Photo-induced oxygen vacancy modulation in solution-processed TiO2/ZnFe2O4 heterointerface for all-oxide dual-mode neuromorphic logic memory","abstract":"Abstract Memristors, with their ability to switch between resistance states and emulate synaptic plasticity, are promising candidates for brain-inspired neuromorphic computing, having the potential to meet the growing demand for energy-efficient, parallel information processing. However, conventional devices are restricted to either electrical or optical operation, limiting multimodal functionality, commonly observed in the nervous systems. Here, we prepared a spin-coated TiO2/ZnFe2O4 heterojunction that effectively mimics synaptic plasticity under electrical and optical stimuli simultaneously, enabling our device to perform configurable logic operations. Ultraviolet photoelectron spectroscopy combined with the X-ray photoelectron spectroscopy studies before and after illumination suggest light-driven oxygen-vacancy modulation dominates the tunneling current through the heterointerface, resulting in persistent photo-response and optical synaptic plasticity. However, the confined filament formation at the heterointerfaces enables plasticity with electrical pulses. Our solution-processed all-oxide heterojunction memristor with multimodal functionality may provide a biologically realistic pathway to a cost-effective, stable alternative in emulating the human brain.","author":[{"family":"Bera","given":"Ashok"},{"family":"Farooq","given":"Faisal"},{"family":"Kaith","given":"Priya"}],"issued":{"date-parts":[[2025]]},"DOI":"10.21203/rs.3.rs-7807556/v1","URL":"https://doi.org/10.21203/rs.3.rs-7807556/v1","source":"europepmc"},{"id":"doi:10.1002/adma.202515532","type":"article-journal","title":"Organic Electrochemical Transistors for Neuromorphic Devices and Applications.","abstract":"Abstract Neuromorphic engineering, an interdisciplinary field bridging bioelectronics and neuroscience, endeavors to address the bottleneck of the von Neumann architecture by constructing hardware‐level artificial neural networks (ANNs) and replicate the complicated architecture and functionality of the human brain, heralding a new era of intelligent sensing, processing, and computing systems. Organic electrochemical transistors (OECTs), which operate via the bulk doping of organic mixed ionic–electronic conductors, are emerging as promising platforms for neuromorphic devices that emulate neuronal and synaptic activities while seamlessly integrating with biological systems. OECTs offer several advantages, including compatibility with flexible and stretchable substrates, tunable ionic and electronic conductivity, multimodal sensing capability, and operation at low voltages. This review aims to provide a comprehensive and state‐of‐the‐art vista of the rapidly advancing field of OECT‐based neuromorphic devices, including organic electrochemical neurons, organic electrochemical synapses, and their integrated devices. Particular emphasis is placed on their ability to perform neuromorphic functions and diverse applications in neuromorphic computing and flexible biointerfaces. Conclusions, remaining challenges, and future prospects for the development of OECT‐based neuromorphic devices are finally outlined.","author":[{"family":"Xiang","given":"Kexin"},{"family":"Song","given":"Jiajun"},{"family":"Liu","given":"Hong"},{"family":"Chen","given":"Junxin"},{"family":"Yan","given":"Feng"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1002/adma.202515532","URL":"https://doi.org/10.1002/adma.202515532","source":"europepmc"},{"id":"doi:10.21203/rs.3.rs-7089051/v1","type":"article-journal","title":"A Superconducting Flux-Quanta Memory Device for Cryogenic Neuromorphic and Probabilistic Computing","abstract":"Abstract As quantum computing progresses toward large-scale implementation, there is a growing demand for cryo-compatible artificial intelligence hardware and computing paradigms that accelerate the quantum control and error correction. Here, we present a superconducting memory device composed of a flux-quanta storage loop and a superconducting quantum interference device (SQUID), functioning as an artificial synapse and a binary stochastic neuron, respectively. Information is encoded in the stochastic behavior of the pulse stream passing through the neuron and is extracted by counting the number of SQUID switching events. In this scheme, sampling precision is flexibly tunable by adjusting the total number of input pulses. This architecture enables both neuromorphic and probabilistic computing, achieving 100% accuracy in nine-pixel image classification and integer factorization tasks by tuning the sampling precision. Furthermore, we found that the fabrication non-uniformities of up to 7.1% can be effectively compensated by an increased coupling strength. These results highlight the potential of superconducting memory devices as scalable and robust building blocks for implementing versatile AI computing paradigms under cryogenic conditions.","author":[{"family":"Chen","given":"Lei"},{"family":"Wang","given":"Yue"},{"family":"Liu","given":"Xu"},{"family":"Zheng","given":"Zengxu"},{"family":"Fan","given":"Xinxin"},{"family":"Liu","given":"Xiaoyu"},{"family":"Wu","given":"Ling"},{"family":"Shi","given":"Weifeng"},{"family":"Zhang","given":"Lu"},{"family":"Peng","given":"Wei"},{"family":"Ren","given":"Jie"},{"family":"Wang","given":"Zhen"}],"issued":{"date-parts":[[2025]]},"DOI":"10.21203/rs.3.rs-7089051/v1","URL":"https://doi.org/10.21203/rs.3.rs-7089051/v1","source":"preprints"},{"id":"doi:10.21203/rs.3.rs-6324848/v1","type":"article-journal","title":"Neuromorphic Computing Using Memristor Synapses and CMOS Neurons","abstract":"Abstract To address the increasing demands of artificial intelligence applications and the limitations of traditional computing, such as high power consumption, limited scalability, and inadequate parallelism, neuromorphic computing systems have been developed. This paper introduces a neuromorphic computing system designed for robust digit pattern recognition. A 20x20 memristor array-based synapse circuit, integrated with a refined Axon-Hillock (A-H) neuron model, forms the foundation to emulate synaptic and neuronal dynamics. A hard-coded approach is employed to adjust the synaptic weights of the memristor array for recognising digit patterns from 1 to 9. The results demonstrate that the neuromorphic computing system can accurately recognise the input patterns. Notably, the system is capable of maintaining its recognition abilities with 5% noise interference.","author":[{"family":"Choo","given":"Jia"},{"family":"Nath","given":"Shibajee"},{"family":"Kumar","given":"TN"}],"issued":{"date-parts":[[2025]]},"DOI":"10.21203/rs.3.rs-6324848/v1","URL":"https://doi.org/10.21203/rs.3.rs-6324848/v1","source":"europepmc"},{"id":"doi:10.21203/rs.3.rs-6838264/v1","type":"article-journal","title":"Artificial Synapse with Tunable Dynamic Range for Neuromorphic Computing with Ion Intercalated Bilayer Graphene","abstract":"Abstract In neuromorphic computing, a tunable dynamic range in artificial synapses is crucial, as it allows devices to emulate the human brain's efficiency in processing complex information with analog programmable states. Here, we introduce an electrochemical random-access memory (ECRAM) based on bilayer graphene. Our device achieves a large and programmable dynamic range through lithium-ion (Li+) intercalation, pulse modulation, and geometric engineering. We systematically investigated how pulse parameters, including amplitude, duty cycle, frequency, and signal type, affect conductance and dynamic range. Our results demonstrate that higher pulse amplitudes and/or longer duty cycles enhance Li+ intercalation efficiency; while lower frequencies pulse trains facilitate ion intercalation, significantly influencing conductance and dynamic range. Additionally, we explored how geometric factors such as channel length in bilayer and multi-layer graphene and the introduction of hole structures may affect intercalation kinetics and subsequently device performance. First-principles density functional theory (DFT) calculations were also performed to support that the interlayer space in bilayer graphene is energetically favorable for Li transport, and that hole structures promote efficient Li intercalation by providing barrierless pathways through the exposed edges of the bilayer. These findings confirm bilayer graphene as a promising material for developing high-performance artificial synapses with tunable characteristics.","author":[{"family":"He","given":"Yuzhi"},{"family":"Cao","given":"Purun"},{"family":"Hashemkhani","given":"Shahin"},{"family":"Liu","given":"Yihan"},{"family":"Vaz","given":"Daniel"},{"family":"Joy","given":"Keya"},{"family":"Youngblood","given":"Nathan"},{"family":"Kubendran","given":"Rajkumar"},{"family":"Anantram","given":"MP"},{"family":"Xiong","given":"Feng"}],"issued":{"date-parts":[[2025]]},"DOI":"10.21203/rs.3.rs-6838264/v1","URL":"https://doi.org/10.21203/rs.3.rs-6838264/v1","source":"europepmc"},{"id":"doi:10.21203/rs.3.rs-5770022/v1","type":"article-journal","title":"NEOSTI: A Neuromorphic Electronic-Opto Spatial-Temporal Hybrid Image Sensor","abstract":"Abstract Image sensors in machine vision systems face significant challenges related to energy efficiency and processing capability when storing, transferring, and processing massive amounts of data. In humans, over 80% of information processed by the brain is obtained through the eyes, which are capable of detecting and synchronously processing information with extremely low overall power consumption. Inspired by the biomimetics, here we propose a Neuromorphic Electronic-Opto Spatial Temporal Imager (NEOSTI), the smallest all-in-one eye size fully integrated vision system enabling acquisition and operation in typical indoor/outdoor non-coherent environments, under both natural and artificial lighting conditions, without any extra requirement of the light source, such as laser or coherent light source. NEOSTI combines processing-pre-sensor (PPS) in optical domain, processing-in-sensor (PIS) with non-linear acquisition capability while optical to electronic converting, and processing-near-sensor (PNS) in electronic domain, enabling parallel data computing capabilities while sensing. NEOSTI also integrates a low complexity Binary Neural Network (BNN) on the chip to process image semantic information. It attains near-human performance in five static and dynamic visual processing tasks.","author":[{"family":"Zhang","given":"Milin"},{"family":"Liu","given":"Tianyi"},{"family":"Huang","given":"Zheng"},{"family":"Wang","given":"Xuecheng"},{"family":"Shi","given":"Wanxin"},{"family":"Chen","given":"Hongwei"}],"issued":{"date-parts":[[2025]]},"DOI":"10.21203/rs.3.rs-5770022/v1","URL":"https://doi.org/10.21203/rs.3.rs-5770022/v1","source":"preprints"},{"id":"doi:10.21203/rs.3.rs-6073810/v1","type":"article-journal","title":"Temporal Hierarchy in Spiking Neural Networks","abstract":"Abstract Taking inspiration from the brain to perform efficient computation is a fascinating challenge, often hampered by the complexity of the brain itself. This work investigates a key feature observed across the cortices of various mammals: the hierarchy of time scales in cortical areas. Experimental evidence suggests that the intrinsic speed of activity in neuronal populations progressively slows down through the cortical hierarchy, forming what we define as a ``temporal hierarchy''. We explore whether this property provides an advantage in artificial computational systems. To test this hypothesis, we analyze SNNs, neural networks inspired by biological computation and with inherent temporal dynamics, and we endow them with temporal hierarchy. We incorporate the temporal hierarchy in various SNN temporal parameters, including their neuronal dynamics, synaptic delays, and recurrent dynamics. Our study evaluates these temporal hierarchies under two settings, (1) as an inductive bias, where SNNs temporal parameters are initialized with a predefined hierarchy to evaluate performance improvements, and (2) through optimizing the temporal parameters of SNNs. Benchmarking these networks on temporal tasks, the Multi-Timescale-XOR, and keyword spotting, we validate the benefits of hierarchy in various temporal mechanisms under both settings. In the inductive bias setting, we show that under iso-parameters settings, (i) classification accuracy correlates positively with the magnitude of temporal hierarchy on all the benchmarks, and (ii) there is an accuracy gain between 2 to 6% across tasks, when introducing temporal hierarchy, compared to a network without hierarchy. Moreover, under iso-accuracy settings, introducing hierarchy reduces the required number of parameters by up to 5 times. In the emergence from the optimization setting, we show that temporal hierarchy is naturally found as an emergent property through gradient descent. Finally, we conduct a detailed mathematical analysis of the temporal processing capabilities of SNNs, showing that a hierarchical arrangement of time constants enables a logarithmic reduction in the number of layers required to process temporal signals with multiple frequency components.","author":[{"family":"Moro","given":"Filippo"},{"family":"Aceituno","given":"Pau"},{"family":"Kriener","given":"Laura"},{"family":"Payvand","given":"Melika"}],"issued":{"date-parts":[[2025]]},"DOI":"10.21203/rs.3.rs-6073810/v1","URL":"https://doi.org/10.21203/rs.3.rs-6073810/v1","source":"preprints"},{"id":"doi:10.1088/1361-6463/ae2edd","type":"article-journal","title":"Enhancing non-volatile memory and neuromorphic computing: integration of PRAM and OTS for scalable, energy-efficient architectures","abstract":"Abstract This paper investigates the integration of phase-change random access memory (PRAM) and ovonic threshold switch (OTS) devices, emphasizing their ability to advance non-volatile memory technologies, neuromorphic computing architectures, and energy-efficient systems. OTS devices’ nonlinear threshold switching effectively mitigates sneak currents in high-density crossbar arrays, while challenges like resistivity drift and structural relaxation are addressed through advanced modeling and experimental analysis. The study highlights innovations in phase-change materials, such as doped Sb 2 Te 3 alloys and Sb 2 Te 3 –GeTe superlattices deposited by magnetron sputtering, which have been reported to improve thermal stability, reduce RESET power, and enhance cycling endurance compared with conventional GST-based phase-change memory devices. PRAM-OTS hybrid systems demonstrate exceptional performance in spiking and multi-layer neural networks, replicating neuronal behaviors such as integrate-and-fire dynamics and spike-timing-dependent plasticity for low-latency, energy-efficient processing in artificial intelligence, robotics, and internet of things applications. These scalable and reliable systems provide a robust framework for next-generation high-performance computational platforms, addressing key challenges in scalability, energy efficiency, and operational longevity.","author":[{"family":"Park","given":"Seoyoung"},{"family":"Koo","given":"Minsuk"},{"family":"Kim","given":"Sungjun"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1088/1361-6463/ae2edd","URL":"https://doi.org/10.1088/1361-6463/ae2edd","source":"crossref"},{"id":"doi:10.1088/1674-4926/24100020","type":"article-journal","title":"Synaptic devices based on silicon carbide for neuromorphic computing","abstract":"Abstract To address the increasing demand for massive data storage and processing, brain-inspired neuromorphic computing systems based on artificial synaptic devices have been actively developed in recent years. Among the various materials investigated for the fabrication of synaptic devices, silicon carbide (SiC) has emerged as a preferred choices due to its high electron mobility, superior thermal conductivity, and excellent thermal stability, which exhibits promising potential for neuromorphic applications in harsh environments. In this review, the recent progress in SiC-based synaptic devices is summarized. Firstly, an in-depth discussion is conducted regarding the categories, working mechanisms, and structural designs of these devices. Subsequently, several application scenarios for SiC-based synaptic devices are presented. Finally, a few perspectives and directions for their future development are outlined.","author":[{"family":"Ye","given":"Boyu"},{"family":"Liu","given":"Xiao"},{"family":"Wu","given":"Chao"},{"family":"Yan","given":"Wensheng"},{"family":"Pi","given":"Xiaodong"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1088/1674-4926/24100020","URL":"https://doi.org/10.1088/1674-4926/24100020","source":"crossref"},{"id":"doi:10.1088/2634-4386/ada989","type":"article-journal","title":"Neuromorphic compliant control facilitates human-prosthetic performance for hand grasp functions","abstract":"Abstract Current bionic hands lack the ability of fine force manipulation for grasping fragile objects due to missing human neuromuscular compliance in control. This incompatibility between prosthetic devices and the sensorimotor system has resulted in a high abandonment rate of hand prostheses. To tackle this challenge, we employed a neuromorphic modeling approach, biorealistic control, to regain human-like grasping ability. The biorealistic control restored muscle force regulation and stiffness adaptation using neuromorphic modeling of the neuromuscular reflex units, which was capable of real-time computing of model outputs. We evaluated the dexterity of the biorealistic control with a set of delicate grasp tasks that simulated varying challenging scenarios of grasping fragile objects in daily activities of life, including the Box and Block Task, the Glass Box Task, and the Potato Chip Task. The performance of the biorealistic control was compared with that of proportional control.&amp;#xD;Results indicated that the biorealistic control with the compliance of the neuromuscular reflex units significantly outperformed the proportional control with more efficient grip forces, higher success rates, fewer break and drop rates. Post-task survey questionnaires revealed that the biorealistic control reduced subjective burdens of task difficulty and improved subjective confidence in control performance significantly. The outcome of the evaluation confirmed that the biorealistic control could achieve superior abilities in fine, accurate, and efficient grasp control for prosthetic users.&amp;#xD;","author":[{"family":"Xie","given":"Anran"},{"family":"Zhang","given":"Zhuozhi"},{"family":"Zhang","given":"Jie"},{"family":"Chen","given":"Weidong"},{"family":"Patton","given":"James"},{"family":"Lan","given":"Ning"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1088/2634-4386/ada989","URL":"https://doi.org/10.1088/2634-4386/ada989","source":"crossref"},{"id":"doi:10.58346/jowua.2026.i2.006","type":"article-journal","title":"Neuromorphic Computing-Enabled Context-Aware Adaptive Mobile Learning Framework for Real-Time Cognitive Load Management","abstract":"The growing use of mobile learning in wireless and ubiquitous computing environments has posed various difficulties in managing the cognitive load of learners in a dynamic environment. Traditional adaptive learning approaches may not be equipped with the capacity to monitor continuously the learner's cognitive state and adapt intelligently to the environmental and behavioral changes. This problem could lead to a lower level of engagement of learners, increased cognitive load, and inefficiency in the learning process. In order to solve such problems, this research paper presents the Neuromorphic Computing-Enabled Context-Aware Adaptive Mobile Learning Framework for Real-Time Cognitive Load Management. This approach consistently monitors contextual factors such as the user's engagement pattern, environmental conditions, and device-level operations to predict the cognitive load and deliver the instruction accordingly. Neuromorphic intelligence was applied for cognitive processes that will be done with low delay and energy consumption in the wireless mobile context. The new learning model was tested with the help of various metrics, namely adaptation accuracy, precision, recall, F1-score, energy efficiency, and response latency. The experimental outcome revealed that the new framework obtained an adaptation accuracy rate of 96.8%, precision of 96.1%, recall of 95.7%, and F1-score of 95.9%, which is superior to traditional mobile learning and deep adaptive learning models. Furthermore, the proposed framework offered a reduction in response delay to 104 ms and an energy efficiency rate of 0.95. Results indicate that the integration of neuromorphic computing technology with context-aware adaptive learning systems can greatly improve learner personalization, cognitive load management, and system reactivity. The suggested framework is expected to contribute towards the design of an intelligent, scalable, and energy-efficient adaptive learning environment for future wireless-based learning platforms.","author":[{"family":"Sidikova","given":"Gulbahor"},{"family":"Karakulov","given":"Nurbol"},{"family":"Tursunov","given":"Mustafo"},{"family":"Kochkarov","given":"Atabek"},{"family":"Oltiboyev","given":"Asqad"},{"family":"Arustamyan","given":"Yana"},{"family":"Toirova","given":"Dilfuza"},{"family":"Madraimov","given":"Abdumajid"}],"issued":{"date-parts":[[2026]]},"DOI":"10.58346/jowua.2026.i2.006","URL":"https://doi.org/10.58346/jowua.2026.i2.006","source":"crossref"},{"id":"doi:10.1088/2634-4386/ae294e","type":"article-journal","title":"Van der Waals integration of 2D materials for advanced intelligent computing","abstract":"Abstract The increasing demand for faster, energy-efficient, and higher bandwidth semiconductor devices has pushed conventional Si-based scaling to its fundamental limits, including mobility degradation, short-channel effects, and high power consumption. To overcome these challenges, three-dimensional integration has emerged as a promising strategy, but wafer-based approaches like through-Si-via face critical limitations in stacking density, mechanical stress, and fabrication complexity. Two-dimensional materials provide a compelling alternative due to their atomically thin structure, superior electrical and mechanical properties, and ability to sustain performance at the atomic scale. Moreover, their van der Waals integration enables heterogeneous, high-density, and efficient assembly of functional layers. This review summarizes recent advances in the preparation and van der Waals integration of 2D materials, including growth, transfer, and direct integration. Their applications in intelligent computing that range from logic to sensor devices and their potential as next-generation electronics are discussed.","author":[{"family":"Kwak","given":"Chaehyeon"},{"family":"Park","given":"Keunpyo"},{"family":"Song","given":"Min"},{"family":"Jang","given":"Ho"},{"family":"Suh","given":"Jun"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1088/2634-4386/ae294e","URL":"https://doi.org/10.1088/2634-4386/ae294e","source":"crossref"},{"id":"doi:10.1126/sciadv.adv6603","type":"article-journal","title":"Neuromorphic ionic computing in droplet interface synapses","abstract":"Ionic devices with memory capabilities can emulate neural functionality, enabling neuromorphic computing and biomedical applications. In this study, we report an ionic spiking synapse based on aqueous droplet interface bilayer assembly. Under stepwise triangular voltages, the device displays coupled memcapacitive-memristive behavior, showing noncrossing pinched hysteretic I - V loops. This hysteretic ion dynamics can be regulated by modifying bilayer components, reconstituting protein channels, or adjusting droplet assembly configuration. Droplet interface synapses (DIS) exhibit fundamental neuromorphic behaviors such as paired-pulse facilitation/depression, spike rate–dependent plasticity, Hebbian learning, and short-term associative learning under classical conditioning. We also used reservoir computing with DIS to implement two learning algorithms: a classification algorithm that recognizes handwritten digits and a reinforcement learning algorithm that learns to play a board game of tic-tac-toe.","author":[{"family":"Li","given":"Zhongwu"},{"family":"Myers","given":"Sydney"},{"family":"Xiao","given":"Jingyi"},{"family":"Li","given":"Yuhao"},{"family":"Noy","given":"Natasha"},{"family":"Leuski","given":"Anton"},{"family":"Noy","given":"Aleksandr"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1126/sciadv.adv6603","URL":"https://doi.org/10.1126/sciadv.adv6603","source":"europepmc"},{"id":"doi:10.1002/aelm.202500515","type":"article-journal","title":"Simultaneous Dual‐Plasticity Organic Synaptic Transistor for Neuromorphic Computing","abstract":"ABSTRACT Neuromorphic computing systems require artificial synaptic devices capable of emulating complex biological neural functions. This study presents a dinaphtho[2,3‐b:2′,3′‐f]thieno[3,2‐b]thiophene (DNTT)‐based organic field‐effect transistor that demonstrates synaptic plasticity under optical stimulation at 200 K. The device exhibits a dual‐mechanism synaptic behavior through charge separation and trapping, where photogenerated holes provide rapid transport while electrons are preferentially captured in deep trap states, creating persistent field modulation. Excitatory postsynaptic current measurements reveal characteristic three‐phase temporal dynamics with rapid activation, exponential decay, and sustained enhancement lasting tens of minutes. Paired‐pulse facilitation demonstrates short‐term plasticity with dual exponential decay constants of 140 and 610 ms, while multi‐pulse stimulation produces remarkable persistent current level enhancement exceeding 10 000% of the initial baseline, reflecting sequential filling of continuous trap state distributions. The device simultaneously implements both short‐term and long‐term plasticity mechanisms in a single component, enabling simultaneous working memory and persistent information storage functions. Neuromorphic functionality is demonstrated through simulated XOR logic operations, showing non‐linearly separable computation capabilities. The 200 K operating temperature aligns favorably with Mars surface conditions, requiring minimal heating compared to terrestrial cooling requirements, making the device particularly promising for space‐based neuromorphic systems where radiation‐hard organic semiconductors provide additional advantages.","author":[{"family":"Vincze","given":"Tomas"},{"family":"Hanic","given":"Michal"},{"family":"Berki","given":"Martin"},{"family":"Weis","given":"Martin"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1002/aelm.202500515","URL":"https://doi.org/10.1002/aelm.202500515","source":"crossref"},{"id":"doi:10.1088/2631-8695/ae1d0c","type":"article-journal","title":"Neuromorphic computing for energy-efficient machine intelligence","abstract":"Abstract Neuromorphic computing has gained a significant amount of attention from industry as well as the research community as a means of overcoming the growing energy demands of machine intelligence. Neuromorphic systems offer promising computation approaches by mimicking the functioning of the human brain’s energy-efficient neural processing. This perspective explores the significance of energy efficiency in artificial intelligence (AI) systems, examines the potential issues associated with traditional AI architectures, and highlights the importance of neuromorphic computing as a sustainable solution. It also provides an overview of the working principle of neuromorphic systems, a spiking neural network (SNN) implementation, and the comparative advantages over conventional computing architectures. This perspective further investigates several real-world applications to assess the potential of neuromorphic computing to perform real-time and energy-efficient operations. Although they offer significant advantages, neuromorphic systems face multiple challenges regarding materials development, fabrication techniques, large-scale deployments, and a lack of standard software tools. Finally, this perspective outlines potential research directions to explore the role of neuromorphic computing in next-generation AI systems.","author":[{"family":"Latif","given":"Shahid"},{"family":"Zafar","given":"Saniya"},{"family":"Ahmad","given":"Jawad"},{"family":"Khan","given":"Muhammad"},{"family":"Ullah","given":"Farhan"},{"family":"Khattak","given":"Aizaz"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1088/2631-8695/ae1d0c","URL":"https://doi.org/10.1088/2631-8695/ae1d0c","source":"crossref"},{"id":"doi:10.3389/fnins.2026.1827009","type":"article-journal","title":"Federated training of spiking neural networks on edge hardware for audio processing.","abstract":"Spiking Neural Networks have caught significant attention recently for their potential for energy-efficient computation on neuromorphic hardware and their event-driven processing. Spiking Neural networks employ spike-based learning paradigms, which require specialized training procedures such as Surrogate Gradient Descent. At the same time, Federated Learning allows collaborative model training on decentralized devices with preservation of data privacy protection. However, to date, few research has examined the suitability of Federated learning with ARM-based hardware. This work primarily investigates whether Federated Spiking Neural Networks training on ARM-based hardware is feasible with the Raspberry Pi 5 as a widely available and low-cost edge computing device for audio signal processing tasks. We perform a comparative analysis of federated Spiking Neural Network and federated convolutional neural networks on ARM processors and evaluate their performance on different data partitioning strategies using Dirichlet-based splits and various federated averaging algorithms. Using Federated learning, this work investigates the impact of data heterogeneity and aggregation strategies on model convergence, communication overhead, and latency in distributed training paradigms. The results provided showcases the important insights into the trade-offs of FL-SNN implementations on Von Neumann architectures and their applications in decentralized neuromorphic computing for audio processing.","author":[{"family":"Kaimal","given":"Swaroop"},{"family":"Jb","given":"Ashwin"},{"family":"Reka","given":"SS"},{"family":"Venugopal","given":"Prakash"}],"issued":{"date-parts":[[2026]]},"DOI":"10.3389/fnins.2026.1827009","URL":"https://doi.org/10.3389/fnins.2026.1827009","source":"europepmc"},{"id":"doi:10.1007/s40820-025-01940-9","type":"article-journal","title":"Multisensory Neuromorphic Devices: From Physics to Integration.","abstract":"Abstract The increasing complexity of intelligent sensing environments, driven by the growth of Internet of Things technologies, has created a strong demand for neuromorphic systems capable of real-time, low-power multisensory perception. Traditional sensory architectures, constrained by single-modal processing and centralized computing, struggle to meet the requirements of diverse and dynamic input conditions. Multisensory neuromorphic devices offer a promising solution by mimicking the distributed, event-driven processing of biological systems. Recent efforts have explored synaptic devices and material systems that respond to various input modalities, including visual, tactile, thermal, and chemical stimuli. However, challenges remain in signal conversion, encoding compatibility, and the fusion of heterogeneous inputs without loss of unisensory information. This review provides a comprehensive overview of the physical mechanisms, device behaviors, and integration strategies that underpin signal processing in neuromorphic hardware. We highlight synaptic mechanisms conducive to cross-modal interaction, analyze representative signal fusion approaches at the device level, and discuss future directions for constructing efficient, scalable, and biologically inspired multisensory neuromorphic systems.","author":[{"family":"Gui","given":"An"},{"family":"Mu","given":"Haoran"},{"family":"Yang","given":"Rong"},{"family":"Zhang","given":"Guangyu"},{"family":"Lin","given":"Shenghuang"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1007/s40820-025-01940-9","URL":"https://doi.org/10.1007/s40820-025-01940-9","source":"europepmc"},{"id":"doi:10.1007/s40820-025-01902-1","type":"article-journal","title":"Two-Dimensional MXene-Based Advanced Sensors for Neuromorphic Computing Intelligent Application.","abstract":"Abstract As emerging two-dimensional (2D) materials, carbides and nitrides (MXenes) could be solid solutions or organized structures made up of multi-atomic layers. With remarkable and adjustable electrical, optical, mechanical, and electrochemical characteristics, MXenes have shown great potential in brain-inspired neuromorphic computing electronics, including neuromorphic gas sensors, pressure sensors and photodetectors. This paper provides a forward-looking review of the research progress regarding MXenes in the neuromorphic sensing domain and discussed the critical challenges that need to be resolved. Key bottlenecks such as insufficient long-term stability under environmental exposure, high costs, scalability limitations in large-scale production, and mechanical mismatch in wearable integration hinder their practical deployment. Furthermore, unresolved issues like interfacial compatibility in heterostructures and energy inefficiency in neuromorphic signal conversion demand urgent attention. The review offers insights into future research directions enhance the fundamental understanding of MXene properties and promote further integration into neuromorphic computing applications through the convergence with various emerging technologies.","author":[{"family":"Lu","given":"Lin"},{"family":"Sun","given":"Bo"},{"family":"Wang","given":"Zheng"},{"family":"Meng","given":"Jialin"},{"family":"Wang","given":"Tianyu"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1007/s40820-025-01902-1","URL":"https://doi.org/10.1007/s40820-025-01902-1","source":"europepmc"},{"id":"doi:10.3390/nano15141130","type":"article-journal","title":"Memristor-Based Spiking Neuromorphic Systems Toward Brain-Inspired Perception and Computing.","abstract":"Threshold-switching memristors (TSMs) are emerging as key enablers for hardware spiking neural networks, offering intrinsic spiking dynamics, sub-pJ energy consumption, and nanoscale footprints ideal for brain-inspired computing at the edge. This review provides a comprehensive examination of how TSMs emulate diverse spiking behaviors—including oscillatory, leaky integrate-and-fire (LIF), Hodgkin–Huxley (H-H), and stochastic dynamics—and how these features enable compact, energy-efficient neuromorphic systems. We analyze the physical switching mechanisms of redox and Mott-type TSMs, discuss their voltage-dependent dynamics, and assess their suitability for spike generation. We review memristor-based neuron circuits regarding architectures, materials, and key performance metrics. At the system level, we summarize bio-inspired neuromorphic platforms integrating TSM neurons with visual, tactile, thermal, and olfactory sensors, achieving real-time edge computation with high accuracy and low power. Finally, we critically examine key challenges—such as stochastic switching origins, device variability, and endurance limits—and propose future directions toward reconfigurable, robust, and scalable memristive neuromorphic architectures.","author":[{"family":"Wang","given":"Xiangjing"},{"family":"Zhu","given":"Yixin"},{"family":"Zhou","given":"Zili"},{"family":"Chen","given":"Xin"},{"family":"Jia","given":"Xiaojun"}],"issued":{"date-parts":[[2025]]},"DOI":"10.3390/nano15141130","URL":"https://doi.org/10.3390/nano15141130","source":"europepmc"},{"id":"doi:10.1186/s43593-025-00087-9","type":"article-journal","title":"Ultrafast neuromorphic computing driven by polariton nonlinearities","abstract":"Abstract Neuromorphic computing offers a promising approach to artificial intelligence by mimicking biological neural networks to perform complex tasks efficiently. While software-based simulations have demonstrated the potential of neuromorphic architectures, a physical platform is crucial to fully realize its computational advantages. Herein, we present the first demonstration of perovskite microcavity exciton polaritons as a platform for reservoir computing-based artificial neural networks. By leveraging the nonlinear response properties of exciton polaritons, we developed a neuromorphic computing architecture capable of performing classification tasks with single-step training, eliminating the need for iterative algorithms like backpropagation. Applying this system to a handwritten digit recognition task, we achieve 92% classification accuracy at room temperature. Notably, we also show that the system is dynamically nonlinear, further enhancing the potential to improve classification efficiency and address more complex tasks. Our findings advocate the promising capabilities of perovskite exciton polaritons as energy-efficient, ultrafast response platforms for artificial intelligence, paving the way for next-generation computational technologies.","author":[{"family":"Gan","given":"Yusong"},{"family":"Shi","given":"Ying"},{"family":"Ghosh","given":"Sanjib"},{"family":"Liu","given":"Haiyun"},{"family":"Xu","given":"Huawen"},{"family":"Xiong","given":"Qihua"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1186/s43593-025-00087-9","URL":"https://doi.org/10.1186/s43593-025-00087-9","source":"crossref"},{"id":"doi:10.1002/aelm.202500713","type":"article-journal","title":"Electrode‐Engineered Dual‐Mode Multifunctional Lead‐Free Perovskite Optoelectronic Memristors for Neuromorphic Computing","abstract":"ABSTRACT Memristive devices based on halide perovskites hold strong promise to provide energy‐efficient systems for the Internet of Things (IoT); however, lead (Pb) element should be minimized or ideally replaced. Herein, we introduce a multifunctional device based on AgBiI 4 Pb‐free perovskite as the active layer in a normal n‐i‐p solar cell configuration that exhibitss multiple mode‐dependent neuromorphic functions. Specifically, through electrode engineering, we show a selection between volatile or non‐volatile memristive switching when the Au top electrode is replaced by Ag. The volatile device having an Au top electrode can be used for emulating threshold‐dependent artificial neuron firing processes, while the non‐volatile device with Ag pads can emulate numerous synaptic protocols (LTP/LTD, PPF, STDP, SRDP, LTM, STM). Both volatile (V TH = −0.86 V) and non‐volatile modes possess switching low voltages &lt;1 V, while the ON/OFF ratio of the non‐volatile system is 10 4 with a cycling endurance of 10 3 cycles and state retention of 10 3 s. Further investigation reveals that electrode type affects the conduction mechanism, as ion charge trapping and detrapping govern threshold switching with activation energy E A ≈ 0.7 eV, while the formation/rupture of Ag filaments is responsible for non‐volatile switching. The response in both modes of devices can be tuned by light, while associative learning by emulating Pavlovian learning is demonstrated. Finally, experimental non‐volatile data were used for MNIST and Fashion‐MNIST classification, achieving 97.11% and 87.56% accuracy, respectively. Our work provides insights into electrode‐engineered mode control in lead‐free perovskite systems capable for a concurrent energy harvesting (PCE ∼1%) toward self‐powered systems for neuromorphic edge computing.","author":[{"family":"Loizos","given":"Michalis"},{"family":"Rogdakis","given":"Konstantinos"},{"family":"Chatzimanolis","given":"Konstantinos"},{"family":"Anagnostou","given":"Katerina"},{"family":"Kymakis","given":"Emmanuel"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1002/aelm.202500713","URL":"https://doi.org/10.1002/aelm.202500713","source":"crossref"},{"id":"doi:10.1007/s40820-025-01705-4","type":"article-journal","title":"Low-Power Memristor for Neuromorphic Computing: From Materials to Applications","abstract":"Abstract As an emerging memory device, memristor shows great potential in neuromorphic computing applications due to its advantage of low power consumption. This review paper focuses on the application of low-power-based memristors in various aspects. The concept and structure of memristor devices are introduced. The selection of functional materials for low-power memristors is discussed, including ion transport materials, phase change materials, magnetoresistive materials, and ferroelectric materials. Two common types of memristor arrays, 1T1R and 1S1R crossbar arrays are introduced, and physical diagrams of edge computing memristor chips are discussed in detail. Potential applications of low-power memristors in advanced multi-value storage, digital logic gates, and analogue neuromorphic computing are summarized. Furthermore, the future challenges and outlook of neuromorphic computing based on memristor are deeply discussed.","author":[{"family":"Xia","given":"Zhipeng"},{"family":"Sun","given":"Xiao"},{"family":"Wang","given":"Zhenlong"},{"family":"Meng","given":"Jialin"},{"family":"Jin","given":"Boyan"},{"family":"Wang","given":"Tianyu"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1007/s40820-025-01705-4","URL":"https://doi.org/10.1007/s40820-025-01705-4","source":"europepmc"},{"id":"doi:10.1515/ntrev-2025-0292","type":"article-journal","title":"Honey-CNT memristive artificial synaptic device for sustainable neuromorphic computing system","abstract":"Abstract Brain-inspired neuromorphic computing systems require hardware components analogous to biological neurons and synapses. Honey based natural organic memristor has demonstrated promising nonvolatile memristive behaviors, with the advantages of sustainability, environmentally friendliness, and low-cost manufacturing. In this study, carbon nanotubes (CNTs) are added in honey to fabricate honey-CNT memristive artificial synaptic devices. Honey-CNT film is characterized by micro-Raman spectroscopy and the distribution of CNT bundles embedded in the honey-CNT composite layer by cross-sectional scanning electron microscopy for the first time. Critical synaptic functions of the honey-CNT memristor, including spike-rate-dependent plasticity, spike voltage dependent plasticity, learn-forget-relearn, and supralinear spatial summation are revealed, which have not been reported by honey based memristive devices before. Paired pulse facilitation with a PPF index as large as 4.5 is observed, indicating the enhancement of synaptic weight by CNTs. Furthermore, honey-CNT memristor based neuromorphic system is evaluated in terms of linearity, accuracy, read/write energy, and overall performance by the image recognition using Stochastic Gradient Descent and Adaptive Moment Estimation learning algorithms and the Modified National Institute of Standards and Technology database.","author":[{"family":"Tanim","given":"Md"},{"family":"Templin","given":"Zoe"},{"family":"Uppaluru","given":"Harshvardhan"},{"family":"Wang","given":"Jinhui"},{"family":"Cheong","given":"Kuan"},{"family":"Zhao","given":"Feng"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1515/ntrev-2025-0292","URL":"https://doi.org/10.1515/ntrev-2025-0292","source":"crossref"},{"id":"doi:10.2139/ssrn.6734323","type":"manuscript","title":"Memristors based on azatriphenylene and its derivatives modified by metal co-ordination for neuromorphic computing","abstract":"Recently, organic small-molecule semiconductors have emerged to serve as the re-sistive switching (RS) layer owing to the well-defined molecular structures and the capacity for precise modulation of transport characteristics through chemical modi-fication. In this work, we have designed and synthesized five organic small mole-cules based on the azatriphenylene (AT) framework, including AT and its derivatives with substituents of methoxy, methyl, chloro, and trifluoromethyl. This provides an opportunity to systematically investigate the relationships between the molecular structure, device configuration and the memristive behaviors. The results demon-strate that different functional groups can modulate the HOMO/LUMO energy levels and the electron distribution, thereby influencing the RS performance of the fabri-cated devices based on the synthesized small-molecule semiconductors. Among them, the 6,11-bis(trifluoromethyl)-2-azatriphenylene (BTFMAT) with the strong electron-withdrawing effect of trifluoromethyl exhibits typical volatile RS behaviors with relatively high ON/OFF ratio. The BTFMAT-based memristor successfully em-ulated biological synaptic long-term potentiation and long-term depression (LTP/LTD) and learning-forgetting behaviors, and can be further employed to con-struct an artificial neural network (ANN), achieving a recognition accuracy of 93.6% for handwritten digit image recognition. Furthermore, by introducing a Cu interlayer into the BTFMAT device, a transition from volatile to non-volatile RS behavior can be realized because Cu2+ can coordinate with N atoms in BTFMAT to form stable complexes and subsequently Cu conductive filaments can be formed under an exter-nal electric field. This work provides a new design strategy for developing high-performance organic small-molecule memristors and exploring their potential applications in neuromorphic computing. Keywords: Neuromorphic computing, Memristor, Organic semiconductor, Metal ion doping, Synaptic plasticity","author":[{"family":"Li","given":"Yuexin"},{"family":"Yang","given":"Lan"},{"family":"Zhang","given":"Didi"},{"family":"Gao","given":"Qinlian"},{"family":"Li","given":"Mei"},{"family":"Tu","given":"Xiushan"},{"family":"He","given":"Yanyan"},{"family":"Zhou","given":"Zhaorong"},{"family":"Shi","given":"Yingbo"},{"family":"Jie","given":"Wenjing"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2139/ssrn.6734323","URL":"https://doi.org/10.2139/ssrn.6734323","source":"crossref"},{"id":"doi:10.33693/2313-223x-2025-12-1-11-16","type":"article-journal","title":"Implementation of Secure Traffic Light Management Using a Neuromorphic Computing Base Based on Fuzzy Graphs","abstract":"The task of safe traffic light management is to normalize traffic. Secure management involves the implementation of the protection of information involved in the management process. The traffic light management process is a complex dynamic process. The control object is a set of traffic lights. The subject of management is a specific management system, which can be autonomous or part of a top-level management system. The implementation of a traffic light based on a neuromorphic computing base allows you to perform some of the control processes in automatic mode. As part of the control process, you have to deal with different types of data (sensor readings, control signals of different levels, etc.). Big data is processed throughout the entire control process, and one of the main tasks is to reduce the dimensionality of the information being processed. The successful solution of this problem directly depends on the model of organization of a set of traffic lights. The article discusses various traffic lights, each of which is equipped with an intelligent controller. The neuromorphic basis of the intelligent controller allows you to expand the capabilities of the computing base at a low level. Fuzzy graphs are used to represent a set of traffic lights and to connect the traffic lights to the control system. This model makes it possible to combine the information and control components of the process of interaction between the object and the subject of management into one whole. The advantages of this representation are the minimization of information necessary for the successful solution of the problem of safe management, and the expansion of the possibilities of the management process through the use of fuzzy information.","author":[{"family":"Volosova","given":"Alexandra"},{"family":"Matyukhina","given":"Ekaterina"},{"family":"Morozov","given":"Egor"}],"issued":{"date-parts":[[2025]]},"DOI":"10.33693/2313-223x-2025-12-1-11-16","URL":"https://doi.org/10.33693/2313-223x-2025-12-1-11-16","source":"crossref"},{"id":"doi:10.1088/2634-4386/ae2155","type":"article-journal","title":"An energy-efficient, CMOS-compatible physical reservoir node with post-fabrication tunable decay dynamics","abstract":"Abstract In reservoir computing, the memory decay rate of physical reservoir nodes governs how quickly past inputs fade, thereby determining their temporal dynamics. Optimising this rate is therefore crucial for effective temporal signal processing. However, in most reported physical reservoirs, it is fixed at the time of fabrication and cannot be altered on demand for different applications. Thus, tailoring a single node for its adaptability across tasks with diverse temporal characteristics remains challenging. In this work, we propose and computationally analyse a CMOS-compatible, tunable-decay, hybrid reservoir node that integrates a subthreshold-operated field-effect transistor (FET) with a programmable ReRAM device (a memristor) and a capacitor connected at its gate terminal. The reservoir output measured as the FET drain current simultaneously captures temporal memory and nonlinear transformation of the input, while the memory decay time constant ( τ = R × C ) can be modulated in real time by adjusting the ReRAM resistance. We demonstrate the effectiveness of the proposed node on two representative benchmark tasks with contrasting τ requirements, namely, the MNIST digit classification with 96% accuracy, and a chaotic Hénon map prediction with a normalised RMS error of 0.0037, matching state-of-the-art hardware reservoirs. Our design achieves ultra-low energy consumption ( ≈ 15.20 pJ/operation), at least an order of magnitude lower than state-of-the-art implementations, while maintaining reliable operation and on-demand τ tunability. This combination of mature silicon technology and adaptive memristive functionality paves the way for energy-efficient, scalable, and reliable temporal learning systems.","author":[{"family":"Chaitanya","given":"Gambali"},{"family":"Arkalgud","given":"Aditya"},{"family":"Pande","given":"Shubham"},{"family":"Arora","given":"Ankit"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1088/2634-4386/ae2155","URL":"https://doi.org/10.1088/2634-4386/ae2155","source":"crossref"},{"id":"doi:10.1002/smll.202412531","type":"article-journal","title":"Neuromorphic Visual Computing with ZnMgO QDs‐Based UV‐Responsive Optoelectronic Synaptic Devices for Image Encryption and Recognition","abstract":"Abstract Retina‐inspired optoelectronic neuromorphic devices integrating optical sensing and computation are the key components in realizing neuromorphic visual computing. In particular, UV‐responsive optoelectronic synaptic devices hold significant value for advanced neuromorphic vision systems, as they can expand human visual perception. Herein, we demonstrate a UV‐responsive optoelectronic synaptic device based on ZnMgO quantum dots (QDs) designed for in‐sensor computing in neuromorphic vision applications. The device demonstrates voltage‐driven short‐term and long‐term synaptic plasticity, as well as multiple photoinduced synaptic functions. Based on this device, an in‐sensor image‐blending encryption method has been designed, which can effectively reduce the risk of data leakage during transmission. Furthermore, an in‐sensor reservoir computing (RC) system with image processing functions is constructed, which integrates a photonic reservoir layer (PRL) for image preprocessing and a multilayer perceptron (MLP) capable of image recognition. The system achieves 98.6% accuracy in recognizing Fashion‐MNIST images and maintains 83% accuracy under 60% random noise, showcasing its robustness. This work introduces a novel approach for developing UV‐responsive optoelectronic synaptic devices equipped with dual‐mode modulation of both electrical and optical signals, offering new perspectives and solutions for integrated applications in neuromorphic vision systems.","author":[{"family":"Guo","given":"Zilong"},{"family":"Kan","given":"Hao"},{"family":"Zhang","given":"Jiaqi"},{"family":"Li","given":"Yang"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1002/smll.202412531","URL":"https://doi.org/10.1002/smll.202412531","source":"crossref"},{"id":"doi:10.3389/fnano.2025.1621554","type":"article-journal","title":"Performance and variability analysis of ALD-grown wafer scale HfO2/Ta2O5-based memristive devices for neuromorphic computing","abstract":"Here, we report a large-scale wafer microfabrication process and in-depth electrical analysis of atomic layer deposition (ALD) grown bilayer (i.e., HfO 2 /Ta 2 O 5 ) memristive devices. The fabricated bilayer devices initially require an electroforming event and show stable bipolar resistive switching responses with some variations in the device switching voltages. These variations are covered in the 15.7%–22.7% range corresponding to the maximum switching voltage of the tested devices. Moreover, time series analysis (TSA) is employed by considering the device switching voltages ( V SET and V RESET ) to predict the device performance and the obtained outcomes are well matched to the experimental data. Furthermore, the least values of coefficient of variability ( C V ) in the device switching voltages are 6.09% ( V SET ) and 3.22% ( V RESET ) in the case of device-to-device (D2D) while 1.76% ( V SET ) and 2.14% ( V RESET ) in the case of cycle-to-cycle (C2C). Furthermore, the fabricated devices efficiently perform the synaptic functionalities in terms of potentiation (P) and depression (D), paired-pulse facilitation (PPF), and paired-pulse depression (PPD), with a least value of nonlinearity (NL) factor of 0.43 in synaptic response, which is close to the ideal value of NL in biological synapses. Therefore, the present work shows that the single ALD system can be an efficient deposition method to deposit high-k oxide materials for memristive arrays over large-scale wafers.","author":[{"family":"Kumar","given":"Sanjay"},{"family":"Yadav","given":"Deepika"},{"family":"Stathopoulos","given":"Spyros"},{"family":"Prodromakis","given":"Themis"}],"issued":{"date-parts":[[2025]]},"DOI":"10.3389/fnano.2025.1621554","URL":"https://doi.org/10.3389/fnano.2025.1621554","source":"crossref"},{"id":"doi:10.1088/2634-4386/ae958f","type":"article-journal","title":"Noise-Robust conceptors for physical reservoir computing: adaptation to perturbations","abstract":"Abstract Conceptors are a powerful extension of reservoir computing (RC) that enable the selective recall and stabilization of internal dynamics. However, their application to physical RC remains challenging because measured reservoir states are inevitably affected by noise and physical perturbations. In this work, we propose a noise-robust cross-trial-correlation-based (CTC) method for computing conceptors in noisy reservoir systems. By exploiting the consistency of the reservoir response across repeated trials, the method suppresses noise contributions that are uncorrelated between measurements. Numerical simulations of leaky echo state networks under additive state noise and parameter drift show that CTC-based conceptors preserve the relevant internal dynamics and extend the operational range of the reservoir compared with standard conceptors and unconstrained reservoirs. In an autonomous-generation task, the CTC-based conceptor maintains predictive capability beyond 50% noise, where the other configurations fail to sustain autonomous generation. Under combined noise and parameter drift, the CTC approach maintains normalized root mean square error values below 0.3, while the alternative approaches considered exceed this error level in the tested conditions. These numerical results establish a concrete step towards adapting conceptor-based control to physical RC platforms.","author":[{"family":"Infantes-Llinares","given":"Gemma"},{"family":"Kang","given":"Hongki"},{"family":"Soriano","given":"Miguel"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/2634-4386/ae958f","URL":"https://doi.org/10.1088/2634-4386/ae958f","source":"crossref"},{"id":"doi:10.1002/adma.202419245","type":"article-journal","title":"Toward Switching and Fusing Neuromorphic Computing: Vertical Bulk Heterojunction Transistors with Multi‐Neuromorphic Functions for Efficient Deep Learning","abstract":"Abstract The combination of artificial neural networks (ANN) and spiking neural networks (SNN) holds great promise for advancing artificial general intelligence (AGI). However, the reported ANN and SNN computational architectures are independent and require a large number of auxiliary circuits and external algorithms for fusion training. Here, a novel vertical bulk heterojunction neuromorphic transistor (VHNT) capable of emulating both ANN and SNN computational functions is presented. TaO x ‐based electrochemical reactions and PDVT‐10/N2200‐based bulk heterojunctions are used to realize spike coding and voltage coding, respectively. Notably, the device exhibits remarkable efficiency, consuming a mere 0.84 nJ of energy consumption for a single multiply accumulate (MAC) operation with excellent linearity. Moreover, the device can be switched to spiking neuron and self‐activation neuron by simply changing the programming without auxiliary circuits. Finally, the VHNT‐based artificial spiking neural network (ASNN) fusion simulation architecture is demonstrated, achieving 95% accuracy for Canadian‐Institute‐For‐Advanced‐ResearchResearch‐10 (CIFARResearch‐10) dataset while significantly enhancing training speed and efficiency. This work proposes a novel device strategy for developing high‐performance, low‐power, and environmentally adaptive AGI.","author":[{"family":"Zou","given":"Yi"},{"family":"Liu","given":"Di"},{"family":"Gan","given":"Xinyan"},{"family":"Yu","given":"Rengjian"},{"family":"Zhang","given":"Xianghong"},{"family":"Gao","given":"Chansong"},{"family":"Chen","given":"Zhenjia"},{"family":"Xu","given":"Chenhui"},{"family":"Ye","given":"Yun"},{"family":"Hu","given":"Yuanyuan"},{"family":"Guo","given":"Tailiang"},{"family":"Chen","given":"Huipeng"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1002/adma.202419245","URL":"https://doi.org/10.1002/adma.202419245","source":"crossref"},{"id":"doi:10.1002/aisy.70419","type":"article-journal","title":"Neuromorphic Denoising with Fully Analog Memristive In‐Memory Computing","abstract":"Analog compute‐in‐memory (CIM) technology utilizes the physical characteristics of memory devices to cancel the repeated data movement between memory and processors. Since this method gets rid of the signal domain conversion, analog CIM is suitable for the post‐sensing processing systems. However, two critical challenges emerge: a lack of area‐efficient analog buffers to complete the analog computing flow and the inevitable accumulation of noise throughout the continuous signal path. Previous works either incur extra power and space costs to mitigate these issues or compromise the analog CIM concept by introducing digital circuits. In this article, we borrow the concepts of episodic memory in human brains to experimentally implement a memristor‐based neuromorphic denoising process. We experimentally demonstrate a homogeneous memristor processing unit for both temporal storage and neural network computation, imitating the synapses in the human brain. Furthermore, based on previous research on functional modeling of episodic memory, we experimentally demonstrate an analog neuromorphic denoising system utilizing the proposed homogeneous memristor cores. Thanks to this neuromorphic design, compared with the latest analog computing neural network works, the proposed method improves the power efficiency by times, saves overall on‐chip analog buffer by times, and achieves area efficiency improvement.","author":[{"family":"Shi","given":"Daijing"},{"family":"Zhang","given":"Teng"},{"family":"Li","given":"Yuqi"},{"family":"Tao","given":"Yaoyu"},{"family":"Yan","given":"Bonan"},{"family":"Yang","given":"Yuchao"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1002/aisy.70419","URL":"https://doi.org/10.1002/aisy.70419","source":"crossref"},{"id":"doi:10.1038/s44335-026-00070-8","type":"article-journal","title":"Fractional-order systems for neuromorphic computing: software and hardware opportunities and challenges","abstract":"Abstract Fractional-order dynamics introduce scale-free memory and non-Markovian behavior into dynamical systems, offering a principled means of modeling long-term temporal dependencies. In this work, we extend the reservoir computing (RC) framework to include fractional-order neurons, yielding fractional-order reservoirs whose dynamics are continuously tunable through the fractional order α . We show that varying α systematically reshapes the reservoir’s information-processing regime: with memory capacity, active information storage, and information transfer each peaking at distinct values of α . This identifies α as a new control parameter of reservoir dynamics, governing a trade-off between memory retention and sensitivity to new inputs. We present a theoretical analysis of fractional-order reservoirs, provide practical software and hardware realizations, and evaluate performance across three benchmark domains: spoken-digit classification, cart-pole control, and diabetes prediction. Together, these results establish fractional-order reservoir computing as a general and effective extension of the RC paradigm.","author":[{"family":"Mastin","given":"Tucker"},{"family":"Anderson","given":"Niklas"},{"family":"Mcneal","given":"Silas"},{"family":"Tubbin","given":"Matthew"},{"family":"Teuscher","given":"Christof"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1038/s44335-026-00070-8","URL":"https://doi.org/10.1038/s44335-026-00070-8","source":"crossref"},{"id":"doi:10.1088/2634-4386/ae537f","type":"article-journal","title":"2D-materials for analog in-memory computing: a device-centric review of advantages and limitations","abstract":"Abstract This review examines the practical advantages of two-dimensional materials for energy-efficient in-memory computing by assembling a curated, experiment-only dataset covering 32 material systems across diverse device structures, mechanisms, and fabrication routes. Energy analysis was standardized using an averaged pulse-based metric, and key figures of merit—switching energy, on/off ratio, endurance, retention, and linearity—were compared against structural and mechanistic factors. Two low-energy-consumption design pathways emerge: ultrathin (&lt;10 nm) two-terminal devices exploiting filament formation for sub- μ s updates and three-terminal heterojunction devices leveraging charge trapping to achieve nA-level programming currents over longer timescales. However, dynamic on/off ratios remain modest and are often overstated by DC sweep data. Endurance improves with shorter switching times, and the most intrinsically linear conductance evolution is observed in three-terminal gate-controlled devices employing charge trapping, Schottky barrier modulation, or ion intercalation. No universal optimum exists, as enhancing one performance metric typically compromises another. Based on the comparative analysis presented in this review, three near-term levers emerge as particularly relevant for translating selective material advantages into reproducible system-level gains: standardized pulsed benchmarking, scalable chemical vapor deposition growth with controlled defects and interfaces, and device–circuit co-design.","author":[{"family":"Shim","given":"Jimin"},{"family":"Yoon","given":"Jayoung"},{"family":"Shin","given":"Sookyung"},{"family":"Hwang","given":"Eunsu"},{"family":"Lee","given":"Dokyoung"},{"family":"Kim","given":"Moon"},{"family":"Kim","given":"Sungho"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/2634-4386/ae537f","URL":"https://doi.org/10.1088/2634-4386/ae537f","source":"crossref"},{"id":"doi:10.1007/s40820-025-01756-7","type":"article-journal","title":"Multifunctional Organic Materials, Devices, and Mechanisms for Neuroscience, Neuromorphic Computing, and Bioelectronics","abstract":"Abstract Neuromorphic computing has the potential to overcome limitations of traditional silicon technology in machine learning tasks. Recent advancements in large crossbar arrays and silicon-based asynchronous spiking neural networks have led to promising neuromorphic systems. However, developing compact parallel computing technology for integrating artificial neural networks into traditional hardware remains a challenge. Organic computational materials offer affordable, biocompatible neuromorphic devices with exceptional adjustability and energy-efficient switching. Here, the review investigates the advancements made in the development of organic neuromorphic devices. This review explores resistive switching mechanisms such as interface-regulated filament growth, molecular-electronic dynamics, nanowire-confined filament growth, and vacancy-assisted ion migration, while proposing methodologies to enhance state retention and conductance adjustment. The survey examines the challenges faced in implementing low-power neuromorphic computing, e.g., reducing device size and improving switching time. The review analyses the potential of these materials in adjustable, flexible, and low-power consumption applications, viz. biohybrid spiking circuits interacting with biological systems, systems that respond to specific events, robotics, intelligent agents, neuromorphic computing, neuromorphic bioelectronics, neuroscience, and other applications, and prospects of this technology.","author":[{"family":"Hoch","given":"Felix"},{"family":"Wang","given":"Qishen"},{"family":"Lim","given":"Kian"},{"family":"Loke","given":"Desmond"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1007/s40820-025-01756-7","URL":"https://doi.org/10.1007/s40820-025-01756-7","source":"europepmc"},{"id":"doi:10.1088/2634-4386/adcbcb","type":"article-journal","title":"Edge neuro-statistical learning for event-based visual motion detection and tracking in roadside safety systems","abstract":"Abstract The Vision Zero Program’s purpose is to reduce traffic-related fatalities and serious injuries while promoting equitable, safe, and healthy mobility for all. Ultimately, the challenge is to detect pedestrians during the day and especially at night to implement safety measures. The current study introduces an award-winning low-power solution employing neuromorphic visual sensing and hybrid neuro-statistical processing developed by the Technische Hochschule Nürnberg team for the TinyML Vision Zero San Jose Competition. The solution proposes a novel neuromorphic edge fusion of spiking neural networks and event-based expectation maximization for the detection and tracking of pedestrians and bicyclists. We provide a deployment-ready evaluation of the detection performance along with robustness, energy footprint, and weatherization while emphasizing the advantages of the neuro-statistical edge solution and its city-level scaling capabilities.","author":[{"family":"Axenie","given":"Cristian"},{"family":"Halilov","given":"Ertan"},{"family":"Main","given":"Julian"},{"family":"Weiss","given":"David"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1088/2634-4386/adcbcb","URL":"https://doi.org/10.1088/2634-4386/adcbcb","source":"crossref"},{"id":"doi:10.1002/aelm.202500250","type":"article-journal","title":"Neuromorphic Computing with Memcapacitors: Advancements, Challenges, and Future Directions","abstract":"Abstract Modern applications demand immense data processing and computational power, yet conventional architectures, constrained by the Von Neumann bottleneck and data presentation, struggle to meet these requirements. This has driven the rise of neuromorphic computing, which mimics the biological nervous system through spike‐encoded data and threshold‐based computations for high energy efficiency. However, traditional hardware (CMOS transistors) designed for continuous computations fails to harness this potential fully, necessitating specialized neuromorphic hardware alternatives. Memristors have emerged as key components for neuromorphic hardware but suffer from high static power consumption, sneak‐path currents, and reliance on selector devices. In contrast, memcapacitors provide a more efficient alternative, leveraging high resistance and charge‐domain computations to overcome these limitations. This review presents a comprehensive analysis of memcapacitors for neuromorphic applications, covering capacitive switching mechanisms and materials, key hardware considerations, and recent advancements. It explores their role in artificial synapses, physical reservoir computing, and crossbar‐based accelerators, highlighting their potential for scalable and low‐power neuromorphic systems. Finally, key challenges and future research directions are discussed, particularly in materials engineering, device fabrication, and large‐scale system integration, positioning memcapacitors as promising candidates for next‐generation neuromorphic computing.","author":[{"family":"Abuhamra","given":"Nada"},{"family":"Khan","given":"Muhammad"},{"family":"Hassan","given":"Eman"},{"family":"Qutayri","given":"Mahmoud"},{"family":"Mohammad","given":"Baker"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1002/aelm.202500250","URL":"https://doi.org/10.1002/aelm.202500250","source":"crossref"},{"id":"doi:10.1088/2634-4386/ae405e","type":"article-journal","title":"Efficient transformer adaptation for analog in-memory computing via low-rank adapters","abstract":"Abstract Analog in-memory computing (AIMC) offers a promising solution to the von Neumann bottleneck. However, deploying transformer models on AIMC remains challenging due to their inherent need for flexibility and adaptability across diverse tasks. For the benefits of AIMC to be fully realized, weights of static vector-matrix multiplications must be mapped and programmed to analog devices in a weight-stationary manner. This poses two challenges for adapting a base network to hardware and downstream tasks: (i) conventional analog hardware-aware (AHWA) training requires retraining the entire model, and (ii) reprogramming analog devices is both time- and energy-intensive. To address these issues, we propose AHWA low-rank adaptation (AHWA-LoRA) training, a novel approach for efficiently adapting transformers to AIMC hardware. AHWA-LoRA training keeps the analog weights fixed as meta-weights and introduces lightweight external LoRA modules for both hardware and task adaptation. We validate AHWA-LoRA training on SQuAD v1.1 and the GLUE benchmark, demonstrate its scalability to larger models, and show its effectiveness in instruction tuning and reinforcement learning. We further evaluate a practical deployment scenario that balances AIMC tile latency with digital LoRA processing using optimized pipeline strategies, with RISC-V-based programmable multi-core accelerators. This hybrid architecture achieves efficient transformer inference with only a 4% per-layer overhead compared to a fully AIMC implementation.","author":[{"family":"Li","given":"Chen"},{"family":"Ferro","given":"Elena"},{"family":"Lammie","given":"Corey"},{"family":"Gallo","given":"Manuel"},{"family":"Boybat","given":"Irem"},{"family":"Rajendran","given":"Bipin"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/2634-4386/ae405e","URL":"https://doi.org/10.1088/2634-4386/ae405e","source":"crossref"},{"id":"doi:10.1088/2634-4386/ae44c6","type":"article-journal","title":"Reservoir computing with a heterogeneous distribution of ionic nanofluidic memristors","abstract":"Abstract Nanofluidic memristive systems exhibit the nonlinear behavior and the short-time plasticity needed for reservoir computing (RC) networks. They use ions as information carriers and operate in an electrochemical environment, in resemblance to the biological synapses. Here we present simulation results of an RC model implementation using a parallel array of memristive nanopores as reservoir. Each nanopore of the array is simulated under distinct chemical conditions using an experimentally justified theoretical model. We demonstrate the potential of the proposed network by performing three different RC tasks: sine wave nonlinear transformation, waveform classification, and forecasting of the Mackey–Glass chaotic time series.","author":[{"family":"Portillo","given":"Sergio"},{"family":"Ramirez","given":"Patricio"},{"family":"Cho","given":"Anthony"},{"family":"Siwy","given":"Zuzanna"},{"family":"Mafe","given":"Salvador"},{"family":"Cervera","given":"Javier"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/2634-4386/ae44c6","URL":"https://doi.org/10.1088/2634-4386/ae44c6","source":"crossref"},{"id":"doi:10.3389/fnins.2025.1511371","type":"article-journal","title":"Evaluation of fluxon synapse device based on superconducting loops for energy efficient neuromorphic computing","abstract":"With Moore’s law nearing its end due to the physical scaling limitations of CMOS technology, alternative computing approaches have gained considerable attention as ways to improve computing performance. Here, we evaluate performance prospects of a new approach based on disordered superconducting loops with Josephson-junctions for energy efficient neuromorphic computing. Synaptic weights can be stored as internal trapped fluxon states of three superconducting loops connected with multiple Josephson-junctions (JJ) and modulated by input signals applied in the form of discrete fluxons (quantized flux) in a controlled manner. The stable trapped fluxon state directs the incoming flux through different pathways with the flow statistics representing different synaptic weights. We explore implementation of matrix–vector-multiplication (MVM) operations using arrays of these fluxon synapse devices. We investigate the energy efficiency of online-learning of MNIST dataset. Our results suggest that the fluxon synapse array can provide ~100× reduction in energy consumption compared to other state-of-the-art synaptic devices. This work presents a proof-of-concept that will pave the way for development of high-speed and highly energy efficient neuromorphic computing systems based on superconducting materials.","author":[{"family":"Kumar","given":"Ashwani"},{"family":"Goteti","given":"Uday"},{"family":"Cubukcu","given":"Ertugrul"},{"family":"Dynes","given":"Robert"},{"family":"Kuzum","given":"Duygu"}],"issued":{"date-parts":[[2025]]},"DOI":"10.3389/fnins.2025.1511371","URL":"https://doi.org/10.3389/fnins.2025.1511371","source":"europepmc"},{"id":"doi:10.4018/979-8-3373-7779-7.ch002","type":"article-journal","title":"Quantum Machine Intelligence","abstract":"Quantum computing has turned out to be a paradigm shift that can address computational problems that are way out of the scope of classical systems. The current hybrid designs tend to view quantum and classical components as non-living, and do not provide the feedback intelligence to control energy, depth and computational balance in real-time. The present paper presents a Quantum Machine Intelligence (QMI) model, which will integrate quantum computation, artificial intelligence as well as reinforcement learning into a self-regulated system. The model is developed on quantum encoding and amplitude-based encoding, a reinforcement-learning controller using hybrid optimization variations, and design to adjust simultaneously the depth of the circuits and the number of qubits that are used to evaluate the performance of a component Empirical analyses show that QMI attains 98.3% accuracy, 17.8% energy saving, and 15 times quick convergence in opposition to the latest DRL, hybrid DL, and LLM-aided baselines.","author":[{"family":"Palit","given":"Shamik"},{"family":"Madanan","given":"Pawan"},{"family":"Srivastava","given":"Shipra"},{"family":"Patil","given":"Ganesh"},{"family":"Tiwari","given":"Mohit"},{"family":"Lourens","given":"Melanie"}],"issued":{"date-parts":[[2026]]},"DOI":"10.4018/979-8-3373-7779-7.ch002","URL":"https://doi.org/10.4018/979-8-3373-7779-7.ch002","source":"crossref"},{"id":"doi:10.1038/s44335-025-00021-9","type":"article-journal","title":"A self-training spiking superconducting neuromorphic architecture","abstract":"Abstract Neuromorphic computing takes biological inspiration to the device level aiming to improve computational efficiency and capabilities. One of the major issues that arises is the training of neuromorphic hardware systems. Typically training algorithms require global information and are thus inefficient to implement directly in hardware. In this paper we describe a set of reinforcement learning based, local weight update rules and their implementation in superconducting hardware. Using SPICE circuit simulations, we implement a small-scale neural network with a learning time of order one nanosecond per update. This network can be trained to learn new functions simply by changing the target output for a given set of inputs, without the need for any external adjustments to the network. Further, this architecture does not require programing explicit weight values in the network, alleviating a critical challenge with analog hardware implementations of neural networks.","author":[{"family":"Schneider","given":"ML"},{"family":"Jué","given":"EM"},{"family":"Pufall","given":"MR"},{"family":"Segall","given":"K"},{"family":"Anderson","given":"CW"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1038/s44335-025-00021-9","URL":"https://doi.org/10.1038/s44335-025-00021-9","source":"crossref"},{"id":"doi:10.1088/2634-4386/ae46d4","type":"article-journal","title":"A scalable hybrid training approach for recurrent spiking neural networks","abstract":"Abstract Recurrent spiking neural networks (RSNNs) can be implemented very efficiently in neuromorphic systems. Nevertheless, training of these models with powerful gradient-based learning algorithms is mostly performed on standard digital hardware using backpropagation through time (BPTT). However, BPTT has substantial limitations. It does not permit online training and its memory consumption scales linearly with the number of computation steps. In contrast, learning methods using forward propagation of gradients operate in an online manner with a memory consumption independent of the number of time steps. These methods enable SNNs to learn from continuous, infinite-length input sequences. In addition, approximate forward propagation algorithms have been developed that can be implemented on neuromorphic hardware. Yet, slow execution speed on conventional hardware as well as inferior performance has hindered their widespread application. In this work, we introduce hybrid propagation (HYPR) that combines the efficiency of parallelization with approximate online forward learning. Our algorithm yields high-throughput online learning through parallelization, paired with constant, i.e. sequence length independent, memory demands. HYPR enables parallelization of parameter update computation over subsequences for RSNNs consisting of almost arbitrary non-linear spiking neuron models. We apply HYPR to networks of spiking neurons with oscillatory subthreshold dynamics. We find that this type of neuron model is particularly well trainable by HYPR, resulting in an unprecedentedly low task performance gap between approximate forward gradient learning and BPTT.","author":[{"family":"Baronig","given":"Maximilian"},{"family":"Bahariasl","given":"Yeganeh"},{"family":"Özdenizci","given":"Ozan"},{"family":"Legenstein","given":"Robert"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/2634-4386/ae46d4","URL":"https://doi.org/10.1088/2634-4386/ae46d4","source":"crossref"},{"id":"doi:10.1088/2634-4386/ae46d5","type":"article-journal","title":"Beyond rate coding: surrogate gradients enable spike timing learning in spiking neural networks","abstract":"Abstract The surrogate gradient descent (SGD) algorithm enabled spiking neural networks (SNNs) to be trained to carry out challenging sensory processing tasks, an important step in understanding how spikes contribute to neural computations. However, it is unclear the extent to which these algorithms fully explore the space of possible spiking solutions to problems. We investigated whether spiking networks trained with SGD can learn to make use of information that is only encoded in the timing and not the rate of spikes. We constructed synthetic datasets with a range of types of spike timing information (interspike intervals, spatio-temporal spike patterns or polychrony, and coincidence codes). We find that SGD training can extract all of these types of information. In more realistic speech-based datasets, both timing and rate information is present. We therefore constructed variants of these datasets in which all rate information is removed, and find that SGD can still perform well. We tested all networks both with and without trainable axonal delays. We find that delays can give a significant increase in performance, particularly for more challenging tasks. To determine what types of spike timing information are being used by the networks trained on the speech-based tasks, we test these networks on time-reversed spikes which perturb spatio-temporal spike patterns but leave interspike intervals and coincidence information unchanged. We find that when axonal delays are not used, networks perform well under time reversal, whereas networks trained with delays perform poorly. This suggests that SNNs with delays are better able to exploit temporal structure. To facilitate further studies of temporal coding, we have released our modified speech-based datasets.","author":[{"family":"Yu","given":"Ziqiao"},{"family":"Sun","given":"Pengfei"},{"family":"Goodman","given":"Dan"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/2634-4386/ae46d5","URL":"https://doi.org/10.1088/2634-4386/ae46d5","source":"crossref"},{"id":"doi:10.1002/advs.77375","type":"article-journal","title":"A Reconfigurable Memristive Spiking Neuron Enabling Advanced Neuromorphic Computing.","abstract":"ABSTRACT Biological neurons exhibit internal complexity that enables a variety of spiking behaviors, complex encoding strategies, and the possibility of advanced networks that underpin cognitive brain functions. Neuromorphic computing, especially using memristors with biologically plausible dynamics, is imperative to realizing next‐generation artificial intelligence. However, current attempts still rely on simple neurons with limited support for novel network algorithms. This work proposes a multi‐mode reconfigurable memristive spiking neuron with a simple transistor‐capacitor feedback circuit that enables fast‐spiking, adaptive spiking, phasic bursting, or single‐spiking. Every spike train parameter in each mode can be freely tuned. With this, a multiplexed encoding strategy in a hazard avoidance application is demonstrated, wherein 3 driving actions and 6 maneuvering control variables are encoded using distinct spiking modes and spike train parameters. Furthermore, long‐short‐term memory spiking neural networks (LSNN) that incorporate either heterogeneous slow time constants or dynamic mode switching are proposed, demonstrating up to 8.0% and 13.7% improvements in accuracy in highly temporal tasks compared with homogeneous LSNNs, while also exhibiting superior generalization capabilities. The ability to support these advanced encoding and network strategies distinguishes the proposed reconfigurable neuron from existing implementations, highlighting its potential to empower more advanced neuromorphic computing.","author":[{"family":"Tiw","given":"Pek"},{"family":"Li","given":"Yuqi"},{"family":"Li","given":"Zhongyuan"},{"family":"Ding","given":"Qihang"},{"family":"Wang","given":"Yuzhe"},{"family":"Wang","given":"Jiarong"},{"family":"Yang","given":"Yuchao"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1002/advs.77375","URL":"https://doi.org/10.1002/advs.77375","source":"europepmc"},{"id":"doi:10.1063/5.0320577","type":"article-journal","title":"Quantum coherence in neuromorphic computing.","abstract":"Quantum effects become significant when hardware computing units scale down to nanoscale dimensions. To maintain reliable performance as neuromorphic computing hardware scales down, researchers must understand how quantum coherence across multiple neurons impacts neural network function. In this study, we model neuromorphic computing with quantum coherence effects using a quantum spiking neural network model. We find that quantum coherence between neural activations can alter the network perception, compared to the incoherent network. Destructive interference between activation signals propagating through different synaptic channels drives this effect at the quantum scale. This quantum effect becomes more prominent with increasing network depth and can be mitigated by increasing the number of input neurons connected to each output neuron.","author":[{"family":"Wang","given":"Yuanheng"},{"family":"Li","given":"Kai"},{"family":"Scholes","given":"Gregory"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1063/5.0320577","URL":"https://doi.org/10.1063/5.0320577","source":"europepmc"},{"id":"doi:10.1364/oe.595963","type":"article-journal","title":"Cavity solitons as a nonlinear substrate for photonic neuromorphic computing.","abstract":"Reservoir computing leverages the nonlinear dynamics of physical systems to process temporal information with minimal training cost. Here, we demonstrate that cavity solitons sustained in a fiber optical cavity provide an optical platform for photonic reservoir computing. Our methodology employs a phase-modulated drive laser to encode the input, while the reservoir states are accessed through a frequency-resolved readout. Numerical simulations indicate that the emission of Kelly waves enriches the dynamics and enhances performance for machine learning tasks. We evaluated the performance of the cavity-soliton reservoir computer on several standard benchmark tasks.","author":[{"family":"Arabieh","given":"Amir"},{"family":"Lupo","given":"Alessandro"},{"family":"Gorza","given":"Simon"},{"family":"Massar","given":"Serge"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1364/oe.595963","URL":"https://doi.org/10.1364/oe.595963","source":"europepmc"},{"id":"doi:10.1002/advs.75989","type":"article-journal","title":"Reconfigurable Selector-Only Memory (SOM) for Scalable Neuromorphic Computing.","abstract":"ABSTRACT Highly scalable reconfigurable neuromorphic devices are critical for addressing continual‐learning challenges in artificial intelligence. However, the scalability of existing reconfigurable devices is severely constrained by limited operating margins and insufficient process maturity. Here, we propose selector‐only memory (SOM) as a scalable device candidate. Its volatile threshold switching and programmable nonvolatile threshold window are operationally decoupled, and it is compatible with in‐line fabrication and 3D stacking. We demonstrate an In‐doped GeSe SOM that enables neuron–synapse reconfigurability within a single cell. By leveraging intrinsic parasitic capacitance, we implement a capacitor‐free leaky integrate‐and‐fire neuron and validate all‐or‐none firing, integrate‐and‐fire dynamics, and input‐controlled firing‐rate modulation using experiments and an equivalent model. For synapses, we propose a one‐shot subthreshold‐conductance readout method. With a unified reverse‐subthreshold pulse scheme, 16 programmed conductance states are obtained through one‐shot subthreshold readout, and most states remain distinguishable over 10 4 s. Finally, SOM‐parameter‐based simulations on a Growing‐When‐Required MNIST task achieve 2.67× higher accuracy with 70% of the nodes and shrink to 58% after rollback. These results indicate that SOM provides a promising selector‐derived device concept for scalable reconfigurable neuromorphic hardware.","author":[{"family":"Wen","given":"Jin‐yu"},{"family":"Yi","given":"Chuan‐qi"},{"family":"Zhang","given":"Ya‐ru"},{"family":"Wang","given":"Bin‐hao"},{"family":"Liu","given":"Zi‐xuan"},{"family":"Zhou","given":"Chun‐yu"},{"family":"Tong","given":"Hao"},{"family":"Miao","given":"Xiang‐shui"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1002/advs.75989","URL":"https://doi.org/10.1002/advs.75989","source":"europepmc"},{"id":"doi:10.1088/1361-6528/ae7381","type":"article-journal","title":"Low Power HfOx/TaOx stacked memristors with nanocolumn electrode for neuromorphic computing.","abstract":"Abstract Emulating biological synaptic behavior using the resistive random access memory (RRAM) is promising for neuromorphic applications. A stacked HfOx/TaOx RRAM model with nanocolumn electrode for low-power neuromorphic computing was constructed, and the finite element method was used to simulate the reset and set processes. As the local electric field was enhanced by the nanocolumn electrode structure, the superior conductive filament control and lower reset/set voltages can be achieved. Meanwhile, the distributions of oxygen vacancy concentration and temperature during switching processes indicate that the nanocolumn electrode significantly reduces the number of programming pulses required for conductance modulation and lower the power consumption of the array. Meantime, the device also exhibits better conductance linearity (long-term potentiation/long-term depression), which is beneficial for improving the accuracy of neural networks. Then the system-level validation was conducted by integrating the device characteristics into a crossbar array and training with the MNIST dataset using backpropagation, achieving 89.83% recognition accuracy, which is superior to that of plane electrode devices (88.67%). This work demonstrates the potential of nanocolumn electrode induced memristors in realizing efficient neuromorphic computing systems, from device physics to system simulation.","author":[{"family":"Yang","given":"Fei"},{"family":"Zhao","given":"Xuanyang"},{"family":"Liu","given":"Junlong"},{"family":"Zhu","given":"Houwei"},{"family":"Shu","given":"Qingsong"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/1361-6528/ae7381","URL":"https://doi.org/10.1088/1361-6528/ae7381","source":"europepmc"},{"id":"doi:10.20944/preprints202606.1857.v1","type":"manuscript","title":"Neuromorphic Computing: Foundations and the Case for Principle-Level Integration in AI Systems <em>Part I of IV</em>","abstract":"Synchronous with technological progress in Artificial Intelligence (AI), demands on underlying computing architectures are increasing rapidly. Neuromorphic computing (NMC) offers novel approaches to energy-efficient processing of time-dependent information and enables low-latency, sensor-proximal interaction between environment, humans, and machines. Yet its relevance for industry and AI development extends beyond dedicated neuromorphic platforms: core neuromorphic principles, including event-driven processing, temporal coding, and co-located memory and computation, are increasingly being absorbed into mainstream AI hardware and software architectures. Rather than emerging as a unified standalone paradigm in the near term, NMC appears poised to reshape AI systems through gradual, principle-level integration and hybridization with classical approaches. This essay, the first in a four-part series, introduces the technical foundations of NMC and develops the above argument through a structured comparison of biological information processing, classical artificial neural networks (ANNs), and spiking neural networks (SNNs). Using visual information processing as a concrete illustrative example, we highlight the distinct operating principles of each paradigm and begin to contextualize their respective limitations and potentials. We discuss the hypothesis that the broader impact of NMC will unfold through hybridization rather than replacement and consider implications for industrial actors and technology transfer. The series aims to make the potential of NMC accessible and tangible for product development and process innovation. It is offered as a contribution to an ongoing cross-disciplinary dialogue, with particular attention to the European research and innovation ecosystem and the strategic opportunity it represents for technology sovereignty and industrial competitiveness.","author":[{"family":"Rotermund","given":"Natalie"},{"family":"Krtil","given":"Alois"},{"family":"Mertes","given":"Jakob"}],"issued":{"date-parts":[[2026]]},"DOI":"10.20944/preprints202606.1857.v1","URL":"https://doi.org/10.20944/preprints202606.1857.v1","source":"europepmc"},{"id":"doi:10.3390/mi17020216","type":"article-journal","title":"Stochastic Neuromorphic Computing Architecture Based on Voltage-Controlled Probabilistic Switching Magnetic Tunnel Junction (MTJ) Devices.","abstract":"As integrated circuits face increasingly stringent demands regarding power consumption, area, and stability, integrating novel spintronic devices with computing architectures has become a crucial direction for breaking through traditional computing paradigms. In the paper, switching mechanism of Magnetic Tunnel Junctions (MTJs) under the synergistic effect of Voltage-Controlled Magnetic Anisotropy (VCMA) and the Spin Hall Effect (SHE) is investigated. VCMA-assisted switching SHE-MTJ device is adopted, and a macrospin approximation model is established based on the Landau-Lifshitz-Gilbert (LLG) equation to systematically analyze its dynamic characteristics. The research demonstrates that applying VCMA voltage pulses with appropriate amplitude and width can significantly reduce the required spin Hall current density and pulse width for switching, thereby effectively minimizing ohmic losses and Joule heating. Furthermore, by incorporating a thermal fluctuation field, voltage-controlled SHE-MTJ device with stochastic switching behavior can be constructed, obtaining an approximately sigmoidal voltage-probability response curve. This provides an ideal physical foundation for stochastic computing and neuromorphic computing. Based on the above established fundamental discovery, an in-memory computing architecture supporting binarized Convolutional Neural Networks (CNNs) is proposed and designed in the paper. Combined with the lightweight network SqueezeNet, this architecture achieves a Top-1 recognition accuracy of 72.49% on the CIFAR-10 dataset, with a parameter count of only 1.25 × 106. This work offers a feasible spintronic implementation scheme for low-power, high-energy-efficiency edge-side intelligent chips.","author":[{"family":"Gao","given":"Liang"},{"family":"Wang","given":"Chenxi"},{"family":"Jiang","given":"Yanfeng"}],"issued":{"date-parts":[[2026]]},"DOI":"10.3390/mi17020216","URL":"https://doi.org/10.3390/mi17020216","source":"europepmc"},{"id":"doi:10.1002/advs.75318","type":"article-journal","title":"WS&lt;sub&gt;2&lt;/sub&gt; Optoelectronic Memristive Reservoir Enabling Ultra-Low-Power, Multi-Task, and Environmentally Stable Neuromorphic Computing.","abstract":"ABSTRACT Energy‐efficient visual and speech processing is essential for edge intelligence, yet conventional silicon‐based chips suffer from high power consumption. Here, we report a WS 2 /Zinc–Tin–Oxide (ZTO)‐based optoelectronic reservoir computing (RC) system that uniquely integrates sensing, memory, and computation within a single compact device to emulate diverse biological functions. The WS 2 /ZTO memristive RC achieves strong performance, with ∼94% accuracy on N‐MNIST, ∼93% in motion perception, and ∼89% in speech recognition within only 30 training epochs, while consuming ultra‐low energy of ∼25.5 fJ/spike. Raw inputs are converted into spike trains to preserve temporal dynamics: motion data from inter‐frame differences, FSDD waveforms reshaped into spike‐like signals, and N‐MNIST reconstructed directly from the address‐event representation format. The system maintains reliable operation under 95% relative humidity, highlighting excellent environmental stability. Distinctively, the WS 2 /ZTO memristor serves simultaneously as sensor and hardware reservoir, exploiting volatile and nonlinear dynamics for direct temporal input decoding. Validation on N‐MNIST further shows 95% accuracy with minimal training energy. In addition, the device demonstrates endurance over 1.5 million cycles and supports synaptic features including excitatory postsynaptic current, short‐term and long‐term plasticity, and photonic paired‐pulse facilitation. This work establishes a humidity‐resilient, ultra‐low‐power WS 2 /ZTO in‐sensor RC platform, advancing neuromorphic processing for next‐generation edge technologies.","author":[{"family":"Kumar","given":"Dayanand"},{"family":"Li","given":"Hanrui"},{"family":"Divyanshu","given":"Divyanshu"},{"family":"Kumbhar","given":"Dhananjay"},{"family":"Rajbhar","given":"Manoj"},{"family":"Singh","given":"Amit"},{"family":"Syed","given":"Abdul"},{"family":"Amara","given":"Selma"},{"family":"Setti","given":"Gianluca"},{"family":"Elatab","given":"Nazek"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1002/advs.75318","URL":"https://doi.org/10.1002/advs.75318","source":"europepmc"},{"id":"doi:10.2478/joeb-2025-0019","type":"article-journal","title":"Using neuromorphic computing in prediction of GABA concentration - a pilot study.","abstract":"Abstract Neuromorphic computing has the potential to facilitate detection of GABA concentration levels in the brain, and offers energy-efficient, real-time machine learning processing possibilities. To study whether neuromorphic computing can be used for GABA concentration detection, dielectric relaxation spectroscopy was used to acquire permittivity data of different concentrations of GABA solution. Thereafter, two different machine learning models were compared (Feedforward neural network (FFNN) and convolutional neural network (CNN)) for accuracy in prediction of GABA concentration from dielectric properties. The CNN model was then converted to spiking Neural Networks (SNNs), which showed promising results for energy efficiency and real-time processing capabilities. The system incorporates Tkinter, a Python interface to the Tcl/Tk GUI toolkit for seamless data transfer between the neuromorphic chip and the measurement system, ensuring flexibility and scalability in a user-friendly system.","author":[{"family":"Hou","given":"Jie"},{"family":"Ali","given":"Abdulkadir"},{"family":"Martinsen","given":"Ørjan"}],"issued":{"date-parts":[[2025]]},"DOI":"10.2478/joeb-2025-0019","URL":"https://doi.org/10.2478/joeb-2025-0019","source":"europepmc"},{"id":"doi:10.1002/asia.202401170","type":"article-journal","title":"Ionic Device: From Neuromorphic Computing to Interfacing with the Brain.","abstract":"Abstract In living organisms, the modulation of ion conductivity in ion channels of neuron cells enables intelligent behaviors, such as generating, transmitting, and storing neural signals. Drawing inspiration from these natural processes, researchers have fabricated ionic devices that replicate the functions of the nervous system. However, this field remains in its infancy, necessitating extensive foundational research in ionic device preparation, algorithm development, and biological interaction. This review summarizes recently developed neuromorphic ionic devices into three categories based on the materials states: liquid, semi‐solid, and solid. The neural network algorithms embedded in these devices for neuromorphic computing are introduced, and future directions for the development of bidirectional human‐computer interaction and hybrid human‐computer intelligence are discussed.","author":[{"family":"Huang","given":"Zijia"},{"family":"Mei","given":"Tingting"},{"family":"Zhu","given":"Xinyi"},{"family":"Xiao","given":"Kai"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1002/asia.202401170","URL":"https://doi.org/10.1002/asia.202401170","source":"europepmc"},{"id":"doi:10.1002/smtd.202500445","type":"article-journal","title":"CMOS-Compatible Protonic Three-Terminal Memristor for Analog Synapse in Neuromorphic Computing.","abstract":"Abstract All‐solid‐state inorganic hydrogen‐based three‐terminal memristors (H‐3TMs) suffer from poor retention, susceptibility to humidity and temperature, and the reliance on wet chemistry during fabrication, hindering their manufacturability within existing foundry processes. To address these, this study presents a CMOS‐compatible H‐3TM based on reversible intercalation and extraction of protons between the SiN x electrolyte and WO x channel. The protons are introduced via a straightforward hydrogen plasma treatment, promoting a compatible fabrication process with back‐end‐of‐line integration. Experimental and simulation results indicate that the low proton transport tendency across the electrolyte/channel interface without an external electric field contributes to high retention performance. Furthermore, the device demonstrates linear potentiation and depression, 512 conductance states with a dynamic range of ≈40, low energy operation (≈73 fJ per write), and excellent overall device‐to‐device variation. Its analog properties are evaluated under the training and inference framework of MNIST and Fashion‐MNIST datasets. The device achieved training and inference accuracies only 0.4% and 0.3% below the ideal benchmark on the F‐MNIST dataset. This work offers a rational approach for future artificial synaptic device design and fabrication.","author":[{"family":"Liu","given":"Lingli"},{"family":"Dananjaya","given":"Putu"},{"family":"Koh","given":"Eng"},{"family":"Tan","given":"Funan"},{"family":"Chen","given":"Ze"},{"family":"Lim","given":"Gerard"},{"family":"Lee","given":"Calvin"},{"family":"Yang","given":"Jin‐lin"},{"family":"Lew","given":"Wen"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1002/smtd.202500445","URL":"https://doi.org/10.1002/smtd.202500445","source":"europepmc"},{"id":"doi:10.1002/advs.202500568","type":"article-journal","title":"High-Performance Synapse Arrays for Neuromorphic Computing via Floating Gate-Engineered IGZO Synaptic Transistors.","abstract":"Abstract Neuromorphic computing emulating the human brain offers a promising alternative to the Von Neumann architecture. Developing artificial synapses is essential for implementing hardware neuromorphic systems. Indium‐gallium‐zinc oxide (IGZO)‐based synaptic transistors using charge trapping have advantages, such as low‐temperature process and complementary metal‐oxide‐semiconductor compatibility. However, these devices face challenges of low charge de‐trapping efficiency and insufficient retention. Here, IGZO synaptic transistors are introduced utilizing an indium‐tin oxide (ITO) floating gate (FG) to overcome these limitations. The ITO FG's higher conductivity and alleviated chemical interactions with the Al 2 O 3 tunneling layer (TL) deposited by atomic layer deposition result in enhanced electrical performance with a smooth FG/TL interface. An 8 × 8 synapse array achieves 100% yield and successful programming without interference using a half‐pulse scheme. Spiking neural network simulations on MNIST and Fashion‐MNIST datasets demonstrate high accuracies of 98.31% and 87.76%, respectively, despite considering device variations and retention. These findings highlight the potential of IGZO synaptic transistors for neuromorphic computing applications.","author":[{"family":"Park","given":"Junhyeong"},{"family":"Yun","given":"Yumin"},{"family":"Bae","given":"Sunyeol"},{"family":"Jang","given":"Yuseong"},{"family":"Shin","given":"Seungyoon"},{"family":"Lee","given":"Soo‐yeon"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1002/advs.202500568","URL":"https://doi.org/10.1002/advs.202500568","source":"europepmc"},{"id":"doi:10.6082/uchicago.13578","type":"article-journal","title":"Biomimetic and AI-Guided Designs for Redox-Active Semiconducting Polymers","abstract":"Bioelectronics serves as an indispensable technology to directly interface with biological tissues for uncovering biological mechanism and applying health diagnosis and therapeutic interventions to the human body. Despite the considerable successes achieved on devices utilizing inorganic electronic materials, their high rigidity and limited stretchability lead to essential discomfort and even potential damage to the tissue, with their low biocompatibility posing significant obstacles for dealing with the foreign body response upon long-term implantation. On the contrary, conjugated polymer-based conductors/semiconductors show great promise for bio-integration due to their much higher ductility and chemical structures that closely resemble biomolecules. Among these, the redox-active semiconducting polymers (RASPs) possess the emerging class of conjugated polymer for their mixed ionic-electronic conduction capabilities, which enable highly efficient signal transduction and amplification by adopting the organic electrochemical transistor (OECT) configuration. To realize such processes at the tissue-electronic interface, however, relies on RASPs with biomimetic mechanical properties, and optimized mixed conduction properties. Unfortunately, these requirements are rarely met within the current category of RASPs. In my Ph.D. research, I focused on the development of RASPs with two major objectives, i.e., imparting tissue-like stretchability and softness on redox-active semiconducting polymers, and achieving high-efficiency auto-screening of high performance RASP films via a self-driving lab (SDL) for accelerating the material exploration and knowledge establishment process. First, I developed two highly stretchable RASPs, i.e., poly(2-(3,3′-bis(2-(2-(2-methoxyethoxy)ethoxy)ethoxy)-[2,2′-bithiophen]-5)yl thiophene) (p(g2T-T)) and poly-[3,3′-bis(2-(2-(2-methoxyethoxy)ethoxy)ethoxy)-2,2′-bithiophene] (p(gT2)), that maintained consistent electrical performance even under 200% strain due to their ability to elongate and dissipate energy. Leveraging these materials, I fabricated intrinsically stretchable OECTs that demonstrated stable on-body electrophysiological signal recording and neuromorphic computing. Second, to address the significant modulus mismatch between RASPs and tissues, I proposed a simple and versatile solvent-exchange strategy for constructing RASP-embedded hydrogel composite, named hydrogel semiconductor (hydro-SC). Notably, the hydro-SC achieves the tissue-level modulus of 81 kPa, which is 3 orders of magnitude lower than the pristine RASP film, while maintaining the high charge-carrier mobility (0.61 cm2 V-1 s-1). In addition, the hydro-SC design also greatly enhances the biocompatibility of RASP for substantially alleviated immune response, and improves semiconductor porosity for more efficient photomodulation and maximized bioreceptor-analyte interaction. Third, to accelerate the development of high-performance RASP films that is mainly hindered by the sophisticated structure-property relationship, I developed a fully automated SDL together with an innovative human-machine collaboration mechanism. This optimized workflow resulted in a 150% improvement in performance compared to spin-coated films and led to the first discovery of a thermodynamically unfavorable polymorph in RASP films, which show strong correlations with volumetric capacitance (C*). All those outcomes were achieved with less than 70 experimental trials and only 10 samples for in-depth characterizations. In the final chapter, I provide summaries and future perspectives on the development of RASPs and OECT sensors for human integration, as well as the improvement of SDL towards the next generation.","author":[{"family":"Dai","given":"Yahao"}],"issued":{"date-parts":[[2024]]},"DOI":"10.6082/uchicago.13578","URL":"https://doi.org/10.6082/uchicago.13578","source":"datacite"},{"id":"doi:10.6082/rr5b2-5m574","type":"article-journal","title":"Biomimetic and AI-Guided Designs for Redox-Active Semiconducting Polymers","abstract":"Bioelectronics serves as an indispensable technology to directly interface with biological tissues for uncovering biological mechanism and applying health diagnosis and therapeutic interventions to the human body. Despite the considerable successes achieved on devices utilizing inorganic electronic materials, their high rigidity and limited stretchability lead to essential discomfort and even potential damage to the tissue, with their low biocompatibility posing significant obstacles for dealing with the foreign body response upon long-term implantation. On the contrary, conjugated polymer-based conductors/semiconductors show great promise for bio-integration due to their much higher ductility and chemical structures that closely resemble biomolecules. Among these, the redox-active semiconducting polymers (RASPs) possess the emerging class of conjugated polymer for their mixed ionic-electronic conduction capabilities, which enable highly efficient signal transduction and amplification by adopting the organic electrochemical transistor (OECT) configuration. To realize such processes at the tissue-electronic interface, however, relies on RASPs with biomimetic mechanical properties, and optimized mixed conduction properties. Unfortunately, these requirements are rarely met within the current category of RASPs. In my Ph.D. research, I focused on the development of RASPs with two major objectives, i.e., imparting tissue-like stretchability and softness on redox-active semiconducting polymers, and achieving high-efficiency auto-screening of high performance RASP films via a self-driving lab (SDL) for accelerating the material exploration and knowledge establishment process. First, I developed two highly stretchable RASPs, i.e., poly(2-(3,3′-bis(2-(2-(2-methoxyethoxy)ethoxy)ethoxy)-[2,2′-bithiophen]-5)yl thiophene) (p(g2T-T)) and poly-[3,3′-bis(2-(2-(2-methoxyethoxy)ethoxy)ethoxy)-2,2′-bithiophene] (p(gT2)), that maintained consistent electrical performance even under 200% strain due to their ability to elongate and dissipate energy. Leveraging these materials, I fabricated intrinsically stretchable OECTs that demonstrated stable on-body electrophysiological signal recording and neuromorphic computing. Second, to address the significant modulus mismatch between RASPs and tissues, I proposed a simple and versatile solvent-exchange strategy for constructing RASP-embedded hydrogel composite, named hydrogel semiconductor (hydro-SC). Notably, the hydro-SC achieves the tissue-level modulus of 81 kPa, which is 3 orders of magnitude lower than the pristine RASP film, while maintaining the high charge-carrier mobility (0.61 cm2 V-1 s-1). In addition, the hydro-SC design also greatly enhances the biocompatibility of RASP for substantially alleviated immune response, and improves semiconductor porosity for more efficient photomodulation and maximized bioreceptor-analyte interaction. Third, to accelerate the development of high-performance RASP films that is mainly hindered by the sophisticated structure-property relationship, I developed a fully automated SDL together with an innovative human-machine collaboration mechanism. This optimized workflow resulted in a 150% improvement in performance compared to spin-coated films and led to the first discovery of a thermodynamically unfavorable polymorph in RASP films, which show strong correlations with volumetric capacitance (C*). All those outcomes were achieved with less than 70 experimental trials and only 10 samples for in-depth characterizations. In the final chapter, I provide summaries and future perspectives on the development of RASPs and OECT sensors for human integration, as well as the improvement of SDL towards the next generation.","author":[{"family":"Dai","given":"Yahao"}],"issued":{"date-parts":[[2024]]},"DOI":"10.6082/rr5b2-5m574","URL":"https://doi.org/10.6082/rr5b2-5m574","source":"datacite"},{"id":"doi:10.21203/rs.3.rs-9782781/v1","type":"article-journal","title":"Spintronic Neuromorphic Hardware Using Domain Wall Based Neurons and Quantized Synapses","abstract":"Abstract In this work, we simulate the functionality of artificial neuron and synapse using spin-orbit torque-based spintronic devices and implemented a fully connected artificial neural netwrok (ANN). These neuro-synaptic devices are emulated using transverse domain wall dynamics in a rectangular magnetic nanotrack comprised of heavy metal/ferromagnet (HM/FM) heterostructures. The ReLU activation function of the neuron has been mimicked using the domain wall motion induced by a 3 ns current pulse. The synapse has been modelled using current-induced domain wall (DW) dynamics through a corrugated HM/FM nanotrack under the influence of a 10 ns current pulse with varying current density. The semicircular corrugations are in the form of notches, which are symmetrically located on both sides of the nanotrack. By applying 10 ns current pulses of varying densities, we achieve controlled DW pinning, revealing a step-like motion caused by temporary pauses at each pinning center. The electrical conductance of the pinned DW across various pinning points, act as stable synaptic weights for our ANN. Furthermore, we observe a threshold-dependent delay effect where each depinning event is influenced by previous ones, successfully mimicking synaptic memory and adaptability in neuromorphic systems. The fully connected ANN has been modeled using the conventional float32 synaptic weights for the MNIST and Fashion MNIST datasets with an accuracy of ∼ 97 % and ∼ 86 % respectively, which serves as a test bed of our neuromorphic simulations. With the aim of implementing a sparse and low memory footprint ANN, we quantize the trained synaptic weights into discrete quantized level and tested the network, which demonstrate an accuracy of ∼95% and ∼62%, for the MNIST and Fashion-MNIST dataset, respectively. Although direct quantization impacted performance, fine-tuning the network fully restored accuracies to near-baseline levels. These findings highlight the potential of engineered DW pinning-depinning dynamics for scalable, adaptive, and hardware-efficient neuromorphic computing.","author":[{"family":"Bandekar","given":"Sakshi"},{"family":"Ganguly","given":"Arnab"},{"family":"Polley","given":"Debanjan"},{"family":"Das","given":"Debasis"}],"issued":{"date-parts":[[2026]]},"DOI":"10.21203/rs.3.rs-9782781/v1","URL":"https://doi.org/10.21203/rs.3.rs-9782781/v1","source":"europepmc"},{"id":"doi:10.21203/rs.3.rs-6505999/v1","type":"article-journal","title":"Data-In-situ Computing with One-Pixel-Multiple-Memristor Architecture for Neuromorphic Sequential Vision","abstract":"Abstract Neuromorphic vision systems based on memristors offer an energy-efficient approach to artificial vision, yet traditional pixel(s)-to-one-memristor architectures remain inefficient in dynamic image processing due to limited temporary storage. Here, inspired by human visual working memory, we propose a novel one-pixel-multiple-memristor (1PnR) architecture with a rolling exposure strategy for fast sequential image acquisition. Furthermore, a data-in-situ computing network for efficient image processing is developed. With network weights mapped to voltage vectors and applied to the image storage memristor array, direct computation is enabled where the image is stored, and the energy-intensive data transmission is eliminated. A hardware prototype of the 1PnR architecture achieved 95.7% recognition accuracy on the Weizmann human action flow dataset. Compared to CMOS-based systems, this architecture is estimated to have a 2000× reduction in latency for image sensing and storage, and a 160× reduction in energy consumption image processing, demonstrating significant potential for future neuromorphic visual systems.","author":[{"family":"Wang","given":"Wei"},{"family":"Sun","given":"Yi"},{"family":"Tong","given":"Peiwen"},{"family":"Shen","given":"Jiangrong"},{"family":"Xu","given":"Hui"},{"family":"Cao","given":"Rongrong"},{"family":"Liu","given":"Chang"},{"family":"Chen","given":"Changlin"},{"family":"Song","given":"Bing"},{"family":"Wang","given":"Yinan"},{"family":"Yang","given":"Yuchao"},{"family":"Li","given":"Qingjiang"}],"issued":{"date-parts":[[2026]]},"DOI":"10.21203/rs.3.rs-6505999/v1","URL":"https://doi.org/10.21203/rs.3.rs-6505999/v1","source":"preprints"},{"id":"doi:10.21203/rs.3.rs-9370190/v1","type":"article-journal","title":"DelRec: learning delays in recurrent spiking neural networks","abstract":"Abstract Biological neurons transmit information with stereotyped electrical impulses called ``spikes'', sensitive to coincident timings. Spiking Neural Networks (SNNs), introduced in the nineties, have gained popularity in AI for their energy efficiency and competitive performance with deep learning. Among them, Recurrent SNNs (RSNNs) are particularly appealing for their ability to learn long-term dependencies and exhibit rich dynamics. In SNNs, each connection can have a weight and a transmission delay, both plastic in the brain. While theory has long suggested that trainable delays enhance a network's expressivity, practical learning methods emerged only recently and remain mostly limited to feedforward delays. Here, we introduce DelRec, the first method to jointly optimize recurrent delays with synaptic weights in RSNNs via surrogate gradient learning, compatible with any spiking neuron model. DelRec works in discrete time, leveraging differentiable interpolation to handle non-integer delays with well-defined gradients at training time, then rounding them for inference. Using simple neurons, DelRec outperforms all baselines on a chaotic time-series prediction task, and sets new state-of-the-art accuracies on two challenging temporal datasets. Analysis of trained networks reveals structured, depth-dependent spatio-temporal receptive fields and delay-weight co-adaptation reshaping temporal selectivity. This work establishes recurrent delay optimization as a promising framework for both biological circuit modeling and neuromorphic computing.","author":[{"family":"Queant","given":"Alexandre"},{"family":"Rancon","given":"Ulysse"},{"family":"Cottereau","given":"Benoit"},{"family":"Masquelier","given":"Timothée"}],"issued":{"date-parts":[[2026]]},"DOI":"10.21203/rs.3.rs-9370190/v1","URL":"https://doi.org/10.21203/rs.3.rs-9370190/v1","source":"preprints"},{"id":"doi:10.1063/5.0285089","type":"article-journal","title":"Learning chaotic dynamics with neuromorphic network dynamics","abstract":"This study investigates how dynamical systems may be learned and modeled with a neuromorphic network, which is itself a dynamical system. The neuromorphic network used in this study is based on a complex electrical circuit comprised of memristive elements that produce neuro-synaptic nonlinear responses to input electrical signals. To determine how computation may be performed using the physics of the underlying system, the neuromorphic network was simulated and evaluated on the autonomous prediction of a multivariate chaotic time series, implemented with a reservoir computing framework. Through manipulating only input electrodes and voltages, optimal nonlinear dynamical responses were found when input voltages maximize the number of memristive components whose internal dynamics explore the entire dynamical range of the memristor model. Increasing the network coverage with the input electrodes was found to suppress other nonlinear responses that are less conducive to learning. These results provide valuable insights into how a physical neuromorphic network device can be feasibly optimized for learning complex dynamical systems using only external control parameters.","author":[{"family":"Xu","given":"Yinhao"},{"family":"Gottwald","given":"Georg"},{"family":"Kuncic","given":"Zdenka"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1063/5.0285089","URL":"https://doi.org/10.1063/5.0285089","source":"crossref"},{"id":"doi:10.1002/aisy.202500223","type":"article-journal","title":"A Field Programmable Gate Array‐Assisted Optoelectronic Emulator for Photonic Neuromorphic Computing","abstract":"As the pace of complementary metal oxide semiconductor (CMOS) scaling slows down and Moore's law approaches its limit, optoelectronics has emerged as a promising solution for next‐generation computing hardware. Optoelectronic systems are appealing due to their fast speed, large bandwidth, low power consumption, and CMOS‐compatible fabrication process. Nevertheless, the absence of efficient design and verification tools in the prefabrication phase poses obstacles to the transition from design to practical applications. To address these hurdles, an optoelectronic emulator is created that combines well‐established optical components with a field programmable gate array. The emulator can execute multiplication operations within the optical domain with a precision of up to 6 bits and at a high frequency reaching 20 GHz. By leveraging this system, various image convolution and inference learning tasks have been effectively carried out. This accomplishment serves as a compelling demonstration of its capability to execute fundamental learning tasks, which is in line with the requirements for hardware simulations of photonic neuromorphic computing. The emulator plays a crucial role in bridging the gap between simulations and real‐world systems. It has the potential to boost the precision and reliability of future optoelectronic system design.","author":[{"family":"Li","given":"Jinxian"},{"family":"Hu","given":"Runyu"},{"family":"Wang","given":"Fengyu"},{"family":"Shen","given":"Jiabin"},{"family":"Cheng","given":"Zengguang"},{"family":"Zhou","given":"Peng"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1002/aisy.202500223","URL":"https://doi.org/10.1002/aisy.202500223","source":"crossref"},{"id":"doi:10.1088/2634-4386/ae7ea7","type":"article-journal","title":"Noise-based reward-modulated learning","abstract":"Abstract The pursuit of energy-efficient and adaptive artificial intelligence (AI) has positioned neuromorphic computing as a promising alternative to conventional computing. However, achieving learning on these platforms requires techniques that prioritize local information while enabling effective credit assignment. Here, we propose noise-based reward-modulated learning (NRL), a novel synaptic plasticity rule that mathematically unifies reinforcement learning and gradient-based optimization with biologically-inspired local updates. NRL addresses the computational bottleneck of exact gradients by approximating them through stochastic neural activity, transforming the inherent noise of biological and neuromorphic substrates into a functional resource. Drawing inspiration from biological learning, our method uses reward prediction errors as its optimization target to generate increasingly advantageous behavior, and eligibility traces to facilitate retrospective credit assignment. Experimental validation on reinforcement tasks, featuring immediate and delayed rewards, shows that NRL achieves performance comparable to baselines optimized using backpropagation, although with slower convergence, while showing significantly superior performance and scalability in multi-layer networks compared to reward-modulated Hebbian learning, the most prominent similar approach. While tested on simple architectures, the results highlight the potential of noise-driven, brain-inspired learning for low-power adaptive systems, particularly in computing substrates with locality constraints. NRL offers a theoretically grounded paradigm well-suited for the event-driven characteristics of next-generation neuromorphic AI.","author":[{"family":"Fernández","given":"Jesús"},{"family":"Ahmad","given":"Nasir"},{"family":"Gerven","given":"Marcel"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/2634-4386/ae7ea7","URL":"https://doi.org/10.1088/2634-4386/ae7ea7","source":"crossref"},{"id":"doi:10.1088/1361-6463/ae01b5","type":"article-journal","title":"Self-compliance and forming-free memristor arrays with a SiO<sub>2</sub> scavenging barrier for energy-efficient neuromorphic computing","abstract":"Abstract In this work, low-power RRAM (Resistive random-access memory) devices were characterized by a SiO 2 layer serving as an oxygen scavenging barrier, which suppresses conductive filament overgrowth and reduces operation current and power consumption. Additionally, the integration of an AlO x /TiO y overshoot suppression layer enabled intrinsic self-compliance and forming-free operation. XPS analysis confirmed the oxygen composition of the oxygen-rich TiO y , and it was also verified that the higher oxygen composition in TiO y suppresses filament formation, which decreases the operation current. Consequently, the SiO 2 layer decreases the LRS current significantly by 10 3 times. Furthermore, the device demonstrates the endurance of 6 × 10 3 cycles and retention over 10 4 s, maintaining analog multi-bit operation. When the proposed device was integrated into an 8 × 8 passive array and programmed with the designated weights, a low mean absolute error (MAE) of 10.8 nA was achieved. Additionally, vector-matrix multiplication operations demonstrated excellent accuracy, with over 99% of results falling within an error margin of 10%. Based on this low MAE, MNIST image classification simulations were conducted, yielding classification accuracy exceeding 96%.","author":[{"family":"Kim","given":"Minki"},{"family":"Kim","given":"Sungjoon"},{"family":"Hwang","given":"Sungmin"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1088/1361-6463/ae01b5","URL":"https://doi.org/10.1088/1361-6463/ae01b5","source":"crossref"},{"id":"doi:10.1088/2634-4386/ae0fc0","type":"article-journal","title":"Benchmarking spiking neurons for linear quadratic regulator control of multi-linked pole on a cart: from single neuron to ensemble","abstract":"Abstract The emerging field of neuromorphic computing for edge control applications poses the need to quantitatively estimate and limit the number of spiking neurons, to reduce network complexity and optimize the number of neurons per core and hence, the chip size, in an application-specific neuromorphic hardware. While rate-encoding for spiking neurons provides a robust way to encode signals with the same number of neurons as an ANN, it often lacks precision. To achieve the desired accuracy, a population of neurons is often needed to encode the complete range of input signals. However, using population encoding immensely increases the total number of neurons required for a particular application, thus increasing the power consumption and on-board resource utilization. A transition from two neurons to a population of neurons for the linear quadratic regulator (LQR) control of a cartpole is shown in this work. The near-linear behavior of a leaky-integrate-and-fire neuron can be exploited to achieve the LQR control of a cartpole system. This has been shown in simulation, followed by a demonstration on a single-neuron hardware, known as Lu.i. The improvement in control performance is then demonstrated by using a population of varying numbers of neurons for similar control in the Nengo neural engineering framework (NEF), on CPU and on Intel’s Loihi neuromorphic chip. Finally, linear control is demonstrated for four multi-linked pendula on cart systems, using a population of neurons in Nengo, followed by an implementation of the same on Loihi. This study compares LQR control in the NEF using 7 control and 7 neuromorphic performance metrics, followed by a comparison with other conventional spiking and non-spiking controllers.","author":[{"family":"Banerjee","given":"Shreyan"},{"family":"Gava","given":"Luna"},{"family":"Rounak","given":"Aasifa"},{"family":"Pakrashi","given":"Vikram"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1088/2634-4386/ae0fc0","URL":"https://doi.org/10.1088/2634-4386/ae0fc0","source":"crossref"},{"id":"doi:10.1088/2634-4386/ae65d6","type":"article-journal","title":"Efficient aspect term extraction using spiking neural network","abstract":"Abstract Aspect term extraction (ATE) identifies aspect terms in review sentences, a key subtask of sentiment analysis. It is a sequence labeling task that aims to identify aspect expressions within opinionated text. While most existing approaches predominantly rely on deep neural networks (DNNs), which achieve strong performance but incur high computational and energy costs. In this work, we propose SpikeATE, an energy-efficient alternative using spiking neural networks (SNNs), which leverage sparse activations and event-driven computation to capture temporal dependencies between words, making them suitable for ATE. SpikeATE employs ternary spiking neurons and direct spike-based training with surrogate gradients to enable efficient end-to-end learning. We evaluate our approach on four SemEval benchmark datasets: Laptop14, Restaurant14, Restaurant15, and Restaurant16. Experimental results show that SpikeATE achieves competitive performance with state-of-the-art methods, obtaining F 1-scores of 84.02, 86.46, 72.25, and 78.19, respectively, with the ternary model. In addition to performance, we provide a theoretical analysis of energy consumption, demonstrating the efficiency of the proposed approach. SpikeATE achieves significantly lower power usage (2.5946 mJ for the ternary model and 1.8943 mJ for the binary model) compared to transformer-based models such as BERT and self-training (≈ 95.5 mJ) and large language models such as GPT-3.5 ( ≈ 3.92 × 10 15 mJ). These results highlight that SpikeATE reduces energy consumption by orders of magnitude while maintaining competitive performance. Overall, this work demonstrates that SNN-based architectures provide a practical and sustainable alternative to conventional DNNs for ATE, achieving a favorable balance between accuracy and computational efficiency.","author":[{"family":"Mishra","given":"Abhishek"},{"family":"Somasundaram","given":"Arya"},{"family":"Das","given":"Anup"},{"family":"Kandasamy","given":"Nagarajan"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/2634-4386/ae65d6","URL":"https://doi.org/10.1088/2634-4386/ae65d6","source":"crossref"},{"id":"doi:10.1063/5.0263232","type":"article-journal","title":"Fully solution-processed ferroelectric thin film transistor based on PZT and its application in neuromorphic computing","abstract":"As a promising alternative to conventional computing paradigms, the neuromorphic computing has been demonstrated by using various artificial synaptic devices. Due to the excellent capability for the conductance modulation, the ferroelectric thin film transistors (FeTFTs) have been shown as one of the promising candidates for artificial synaptic devices. In this work, the FeTFTs based on the lead zirconate titanate (PZT) thin films were integrated by the fully solution process. Prior to the integration of the FeTFTs, a lanthanum nickelate (LNO) thin film was prepared as the seed layer. The introduction of the LNO has been demonstrated to improve the crystallinity of the PZT thin films. It is confirmed that the channel conductance of the FeTFTs can be precisely modulated by adjusting the amplitude, duration, and number of the pulses. The potentiation and depression (P-D) characteristics of the FeTFTs have been demonstrated, and the P-D curve shows low nonlinearity and small cycle-to-cycle variations. Based on the P-D characteristics of the FeTFTs, an artificial neural network has been constructed for the pattern recognition, and a recognition accuracy of 93.1% has been achieved. These results suggest that the fully solution-processed FeTFTs based on PZT are the promising candidate for the artificial synaptic devices.","author":[{"family":"Dong","given":"Yao"},{"family":"Miao","given":"Guangtan"},{"family":"Xiao","given":"Wenlan"},{"family":"You","given":"Chunyan"},{"family":"Liu","given":"Guoxia"},{"family":"Shan","given":"Fukai"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1063/5.0263232","URL":"https://doi.org/10.1063/5.0263232","source":"crossref"},{"id":"doi:10.3390/technologies13080326","type":"article-journal","title":"Trigger-Based Systems as a Promising Foundation for the Development of Computing Architectures Based on Neuromorphic Materials","abstract":"It is demonstrated that neuromorphic materials designed for computational tasks can be effectively implemented by drawing an analogy with trigger-based systems built upon classical binary elements. Among the most promising approaches in this context are systems that perform computations based on the Residue Number System (RNS). A specific implementation of a trigger-based adder employing the proposed methodology is presented and tested through simulation modeling. This adder utilizes the representation of natural numbers as elements of a subtraction ring modulo P, where P is the product of Mersenne prime numbers. This configuration enables component-wise, independent execution of arithmetic operations. It is further shown that analogous trigger-based systems can be realized using recurrent neural network analogs, particularly those implemented with neuromorphic materials. The study emphasizes that it is possible to construct a neural network, especially one based on neuromorphic substrates, that can perform logical operations equivalent to those executed by conventional binary circuitry. A key challenge in the proposed approach lies in implementing an operation analogous to the carry mechanism employed in classical binary adders. An algorithm addressing this issue is proposed, based on the transition from computations modulo P to computations modulo 2P.","author":[{"family":"Shaltykova","given":"Dina"},{"family":"Kadyrzhan","given":"Kaisarali"},{"family":"Caiko","given":"Jelena"},{"family":"Vitulyova","given":"Yelizaveta"},{"family":"Suleimenov","given":"Ibragim"}],"issued":{"date-parts":[[2025]]},"DOI":"10.3390/technologies13080326","URL":"https://doi.org/10.3390/technologies13080326","source":"crossref"},{"id":"doi:10.1063/5.0260692","type":"article-journal","title":"Interface engineering modulation of ferroelectric synapses for high-precision neuromorphic computing","abstract":"Nb:SrTiO3 (NSTO) are commonly employed as substrate and electrode for BaTiO3-based ferroelectric memristors. These substrates are available in two types. The first one is with the hybrid-terminated interface, which consists of alternating SrO and TiO2 planes, while the second one is a TiO2-terminated interface. The interaction between surfaces and interfaces plays a crucial role in determining the overall performance of synapses. This paper reports a ferroelectric synapse whose neuromorphic performance can be regulated by the terminated interface of NSTO substrate. Compared with hybrid-terminated devices, the TiO2-terminated devices exhibit a 0.38 eV increase in barrier height, a 38.2% reduction in dislocation density, an approximately 10-fold enhancement in the on/off ratio, a 47.1% improvement in the linearity of long-term potentiation, and a 57.1% improvement in the linearity of long-term depression. This is due to their lower surface state density and atomically flat surface topography. In addition, the TiO2-terminated devices accurately emulate the characteristics of artificial synapses, and the neural network developed based on the weight update characteristics of the memristor achieves an image recognition accuracy of 96.1% on the National Institute of Standards and Technology handwritten digit dataset.","author":[{"family":"Liu","given":"Hao"},{"family":"Wang","given":"Yan"},{"family":"Wu","given":"Wenshuo"},{"family":"Zhang","given":"Minghao"},{"family":"Su","given":"Jie"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1063/5.0260692","URL":"https://doi.org/10.1063/5.0260692","source":"crossref"},{"id":"doi:10.1088/2634-4386/ade622","type":"article-journal","title":"SpinONN: energy efficient brain-inspired spintronics-based Hopfield oscillatory neural network for image denoising","abstract":"Abstract This work proposes a Spintronics-based Hopfield oscillatory neural network (HONN) that leverages dynamic frequency-encoded electrical synchronization between two spin-torque vortex nano-oscillators (SVNOs) as oscillatory neurons, with a non-volatile memristor as a coupling element (synaptic connection). The frequency synchronization mechanism, inspired by the brain’s oscillatory dynamics, enables the synchronization of SVNOs, facilitating efficient information processing of the dynamic oscillatory signals within the network. This coupling mechanism has been investigated to design SVNOs-based neural circuit design topology for enhanced frequency-encoded computing using SVNOs neurons and memristive coupling synapses. The proposed transmission gate-based SVNO oscillatory neural circuit has been implemented, offering efficient frequency synchronization, non-linearity, and a less complex neural circuit design. Further, a hybrid Spintronic/complementary metal oxide semiconductor 16-SVNOs HONN is designed, and circuit-based simulations are performed, which offer a promising solution for building robust and scalable HONNs. We achieve fast computation (∼4 ns) and offer significantly lower energy consumption (∼24 fJ/neuron) as compared to VO 2 -based ONN architectures (8× faster and 4× reduced power/neuron). Finally, we demonstrate an image denoising application on the proposed SVNO-based HONN hardware-compatible accelerator using an image-splitting approach with parallel processing. The 32 × 32 street view house number image dataset is efficiently split into blocks and processed through the 16-SVNOs HONN design, dividing the image into 4 × 4 blocks. Lastly, we examined the peak signal-to-noise ratio and structural similarity index measure for denoising the images with an efficient splitting approach for scalability. The network effectively denoises images while maintaining image quality, demonstrating the potential of the HONN hardware-compatible architecture for large-scale and real-time applications.","author":[{"family":"Soni","given":"Sandeep"},{"family":"Rezaeiyan","given":"Yasser"},{"family":"Boehnert","given":"Tim"},{"family":"Farkhani","given":"Hooman"},{"family":"Ferreira","given":"Ricardo"},{"family":"Kaushik","given":"Brajesh"},{"family":"Moradi","given":"Farshad"},{"family":"Shreya","given":"Sonal"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1088/2634-4386/ade622","URL":"https://doi.org/10.1088/2634-4386/ade622","source":"crossref"},{"id":"doi:10.1002/adom.202501078","type":"article-journal","title":"High Sensitivity Optoelectronic Artificial Synapse Based on GaN Porous Nanocone Array for Neuromorphic Computing","abstract":"Abstract Neuromorphic computing architecture, inspired by biological systems, is one of the key solutions to overcoming the von Neumann bottleneck. In particular, as a core component of artificial visual perception systems, optoelectronic synapses with high sensitivity and long memory retention hold significant potential applications. Herein, a two‐terminal GaN porous nanocone array (PNA) optoelectronic artificial synapse is designed and fabricated. The unique structure of the GaN PNA significantly enhances light absorption, achieving a responsivity of up to 2.07 × 10 6 A W −1 and a specific detectivity ( D* ) of 3.50 × 10 15 Jones. The persistent photoconductivity (PPC) effect, originated from surface states, exhibits a remarkably long decay characteristic time of up to 1204 s. This enables the synapses to emulate various key synaptic functions, including paired‐pulse facilitation (PPF), memory‐forgetting‐relearning processes, and the transition from short‐term potentiation (STP) to long‐term potentiation (LTP). Moreover, the GaN PNA optoelectronic artificial synapse demonstrates exceptional performance in artificial visual neural networks with a recognition accuracy approaching 93%. This work presents a novel solution for enhancing computational efficiency in artificial visual neural networks.","author":[{"family":"Chen","given":"Jiawei"},{"family":"Huang","given":"Yuqing"},{"family":"Wen","given":"Hui"},{"family":"Wang","given":"Yujing"},{"family":"Li","given":"Huiying"},{"family":"Zheng","given":"Xinyuan"},{"family":"Wang","given":"Xin"},{"family":"Ma","given":"Zhanhong"},{"family":"Wang","given":"Ting"},{"family":"Yan","given":"Sen"},{"family":"Wang","given":"Kaiyou"},{"family":"Zhao","given":"Lixia"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1002/adom.202501078","URL":"https://doi.org/10.1002/adom.202501078","source":"crossref"},{"id":"doi:10.1063/5.0275455","type":"article-journal","title":"Enhancement of spin–orbit torque in sputtered BiSb-based perpendicular magnetic tunnel junctions for neuromorphic computing applications","abstract":"Topological insulators offer unique properties for generating high spin–orbit torque (SOT), promising to revolutionize magnetoresistive random-access memory with a low power consumption. In this work, BiSb is integrated into perpendicular magnetic tunnel junctions (pMTJs) to enable efficient SOT switching. By optimizing the BiSb thickness and introducing a Ta buffer layer, a threefold enhancement in damping-like SOT efficiency and a 60% reduction in switching current are achieved compared to the BiSb-free sample. X-ray diffraction measurements confirm the improved crystalline quality with increasing BiSb thickness, contributing to the enhanced spin current generation. The fabricated BiSb-pMTJs exhibit key neuromorphic functionalities, including gradual long-term potentiation/depression and sigmoidal resistance modulation under pulsed current. Utilizing these features, a three-layer artificial neural network is implemented based on experimentally extracted device behavior, achieving over 90% accuracy in handwritten digit recognition.","author":[{"family":"Wu","given":"S"},{"family":"Lim","given":"GJ"},{"family":"Tan","given":"FN"},{"family":"Jin","given":"TL"},{"family":"Ang","given":"CCI"},{"family":"Koh","given":"EK"},{"family":"Lee","given":"SH"},{"family":"Cheng","given":"KJ"},{"family":"Lew","given":"WS"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1063/5.0275455","URL":"https://doi.org/10.1063/5.0275455","source":"crossref"},{"id":"doi:10.1002/aisy.202500506","type":"article-journal","title":"Cryogenic Neuromorphic Synaptic Behavior in 180 nm Silicon Transistors for Emerging Computing Systems","abstract":"With the advancement of artificial intelligence (AI), there is an increasing demand for high‐speed, energy‐efficient hardware capable of running complex machine learning algorithms. Traditional hardware is constrained by the Von Neumann bottleneck, resulting in high power consumption and slower speeds. Inspired by the human brain, bio‐mimicking the dynamic synaptic plasticity of the biological synapse using synaptic transistors is crucial to building the next generation of high‐performance computing hardware‐based neural networks. This study investigates neuromorphic behavior in 180 nm bulk complementary metal oxide semiconductor (CMOS) devices at 4 K, emphasizing memory properties and synapse‐like characteristics. These findings position bulk CMOS as a scalable, energy‐efficient, cryo‐compatible platform for neuromorphic and quantum computing use. Gated‐pulse measurements are used to study potentiation–depression behavior by quantifying conductance changes as functions of pulse amplitude and width. These results closely resemble biological synaptic plasticity, laying the groundwork for integrating cryo‐CMOS technology into neuromorphic computing. The work reported here aims to work toward the development of hybrid computational systems by bridging the gap between conventional CMOS devices and emerging cryogenic technology, offering new avenues for scalable, energy‐efficient, and high‐performance cryogenic neuromorphic technologies.","author":[{"family":"Imroze","given":"Fiheon"},{"family":"Yalagala","given":"Bhavani"},{"family":"Kumar","given":"Naveen"},{"family":"Elsayed","given":"Mostafa"},{"family":"Ahmad","given":"Meraj"},{"family":"Graham","given":"Robert"},{"family":"Georgiev","given":"Vihar"},{"family":"Heidari","given":"Hadi"},{"family":"Weides","given":"Martin"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1002/aisy.202500506","URL":"https://doi.org/10.1002/aisy.202500506","source":"crossref"},{"id":"doi:10.1002/admt.202500786","type":"article-journal","title":"Toward Advancement of Fabrication Techniques of Neuromorphic Computing Devices Based on 2D Materials","abstract":"Abstract The growing necessity for power‐efficient and cognitive computation mechanisms has driven progress in neuromorphic computing which seeks to imitate the synaptic mechanisms underlying human brain functionality. The drawbacks of traditional computation paradigms which involve a large amount of power utilization and restricted data communication drive the quest for alternative materials and technologies. In this context, 2D materials have proven themselves an especially valuable class of materials for new‐generation neuromorphic devices because of their atomic thickness and distinct electronic attributes. The present review gives a detailed account of advanced techniques enabling the fabrication of neuromorphic devices using 2D materials with a focus on deposition methods and device engineering strategies to enhance synaptic functionalities for energy‐efficient signal processing. This article also explores the role of 2D materials in establishing effective synthetic synapses which are predominant in supporting key functions such as short‐term plasticity (STP) and long‐term plasticity (LTP). This article further addresses key fabrication challenges such as scalability, contact/interface issues, and variability, along with emerging solutions like atomic‐thickness control and heterostructure integration. Through a carefully designed roadmap, this article attempts to blend fabrication processes and 2D material's neuromorphic device physics hence presenting valuable insights in constructing brain inspired computational devices.","author":[{"family":"Gupta","given":"Shubham"},{"family":"Patel","given":"Malkeshkumar"},{"family":"Kumar","given":"Naveen"},{"family":"Pohl","given":"László"},{"family":"Park","given":"Min‐joon"},{"family":"Youn","given":"Sung‐min"},{"family":"Jeong","given":"Chaewhan"},{"family":"Kim","given":"Joondong"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1002/admt.202500786","URL":"https://doi.org/10.1002/admt.202500786","source":"crossref"},{"id":"doi:10.1038/s41377-025-01773-6","type":"article-journal","title":"Versatile optoelectronic memristor based on wide-bandgap Ga2O3 for artificial synapses and neuromorphic computing","abstract":"Abstract Optoelectronic memristors possess capabilities of data storage and mimicking human visual perception. They hold great promise in neuromorphic visual systems (NVs). This study introduces the amorphous wide-bandgap Ga 2 O 3 photoelectric synaptic memristor, which achieves 3-bit data storage through the adjustment of current compliance ( I cc ) and the utilization of variable ultraviolet (UV-254 nm) light intensities. The “AND” and “OR” logic gates in memristor-aided logic (MAGIC) are implemented by utilizing voltage polarity and UV light as input signals. The device also exhibits highly stable synaptic characteristics such as paired-pulse facilitation (PPF), spike-intensity dependent plasticity (SIDP), spike-number dependent plasticity (SNDP), spike-time dependent plasticity (STDP), spike-frequency dependent plasticity (SFDP) and the learning experience behavior. Finally, when integrated into an artificial neural network (ANN), the Ag/Ga 2 O 3 /Pt memristive device mimicked optical pulse potentiation and electrical pulse depression with high pattern accuracy (90.7%). The single memristive cells with multifunctional features are promising candidates for optoelectronic memory storage, neuromorphic computing, and artificial visual perception applications.","author":[{"family":"Cui","given":"Dongsheng"},{"family":"Pei","given":"Mengjiao"},{"family":"Lin","given":"Zhenhua"},{"family":"Zhang","given":"Hong"},{"family":"Kang","given":"Mengyang"},{"family":"Wang","given":"Yifei"},{"family":"Gao","given":"Xiangxiang"},{"family":"Su","given":"Jie"},{"family":"Miao","given":"Jinshui"},{"family":"Li","given":"Yun"},{"family":"Zhang","given":"Jincheng"},{"family":"Hao","given":"Yue"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1038/s41377-025-01773-6","URL":"https://doi.org/10.1038/s41377-025-01773-6","source":"europepmc"},{"id":"doi:10.1002/rpm.20240038","type":"article-journal","title":"Artificial optoelectronic synapses based on Ga\n                    <sub>2</sub>\n                    O\n                    <sub>3</sub>\n                    metal–semiconductor–metal solar‐blind ultraviolet photodetectors with asymmetric electrodes for neuromorphic computing","abstract":"Abstract Research on optoelectronic synapses that can integrate both detection and processing functions is essential for the development of efficient neuromorphic computing. Here, we experimentally demonstrated an Ga 2 O 3 ‐based metal–semiconductor–metal (MSM) solar‐blind ultraviolet (UV) photodetector (PD) with asymmetric interdigital electrodes. The Ga 2 O 3 PD exhibits a responsivity of 732 A/W under a forward bias of 6 V. The tunable conductance properties of PDs provide a novel approach to synaptic performance. The proposed PDs as artificial synapse realized several essential synaptic function, including excitatory postsynaptic current, paired‐pulse facilitation, long‐term potentiation, the transition from short‐term memory to long‐term memory, and learning experience behaviors successfully. At a reverse bias, an ultra‐low energy consumption of 140 fJ was achieved. In addition, the optoelectronic synapses demonstrated a recognition accuracy of over 95% in the MNIST handwritten number recognition task. These results suggest that Ga 2 O 3 MSM solar‐blind UV PDs have high potential for efficient optoelectronic neuromorphic computing applications.","author":[{"family":"Sun","given":"Huazhen"},{"family":"Ye","given":"Bingjie"},{"family":"Ge","given":"Mei"},{"family":"Gong","given":"Biao"},{"family":"Qian","given":"Leyang"},{"family":"Parkhomenko","given":"Irina"},{"family":"Komarov","given":"Fadei"},{"family":"Liu","given":"Yu"},{"family":"Yang","given":"Guofeng"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1002/rpm.20240038","URL":"https://doi.org/10.1002/rpm.20240038","source":"crossref"},{"id":"doi:10.1021/acsaelm.6c01218","type":"article-journal","title":"MoS2-Based Synaptic Transistor for Neuromorphic Computing and Encoded Optical Communication","abstract":"Abstract Two-dimensional (2D) transition metal dichalcogenides (TMDs) have attracted considerable attention for next-generation electronic and optoelectronic devices due to their atomically thin structure, tunable bandgap, and strong light-matter interactions. In this work, we present a facile approach by employing 2D molybdenum disulfide (MoS2) to fabricate an optoelectronic artificial synapse, emphasizing its potential for neuromorphic computing and encoded optical communication. The MoS2-based field-effect transistor (FET) exhibits n-type semiconducting behavior with a current on/off ratio (ION/IOFF) of ∼2.5 × 108 and a large clockwise hysteresis in its transfer characteristics arising from charge trapping defects at the MoS2/SiO2 interface. A memory window of ∼41 V is observed under a gate voltage sweep range of ±60 V, and the memory effect can be gradually modulated from 1 V to 41 V by varying the sweeping range of the Si gate from ±10 to ±60 V. The device demonstrates optically controllable synaptic plasticity under 365 nm UV stimulation, exhibiting key neuromorphic functionalities such as paired-pulse facilitation (PPF), excitatory postsynaptic current (EPSC), spike-number-dependent plasticity (SNDP), and spike-time-dependent plasticity (STDP). Furthermore, the proposed MoS2 synaptic transistor enables encoded optical communication based on temporal pulse-width modulation, where optically encoded signals are directly decoded through distinct EPSC responses. This capability highlights the robustness of the device for optically encoded information processing and its potential for low-power encoded communication, neuromorphic photonic interfaces, and brain-inspired computing systems.","author":[{"family":"Lee","given":"Jaehyeop"},{"family":"Kim","given":"Minsu"},{"family":"Nasim","given":"Muhammad"},{"family":"Kim","given":"Chiyoung"},{"family":"Khan","given":"Muhammad"},{"family":"Shin","given":"Jae"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1021/acsaelm.6c01218","URL":"https://doi.org/10.1021/acsaelm.6c01218","source":"crossref"},{"id":"doi:10.1063/5.0256082","type":"article-journal","title":"Visible light-driven synaptic transistors based on bilayer InGaZnO homojunction for neuromorphic computing","abstract":"The development of photoelectric synaptic transistors (PSTs) using visible light-driven mimicking synaptic behaviors represents a key advancement toward biomimetic visual systems. This study proposes a PST based on bilayer indium-gallium-zinc-oxide (IGZO) homojunctions with tunable gallium ratios. By optimizing the gallium content, oxygen vacancy concentrations in the channel were precisely controlled, suppressing deionization processes and enhancing device performance. The IGZO homojunction PST demonstrated outstanding electrical characteristics (Ion/Ioff = 1.2 × 107, μ = 3.88 cm2/Vs, Vth = 0 V) and exhibited high photocurrent and robust persistent photoconductivity under visible light. The device mimicked various synaptic behaviors, including excitatory postsynaptic current, paired-pulse facilitation, the transition from short-term plasticity to long-term plasticity, spiking-rate-dependent plasticity, and spike-timing-dependent plasticity. Furthermore, leveraging the potentiation and depression behaviors of the IGZO homojunction PST, a triple-layer neural network achieved 96.8% accuracy in pattern recognition tasks. These results underscore the IGZO homojunction PST's immense potential for advancing artificial vision systems.","author":[{"family":"Yin","given":"Zezhong"},{"family":"Shan","given":"Liuyue"},{"family":"Ci","given":"Ranran"},{"family":"Hao","given":"Dandan"},{"family":"Miao","given":"Guangtan"},{"family":"Tian","given":"Likun"},{"family":"Liu","given":"Guoxia"},{"family":"Shan","given":"Fukai"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1063/5.0256082","URL":"https://doi.org/10.1063/5.0256082","source":"crossref"},{"id":"doi:10.1088/2634-4386/ae9215","type":"article-journal","title":"NeuRehab: a reinforcement learning and spiking neural network-based rehab automation framework","abstract":"Abstract Recent advancements in robotic rehabilitation therapy have provided modular exercise systems for post-stroke muscle recovery with basic control schemes. But these systems struggle to adapt to patients’ complex and ever-changing behaviour, and to operate within mobile settings, such as heat and power constraints. To aid this, we present NeuRehab: an end-to-end framework consisting of a training and inference pipeline with AI-based automation, co-designed with neuromorphic computing-based control systems that balance action performance, power consumption, and observed latency. The framework consists of 2 partitions. One is designated for the rehabilitation device based on ultra-low-power spiking networks deployed on dedicated neuromorphic hardware. The other resides on stationary hardware that can accommodate computationally intensive hardware for fine-tuning on a per-patient basis. By maintaining a communication channel between both the modules and splitting the algorithm components, the power and latency requirements of the movable system have been optimised, while retaining the learning performance advantages of compute- and power-hungry hardware on the stationary machine. As part of the theoretical framework definition, we propose (a) the split machine learning processes for efficiency in architectural utilisation, and (b) task-specific temporal optimisations to lower edge-inference control latency. This paper evaluates the proposed optimisation methods on a reference stepper motor-based shoulder exercise in an assist-as-needed scenario, which is presented as a challenging use case for a spiking neural network-based control system. Overall, these methods offer comparable performance uplifts over the State-of-the-art for neuromorphic deployment in terms of spike counts and time steps, leading to theoretical gains in power and latency.","author":[{"family":"Kambhampati","given":"Phani"},{"family":"Gautam","given":"Chainesh"},{"family":"Rao","given":"Madhav"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/2634-4386/ae9215","URL":"https://doi.org/10.1088/2634-4386/ae9215","source":"crossref"},{"id":"doi:10.1002/aelm.202500370","type":"article-journal","title":"Solution‐Processed Bi\n                    <sub>2</sub>\n                    S\n                    <sub>3</sub>\n                    Nanostructures for Flexible Memory and Neuromorphic Computing","abstract":"Abstract The rapid advancement of wearable computing and edge AI technologies is driving the need for low‐temperature, flexible, and neuromorphic‐compatible electronic materials. In this work, the successful low‐temperature deposition of Bi 2 S 3 thin films via an in situ solvothermal method using the single‐source precursor (SSP) [Bi(S 2 P(OC 3 H₇) 2 ) 3 ] is reported. This solution‐processed approach enables the formation of high‐quality, crystalline, and stoichiometric Bi 2 S 3 films over a broad temperature window (140–200 °C), compatible with a range of substrates including silicon, polyimide, and PET. Leveraging this deposition technique, Bi 2 S 3 ‐based memristors are fabricated on both rigid and flexible substrates. The devices exhibit stable resistive switching behavior and demonstrate mechanical and electrical robustness under stress conditions. Furthermore, the memristors effectively emulate long‐term synaptic plasticity, achieving high learning accuracy. These findings establish SSP‐derived Bi 2 S 3 films as a promising material platform for next‐generation flexible neuromorphic computing and memory technologies.","author":[{"family":"Harke","given":"Sayali"},{"family":"Kapur","given":"Omesh"},{"family":"Dai","given":"Peng"},{"family":"Zhang","given":"Tongjun"},{"family":"Ding","given":"Bingkai"},{"family":"Ding","given":"Bohao"},{"family":"Huang","given":"Ruomeng"},{"family":"Gurnani","given":"Chitra"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1002/aelm.202500370","URL":"https://doi.org/10.1002/aelm.202500370","source":"crossref"},{"id":"doi:10.1088/2634-4386/ae7ab5","type":"article-journal","title":"Temporal hierarchy in spiking neural networks","abstract":"Abstract Taking inspiration from the brain to design efficient computational systems remains challenging due to its complexity. This work investigates a key biological feature observed across mammals: a hierarchy of time scales in cortical areas. Experimental evidence shows that intrinsic neural dynamics slow down across the cortical hierarchy, forming a temporal hierarchy. We examine whether this property benefits artificial systems by introducing temporal hierarchy into spiking neural networks (SNNs), which inherently process information over time. We implement hierarchical time scales across neuronal, synaptic, and recurrent dynamics, and evaluate their effect under two settings: (1) as an inductive bias, and (2) as an emergent property through optimization. On temporal benchmarks such as multi-timescale-XOR and keyword spotting, hierarchical SNNs consistently outperform non-hierarchical ones, achieving 2%–6% higher accuracy and up to 5 × parameter reduction under iso-accuracy conditions. Moreover, when trained freely, temporal hierarchy emerges spontaneously through gradient descent. Finally, our theoretical analysis shows that hierarchical time constants enable processing of multi-frequency temporal signals with only log N layers—compared to N layers for non-hierarchical systems—highlighting hierarchy as a key organizational principle for efficient temporal computation.","author":[{"family":"Moro","given":"Filippo"},{"family":"Aceituno","given":"Pau"},{"family":"Kriener","given":"Laura"},{"family":"Payvand","given":"Melika"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/2634-4386/ae7ab5","URL":"https://doi.org/10.1088/2634-4386/ae7ab5","source":"crossref"},{"id":"doi:10.1002/idm2.12244","type":"article-journal","title":"High‐Performance Memristors Based on Ordered Imine‐Linked Two‐Dimensional Covalent Organic Frameworks for Neuromorphic Computing","abstract":"ABSTRACT Covalent organic frameworks (COFs) have emerged as highly promising materials for high‐performance memristors due to their exceptional stability, molecular design flexibility, and tunable pore structures. However, the development of COF memristors faces persistent challenges stemming from the structural disorder and quality control of COF films, which hinder the effective regulation of active metal ion migration during resistive switching. Herein, we report the synthesis of high‐quality, long‐range ordered, imine‐linked two‐dimensional (2D) COF TP‐TD film via the innovative surface‐initiated polymerization (SIP) strategy. The long‐range ordered one‐dimensional (1D) nanochannels within 2D COF TP‐TD film facilitate the stable and directed growth of conductive filaments (CFs), further enhanced by imine–CFs coordination effects. As a result, the fabricated memristor devices exhibit exceptional multilevel nonvolatile memory performance, achieving an ON/OFF ratio of up to 10 6 and a retention time exceeding 2.0 × 10 5 s, marking a significant breakthrough in porous organic polymer (POP) memristors. Furthermore, the memristors demonstrate high‐precision waveform data recognition with an accuracy of 92.17%, comparable to software‐based recognition systems, highlighting its potential in advanced signal processing tasks. This study establishes a robust foundation for the development of high‐performance COF memristors and significantly broadens their application potential in neuromorphic computing.","author":[{"family":"Huo","given":"Da"},{"family":"Gu","given":"Zhangjie"},{"family":"Song","given":"Bailing"},{"family":"Yu","given":"Yimeng"},{"family":"Wang","given":"Mengqi"},{"family":"Qin","given":"Lanhao"},{"family":"Li","given":"Huicong"},{"family":"Ouyang","given":"Decai"},{"family":"Xiao","given":"Shikun"},{"family":"Hu","given":"Wenhua"},{"family":"Wu","given":"Jinsong"},{"family":"Li","given":"Yuan"},{"family":"Chi","given":"Xiaodong"},{"family":"Zhai","given":"Tianyou"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1002/idm2.12244","URL":"https://doi.org/10.1002/idm2.12244","source":"crossref"},{"id":"doi:10.1088/2634-4386/adcbcc","type":"article-journal","title":"Mixed photonic/electronic neural network based on microLED arrays","abstract":"Abstract We present a novel approach for the implementation of a mixed photonic/electronic neural network using gallium nitride based microLEDs as an efficient key element. The system is fully analog, with the synaptic fan-out and weighting implemented in free-space optics, while the fan-in and nonlinear activation functions are realized by optoelectronic components. Activations are represented by patterns of incoherent light generated by microLED arrays. Spatial light modulators represent the weighting matrices in the optical path, and the summation of the weighted activations is realized by photodetectors. We practically demonstrate our approach with a prototype, implementing a single-layer artificial neural network for the classification of the MNIST dataset. A test accuracy of 86.8 % is achieved. A key feature of this photonic/electronic implementation is that an efficient conversion between the electrical and optical domains is realized by microLEDs, combining the advantages of the efficiency of synaptic connections through optics and nonlinear activation functions in electronics, which have the potential for miniaturization and integration. We estimate, that an upscaled version of our prototype can achieve an energy efficiency of around 14 fJ per multiply-accumulate operation using state-of-the-art microLEDs. Finally we present a scalability analysis, showing that for a fully connected layer, for which a total of N multiply-accumulate operations have to be performed, in the average case the energy requirement scales with O ( N ) , while the energy requirement of conventional computers, as well as digital neuromorphic computing approaches, scales with O ( N ) . Therefore, we suggest to follow this approach in order to fully exploit the potential of mixed optical/electrical neuromorphic systems and possibly demonstrate superiority over conventional GPUs.","author":[{"family":"Müller","given":"M"},{"family":"Kraneis","given":"R"},{"family":"Kälin","given":"N"},{"family":"Malm","given":"NV"},{"family":"Waag","given":"A"},{"family":"Werner","given":"C"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1088/2634-4386/adcbcc","URL":"https://doi.org/10.1088/2634-4386/adcbcc","source":"crossref"},{"id":"doi:10.64189/ssc.25212","type":"article-journal","title":"The Unified Neuromorphic Assembly Layer for Hardware-Agnostic Compilation in Neuromorphic Computing","abstract":"The programming of neuromorphic assembly has advanced steadily, providing essential tools and paradigms to help connect the gap between abstract Spiking Neural Network (SNN) models and brain-inspired computing hardware. This work presents UNAL (Unified Adaptive, Hardware-Agnostic Neuromorphic Assembly Layer). This compilation framework translates high-level Spiking Neural Network (SNN) models into portable, spike-level assembly across heterogeneous neuromorphic platforms. UNAL introduces a unified intermediate representation (UNAL-IR), a compact instruction set, and an optimization-driven mapping pipeline that jointly addresses latency, energy efficiency, routing congestion, and adaptability. Quantitative evaluation on standard SNN benchmarks (DVS Gesture and CIFAR-10 SNN) mapped to Intel Loihi 2 demonstrates 18–32% latency reduction, 21–38% energy savings, and 25–40% lower routing congestion compared to Loihi-native and platform-specific tool chains. A smart-city surveillance case study further validates the deployment of real-time edge computing. These results establish UNAL as a scalable and future-ready neuromorphic compiler infrastructure.","author":[{"family":"Jadhav","given":"Ganesh"},{"family":"Dagade","given":"Rahul"},{"family":"Jakhade","given":"Sushant"},{"family":"Jadhav","given":"Kshitij"},{"family":"Hinge","given":"Rutu"},{"family":"Joshi","given":"Swarada"}],"issued":{"date-parts":[[2025]]},"DOI":"10.64189/ssc.25212","URL":"https://doi.org/10.64189/ssc.25212","source":"crossref"},{"id":"doi:10.1109/emergin67762.2025.11450669","type":"article-journal","title":"Efficient AI Systems through Neuromorphic Computing: Bridging Biological Intelligence and Machine Learning","abstract":"The growing computational requirement of artificial intelligence (AI) mock-ups creates challenges in energy efficacy, scalability, and inactivity. Traditional AI architectures built on von Neumann structures and experience integral restrictions in imitating the human brain's competence. Neuromorphic computing, encouraged by biological neuronal structures, appears as a promising substitute for developing less-power, highly efficient AI systems. This paper explores neuromorphic computing architectures and their application to machine learning tasks using spiking neural networks (SNNs). This paper provides a thorough review on the present work along with the limitations. At the end paper also provided the details of future scope of neuromorphic computing in developing AI systems.","author":[{"family":"Sharma","given":"Aparna"},{"family":"Sharma","given":"Bhavesh"},{"family":"Bhushan","given":"Shashi"},{"family":"Nagar","given":"Vishvendra"},{"family":"Kumar","given":"Virendra"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1109/emergin67762.2025.11450669","URL":"https://doi.org/10.1109/emergin67762.2025.11450669","source":"crossref"},{"id":"doi:10.1063/5.0235267","type":"article-journal","title":"Recent advances in fluidic neuromorphic computing","abstract":"Human brain is capable of optimizing information flow and processing without energy-intensive data shuttling between processor and memory. At the core of this unique capability are billions of neurons connected through trillions of synapses—basic processing units of the brain. The action potentials or “spikes” based temporal processing using the regulated flow of ions across ion channels in neuron cells allows sparse and efficient transmission of data in the brain. Emerging systems based on confined fluidic systems have provided a framework for a new type of neuromorphic computing with lower energy consumption, hardware-level plasticity, and multiple information carriers that emulate natural processes and mechanisms of human brain. These systems mimic neuronal architectures by harnessing and modulating ion transport along artificial channels. The spikes-induced ion-to-surface interactions within these fluidic systems enables the control of ionic conductivity to achieve synaptic plasticity for the realization of brain-inspired functionalities such as memory effect and signal transmission. Herein, this review provides an overview of recent advances in fluidic devices such as memristors and other computing components, covering their basic operations, materials and architectures, as well as applications in neuromorphic computing. The review concludes with a brief outline of the challenges that these emerging technologies face and an outlook for the development of fluidic-based brain-inspired computing.","author":[{"family":"Law","given":"Cheryl"},{"family":"Wang","given":"Juan"},{"family":"Nielsch","given":"Kornelius"},{"family":"Abell","given":"Andrew"},{"family":"Bisquert","given":"Juan"},{"family":"Santos","given":"Abel"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1063/5.0235267","URL":"https://doi.org/10.1063/5.0235267","source":"crossref"},{"id":"doi:10.1088/2634-4386/ae8626","type":"article-journal","title":"Feedforward spiking neural networks are not transformers (yet): a learning-theoretic framework for long-range dependencies and biological efficiency","abstract":"Abstract Spiking neural networks offer a promising route toward low-power sequence computation on neuromorphic hardware, but they continue to lag behind attention-based artificial neural networks on long-context tasks. A central open question is whether this gap reflects only implementation and optimization limitations, or whether architectural features of spiking computation impose unfavorable learnability constraints as sequence length increases. Here, we address this question using a covering-number analysis of feedforward non-leaky integrate-and-fire (nLIF) networks in the probably approximately correct framework. Building on causal-piece decompositions and local Lipschitz continuity, we derive a global sensitivity bound for feedforward nLIF networks and extend it from single-token inputs to multi-token spike sequences. For fixed architectures under stated boundedness and margin assumptions, the resulting sufficient worst-case sample requirement has leading quadratic dependence on sequence length. This dependence arises from cumulative causal participation across time and depth, which increases global sensitivity along active spike paths. We then test the mechanistic implications of this theory using finite-sample cue-recall and teacher–student benchmarks across spiking, recurrent, and attention-based model classes. In cue-recall, an early cue must be retained across distractors and reported at a final query token; in teacher–student, labels are generated by a fixed nLIF teacher, separating representability from finite-sample learnability. Unconstrained feedforward spiking models show sequence-length sensitivity, elevated hidden spike-participation density, and increased samples-to-threshold burden. Post-spike refractoriness, leak-mediated forgetting, learned lateral inhibition, and activity-constrained winner-take-all competition reduce hidden participation and improve empirical robustness in task- and regime-dependent ways. Together, these results identify diffuse causal-set growth as a fundamental architectural bottleneck for baseline feedforward spiking sequence models and suggest that scalable neuromorphic sequence architectures will require circuit mechanisms that explicitly constrain temporal accumulation and effective spike participation.","author":[{"family":"Fishell","given":"William"},{"family":"Fishell","given":"Gord"},{"family":"Honnuraiah","given":"Suraj"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/2634-4386/ae8626","URL":"https://doi.org/10.1088/2634-4386/ae8626","source":"crossref"},{"id":"doi:10.1088/2634-4386/ae4648","type":"article-journal","title":"Spatiotemporal radar gesture recognition with hybrid spiking neural networks: balancing accuracy and efficiency","abstract":"Abstract Radar-based human activity recognition (HAR) offers privacy and robustness over camera-based methods, yet remains computationally demanding for edge deployment. We present the first application of spiking neural networks (SNNs) for radar-based HAR on aircraft marshaling signal classification. Our novel hybrid architecture combines a pre-trained convolutional backbone for spatial feature extraction and leaky integrate-and-fire neurons for temporal processing, inherently capturing gesture dynamics. The model reduces trainable parameters by 88% with under 1% accuracy loss compared to existing state of the art methods, and generalizes well to the Soli gesture dataset. Through systematic comparisons with other three artificial neural network architectures, we demonstrate the trade-offs of spiking computation in terms of accuracy, latency, memory, and energy, establishing SNNs as an efficient and competitive solution for radar-based HAR.","author":[{"family":"Mazzieri","given":"Riccardo"},{"family":"Cicciarella","given":"Eleonora"},{"family":"Pegoraro","given":"Jacopo"},{"family":"Corradi","given":"Federico"},{"family":"Rossi","given":"Michele"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/2634-4386/ae4648","URL":"https://doi.org/10.1088/2634-4386/ae4648","source":"crossref"},{"id":"doi:10.11591/ijai.v14.i2.pp1000-1021","type":"article-journal","title":"Adaptive silicon synapse and CMOS neuron for neuromorphic VLSI computing","abstract":"The design of a fully integrated adaptive modified complementary metal-oxide-semiconductor (CMOS) synapse circuit is presented. By using multiple-gated transistor configuration in the modified CMOS synapse an additional branch provide control where the synaptic output current time-constant is tuned. The effect of changing the multiple-gated transistor bias voltage from 0.25 to 0.45 V tunes the spiking output current exponential time-constant range by 200 ms as shown in simulation results. Moreover, a fully-integrated adaptive quadratic integrate-and-fire (QIF) CMOS neuron circuit is presented as well. A differential pair with variable capacitor integrator and a tunable schmitt trigger threshold detector circuit are integrated in the CMOS neuron that can be tuned varying its spiking frequency. The proposed adaptive quadratic integrate-and-fire (AQIF) neuron has the ability to adjust the spiking frequency without changing the input current. The simulation results show the proposed CMOS neuron circuit spiking frequency can be tuned from 58.4 to 312.5 Hz and its spiking period from 17.1 to 3.2 ms with tuning the bias voltage of variable capacitor integrator. Having a peak voltage Vpeak=0.95 V, a reset voltage Vreset=-0.75 V and a voltage threshold of 0.35 V with a membrane potential range of 1.5 V. The proposed CMOS neuron circuit is designed in 130 nm process with a supply voltage of 1.8 V and a total power dissipation of 1.8 mW.","author":[{"family":"El-Khatib","given":"Ziad"},{"family":"Moussa","given":"Sherif"},{"family":"Kamalov","given":"Firuz"},{"family":"Yagoub","given":"Mustapha"}],"issued":{"date-parts":[[2025]]},"DOI":"10.11591/ijai.v14.i2.pp1000-1021","URL":"https://doi.org/10.11591/ijai.v14.i2.pp1000-1021","source":"crossref"},{"id":"doi:10.1088/2634-4386/ae65d4","type":"article-journal","title":"Solving Sudoku using oscillatory neural networks","abstract":"Abstract We explore the capabilities of physical computing with oscillatory neural networks (ONNs) to solve combinatorial optimization problems. To solve Sudokus with ONNs, we define a novel mapping strategy that utilizes the unique characteristics of the computation paradigm. The problem is encoded through a puzzle specific graph-embedding, which implements the constraints through different subgraphs. These subgraphs are then combined into a single adjacency matrix , which allows the natural dynamics of the phases of coupled oscillators to find a solution to the puzzle. We model the phase dynamics of the ONN by means of the Kuramoto differential equation. This novel approach is then compared to the well-established iterative method to solve Sudoku already used in binary Hopfield neural networks (HNNs). Solving optimization problems typically requires a large amount of energy to solve on conventional hardware. Therefore, we are motivated to explore the mapping of Sudoku from a theoretical point of view to establish the validity of this approach. The simulation results show that the novel ONN mapping outperforms the established HNN methodology.","author":[{"family":"Haverkort","given":"Bram"},{"family":"Sbravati","given":"Federico"},{"family":"Porfir","given":"Stefan"},{"family":"Todri-Sanial","given":"Aida"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/2634-4386/ae65d4","URL":"https://doi.org/10.1088/2634-4386/ae65d4","source":"crossref"},{"id":"doi:10.5772/acrt.deposit.c.8440846","type":"article-journal","title":"Neuromorphic Computing in Outer Space: Intel Loihi Deployed On-Satellite for Spike Based Software Defined Communication","abstract":"&lt;table&gt;&lt;tr&gt;&lt;td&gt;&lt;p dir=\"ltr\"&gt;In space exploration and communication technology, software-defined networking (SDN) architectures significantly improve communication throughput and latency. This study developed a Q-learning-based continuous learning capable cognitive agent to select optimal routing options for a dynamically changing networking environment. With the dynamically changing network link latency, the cognitive agent achieved a 90% success rate in routing packets. The agent was executed on neuromorphic hardware called Intel Loihi. However, due to the system's power limitations, a simplified version of the agent was engineered to launch into space aboard a CubeSat. The CubeSat was launched in January 2022, making a historic milestone as the first launch of a neuromorphic system into space. The developed applications were successfully executed in space.&lt;/p&gt;&lt;/td&gt;&lt;/tr&gt;&lt;/table&gt;","author":[{"family":"Rahman","given":"Nayim"},{"family":"Yakopcic","given":"Chris"},{"family":"Taha","given":"Tarek"},{"family":"Lent","given":"Ricardo"},{"family":"Briones","given":"Janette"},{"family":"Chelmins","given":"David"},{"family":"Dudukovitch","given":"Rachel"},{"family":"Smith","given":"Aaron"},{"family":"Gannon","given":"Adam"},{"family":"Lowry","given":"Michael"},{"family":"Murbach","given":"Marcus"},{"family":"Salas","given":"Alejandro"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5772/acrt.deposit.c.8440846","URL":"https://doi.org/10.5772/acrt.deposit.c.8440846","source":"crossref"},{"id":"doi:10.2139/ssrn.7202153","type":"manuscript","title":"Dual-Layer PMMA/CsPbBr3 Regulation of Ag Filament Spatiotemporal Dynamics in SnO2 Quantum Dots Memristors for Neuron–Synapse Integrated Neuromorphic Computing","abstract":"Precise control over filament dynamics is critical for neuromorphic devices with neuronal and synaptic functions. However, most existing strategies regulate either spatial localization or temporal stability, lacking simultaneous and programmable control. Here, a dual-layer synergistic strategy is demonstrated in solution-processed SnO2 quantum dots (QDs) memristors for programmable spatiotemporal regulation of Ag filament formation. A CsPbBr3 layer serves as a nucleation platform to spatially confine filament growth, while a PMMA layer regulates Ag⁺ injection kinetics and filament evolution. By tuning PMMA thickness in Ag/(PMMA)ₙ/SnO2 QDs/CsPbBr3/ITO devices, a reversible transition between volatile and nonvolatile switching is achieved. The Ag/(PMMA)ₙ=5/SnO2 QDs/CsPbBr3/ITO volatile device exhibits an operating voltage of 0.5 V, an ON/OFF ratio of 105, endurance over 105 cycles, and leaky integrate-and-fire neuronal behavior. The Ag/(PMMA)ₙ=1/SnO2 QDs/CsPbBr3/ITO nonvolatile device shows an ON/OFF ratio of 103, endurance over 104 cycles, and data retention exceeding 104 s. Based on these two modes, a hybrid neuron–synapse system is constructed, achieving 93.3% MNIST recognition accuracy. This work provides a programmable filament dynamics paradigm for scalable neuromorphic computing hardware.","author":[{"family":"Yu","given":"Qiuxu"},{"family":"Li","given":"Shuyi"},{"family":"Mi","given":"Wei"},{"family":"Wang","given":"Di"},{"family":"He","given":"Lin&apos;an"},{"family":"Cai","given":"Gangri"},{"family":"Zhou","given":"Liwei"},{"family":"Zhao","given":"Jinshi"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2139/ssrn.7202153","URL":"https://doi.org/10.2139/ssrn.7202153","source":"crossref"},{"id":"doi:10.1002/aidi.202500053","type":"article-journal","title":"Advances in Organic In‐Sensor Neuromorphic Computing: from Material Mechanisms to Applications","abstract":"In‐sensor neuromorphic computing integrates sensing and processing within a single material system, enabling real‐time, ultralow‐power computation for biomedical signal analysis, artificial skin, and brain‐machine interfaces. Organic sensors, organic electrochemical transistors, and memory‐based synaptic devices serve as key components, utilizing ion–electron coupling and nonvolatile synaptic weight modulation for energy‐efficient neuromorphic functions. Reflecting the potential of the aforementioned in‐sensor computing, this review provides a comprehensive analysis of organic‐based in‐sensor computing for wearable and bioelectronic applications. It explores the design and mechanisms of organic synaptic devices, with a focus on memory‐based and organic electrochemical transistor‐based architectures. The fundamental principles of neuromorphic computing are examined, highlighting various organic neuromorphic computing devices and their operational characteristics. Through an in‐depth discussion of recent advancements, challenges, and future perspectives, this review aims to offer valuable insights into the potential of organic electronics in advancing intelligent systems.","author":[{"family":"Lee","given":"Dong"},{"family":"Kim","given":"Woojo"},{"family":"Lee","given":"Eun"},{"family":"Yoo","given":"Hocheon"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1002/aidi.202500053","URL":"https://doi.org/10.1002/aidi.202500053","source":"crossref"},{"id":"doi:10.1149/ma2025-01633064mtgabs","type":"article-journal","title":"Revealing Dual Functionality of Graphene Memristor Circuit for Advanced Neuromorphic Computing","abstract":"Neuromorphic computing [1], inspired by the energy-efficient and parallelized architecture of the human brain, represents a paradigm shift in computing. This approach not only drives innovative hardware designs but also inspires novel network architectures like spiking neural networks (SNNs), which process information as discrete spike events. However, scalability in neuromorphic systems is constrained by the large hardware footprint. Realizing multimodal functionality using the same set of hardware primitives can offer a promising solution to these limitations. In this work, we present a dual encoder-neuron circuit based on a nanoporous graphene (NPG) memristor device [2]. The graphene leaky integrate-and-fire (GLIF) circuit has a versatile multimodal design that combines the roles of a LIF neuron and a spike encoder within the same architecture. This design mimics the dual functionality of biological neurons, which encode and process signals locally. The spike encoding capability of the circuit is further revealed using single and double layer SNN networks for pattern recognition tasks using the Modified National Institute of Standards and Technology (MNIST) dataset. Inset of Figure 1(a) shows the schematic of the fabricated NPG device with a channel length of 100 µm. The device has a lateral structure with NPG channel and gold (Au) electrodes. Figure 1(a) shows the current-voltage (I-V) characteristics of the fabricated NPG device which shows a threshold switching behavior with a wide hysteresis window and a threshold voltage (V th ) of 4.9 V. A behavioral SPICE model is developed based on a voltage-controlled switch model. It is seen from Figure 1(a) that the I-V curves simulated using the SPICE model matches the experimentally observed device switching characteristics. This reveals that the simplified SPICE model can be a good substitute for complex Verilog based models. Based on the SPICE model, we designed a simple neuron circuit with a resistor (R s , 12 kΩ) in series with a parallel combination of a capacitor (C m , 10 nF) and the NPG device (SPICE model). Figure 1(b) shows the variation of the membrane potential (V m ) and the output spike current in response to the input pulse voltage (V in , 6 V). V m shows a clear leaky integration behavior and full reset to zero voltage after the spike response. These are essential attributes of a bioplausible LIF neuron circuit. In order to develop the multimodal GLIF circuit, the output spike frequency of the LIF neuron circuit is plotted as a function of the R s values. This mapping allows pixel values from MNIST input images to be directly translated into normalized conductance values (G/G 0 ). Hence, each pixel is represented as a normalized conductance value, establishing a clear correlation between pixel intensity and spike frequency. This core idea is further used to develop the GLIF encoder based on the same neuron circuit as before. Based on the dual encoder-neuron circuit, we designed a single layer SNN for recognising the handwritten digits from the MNIST dataset. Figure 1(c) shows the spike-encoded images for the MNIST digits “2” and “5”. The probability of spiking in each timestep is proportional to the spiking frequency. The spike encoded images at each time step reveal the contours of the original input images using sparse representation. These sparse matrices highlight the energy efficiency and computational advantages of the GLIF system. The performance of the dual encoder-neuron system is evaluated by training the single layer SNN using both the GLIF encoder and neuron circuit as integral components. The single layer SNN recorded a high pattern recognition accuracy of 90.77% which is comparable to the purely software-based implementation (92.69%). To explore the scalability of the GLIF system, we implemented a double-layer SNN using the GLIF system, achieving a high recognition accuracy of 97.37%. The GLIF circuit exemplifies how reconfigurable primitive compon","author":[{"family":"Mohanan","given":"Kannan"},{"family":"Kim","given":"Chang"},{"family":"Sattari-Esfahlan","given":"Seyed"},{"family":"Cho","given":"Eou"},{"family":"Kymissis","given":"Ioannis"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1149/ma2025-01633064mtgabs","URL":"https://doi.org/10.1149/ma2025-01633064mtgabs","source":"crossref"},{"id":"doi:10.1002/adma.202501813","type":"article-journal","title":"Neuromorphic Light‐Responsive Organic Matter for\n                    <i>in Materia</i>\n                    Reservoir Computing","abstract":"Abstract Materials able to sense and respond to external stimuli by adapting their internal state to process and store information, represent promising candidates for implementing neuromorphic functionalities and brain‐inspired computing paradigms. In this context, neuromorphic systems based on light‐responsive materials enable the use of light as information carrier, allowing to emulate basic functions of the human retina. In this work it is demonstrated that optically‐induced molecular dynamics in azopolymers can be exploited for neuromorphic‐type of data processing in the analog domain and for computing at the matter level (i.e., in materia ). Besides showing that azopolymers can be exploited for data storage, it is demonstrated that the adaptiveness of these materials enables the implementation of synaptic functionalities including short‐term memory, long‐term memory, and visual memory. Results show that azopolymers allow event detection and motion perception, enabling physical implementation of information processing schemes requiring real‐time analysis of spatio‐temporal inputs. Furthermore, it is shown that light‐induced dynamics can be exploited for the in materia implementation of the unconventional computing paradigm denoted as reservoir computing. This work underscores the potential of azopolymers as promising materials for developing adaptive, intelligent photo‐responsive systems that mimic some of the complex processing abilities of biological systems.","author":[{"family":"Lupi","given":"Federico"},{"family":"Roserorealpe","given":"Mateo"},{"family":"Ocarino","given":"Antonio"},{"family":"Frascella","given":"Francesca"},{"family":"Milano","given":"Gianluca"},{"family":"Angelini","given":"Angelo"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1002/adma.202501813","URL":"https://doi.org/10.1002/adma.202501813","source":"crossref"},{"id":"doi:10.1088/2634-4386/ae7f47","type":"article-journal","title":"A review of event-based vision sensor fusion: architectures, evaluation and robustness challenges","abstract":"Abstract Event-based vision is a neuromorphic sensing approach that enables low-latency, energy-aware perception by encoding brightness changes as sparse, asynchronous events rather than dense image frames. Event cameras provide microsecond-level timestamps and high dynamic range, supporting perception during fast motion and challenging illumination. In autonomy, event streams are often fused with complementary sensors such as inertial measurement units, color cameras, light-detection-and-ranging sensors, or radar to improve scale estimation, robustness, and redundancy. Event-centric fusion is challenging because sensor clocks and noise models differ across modalities, event rates are activity-dependent, event representations trade timing fidelity for computational convenience, and hardware cost is often dominated by communication and memory movement. Robustness to missing or degraded modalities is common in deployment but remains inconsistently evaluated. This topical review synthesizes event-based vision fusion around three coupled design decisions: fusion stage, temporal coupling, and event representation. We show how these choices determine whether systems preserve event-driven efficiency or revert to dense processing. As a case study, we review event-based depth estimation, consolidate benchmark results, and analyze missing-modality behavior. The evidence suggests that many gains are reported primarily as accuracy improvements, with limited analysis of modality dropout, event-rate shifts, timing mismatch, preprocessing cost, and efficiency. We relate recurring patterns to optical flow and semantic perception, connect them to neuromorphic hardware and hardware-algorithm co-design, and conclude with benchmarking recommendations and open problems in robustness, scalability, and training of spiking and hybrid models.","author":[{"family":"Devulapally","given":"Anusha"},{"family":"Tumpa","given":"Sadia"},{"family":"Narayanan","given":"Vijaykrishnan"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/2634-4386/ae7f47","URL":"https://doi.org/10.1088/2634-4386/ae7f47","source":"crossref"},{"id":"doi:10.1088/2634-4386/ae978d","type":"article-journal","title":"Learning in spiking neural networks with a calcium-based Hebbian rule for spike timing-dependent plasticity","abstract":"Abstract Understanding how biological neural networks are shaped via local plasticity mechanisms can lead to energy-efficient and self-adaptive information processing systems, which promises to mitigate some of the&amp;#xD;current roadblocks in edge computing systems. While biology makes use of spikes to seamless use both spike timing and mean firing rate to modulate synaptic strength, most models focus on one of the two. In this work, we present a Hebbian local learning rule that models synaptic modification as a function of calcium traces tracking neuronal activity. We show how the rule reproduces results from spike time and spike rate protocols from neuroscientific studies. Moreover, we use the model to train spiking neural networks on MNIST digit recognition to show and explain what sort of mechanisms are needed to learn real-world patterns. We show how our model is sensitive to correlated spiking activity and how this enables it to modulate the learning rate of the network without altering the mean firing rate of the neurons nor the hyparameters of the learning rule. To the best of our knowledge, this is the first work that showcases how spike timing and rate can be complementary in their role of shaping the connectivity of spiking neural networks.","author":[{"family":"Girāo","given":"Willian"},{"family":"Risi","given":"Nicoletta"},{"family":"Geisler","given":"Caroline"},{"family":"Chicca","given":"Elisabetta"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/2634-4386/ae978d","URL":"https://doi.org/10.1088/2634-4386/ae978d","source":"crossref"},{"id":"doi:10.1088/2634-4386/add0db","type":"article-journal","title":"NeuroPong: the event-based camera driven embedded neuromorphic system","abstract":"Abstract Neuromorphic computing is a novel style of computing that features low-power spiking neural networks (SNNs) as the main compute components. It is an event-driven computational paradigm that naturally pairs with event-based cameras and their asynchronous event output. In this work, we present NeuroPong, a novel closed-loop neuromorphic hardware system composed of an event-based camera, a neuromorphic system, and an Atari 2600 console. The system facilitates the implementation of SNN Atari agents capable of playing Atari games in real time using event camera capture as input. We perform a small parameter optimization experiment to examine how software agents translate to hardware, discuss some of the challenges intrinsic to the hardware system, and propose some future improvements of the system and its components.","author":[{"family":"Rizzo","given":"Charles"},{"family":"Gullett","given":"Bryson"},{"family":"Crumley","given":"Alex"},{"family":"Marcum","given":"Maxwell"},{"family":"Hyman","given":"Mason"},{"family":"Earheart-Brown","given":"Carter"},{"family":"Steed","given":"Julia"},{"family":"Standaert","given":"Frank"},{"family":"Schuman","given":"Catherine"},{"family":"Plank","given":"James"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1088/2634-4386/add0db","URL":"https://doi.org/10.1088/2634-4386/add0db","source":"crossref"},{"id":"doi:10.47392/irjaeh.2025.0640","type":"article-journal","title":"Neuromorphic Computing for Edge AI","abstract":"The Neuromorphic computing and Edge AI (Artificial intelligence) are two inter related concepts that have left a lasting impression in recent years. As a result of the ability of neuromorphic computing to imitate the capability of the brain for processing information in an energy saving and extremely parallel way. Similarly, in the same way edge AI refers to employing AI algorithms or models straightly on servers or cloud platforms. Accordingly, to achieve this, edge AI and Neuromorphic computing are merged due to the parallelism of neuromorphic computing which goes hand in hand with edge AI applications. As a result, the present SLR concentrates on examining studies highlighting neuromorphic computing for edge AI, dissimilarities in traditional and neuromorphic computing, different chips utilized for neuromorphic computing and application of neuromorphic computing for edge AI.","author":[{"family":"Ghanti","given":"Dr"},{"family":"Patil","given":"Nikita"},{"family":"Nikhitasalgar"}],"issued":{"date-parts":[[2025]]},"DOI":"10.47392/irjaeh.2025.0640","URL":"https://doi.org/10.47392/irjaeh.2025.0640","source":"crossref"},{"id":"doi:10.1088/2634-4386/addb6d","type":"article-journal","title":"Hardware-friendly implementation of physical reservoir computing with CMOS-based time-domain analog spiking neurons","abstract":"Abstract This paper introduces an analog spiking neuron that utilizes time-domain information, i.e. a time interval of two signal transitions and a pulse width, to construct a spiking neural network (SNN) for a hardware-friendly physical reservoir computing (RC) on a complementary metal-oxide-semiconductor platform. A neuron with leaky integrate-and-fire is realized by employing two voltage-controlled oscillators with opposite sensitivities to the internal control voltage, and the neuron connection structure is restricted by the use of only 4 neighboring neurons on a 2-dimensional plane to feasibly construct a regular network topology. Such a system enables us to compose an SNN with a counter-based readout circuit, which simplifies the hardware implementation of the SNN. Moreover, another technical advantage thanks to the bottom–up integration is the capability of dynamically capturing every neuron state in the network, which can significantly contribute to finding guidelines on how to enhance the performance for various computational tasks in temporal information processing. Diverse nonlinear physical dynamics needed for RC can be realized by collective behavior through dynamic interaction between neurons, like coupled oscillators, despite the simple network structure. With behavioral system-level simulations, we demonstrate physical RC through short-term memory and exclusive OR tasks, and the spoken digit recognition task with an accuracy of 97.7% as well. Our system is considerably feasible for practical applications and can also be a useful platform for studying the mechanism of physical RC.","author":[{"family":"Kimura","given":"Nanako"},{"family":"Duran","given":"Ckristian"},{"family":"Byambadorj","given":"Zolboo"},{"family":"Nakane","given":"Ryosho"},{"family":"Iizuka","given":"Tetsuya"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1088/2634-4386/addb6d","URL":"https://doi.org/10.1088/2634-4386/addb6d","source":"crossref"},{"id":"doi:10.1088/2634-4386/adc0b8","type":"article-journal","title":"Probabilistic computing with percolating nanoparticle networks using experimental data with signatures of criticality","abstract":"Abstract Percolating networks of nanoparticles (PNNs) are promising systems for neuromorphic computing due to their brain-like network structure and dynamics. In particular, electrical spiking in PNNs meets criteria for criticality, which is thought to be the operating point for biological brains and associated with optimal computation. Previous work showed through simulations that spiking PNNs can be used as the core stochastic component in a probabilistic computing scheme. Here, we demonstrate a route to experimental implementation of an integer factorization algorithm. We outline an important modification to the algorithm previously used and demonstrate factorization of up to six-digit integers. Finally, we explore the effect of criticality in the context of the integer factorization task by comparing critical and non-critical systems. We show significant differences in the probability distribution of states generated by the critical and non-critical systems, though for the task considered, critical systems provide no advantage.","author":[{"family":"Studholme","given":"Sofie"},{"family":"Mallinson","given":"Joshua"},{"family":"Steel","given":"Jamie"},{"family":"Brown","given":"Simon"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1088/2634-4386/adc0b8","URL":"https://doi.org/10.1088/2634-4386/adc0b8","source":"crossref"},{"id":"doi:10.1088/2634-4386/ae441f","type":"article-journal","title":"RCbench: a unified framework for benchmarking reservoir computing systems","abstract":"Abstract Reservoir computing (RC) is a computational framework where a fixed dynamical reservoir projects an input into a higher-dimensional state that is then analyzed by a readout , which is trained to map the reservoir state into the desired output. While the conventional RC paradigm is based on dynamics of in-silico implemented recurrent neural networks, this computing paradigm can be efficiently implemented in hardware by exploiting dynamics of a wide range of physical systems in a paradigm denoted as Physical RC (PRC), attracting interest from a broader research community spanning from computer scientists to physicists, and material scientists. Here, we present RCbench, an open-source RC benchmark toolkit that implements a standardized and comprehensive suite for benchmarking computational reservoir models and physical implementations of RC. RCbench integrates widely recognized metrics such as Memory capacity, Nonlinear autoregressive moving average of order N, Kernel rank , and generalization rank, along with nonlinear transformation tasks. It also allows testing and comparing different readout algorithms, the evaluation of computational capabilities with diverse accuracy metrics, and includes feature selection methods to unravel the effect of specific reservoir outputs on computational performances. In particular, the toolkit enables easy benchmarking of PRC systems, providing a comprehensive benchmark tool that can be easily integrated with experimental data acquisition processes. By standardizing performance assessments, RCbench aims to facilitate inter-study comparisons and to accelerate the exploration, characterization and optimization of RC systems.","author":[{"family":"Pilati","given":"Davide"},{"family":"Ceni","given":"Andrea"},{"family":"Michieletti","given":"Fabio"},{"family":"Gallicchio","given":"Claudio"},{"family":"Ricciardi","given":"Carlo"},{"family":"Milano","given":"Gianluca"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/2634-4386/ae441f","URL":"https://doi.org/10.1088/2634-4386/ae441f","source":"crossref"},{"id":"doi:10.1088/2634-4386/add293","type":"article-journal","title":"Enhancing temporal learning in recurrent spiking networks for neuromorphic applications","abstract":"Abstract Training Recurrent Spiking Neural Networks (RSNNs) with binary spikes for tasks of extended time scales presents a challenge due to the amplified vanishing gradient problem during back propagation through time. This paper introduces three crucial elements that significantly enhance the memory and capabilities of RSNNs, with a strong emphasis on compatibility with hardware and neuromorphic systems. Firstly, we incorporate neuron-level synaptic delays, which not only allow the gradient to skip time steps but also reduce the overall neuron population’s firing rate. Subsequently, we apply a biologically inspired branching factor regularization rule to stabilize the network’s dynamics and make training easier by incorporating a time-local error in the loss function. Lastly, we modify a commonly used surrogate gradient function by increasing its support to facilitate learning over longer timescales when using binary spikes. By integrating these three innovative elements, we not only resolve several complex benchmarks but also achieve state-of-the-art results on the spiking permuted sequential MNIST task (psMNIST), showcasing the practicality and relevance of our approach for digital and analog neuromorphic systems.","author":[{"family":"Balafrej","given":"Ismael"},{"family":"Bahadi","given":"Soufiyan"},{"family":"Rouat","given":"Jean"},{"family":"Alibart","given":"Fabien"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1088/2634-4386/add293","URL":"https://doi.org/10.1088/2634-4386/add293","source":"crossref"},{"id":"doi:10.70177/jsca.v3i3.3331","type":"article-journal","title":"COMPUTING AT THE EDGE: THE ROLE OF NEUROMORPHIC CHIPS IN INTELLIGENT ROBOTICS","abstract":"The deployment of autonomous mobile robots in resource-constrained environments is currently impeded by the excessive power consumption and latency bottlenecks of traditional Von Neumann architectures. This study investigates the efficacy of neuromorphic computing as a hardware solution for low-power, low-latency edge intelligence, specifically focusing on obstacle avoidance and navigational endurance. A quantitative comparative analysis was conducted benchmarking a Spiking Neural Network (SNN) based control architecture against standard embedded GPU solutions, utilizing event-based vision sensors to evaluate energy efficiency, inference latency, and task success rates. Empirical results demonstrate that the neuromorphic architecture achieved a twenty-fold reduction in power consumption (0.25 W) and sub-millisecond latency, significantly outperforming synchronous baselines while maintaining a 98.2% navigational success rate. The findings validate event-driven processing as a superior paradigm for edge robotics, offering a sustainable path toward \"Green Robotics\" with extended operational autonomy independent of cloud connectivity.","author":[{"family":"Keolavong","given":"Manivone"},{"family":"Vong","given":"Soneva"},{"family":"Phoutthavong","given":"Thipphavone"}],"issued":{"date-parts":[[2026]]},"DOI":"10.70177/jsca.v3i3.3331","URL":"https://doi.org/10.70177/jsca.v3i3.3331","source":"crossref"},{"id":"doi:10.1007/s40820-026-02322-5","type":"article-journal","title":"Analysis and Applications of Neuromorphic Memristors in Artificial Intelligence Computing","abstract":"Abstract Artificial intelligence (AI) has advanced rapidly in recent years and has been widely applied in healthcare, intelligent sensing, machine perception, and image recognition. Neuromorphic computing, inspired by the structure and operating principles of the human brain, has emerged as a promising paradigm for building efficient, low-power, and adaptive information processing systems. In this review, we summarize the development of neuromorphic memristors from the perspectives of biological inspiration, representative material systems, device architectures, performance metrics, and artificial intelligence-oriented applications. Different from previous reviews that mainly focus on memristor materials, switching mechanisms, or neuromorphic functions separately, this article further emphasizes the relationships between memristor characteristics and distinct AI-oriented tasks, including AI acceleration, neuromorphic computing, intelligent sensing, and human–machine interaction. Finally, the major challenges and future opportunities of neuromorphic memristors are discussed from the viewpoints of device optimization, system integration, scalability, and practical application.","author":[{"family":"Wang","given":"Jianhui"},{"family":"Chen","given":"Qingxin"},{"family":"Zhang","given":"Jinhao"},{"family":"Meng","given":"Jialin"},{"family":"Wang","given":"Tianyu"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1007/s40820-026-02322-5","URL":"https://doi.org/10.1007/s40820-026-02322-5","source":"europepmc"},{"id":"doi:10.1088/2634-4386/ae4a47","type":"article-journal","title":"The more the merrier: running multiple neuromorphic components on-chip for robotic control","abstract":"Abstract It has long been realized that neuromorphic hardware offers benefits for the domain of robotics such as low energy, low latency, as well as unique methods of learning. In aiming for more complex tasks, especially those incorporating multimodal data, one hurdle continuing to prevent their realization is an inability to orchestrate multiple networks on neuromorphic hardware without resorting to off-chip process management logic. To address this, we show a first example of a pipeline for vision-based robot control in which numerous complex networks can be run entirely on hardware via the use of a spiking neural state machine for process orchestration. The pipeline is validated on the Intel Loihi 2 research chip. We show that all components can run concurrently on-chip in the milliwatt regime at latencies competitive with the state-of-the-art. An equivalent network on simulated hardware is shown to accomplish robotic arm plug insertion in simulation, and the core elements of the pipeline are additionally tested on a real robotic arm.","author":[{"family":"Eames","given":"Evan"},{"family":"Kannan","given":"Priyadarshini"},{"family":"Sangouard","given":"Ronan"},{"family":"Plank","given":"Philipp"},{"family":"Hajizada","given":"Elvin"},{"family":"Palinauskas","given":"Gintautas"},{"family":"Amaya","given":"Lana"},{"family":"Neumeier","given":"Michael"},{"family":"Sharma","given":"Sai"},{"family":"Toth","given":"Marcella"},{"family":"Sarkar","given":"Prottush"},{"family":"Arnim","given":"Axel"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/2634-4386/ae4a47","URL":"https://doi.org/10.1088/2634-4386/ae4a47","source":"crossref"},{"id":"doi:10.1088/2634-4386/ae3b6c","type":"article-journal","title":"Neuromorphic hardware based on memristive nanodevices for seizure detection and recovery","abstract":"Abstract During the last decades, neuromorphic engineers have developed specific hardware designed to build efficient computing systems inspired by the structure of the human brain. The emergence of nanoscale memristors provided these systems with a new component which can approximately emulate the behavior of synaptic connections, improving the capability to implement in-situ learning algorithms like spike-timing-dependent plasticity. Meanwhile, neuro-inspired biomimetic platforms have been developed to directly interface with biological neurons, allowing to record and process neural signals like local field potentials (LFP). Combining both technologies, it would be possible to implant intracraneal electroencephalography electrodes with a neuromorphic chip which could sense signals from epileptic tissues and provide stimulation to prevent seizures in a closed-loop setup. In this work, we use a neuromorphic hardware platform with memristors to process LFP activity generated by an artificial neural mass model (ANMM) of the hippocampal loop implemented on a microcontroller for real-time operation, showing that the memristor system can learn correlations between neurons to detect seizures and eventually prevent them. This closed-loop ANMM-memristor crossbar interaction demonstration paves the way for trying a similar setup, replacing the ANMM with biological epileptic tissues.","author":[{"family":"Díez-De-Los-Ríos","given":"Iván"},{"family":"Farsani","given":"Javad"},{"family":"Ricci","given":"Saverio"},{"family":"Bridarolli","given":"Davide"},{"family":"Camuñas-Mesa","given":"Luis"},{"family":"Subramaniyam","given":"Narayan"},{"family":"Tanskanen","given":"Jarno"},{"family":"Hyttinen","given":"Jari"},{"family":"Ielmini","given":"Daniele"},{"family":"Serrano-Gotarredona","given":"Teresa"},{"family":"Linares-Barranco","given":"Bernabé"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/2634-4386/ae3b6c","URL":"https://doi.org/10.1088/2634-4386/ae3b6c","source":"crossref"},{"id":"doi:10.70177/scientia.v2i4.2385","type":"article-journal","title":"FERROELECTRIC THIN FILMS FOR NEUROMORPHIC COMPUTING: SYNTHESIS, CHARACTERIZATION, AND DEVICE INTEGRATION","abstract":"The limitations of conventional von Neumann computing architectures in handling complex, data-intensive tasks have spurred significant interest in brain-inspired neuromorphic computing. A critical challenge in this field is the development of hardware that can efficiently emulate the synaptic plasticity of biological neurons. This study focuses on the synthesis, characterization, and integration of ferroelectric thin films, specifically hafnium zirconium oxide (HZO), as a promising material platform for creating artificial synaptic devices. The primary objective was to fabricate high-quality HZO thin films and demonstrate their capacity to mimic key synaptic functions. HZO films were synthesized using pulsed laser deposition, followed by comprehensive characterization of their structural, ferroelectric, and electrical properties using XRD, PFM, and I-V measurements. The optimized films were then integrated into two-terminal memristive device structures. The resulting devices successfully exhibited essential synaptic behaviors, including potentiation, depression, and spike-timing-dependent plasticity (STDP), with low energy consumption per synaptic event. The gradual and controllable modulation of ferroelectric domain switching was identified as the core mechanism enabling this analog-like resistance modulation.","author":[{"family":"Huda","given":"Nurul"},{"family":"Zaki","given":"Amin"},{"family":"Chai","given":"Nong"},{"family":"Shofiah","given":"Siti"}],"issued":{"date-parts":[[2026]]},"DOI":"10.70177/scientia.v2i4.2385","URL":"https://doi.org/10.70177/scientia.v2i4.2385","source":"crossref"},{"id":"doi:10.1088/1674-4926/24110006","type":"article-journal","title":"Revolutionizing neuromorphic computing with memristor-based artificial neurons","abstract":"Abstract As traditional von Neumann architectures face limitations in handling the demands of big data and complex computational tasks, neuromorphic computing has emerged as a promising alternative, inspired by the human brain's neural networks. Volatile memristors, particularly Mott and diffusive memristors, have garnered significant attention for their ability to emulate neuronal dynamics, such as spiking and firing patterns, enabling the development of reconfigurable and adaptive computing systems. Recent advancements include the implementation of leaky integrate-and-fire neurons, Hodgkin−Huxley neurons, optoelectronic neurons, and time-surface neurons, all utilizing volatile memristors to achieve efficient, low-power, and highly integrated neuromorphic systems. This paper reviews the latest progress in volatile memristor-based artificial neurons, highlighting their potential for energy-efficient computing and integration with artificial synapses. We conclude by addressing challenges such as improving memristor reliability and exploring new architectures to advance memristor-based neuromorphic computing.","author":[{"family":"Chen","given":"Yanning"},{"family":"Zhang","given":"Guobin"},{"family":"Liu","given":"Fang"},{"family":"Wu","given":"Bo"},{"family":"Deng","given":"Yongfeng"},{"family":"Gao","given":"Dawei"},{"family":"Zhang","given":"Yishu"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1088/1674-4926/24110006","URL":"https://doi.org/10.1088/1674-4926/24110006","source":"crossref"},{"id":"doi:10.5772/acrt.deposit.c.8440846.v1","type":"article-journal","title":"Neuromorphic Computing in Outer Space: Intel Loihi Deployed On-Satellite for Spike Based Software Defined Communication","abstract":"&lt;table&gt;&lt;tr&gt;&lt;td&gt;&lt;p dir=\"ltr\"&gt;In space exploration and communication technology, software-defined networking (SDN) architectures significantly improve communication throughput and latency. This study developed a Q-learning-based continuous learning capable cognitive agent to select optimal routing options for a dynamically changing networking environment. With the dynamically changing network link latency, the cognitive agent achieved a 90% success rate in routing packets. The agent was executed on neuromorphic hardware called Intel Loihi. However, due to the system's power limitations, a simplified version of the agent was engineered to launch into space aboard a CubeSat. The CubeSat was launched in January 2022, making a historic milestone as the first launch of a neuromorphic system into space. The developed applications were successfully executed in space.&lt;/p&gt;&lt;/td&gt;&lt;/tr&gt;&lt;/table&gt;","author":[{"family":"Rahman","given":"Nayim"},{"family":"Yakopcic","given":"Chris"},{"family":"Taha","given":"Tarek"},{"family":"Lent","given":"Ricardo"},{"family":"Briones","given":"Janette"},{"family":"Chelmins","given":"David"},{"family":"Dudukovitch","given":"Rachel"},{"family":"Smith","given":"Aaron"},{"family":"Gannon","given":"Adam"},{"family":"Lowry","given":"Michael"},{"family":"Murbach","given":"Marcus"},{"family":"Salas","given":"Alejandro"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5772/acrt.deposit.c.8440846.v1","URL":"https://doi.org/10.5772/acrt.deposit.c.8440846.v1","source":"crossref"},{"id":"doi:10.64091/aticl.2025.000232","type":"article-journal","title":"Harnessing Synaptic Plasticity for Real-Time Edge Processing in Neuromorphic Computing Architectures","abstract":"Brain-inspired neuromorphic computing, which is based on the neuronal structure of the brain, provides a revolutionary paradigm for real-time edge computing that is efficient in terms of energy consumption. The purpose of this work is to investigate the role of synaptic plasticity, specifically spike-timing-dependent plasticity (STDP), in increasing computational efficiency in neuromorphic edge device topologies. A new architecture is shown here that uses STDP to facilitate on-chip learning and adaptation to a wide range of sensory inputs. Additionally, this design features reduced cloud processing, lower latency, and improved energy efficiency. The classification of streams online is the primary focus of our efforts. This is a fundamental operation performed in most edge operations, including autonomous navigation and health monitoring on wearable devices. Our model is trained and evaluated in a Python-based simulation environment, and Brian2 is used to simulate neuronal dynamics. The performance of the novel architecture is evaluated using the Spiking Heidelberg Digits (SHD) benchmark, an appropriate metric for spike-based classification of auditory samples. This architecture demonstrates higher processing speed and energy efficiency than traditional von Neumann architectures, achieving 96.4% classification accuracy and a mean power consumption of 2.7 milliwatts.","author":[{"family":"Shah","given":"Jashkumar"},{"family":"Desai","given":"Aashna"},{"family":"Gramopadhye","given":"Rugved"},{"family":"Gada","given":"Tina"},{"family":"Das","given":"Debabrata"},{"family":"Rajest","given":"SS"}],"issued":{"date-parts":[[2026]]},"DOI":"10.64091/aticl.2025.000232","URL":"https://doi.org/10.64091/aticl.2025.000232","source":"crossref"},{"id":"doi:10.1088/2634-4386/adcf46","type":"article-journal","title":"Range and angle estimation with spiking neural resonators for FMCW radar","abstract":"Abstract Automotive radar systems face the challenge of managing high sampling rates and large data bandwidth while complying with stringent real-time and energy efficiency requirements. Neuromorphic computing offers promising solutions because of its inherent energy efficiency and parallel processing capacity. Yet, most sensor systems, such as radars, do not produce suitable data for further neuromorphic processing of the data. This research presents a novel spiking neuron model for signal processing of frequency-modulated continuous wave (FMCW) radars that outperforms the state-of-the-art spectrum analysis algorithms in latency and data bandwidth. These spiking neural resonators are based on the resonate-and-fire neuron model and optimized to dynamically process raw radar data while simultaneously emitting an output in the form of spikes. We designed the first neuromorphic neural network consisting of these spiking neural resonators that estimates range and angle from FMCW radar data, evaluated the network on simulated automotive datasets and compared the results with a state-of-the-art pipeline for radar processing. The proposed neuron model significantly reduces the processing latency compared to traditional frequency analysis algorithms, such as the Fourier transformation (FT), which needs to sample and store entire data frames before processing. The evaluations demonstrate that these spiking neural resonators achieve state-of-the-art detection accuracy while emitting spikes simultaneously to processing and transmitting only 0.02% of the data compared to a float-32 FT. The results showcase the potential for neuromorphic signal processing for FMCW radar systems and pave the way for designing neuromorphic radar sensors.","author":[{"family":"Reeb","given":"Nico"},{"family":"Lopez-Randulfe","given":"Javier"},{"family":"Dietrich","given":"Robin"},{"family":"Knoll","given":"Alois"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1088/2634-4386/adcf46","URL":"https://doi.org/10.1088/2634-4386/adcf46","source":"crossref"},{"id":"doi:10.1088/2634-4386/ae98af","type":"article-journal","title":"Why temporal spike order reversal drops spiking network accuracy and how to partially mitigate it","abstract":"Abstract Spiking neural networks (SNNs) offer an energy-efficient alternative to traditional neural networks by utilizing discrete, temporally precise spike events. However, this study identifies a critical vulnerability in SNNs on recently established bit-based codes: consistent performance degradation when temporal spike encoding orders are reversed, such as using least-significant-bit ordering instead of most-significant-bit. We theoretically formalize this phenomenon as premature state annihilation, wherein early noisy spikes in information-discordant encodings trigger hard resets in leaky integrate-and-fire (LIF) neurons. These resets erase accumulated membrane state and, because the effective temporal influence of an input is largest for early timesteps, leave the backpropagated learning signal concentrated where the information is not. We measure the per-timestep class-mutual-information profile of six encodings directly, without reference to network accuracy, and show that the resulting concordance ordering predicts the observed degradation. While dense codes like weighted phase encoding suffer catastrophic drops (up to 48 % ), sparse codes like time-to-first-spike remain robust, and exchangeable rate codes are provably invariant to reversal. We evaluate several mitigation strategies, finding that parametric LIF (PLIF) neurons and aggressive membrane leakage significantly recover performance by adapting to or suppressing early noise.","author":[{"family":"Luu","given":"Nhan"},{"family":"Luu","given":"Duong"},{"family":"Pham","given":"Nam"},{"family":"Truong","given":"Thang"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/2634-4386/ae98af","URL":"https://doi.org/10.1088/2634-4386/ae98af","source":"crossref"},{"id":"doi:10.26599/nr.2026.94908881","type":"article-journal","title":"Robust Sb\n                    <sub>2</sub>\n                    Se\n                    <sub>3</sub>\n                    memristors via pressure-modulated growth for noise-resilient neuromorphic computing","abstract":"Abstract High-fidelity neuromorphic computing requires synaptic hardware that balances analog precision with array-level noise immunity, yet suppressing leakage currents in chalcogenide crossbars often necessitates complex interface engineering. Here, a robust Ag/Sb2Se3/ITO synapse is reported, fabricated via a pressure-modulated rapid thermal evaporation (RTE) strategy that targets intrinsic defect control. Crucially, this thermodynamic optimization preserves the ultralow intrinsic carrier concentration (~1014 cm-3) of the Sb2Se3 functional layer, physically prohibiting background leakage pathways without requiring additional buffer layers. Consequently, the device demonstrates highly uniform analog switching behavior, a substantial ON/OFF ratio (&gt;105), and low set/reset voltages. These characteristics effectively suppress sneak path currents and maximize the sensing margins within the crossbar array. At the system level, physics-based neural networks achieve 96.3% accuracy on MNIST, maintaining exceptional robustness against severe salt-and-pepper noise. Furthermore, we demonstrate that the hardware can execute complex motion perception algorithms, successfully extracting clear motion edges in dynamic spatiotemporal scenarios. This work establishes intrinsic carrier concentration modulation as a minimalist yet powerful paradigm for next-generation noise-resilient edge intelligence.","author":[{"family":"Wang","given":"Zining"},{"family":"Song","given":"Chensi"},{"family":"Chen","given":"Huanyu"},{"family":"Liu","given":"Xinsheng"},{"family":"Zhang","given":"Xi"},{"family":"Zhu","given":"Jichun"},{"family":"Li","given":"Huilin"}],"issued":{"date-parts":[[2026]]},"DOI":"10.26599/nr.2026.94908881","URL":"https://doi.org/10.26599/nr.2026.94908881","source":"crossref"},{"id":"doi:10.1088/2634-4386/adad10","type":"article-journal","title":"Towards neuromorphic compression based neural sensing for next-generation wireless implantable brain machine interface","abstract":"Abstract This work introduces a neuromorphic compression based neural sensing architecture with address-event representation inspired readout protocol for massively parallel, next-gen wireless implantable brain machine interface (iBMI). The architectural trade-offs and implications of the proposed method are quantitatively analyzed in terms of compression ratio (CR) and spike information preservation. For the latter, we used metrics such as root-mean-square error and correlation coefficient (CC) between the original and recovered signals to assess the effect of neuromorphic compression on the spike shape. Furthermore, we use accuracy, sensitivity, and false detection rate to understand the effect of compression on downstream iBMI tasks, specifically, spike detection. We demonstrate that a data CR of 15–265 per channel can be achieved by transmitting address-event pulses for two different biological datasets. The CR further increases to 200– 50 K per channel, 50 × more than in prior works, by the selective transmission of event pulses corresponding to neural spikes. A CC of ≈0.9 and spike detection accuracy of over 90% were obtained for the worst-case analysis involving 10 K -channel simulated recording and typical analysis using 100 or 384-channel real neural recordings. We also analyzed the collision handling capability for up to 10K channels and observed no significant error, indicating the scalability of the proposed pipeline. We also present initial results to show the ability of intention decoders to work directly on the events generated by the neuromorphic front-end.","author":[{"family":"Mohan","given":"Vivek"},{"family":"Tay","given":"Wee"},{"family":"Basu","given":"Arindam"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1088/2634-4386/adad10","URL":"https://doi.org/10.1088/2634-4386/adad10","source":"crossref"},{"id":"doi:10.1088/1674-4926/25120015","type":"article-journal","title":"High hall mobility carbonized steamed buns-based polymer memristor for neuromorphic computing and image recognition","abstract":"Abstract The development of new n-type semiconductors is crucial for the further advancement of electronic and optoelectronic devices. Steamed buns, anciently known as \"man tou\", mainly made of wheat flour and are one of the staple foods for Chinese people. After being subjected to high-temperature treatment, the steamed buns transformed into carbonized steamed buns (CSB) with porous nanostructures, which exhibit a Hall mobility of up to 1.62 cm 2 /(V·s), far greater than C 60 (1.5 × 10 −3 −2.5 × 10 −2 cm 2 /(V·s)), PCBM (2.0 × 10 −7 cm 2 /(V·s)) and many polymer semiconductors (~10 −6 −10 −2 cm 2 /(V·s)). A CSB-based bulk heterojunction memristor with a configuration of ITO/the CSB: PVK blends/Al is successfully fabricated. The device shows outstanding history dependent memristive switching performance, with 35 distinguishable conductance states, at a small sweep voltage range of ±1 V. An achieved production yield reaches up to 89%. Upon being subjected to consecutive positive or negative voltage sweeps, the current flowing through the device can be modulated continuously. When the 15 consecutive pulse voltages (pulse amplitude: 0.1 V; pulse width:10 μ s, pulse period: 20 μ s) were applied to the device, the observed total power consumption was about 7.63 nJ, suggesting a potential in low-energy neuromorphic computing applications. As expected, both the CSB and PVK do not exhibit any memristive effect under the same experimental condition. Utilizing the characteristic that the device can linearly adjust the weights, a simple convolutional neural network for traffic sign recognition was successfully constructed. After 300 rounds of training, the achieved recognition accuracy rate reached 88.77%. This work not only provides a new approach for developing low-cost and readily available organic semiconductors with high Hall mobility, but also offers a new idea for the subsequent development of high-performance artificial synapses and optoelectronic devices using carbonized steamed buns.","author":[{"family":"Zhang","given":"Chenjian"},{"family":"Liu","given":"Jiaxuan"},{"family":"Zhang","given":"Dongliang"},{"family":"Qin","given":"Tianhao"},{"family":"Wang","given":"Kexin"},{"family":"He","given":"Haidong"},{"family":"Chen","given":"Yu"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/1674-4926/25120015","URL":"https://doi.org/10.1088/1674-4926/25120015","source":"crossref"},{"id":"doi:10.1088/2634-4386/ae6f46","type":"article-journal","title":"A spiking neural network implementation of Gaussian belief propagation","abstract":"Abstract Bayesian inference offers a principled account of information processing in natural agents. However, it remains an open question how neural mechanisms perform their abstract operations. We investigate a hypothesis where a distributed form of Bayesian inference, namely message passing on factor graphs, is performed by a simulated network of leaky-integrate-and-fire neurons. Specifically, we perform Gaussian belief propagation by encoding messages that come into factor nodes as spike-based signals, propagating these signals through a spiking neural network and decoding the spike-based signal back to an outgoing message. Three core linear operations, equality (branching), addition, and multiplication, are realized in networks of leaky integrate-and-fire models. Validation against the standard sum-product algorithm shows accurate message updates, while applications to Kalman filtering and Bayesian linear regression demonstrate the framework’s potential for both static and dynamic inference tasks. Our results provide a step toward biologically grounded, neuromorphic implementations of probabilistic reasoning.","author":[{"family":"Adamiat","given":"Sepideh"},{"family":"Kouw","given":"Wouter"},{"family":"Vries","given":"Bert"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/2634-4386/ae6f46","URL":"https://doi.org/10.1088/2634-4386/ae6f46","source":"crossref"},{"id":"doi:10.1088/2634-4386/ae0aee","type":"article-journal","title":"High-speed, ultra-low-power, and robust superconductive neuron with ReLU activation","abstract":"Abstract We propose a novel ultra-high-speed neuron device utilizing a superconductive single flux quantum (SFQ) circuit to realize an ideal rectified linear unit (ReLU) activation function. This circuit generates quantum-accurate voltage output through frequency conversion within the SFQ digital circuit. A significant advantage of this design is its combination of high-speed and ultra-low-power operation with inherent tolerance to device parameter variations. This crucial feature mitigates performance degradation often observed in large-scale neural networks that rely on analog neuron circuits susceptible to characteristic variation of neuron devices. We designed and implemented the proposed neuron circuit using a 10 kA cm −2 Nb four-layer 1.0 μ m fabrication process. Experimental measurements at 4.2 K confirmed correct operation up to approximately 41.2 GHz input. Results from multiple chips successfully demonstrated ideal ReLU input–output characteristics, showcasing both the high-speed nature of the device and the scalability and robustness of our neuron circuits for next-generation artificial neural network hardware.","author":[{"family":"Ueno","given":"Yuto"},{"family":"Hironaka","given":"Yuki"},{"family":"Yoshikawa","given":"Nobuyuki"},{"family":"Yamanashi","given":"Yuki"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1088/2634-4386/ae0aee","URL":"https://doi.org/10.1088/2634-4386/ae0aee","source":"crossref"},{"id":"doi:10.5772/acrt.deposit.32253528","type":"article-journal","title":"Neuromorphic Computing in Outer Space: Intel Loihi Deployed On-Satellite for Spike-Based Software-Defined Communication - Presentation","abstract":"&lt;p dir=\"ltr\"&gt;In space exploration and communication technology, software-defined networking architectures significantly improve the communications throughput and latency. This study developed a cognitive agent based on Q-learning with continuous learning capabilities to select optimal routing options in a dynamically changing networking environment. This proposed method achieves up to 90% reduction in measured Internet Control Message Protocol (ICMP) round-trip time compared to the Dijkstra shortest-path baseline under identical test conditions. Under dynamically changing network link latency conditions, the cognitive agent achieved a 90% success rate in routing packets in laboratory experiments. The agent was executed on neuromorphic hardware called Intel Loihi. However, due to the system’s power limitations, a simplified version of the agent was engineered to launch into space aboard a CubeSat. The CubeSat was launched in January 2022, making a historic milestone as the first launch of a neuromorphic system into space. The developed applications were successfully executed in space.&lt;/p&gt;","author":[{"family":"Rahman","given":"Nayim"},{"family":"Yakopcic","given":"Chris"},{"family":"Taha","given":"Tarek"},{"family":"Lent","given":"Ricardo"},{"family":"Briones","given":"Janette"},{"family":"Chelmins","given":"David"},{"family":"Dudukovich","given":"Rachel"},{"family":"Smith","given":"Aaron"},{"family":"Gannon","given":"Adam"},{"family":"Lowry","given":"Michael"},{"family":"Murbach","given":"Marcus"},{"family":"Salas","given":"Alejandro"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5772/acrt.deposit.32253528","URL":"https://doi.org/10.5772/acrt.deposit.32253528","source":"crossref"},{"id":"doi:10.5772/acrt.deposit.32253477","type":"article-journal","title":"Neuromorphic Computing in Outer Space: Intel Loihi Deployed On-Satellite for Spike-Based Software-Defined Communication - Python Code","abstract":"&lt;p dir=\"ltr\"&gt;In space exploration and communication technology, software-defined networking architectures significantly improve the communications throughput and latency. This study developed a cognitive agent based on Q-learning with continuous learning capabilities to select optimal routing options in a dynamically changing networking environment. This proposed method achieves up to 90% reduction in measured Internet Control Message Protocol (ICMP) round-trip time compared to the Dijkstra shortest-path baseline under identical test conditions. Under dynamically changing network link latency conditions, the cognitive agent achieved a 90% success rate in routing packets in laboratory experiments. The agent was executed on neuromorphic hardware called Intel Loihi. However, due to the system’s power limitations, a simplified version of the agent was engineered to launch into space aboard a CubeSat. The CubeSat was launched in January 2022, making a historic milestone as the first launch of a neuromorphic system into space. The developed applications were successfully executed in space.&lt;/p&gt;","author":[{"family":"Rahman","given":"Nayim"},{"family":"Yakopcic","given":"Chris"},{"family":"Taha","given":"Tarek"},{"family":"Lent","given":"Ricardo"},{"family":"Briones","given":"Janette"},{"family":"Chelmins","given":"David"},{"family":"Dudukovich","given":"Rachel"},{"family":"Smith","given":"Aaron"},{"family":"Gannon","given":"Adam"},{"family":"Lowry","given":"Michael"},{"family":"Murbach","given":"Marcus"},{"family":"Salas","given":"Alejandro"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5772/acrt.deposit.32253477","URL":"https://doi.org/10.5772/acrt.deposit.32253477","source":"crossref"},{"id":"doi:10.1063/5.0246384","type":"article-journal","title":"Resistive switching and synaptic characteristics in ZnO@β-SiC\n                    composite-based RRAM for neuromorphic computing","abstract":"The advancement of neuromorphic computing in resistive random-access memory (RRAM) is crucial for the rapid expansion of artificial intelligence. Conventional metal oxide-based RRAM faces challenges in mimicking synaptic activity, leading to the exploration of new resistive switching (RS) materials. This study introduces a ZnO@β-SiC composite-based RRAM device that exhibits biological synapse-like functionality. The device shows self-compliance and forming-free RS at ∼0.8 V, where it also mimics synaptic responses such as potentiation, depression, and paired-pulse facilitation at low voltage stimuli (∼0.6 V, 40 ms) with learning and forgetting behavior. Moreover, the synaptic plasticity is analyzed through spike rate dependent plasticity, spike number dependent plasticity, and spike time dependent plasticity. Further, the transition from short-term plasticity to long-term plasticity is observed under more training pulses and lower interval stimuli. The observed RS mechanism and synaptic functionalities are explained by the electric field-driven formation and dissolution of conducting filaments of oxygen vacancies. The chemical properties and local electronic structure have been examined by x-ray photoelectron spectroscopy and x-ray absorption spectroscopy. To elucidate the atomistic memristive behavior and the contribution of different electrical parameters in RRAM, detailed conductive atomic-force microscopy and impedance analysis have been carried out.","author":[{"family":"Santra","given":"Bisweswar"},{"family":"Das","given":"Gangadhar"},{"family":"Aquilanti","given":"Giuliana"},{"family":"Kanjilal","given":"Aloke"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1063/5.0246384","URL":"https://doi.org/10.1063/5.0246384","source":"crossref"},{"id":"doi:10.1088/2634-4386/ae688e","type":"article-journal","title":"FeNN-DMA: A RISC-V system-on-chip for spiking neural network acceleration","abstract":"Abstract Spiking neural networks (SNNs) are a promising, energy-efficient alternative to standard artificial neural networks (ANNs) and are particularly well-suited to spatio-temporal tasks such as keyword spotting and video classification. However, SNNs have a much lower arithmetic intensity than ANNs and are therefore not well-matched to standard accelerators like Graphical Processing Units and Tensor Processing Units (TPUs). Field Programmable Gate Arrays (FPGAs) are designed for such memory-bound workloads, and here we present a novel, system-on-chip design using a RISC-V softcore with a vector co-processor (FeNN-DMA), tailored to simulating SNNs on modern UltraScale+ FPGAs. We show that FeNN-DMA has comparable resource usage and energy requirements to state-of-the-art fixed-function SNN accelerators, yet it supports more complex neuron models and network topologies, and can simulate up to 16 thousand neurons and 256 million synapses per core. Using this functionality, we demonstrate state-of-the-art classification accuracy on the Spiking Heidelberg Digits, Neuromorphic MNIST and Braille tactile classification tasks.","author":[{"family":"Aizaz","given":"Zainab"},{"family":"Knight","given":"James"},{"family":"Nowotny","given":"Thomas"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/2634-4386/ae688e","URL":"https://doi.org/10.1088/2634-4386/ae688e","source":"crossref"},{"id":"doi:10.1063/5.0308306","type":"article-journal","title":"Dynamic capacitive analysis and physical modeling on ZnO resistive random access memory (RRAM) for enabling neuromorphic computing","abstract":"With the increasing demand for large data storage and artificial intelligence, resistive random-access memory (RRAM) thrives as one of the applicable candidates for the next-generation nonvolatile memory, owing to its simple structure, high scalability, high speed, low power, and tunable conductance. Among oxide-based RRAM, ZnO shows unique optical and electrical properties toward the future heterogeneous integration and low power memory-in-computing systems. In this study, we present a ZnO RRAM manufactured under earth gravity and in-space through inkjet printing. Memory devices with various fabrication environments and conditions include methanol ground, methanol flight, ethanol ground, to ethanol flight. The device fabricated under the microgravity shows a significantly reduced forming voltage and improved reliability. To investigate the filamentary formation in the ZnO RRAM, activation energy was extracted from Arrhenius equations on temperature modulations testing schemes for a comprehensive filament modeling. The capacitive models have concluded oxygen migration conduction dominated on this ZnO RRAM. Finally, the devices' conductance was modulated by AC potentiation and depression with an optimized linearity (R2 = 98%) toward a good training accuracy of 90% on the MNIST data set training toward neuromorphic computing.","author":[{"family":"Huang","given":"Yujian"},{"family":"Maddineni","given":"Sai"},{"family":"Chen","given":"Daphne"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1063/5.0308306","URL":"https://doi.org/10.1063/5.0308306","source":"crossref"},{"id":"doi:10.5772/acrt.deposit.32253477.v1","type":"article-journal","title":"Neuromorphic Computing in Outer Space: Intel Loihi Deployed On-Satellite for Spike-Based Software-Defined Communication - Python Code","abstract":"&lt;p dir=\"ltr\"&gt;In space exploration and communication technology, software-defined networking architectures significantly improve the communications throughput and latency. This study developed a cognitive agent based on Q-learning with continuous learning capabilities to select optimal routing options in a dynamically changing networking environment. This proposed method achieves up to 90% reduction in measured Internet Control Message Protocol (ICMP) round-trip time compared to the Dijkstra shortest-path baseline under identical test conditions. Under dynamically changing network link latency conditions, the cognitive agent achieved a 90% success rate in routing packets in laboratory experiments. The agent was executed on neuromorphic hardware called Intel Loihi. However, due to the system’s power limitations, a simplified version of the agent was engineered to launch into space aboard a CubeSat. The CubeSat was launched in January 2022, making a historic milestone as the first launch of a neuromorphic system into space. The developed applications were successfully executed in space.&lt;/p&gt;","author":[{"family":"Rahman","given":"Nayim"},{"family":"Yakopcic","given":"Chris"},{"family":"Taha","given":"Tarek"},{"family":"Lent","given":"Ricardo"},{"family":"Briones","given":"Janette"},{"family":"Chelmins","given":"David"},{"family":"Dudukovich","given":"Rachel"},{"family":"Smith","given":"Aaron"},{"family":"Gannon","given":"Adam"},{"family":"Lowry","given":"Michael"},{"family":"Murbach","given":"Marcus"},{"family":"Salas","given":"Alejandro"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5772/acrt.deposit.32253477.v1","URL":"https://doi.org/10.5772/acrt.deposit.32253477.v1","source":"crossref"},{"id":"doi:10.1088/1361-6463/ae1241","type":"article-journal","title":"Recent progress in neuromorphic computing based on spin–orbit torque devices","abstract":"Abstract Neuromorphic computing, inspired by the structure and functionality of the biological brain, aims to simulate brain processes through the design of innovative devices, algorithms, and architectures. Neuromorphic devices constitute the foundational hardware components essential for the realization of neuromorphic computing. In recent years, spintronic devices based on the spin–orbit torque (SOT) effect have emerged as the central focus due to their exceptional durability, rapid response times, and low energy consumption. By designing SOT devices with diverse structures and functionalities, researchers have successfully emulated the roles of synapses and neurons in the human brain, thereby enabling the execution of neuromorphic computing tasks. This paper presents a comprehensive review of recent advancements in the application of SOT spintronic devices for neuromorphic computing. We first introduce the underlying mechanisms and testing methods of SOT spintronic devices, then commence with an introduction to biological neural networks in the human brain, followed by an exposition of widely adopted algorithmic architectures and hardware requirements for neuromorphic computing based on SOT devices. After that, we discuss the practical applications of four primary types of SOT devices: Domain Wall (DW)-SOT, Skyrmion-SOT, Nucleation-SOT, and spin Hall nano-oscillators in neuromorphic computing. Finally, we address the current challenges in the field and proposes potential solutions for the future.","author":[{"family":"Liang","given":"Qi"},{"family":"Huang","given":"Yujie"},{"family":"Tan","given":"Yinlong"},{"family":"Tang","given":"Yuhua"},{"family":"Xie","given":"Xiangnan"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1088/1361-6463/ae1241","URL":"https://doi.org/10.1088/1361-6463/ae1241","source":"crossref"},{"id":"doi:10.1149/ma2025-01633076mtgabs","type":"article-journal","title":"Tuning Synaptic Plasticity in WO<sub>3</sub> Ion-Gated Transistors with Aqueous Electrolytes for Neuromorphic Computing","abstract":"Ion-Gated Transistors (IGTs) are a promising technology for neuromorphic computing, offering processing rates and energy efficiency. These devices support low-power training and operation of neural network algorithms while integrating both long-term and short-term modulation within a single system. The suitability of IGTs for neuromorphic applications depends critically on their time-resolved behaviour, governed by the doping mechanism. Specifically, the ionic permeability of the semiconducting channel dictates whether the device operates via electrochemical (three-dimensional) or electrostatic (two-dimensional) doping, leading to varying response times. This work focuses on controlling the response time of WO 3 -based IGTs using aqueous electrolytes—Li 2 SO 4 , Na 2 SO 4 , and K 2 SO 4 —as the gating media. We systematically study the effects of gate-source voltage (V gs ) pulse parameters (frequency, duration, and number) and sampling time on synaptic behaviour. These parameters are tuned to optimize ion intercalation and channel conductivity, enabling precise modulation of synaptic plasticity. Our findings demonstrate the tunability of WO 3 -based IGTs for neuromorphic applications, providing insights into the interplay between ion gating medium properties and channel dynamics. This work highlights the potential of these devices for energy-efficient and adaptive artificial synapses.","author":[{"family":"Bhaskaran","given":"Kesawarthini"},{"family":"Azari","given":"Ramin"},{"family":"Manirakiza","given":"Melchiade"},{"family":"Santato","given":"Clara"},{"family":"Soavi","given":"Francesca"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1149/ma2025-01633076mtgabs","URL":"https://doi.org/10.1149/ma2025-01633076mtgabs","source":"crossref"},{"id":"doi:10.26215/heal.uoa.11720","type":"article-journal","title":"Μελέτη νευρομορφικών φωτονικών αρχιτεκτονικών για βιοϊατρική και συμβατική επεξεργασία εικόνων","abstract":"Η διδακτορική αυτή διατριβή μελετά την ανάπτυξη μιας υψίρυθμης, ενεργειακά αποδοτικής, ολοκληρωμένης νευρομορφικής φωτονικής πλατφόρμας ως ένα υπολογιστικό ρεζερβουάρ (RC) και ως έναν επιταχυντή συνελικτικών νευρωνικών δικτύων (CNNs), για προβλήματα ταξινόμησης εικόνων σε βιοϊατρικές αλλά και συμβατικές εφαρμογές. Οι αρχιτεκτονικές RC περιλαμβάνουν τυχαίες, σταθερές διασυνδέσεις μεταξύ των επεξεργαστικών μονάδων (νευρώνων), προβάλλοντας τα σήματα στην είσοδό τους σ’ έναν παραμετρικό χώρο υψηλότερης διάστασης, ώστε να επιτρέψουν την ταξινόμηση τους μέσω ενός απλού επίπεδου ψηφιακής επεξεργασίας στην έξοδο. Από την άλλη πλευρά, τα CNNs είναι μια υποκατηγορία τεχνητών νευρωνικών δικτύων με πολλαπλά επίπεδα επεξεργασίας που εκτελούν συνελικτικές πράξεις σε τουλάχιστον ένα απ’ αυτά. Η υλοποίηση του συνελικτικού σταδίου ωστόσο σε ολοκληρωμένες φωτονικές πλατφόρμες απαιτεί μία σειρά από πολλαπλασιασμούς πινάκων με διανύσματα, οδηγώντας σε πολλές επαναχρησιμοποιήσεις του ίδιου τσίπ και κατ’ επέκταση σε μία αύξηση του συνολικού χρόνου επεξεργασίας καθώς και της συνολικής κατανάλωση ισχύος. Στη διατριβή αυτή, παρουσιάζεται μια εναλλακτική προσέγγιση βασισμένη σε μία τεχνική κατάτμησης του οπτικού φάσματος (OSS) μέσω πολλαπλών παράλληλων κόμβων αποτελούμενων από παθητικά οπτικά φίλτρα. Καθένας εξ’ αυτών των κόμβων αλληλεπιδρά με μία διαφορετική φασματική περιοχή του οπτικού σήματος, με την έξοδό τους να ανιχνεύεται μέσω ενός φωτοδέκτη και να ψηφιοποιείται, πριν οδηγηθούν σ’ ένα απλό ψηφιακό νευρωνικό δίκτυο για ταξινόμηση. Μέσω της τεχνικής αυτής, αναλογικές συνελίξεις πραγματοποιούνται στο πεδίο του χρόνου μέσω της αλληλεπίδρασης ενός διαμορφωμένου οπτικού σημάτος και των OSS κόμβων, χαρακτηριζόμενων από διαφορετικές κρουστικές αποκρίσεις, με τις εξόδους τους να υπόκεινται επιπλέον σε μη γραμμικότητες και μια διαδικασία εξαγωγής ενός μέσου όρου κατά τη φωτοανίχνευση. Αριθμητικές προσομοιώσεις της επεξεργασίας OSS με 7 κόμβους συντονιζόμενων μικροδακτυλίων πρίν από ενός απλό νευρωνικό δίκτυο για ένα πρόβλημα ταξινόμησης εικόνων χειρόγραφων ψηφίων, ανέδειξαν μία βελτίωση στην ακρίβεια έως και 5.4%. Η επιβεβαιώση για αυτό το πρόβλημα επήλθε και πειραματικά μέσω χρήσης ενός επαναπρογραμματιζόμενου ολοκληρωμένου φωτονικού τσίπ για το OSS στάδιο. Η σύγκριση αυτής της τεχνικής με ένα αντίστοιχο ψηφιακό CNN, ανέδειξε μία μείωση κατά 30% στη συνολική κατανάλωση ενέργειας με συγκρίσιμη ακρίβεια, ενώ μια συγκριτική ανάλυση με σύχρονους ηλεκτρονικούς και φωτονικούς επιταχυντές CNN, παρουσίασε μία έως και 13 φορές καλύτερη απόδοση ισχύος καθώς και μία επίσης υψηλή υπολογιστική πυκνότητα. Σ’ ένα πρόβλημα ταξινόμησης δεδομένων ενός νευρομορφικού συστήματος απεικόνισης κυτταρομετρίας ροής (IFC), η επιτάχυνση OSS βελτίωσε την ακρίβεια κατά 0.4%, μειώνοντας ταυτόχρονα τις εκπαιδεύσιμες παραμέτρους και τις πράξεις κινητής υποδιαστολής του ψηφιακού δικτύου ταξινόμησης στην έξοδο από 9 έως και 15 φορές, διατηρώντας επίσης παρόμοια ακρίβεια με ένα σύγχρονο GRU-RNN με 22 φορές λιγότερες παραμέτρους. Τέλος, για ένα πρόβλημα ταξινόμησης σημάτων ενός συστήματος IFC βασισμένο στην τεχνική του STEAM, η επιτάχυνση OSS βελτίωσε αντίστοιχα την ακρίβεια κατά 0.2%, ενώ μείωσε τις ψηφιακές παραμέτους έως και 13 φορές. Διερευνώντας μία αναδραστική RC τοπολογία στην οπτική έξοδο του ίδιου συστήματος βασισμένη σε OSS κόμβους, η ακρίβεια του νευρωνικού δικτύου στην έξοδο αυξήθηκε κατά 3%, ενώ μειώθηκαν οι απαιτήσεις ως προς το υλικό για φωτοανίχνευση και ψηφιοποίηση κατά 10 και 5 φορές αντίστοιχα.","author":[{"family":"Τσιριγώτης","given":"Άρης"}],"issued":{"date-parts":[[2024]]},"DOI":"10.26215/heal.uoa.11720","URL":"https://doi.org/10.26215/heal.uoa.11720","source":"datacite"},{"id":"doi:10.26215/heal.uoa.11707","type":"article-journal","title":"Photonic neuromorphic processors based on semiconductor lasers' dynamics for reservoir computing and spiking neural networks","abstract":"Η παρούσα διδακτορική διατριβή μελετά την ανάπτυξη φωτονικών νευρομορφικών επεξεργαστών (ΝΕ) που απευθύνονται στις εξής τρεις κ��ίσιμες προκλήσεις των Big Data και του Internet of Things: την υψηλή ταχύτητα επεξεργασίας, την χαμηλή κατανάλωση ισχύος και την ελαχιστοποίηση του υλικού. Η αντιμετώπιση των εν λόγω προκλήσεων ανατέθηκε στα βιοεμπνευσμένα Τεχνητά Νευρωνικά Δίκτυα. Ωστόσο, η ταυτόχρονη επίτευξη δύο βασικών, αλλά αντικρουόμενων στόχων περιόρισε την αποτελεσματικότητά τους, καθώς ήταν αναγκαία η αύξηση του αριθμού των νευρώνων και των συνάψεων παράλληλα με τον περιορισμό της καταναλισκόμενης ισχύος και της πολυπλοκότητας. Σε αυτό το πλαίσιο, επιλέχθηκαν οι χαμηλής κατανάλωσης ΝΕ, οι οποίοι όμως αδυνατούσαν να μειώσουν την αυξανόμενη πολυπλοκότητα και συνδεσιμότητα. Για το λόγο αυτό, οι ΝΕ υιοθέτησαν αρχές των Time Delay Network (TDN) που υλοποιούν χρονικά πολυπλεγμένους νευρώνες και συνάψεις. Μειώθηκαν, έτσι, οι απαιτήσεις του υλικού, αλλά και ο ρυθμός επεξεργασίας. Στόχος της παρούσας διατριβής είναι η αξιοποίηση της υψηλής ταχύτητας των φωτονικών δομών, ώστε να ενισχυθούν οι δυνατότητες επεξεργασίας των TDN ΝΕ. Κατ’ αρχάς αναλύθηκε ένας φωτονικός TDN ΝΕ βασισμένος σε ένα spin VCSEL κβαντικών τελειών. Το προταθέν σύστημα επιτυγχάνει τον τετραπλασιασμό του αριθμού των νευρώνων συγκλίνοντας σε επιδόσεις με τους σύγχρονους επεξεργαστές με τη χρήση λιγότερων νευρώνων συγκριτικά με λοιπές παρόμοιες διατάξεις. Επιπλέον, η χρήση πολλαπλών μασκών και η έγχυση των μη διαμορφωμένων πεδίων βελτιστοποιεί τις επιδόσεις του συστήματος και ελαχιστοποιεί τις απαιτήσεις του υλικού. Συνεχίζοντας, ερευνήθηκε πειραματικά και αριθμητικά ένας επαναδιαμορφώσιμος φωτονικός TDN ΝΕ ερειδόμενος σε Fabry-Perot (FP) λέιζερ. Επιβεβαιώθηκε πειραματικά η δυνατότητα αύξησης του αριθμού των εικονικών νευρώνων και των επιδόσεων του FP-TDN αξιοποιώντας την πολύτροπη εκπομπή του FP. Μετέπειτα απεδείχθη αριθμητικά η αύξηση του ρυθμού επεξεργασίας του FP-TDN χωρίς να υπάρξει μείωση της ακρίβειας του επεξεργαστή μέσω μιας νέας τεχνικής Φασματοχρονικής Πολυπλεξίας (SPTM). O FP-TDN ΝΕ σε συνδυασμό με το SPTM κατέγραψε ρυθμούς σφάλματος ίσους με 10-4 χρησιμοποιώντας μήκος ανάδρασης 140 ps, ενώ η HD-FEC συμβατότητα του συστήματος επιτεύχθηκε ακόμη και για χρόνους των 60 ps. Σημειώνεται ότι ο FP-TDN κατέγραψε ακρίβεια 95.95% στην κατηγοριοποίηση των εικόνων του MNIST επιτυγχάνοντας ρυθμό 255.1Mimages/sec. Στο επόμενο κεφάλαιο παρατίθεται ένα χρονικά πολυπλεγμένο δίκτυο με spiking νευρώνες. Ειδικότερα, παρουσιάστηκε ένα βαθύ φωτονικό spiking συνελικτικό νευρωνικό δίκτυο βασισμένο σε VCSEL νευρώνες, το οποίο εκμεταλλευόμενο τη χρονική πολυπλεξία δύναται να ρυθμίσει τον αριθμό των πραγματικών νευρώνων από 62 έως 2020 ανάλογα με τις απαιτήσεις του ρυθμού επεξεργασίας. Το παρόν δίκτυο επέδειξε δυνατότητες μάθησης χωρίς επιτήρηση, υψηλή ανοχή στο θόρυβο και ρυθμούς επεξεργασίας από 5 ns έως 720 ns ανά εικόνα. Τέλος, αναπτύχθηκε ένα spiking TDN βασισμένο σε VCSEL, το οποίο δύναται να κατηγοριοποιήσει με υψηλή ακρίβεια δεδομένα από τις βάσεις δεδομένων Iris, MNIST, αλλά και από πειραματικές εικόνες κυτταρομετρίας καταναλίσκοντας μόλις 3.23pJ ανά εικόνα. Ο συγκεκριμένος ΝΕ ενoποιεί για πρώτη φορά τα πεδία των ΝΕ και των νευρομορφικών αισθητήρων σε μια κοινή πλατφόρμα, ανοίγοντας το δρόμο για τη μελέτη κλιμακούμενων και ενεργειακά αποδοτικών ΝΕ για εφαρμογές πραγματικού χρόνου","author":[{"family":"Skontranis","given":"Menelaos"},{"family":"Σκοντράνης","given":"Μενέλαος"}],"issued":{"date-parts":[[2024]]},"DOI":"10.26215/heal.uoa.11707","URL":"https://doi.org/10.26215/heal.uoa.11707","source":"datacite"},{"id":"doi:10.48550/arxiv.2312.12264","type":"manuscript","title":"Exploring Non-Steady-State Charge Transport Dynamics in Information Processing: Insights from Reservoir Computing","abstract":"Exploring nonlinear chemical dynamic systems for information processing has emerged as a frontier in chemical and computational research, seeking to replicate the brain's neuromorphic and dynamic functionalities. We have extensively explored the information processing capabilities of a nonlinear chemical dynamic system through theoretical modeling by integrating a non-steady-state proton-coupled charge transport system into reservoir computing (RC) architecture. Our system demonstrated remarkable success in tasks such as waveform recognition, voice identification and chaos system prediction. More importantly, through a quantitative study, we revealed the key role of the alignment between the signal processing frequency of the RC and the characteristic time of the dynamics of the nonlinear system, which dictates the efficiency of RC task execution, the reservoir states and the memory capacity in information processing. The system's information processing frequency range was further modulated by the characteristic time of the dynamic system, resulting in an implementation akin to a 'chemically-tuned band-pass filter' for selective frequency processing. Our study thus elucidates the fundamental requirements and dynamic underpinnings of the non-steady-state charge transport dynamic system for RC, laying a foundational groundwork for the application of dynamic molecular devices for in-materia computing.","author":[{"family":"Li","given":"Zheyang"},{"family":"Yu","given":"Xi"}],"issued":{"date-parts":[[2023]]},"DOI":"10.48550/arxiv.2312.12264","URL":"https://doi.org/10.48550/arxiv.2312.12264","source":"datacite"},{"id":"doi:10.5281/zenodo.20277956","type":"article-journal","title":"Neuromorphic tissues: Soft Biomolecular Networks for Brain-Inspired Temporal Computing","abstract":"This repository contains all MATLAB scripts and example input data required to reproduce the main computational results presented in the companion manuscript (see Connected_Reservoir_Paper_Science.pdf). The package implements and compares two physical-model reservoir computing systems built from droplet-interface-bilayer (DIB) ionic devices: Connected Reservoir Network – simulates a spatially connected droplet lattice driven by masked current inputs. Parallel Reservoir Network – simulates independent (uncoupled) memristive nodes in parallel as a reference system. Both reservoirs are modeled with the Alamethicin-type Richard’s differential equation (RDE) pore-state dynamics, solved numerically using MATLAB’s stiff ODE solvers (ode15s). Main capabilities NARMA tasks: Generates masked input sequences and trains linear readouts to predict NARMA orders 2–10; reports train/test NRMSE. Lorenz-63 task: Demonstrates prediction of all three Lorenz-63 chaotic states using the optimized 43-node connected reservoir. Hyperparameter sweeps: Grid-search scripts for input amplitude, mask length, and node count for performance benchmarking. Reproducible I/O: Includes functions for input-mask generation, state sampling after each pulse, and standard feature-matrix construction. How to reproduce figures/results ParallelReservoirNARMA_Optimized.m — reproduces the main NARMA benchmarking shown for the uncoupled network. HyperParameterTuningOptimized.m — performs grid search and NARMA evaluation for the connected reservoir. Lorenz63_Prediction.m — trains on Lorenz-63 time series and plots the 3-D attractors for training vs. testing datasets. Sub-functions in /src/Codes… provide ODE models, input masking (binary and weighted), and De Bruijn-based temporal masks. All random seeds, pulse widths, and parameter values are pre-set to match the paper’s reported results. The scripts save intermediate .mat files for sequences, states, and readout predictions so figures and error metrics can be regenerated without code modification. Requirements MATLAB R2023a or later (tested up to R2024b). No external toolboxes beyond MATLAB’s built-in ODE suite and Statistics & Machine Learning Toolbox (for fitlm).","author":[{"family":"Armendarez","given":"Nicholas"},{"family":"Mohamed","given":"Ahmed"},{"family":"Hasan","given":"Md"},{"family":"Najem","given":"Joseph"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20277956","URL":"https://doi.org/10.5281/zenodo.20277956","source":"datacite"},{"id":"doi:10.5281/zenodo.17296787","type":"article-journal","title":"Neuromorphic tissues: Soft Biomolecular Networks for Brain-Inspired Temporal Computing","abstract":"This repository contains all MATLAB scripts and example input data required to reproduce the main computational results presented in the companion manuscript (see Connected_Reservoir_Paper_Science.pdf). The package implements and compares two physical-model reservoir computing systems built from droplet-interface-bilayer (DIB) ionic devices: Connected Reservoir Network – simulates a spatially connected droplet lattice driven by masked current inputs. Parallel Reservoir Network – simulates independent (uncoupled) memristive nodes in parallel as a reference system. Both reservoirs are modeled with the Alamethicin-type Richard’s differential equation (RDE) pore-state dynamics, solved numerically using MATLAB’s stiff ODE solvers (ode15s). Main capabilities NARMA tasks: Generates masked input sequences and trains linear readouts to predict NARMA orders 2–10; reports train/test NRMSE. Lorenz-63 task: Demonstrates prediction of all three Lorenz-63 chaotic states using the optimized 43-node connected reservoir. Hyperparameter sweeps: Grid-search scripts for input amplitude, mask length, and node count for performance benchmarking. Reproducible I/O: Includes functions for input-mask generation, state sampling after each pulse, and standard feature-matrix construction. How to reproduce figures/results ParallelReservoirNARMA_Optimized.m — reproduces the main NARMA benchmarking shown for the uncoupled network. HyperParameterTuningOptimized.m — performs grid search and NARMA evaluation for the connected reservoir. Lorenz63_Prediction.m — trains on Lorenz-63 time series and plots the 3-D attractors for training vs. testing datasets. Sub-functions in /src/Codes… provide ODE models, input masking (binary and weighted), and De Bruijn-based temporal masks. All random seeds, pulse widths, and parameter values are pre-set to match the paper’s reported results. The scripts save intermediate .mat files for sequences, states, and readout predictions so figures and error metrics can be regenerated without code modification. Requirements MATLAB R2023a or later (tested up to R2024b). No external toolboxes beyond MATLAB’s built-in ODE suite and Statistics & Machine Learning Toolbox (for fitlm).","author":[{"family":"Armendarez","given":"Nicholas"},{"family":"Mohamed","given":"Ahmed"},{"family":"Hasan","given":"Md"},{"family":"Najem","given":"Joseph"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.17296787","URL":"https://doi.org/10.5281/zenodo.17296787","source":"datacite"},{"id":"doi:10.1063/5.0257237","type":"article-journal","title":"Controlled crystallization of thermal evaporated GST-on-SOI for photonic neuromorphic application","abstract":"Neuromorphic computing is inevitable in addressing the computing challenges faced by the current computing architecture. A nonvolatile process is one of the key elements in realizing a neuromorphic architecture. Realization of a photonic nonvolatile state has been a challenge. Germanium antimony telluride (GST) offers potential electrical and optical characteristics to realize an optical nonvolatile state through a material phase transition. In this work, we demonstrate controlled phase tuning of a thermally evaporated GST-integrated silicon ring-resonator for potential neuromorphic application. We present material characterization and device fabrication and correlate the electrical and photonic phase transitions of ring-integrated GST. The temperature cycling in temperature-dependent resistance measurement shows the feasibility of pinning the phase of the transition region, which could be exploited for multiple non-volatile states. Using a simple device architecture, we have exploited the non-volatile and gradual transition of thermally evaporated GST to access intermediate states, which can be realized as synaptic weights in neural networks.","author":[{"family":"Kallega","given":"Rakshitha"},{"family":"Shekhawat","given":"Roopali"},{"family":"Ramesh","given":"K"},{"family":"Selvaraja","given":"Shankar"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1063/5.0257237","URL":"https://doi.org/10.1063/5.0257237","source":"crossref"},{"id":"doi:10.21474/jncs01/126","type":"article-journal","title":"NEUROMORPHIC COMPUTING FOR NEXT-GENERATION ARTIFICIAL INTELLIGENCE: ARCHITECTURES, APPLICATIONS, AND FUTURE RESEARCH DIRECTIONS","abstract":"The growing computational demands of modern Artificial Intelligence (AI) have exposed the limitations of conventional von Neumann computing architectures in terms of energy efficiency, latency, and scalability. Neuromorphic computing has emerged as a revolutionary computing paradigm inspired by the structure and functioning of the human brain. By integrating specialized hardware with brain-inspired neural processing models, neuromorphic systems aim to perform intelligent computation with significantly lower power consumption and faster response times. This paper provides a comprehensive review of neuromorphic computing, including its architecture, fundamental principles, enabling technologies, real-world applications, implementation challenges, and future research opportunities. The study also discusses the integration of neuromorphic processors with edge computing, robotics, the Internet of Things (IoT), and autonomous systems. The findings indicate that neuromorphic computing has the potential to transform intelligent computing by enabling energy-efficient, adaptive, and real-time AI systems.","author":[{"family":"Foster","given":"Amelia"},{"family":"Khalid","given":"Yusuf"},{"family":"Romero","given":"Diego"}],"issued":{"date-parts":[[2026]]},"DOI":"10.21474/jncs01/126","URL":"https://doi.org/10.21474/jncs01/126","source":"crossref"},{"id":"doi:10.3390/jlpea15020016","type":"article-journal","title":"2D Spintronics for Neuromorphic Computing with Scalability and Energy Efficiency","abstract":"The demand for computing power has been growing exponentially with the rise of artificial intelligence (AI), machine learning, and the Internet of Things (IoT). This growth requires unconventional computing primitives that prioritize energy efficiency, while also addressing the critical need for scalability. Neuromorphic computing, inspired by the biological brain, offers a transformative paradigm for addressing these challenges. This review paper provides an overview of advancements in 2D spintronics and device architectures designed for neuromorphic applications, with a focus on techniques such as spin-orbit torque, magnetic tunnel junctions, and skyrmions. Emerging van der Waals materials like CrI3, Fe3GaTe2, and graphene-based heterostructures have demonstrated unparalleled potential for integrating memory and logic at the atomic scale. This work highlights technologies with ultra-low energy consumption (0.14 fJ/operation), high switching speeds (sub-nanosecond), and scalability to sub-20 nm footprints. It covers key material innovations and the role of spintronic effects in enabling compact, energy-efficient neuromorphic systems, providing a foundation for advancing scalable, next-generation computing architectures.","author":[{"family":"Plummer","given":"Douglas"},{"family":"Dalessandro","given":"Emily"},{"family":"Burrowes","given":"Aidan"},{"family":"Fleischer","given":"Joshua"},{"family":"Heard","given":"Alexander"},{"family":"Wu","given":"Yingying"}],"issued":{"date-parts":[[2025]]},"DOI":"10.3390/jlpea15020016","URL":"https://doi.org/10.3390/jlpea15020016","source":"crossref"},{"id":"doi:10.56553/popets-2025-0060","type":"article-journal","title":"Are Neuromorphic Architectures Inherently Privacy-preserving? An Exploratory Study","abstract":"While machine learning (ML) models are becoming mainstream, including in critical application domains, concerns have been raised about the increasing risk of sensitive data leakage. Various privacy attacks, such as membership inference attacks (MIAs), have been developed to extract data from trained ML models, posing significant risks to data confidentiality. While the predominant work in the ML community considers traditional Artificial Neural Networks (ANNs) as the default neural model, neuromorphic architectures, such as Spiking Neural Networks (SNNs), have recently emerged as an attractive alternative mainly due to their significantly low power consumption. These architectures process information through discrete events, i.e., spikes, to mimic the functioning of biological neurons in the brain. While the privacy issues have been extensively investigated in the context of traditional ANNs, they remain largely unexplored in neuromorphic architectures, and little work has been dedicated to investigating their privacy-preserving properties. In this paper, we investigate the question of whether SNNs have inherent privacy-preserving advantages. Specifically, we investigate SNNs’ privacy properties through the lens of MIAs across diverse datasets, in comparison with ANNs. We explore the impact of different learning algorithms (surrogate gradient and evolutionary learning), programming frameworks (snnTorch, TENNLab, and LAVA), and various parameters on the resilience of SNNs against MIA. Our experiments reveal that SNNs demonstrate consistently superior privacy preservation compared to ANNs, with evolutionary algorithms further enhancing their resilience. For example, on the CIFAR-10 dataset, SNNs achieve an AUC as low as 0.59 compared to 0.82 for ANNs, and on CIFAR-100, SNNs maintain a low AUC of 0.58, whereas ANNs reach 0.88. Furthermore, we investigate the privacy-utility trade-off through Differentially Private Stochastic Gradient Descent (DPSGD), observing that SNNs incur a notably lower accuracy drop than ANNs under equivalent privacy constraints.","author":[{"family":"Moshruba","given":"Ayana"},{"family":"Alouani","given":"Ihsen"},{"family":"Parsa","given":"Maryam"}],"issued":{"date-parts":[[2025]]},"DOI":"10.56553/popets-2025-0060","URL":"https://doi.org/10.56553/popets-2025-0060","source":"crossref"},{"id":"doi:10.1038/s42005-025-02257-0","type":"article-journal","title":"Magnetic tunnel junctions driven by hybrid optical-electrical signals as a flexible neuromorphic computing platform","abstract":"Abstract Magnetic tunnel junctions (MTJs) offer a promising pathway toward energy-efficient neuromorphic computing due to their nanoscale footprint, nonvolatile switching, and intrinsic nonlinear dynamics that emulate synaptic behavior. However, generating large thermoelectric voltages with bias-tunable nonlinearities for neuromorphic use remains largely unexplored. Here, we introduce a hybrid opto-electrical excitation scheme—combining pulsed laser heating with DC bias—to drive MTJs into the nonlinear bias-enhanced tunnel magneto-Seebeck regime. This regime yields thermoelectric voltages in the tens of millivolts with a strong contrast between magnetic states, while also revealing spiking and double-switching behavior linked to vortex dynamics and fixed-layer depinning. The thermovoltage exhibits cubic dependence on bias current, enabling tunable synaptic weights. We simulate a single-layer neuromorphic network using optically encoded inputs and achieve 93.7% classification accuracy on handwritten digits. These results establish hybrid-driven MTJs as a compact, CMOS-compatible platform for neuromorphic computing, integrating optical input with spintronic functionality.","author":[{"family":"Oberbauer","given":"Felix"},{"family":"Winkel","given":"Tristan"},{"family":"Böhnert","given":"Tim"},{"family":"Wanjura","given":"Clara"},{"family":"Claro","given":"Marcel"},{"family":"Benetti","given":"Luana"},{"family":"Çaha","given":"Ihsan"},{"family":"Deepak","given":"Francis"},{"family":"Moradi","given":"Farshad"},{"family":"Ferreira","given":"Ricardo"},{"family":"Münzenberg","given":"Markus"},{"family":"Parvini","given":"Tahereh"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1038/s42005-025-02257-0","URL":"https://doi.org/10.1038/s42005-025-02257-0","source":"crossref"},{"id":"doi:10.55041/ijsrem57837","type":"article-journal","title":"Real-Time Disaster Response and Recovery by using Neuromorphic Computing-Enabled Autonomous Agents","abstract":"Abstract - This paper presents a practical approach to disaster response and recovery through the integration of neuromorphic computing, autonomous agents, and energy-efficient processing systems. Neuromorphic computing, inspired by the human brain's architecture, enables ultralow power consumption while maintaining high-speed, real-time processing capabilities essential for disaster scenarios. The proposed system deploys Spiking Neural Networks (SNNs) running on neuromorphic hardware platforms like Brain Chip Akida, combined with event-driven sensors including Dynamic Vision Sensors (DVS) and Silicon Cochlea for audio processing. These autonomous agents can operate independently for extended periods without external power sources or internet connectivity, making them ideal for disaster environments where traditional infrastructure fails. The system integrates multi-agent coordination protocols, explainable AI for trustworthy decision-making, and real-time adaptation through spike-timing-dependent plasticity (STDP). The neuromorphic approach addresses critical limitations of traditional disaster response systems including high power consumption, infrastructure dependency, and limited operational duration. This research contributes to advancing sustainable AI for humanitarian applications, potentially revolutionizing emergency response capabilities and saving lives through more effective, autonomous disaster management systems. Key Words: Neuromorphic Computing, Spiking Neural Networks, Autonomous Agents, Disaster Management Response, Multi-agent Systems, Sustainable AI.","author":[{"family":"Gokilavani","given":"G"},{"family":"Vidhya","given":"Ct"}],"issued":{"date-parts":[[2026]]},"DOI":"10.55041/ijsrem57837","URL":"https://doi.org/10.55041/ijsrem57837","source":"crossref"},{"id":"doi:10.3390/astronautics1030011","type":"article-journal","title":"Hybrid Neuromorphic Edge Computing and Quantum Cloud Optimization for Martian Swarm Robot Survival and Map Recovery","abstract":"Martian dust storms cut off communication and break standard robot navigation. We built a hybrid system that keeps robot swarms alive during these blackouts and recovers their data quickly. Our rovers use Spiking Neural Networks (SNNs) on their own edge processors to navigate without a signal. Once the storm passes, we use the Quantum Approximate Optimization Algorithm (QAOA) on a cloud platform to merge the fragmented maps the rovers collected while they were offline. We tested this system in a Robot Operating System 2 (ROS 2) and Gazebo environment using a simulated 10-rover Martian deployment. During the simulated blackout, our SNN edge navigation achieved a 92.0% survival rate, outperforming traditional planners like Dynamic Window Approach (DWA) (29.0%) and Timed Elastic Band (TEB) (24.3%). The neuromorphic approach also reduced overall system power consumption by 80.0% compared to a traditional unoptimized Graphics Processing Unit (GPU)-based Simultaneous Localization and Mapping (SLAM) baseline. For the map recovery phase, our simulated QAOA proof-of-concept evaluated the map constraints in just 1.2 ms, compared to 50.0 ms for a classical Generalized Iterative Closest Point (G-ICP) and g2o pose-graph approach. Despite the noisy sensor data collected during the blackout, the final quantum-stitched map achieved an 8.54 cm Root Mean Square Error (RMSE). These results show that combining edge-based neuromorphic processing with quantum cloud computing secures swarm survival and accelerates post-disaster data recovery for deep-space missions.","author":[{"family":"Sheikder","given":"Chandan"},{"family":"Zhang","given":"Weimin"},{"family":"Chen","given":"Xiaopeng"},{"family":"Fan","given":"Shicheng"},{"family":"Li","given":"Tairan"},{"family":"He","given":"Haotong"}],"issued":{"date-parts":[[2026]]},"DOI":"10.3390/astronautics1030011","URL":"https://doi.org/10.3390/astronautics1030011","source":"crossref"},{"id":"doi:10.1038/s43246-024-00707-w","type":"article-journal","title":"Integrating molecular photoswitch memory with nanoscale optoelectronics for neuromorphic computing","abstract":"Abstract Photonic solutions are potentially highly competitive for energy-efficient neuromorphic computing. However, a combination of specialized nanostructures is needed to implement all neuro-biological functionality. Here, we show that donor-acceptor Stenhouse adduct dyes integrated with III-V semiconductor nano-optoelectronics have combined excellent functionality for bio-inspired neural networks. The dye acts as synaptic weights in the optical interconnects, while the nano-optoelectronics provide neuron reception, interpretation and emission of light signals. These dyes can reversibly switch from absorbing to non-absorbing states, using specific wavelength ranges. Together, they show robust and predictable switching, low energy thermal reset and a memory dynamic range from days to sub-seconds that allows both short- and long-term memory operation at natural timescales. Furthermore, as the dyes do not need electrical connections, on-chip integration is simple. We illustrate the functionality using individual nanowire photodiodes as well as arrays. Based on the experimental performance metrics, our on-chip solution is capable of operating an anatomically validated model of the insect brain navigation complex.","author":[{"family":"Alcer","given":"David"},{"family":"Zaiats","given":"Nelia"},{"family":"Jensen","given":"Thomas"},{"family":"Philip","given":"Abbey"},{"family":"Gkanias","given":"Evripidis"},{"family":"Ceberg","given":"Nils"},{"family":"Das","given":"Abhijit"},{"family":"Flodgren","given":"Vidar"},{"family":"Heinze","given":"Stanley"},{"family":"Borgström","given":"Magnus"},{"family":"Webb","given":"Barbara"},{"family":"Laursen","given":"Bo"},{"family":"Mikkelsen","given":"Anders"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1038/s43246-024-00707-w","URL":"https://doi.org/10.1038/s43246-024-00707-w","source":"crossref"},{"id":"doi:10.1002/aisy.202500351","type":"article-journal","title":"Harnessing Nonidealities in Analog In‐Memory Computing Circuits: A Physical Modeling Approach for Neuromorphic Systems","abstract":"Large‐scale deep learning models are increasingly constrained by their immense energy consumption, which limits their scalability and applicability for edge intelligence. In‐memory computing (IMC) offers a promising solution by addressing the von Neumann bottleneck inherent in traditional deep learning accelerators, significantly reducing energy consumption. However, the analog nature of IMC introduces hardware nonidealities that degrade model performance and reliability. This article presents a novel approach to directly train physical models of IMC, formulated as ordinary differential equation (ODE)‐based physical neural networks (PNNs). To enable the training of large‐scale networks, a technique called differentiable spike‐time discretization is proposed, which reduces the computational cost of ODE‐based PNNs by up to 20 times in speed and 100 times in memory. Such large‐scale networks enhance learning performance by exploiting hardware nonidealities on the CIFAR‐10 dataset. The proposed bottom‐up methodology is validated through post‐layout SPICE simulations on the IMC circuit with nonideal characteristics using the sky130 process. The proposed PNN approach reduces the discrepancy between model behavior and circuit dynamics by at least an order of magnitude. This work paves the way for leveraging nonideal physical devices, such as nonvolatile resistive memories, for energy‐efficient deep learning applications.","author":[{"family":"Sakemi","given":"Yusuke"},{"family":"Okamoto","given":"Yuji"},{"family":"Morie","given":"Takashi"},{"family":"Nobukawa","given":"Sou"},{"family":"Hosomi","given":"Takeo"},{"family":"Aihara","given":"Kazuyuki"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1002/aisy.202500351","URL":"https://doi.org/10.1002/aisy.202500351","source":"crossref"},{"id":"doi:10.1088/2634-4386/ae9be4","type":"article-journal","title":"Node perturbation can effectively train multi-layer neural networks","abstract":"Abstract Backpropagation (BP) remains the dominant and most successful method for training parameters of deep neural network models. However, BP relies on two computationally distinct phases, does not provide a satisfactory explanation of biological learning, and can be challenging to apply for training of networks with discontinuities or noisy node dynamics. By comparison, node perturbation (NP), also known as activity-perturbed forward gradients, proposes learning by the injection of noise into network activations, and subsequent measurement of the induced loss change. NP relies on two forward (inference) passes, does not make use of network derivatives, and has been proposed as a model for learning in biological systems. However, standard NP is highly data inefficient and can be unstable due to its unguided noise-based search process. In this work, we develop a modern perspective on NP by relating it to the directional derivative and incorporating input decorrelation. We find that a closer alignment with directional derivatives together with input decorrelation at every layer theoretically and practically enhances performance of NP learning with large improvements in parameter convergence and much higher performance on the test data, approaching that of BP. Furthermore, our novel formulation allows for application to noisy systems in which the noise process itself is inaccessible, which is of particular interest for on-chip learning in neuromorphic systems.","author":[{"family":"Dalm","given":"Sander"},{"family":"Gerven","given":"Marcel"},{"family":"Ahmad","given":"Nasir"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/2634-4386/ae9be4","URL":"https://doi.org/10.1088/2634-4386/ae9be4","source":"crossref"},{"id":"doi:10.1101/2025.07.10.25331024","type":"article-journal","title":"Event-based seizure detection in human iEEG with neuromorphic hardware","abstract":"Abstract Background Epilepsy is a neurological disorder that affects approximately 1% of the global population. The current method for seizure monitoring, seizure diaries, is often inaccurate, making precise monitoring challenging. A monotonic descending “chirp” pattern in intracranial EEG (iEEG) is a specific marker of seizure onset and can be used for automatic detection. Objective To determine whether a spiking neural network (SNN) implemented on the DYNAP-SE1 neuromorphic chip can detect chirps/seizures. Methods We analysed 48 h of continuous bipolar iEEG from one patient (40 seizures). The signal was filtered into six 10 Hz sub-bands between 0 and 40 Hz, then encoded into UP/DOWN events with software asynchronous delta modulation (ADM). The encoded signal was used as input in the hardware-implemented SNN with 34 adaptive-exponential neurons (3.3 % of the 1024 neurons on one chip). Hierarchical inhibition enforced the required high-to-low band sequence, and a disinhibition unit suppressed isolated low-frequency bursts. We implemented the same 28-neuron SNN (without the dis-inhibition population) on software. Results Chirps appeared at the onset of all 40 seizures (100 %). The hardware SNN detected every seizure (sensitivity = 100 %) and produced one false alarm in 48 h (falsealarm rate = 0.021 h -1 ). Mean processing time was 4 h 55 s ± 42 s for each 4-h data block, showing real-time operation. In the software SNN implementation, we analysed the same 48-h iEEG bipolar channel recording, and detected 32/40 seizures (80 % sensitivity) with 9 false alarms (false-alarm rate = 0.19 h -1 ). Conclusion A 34-neuron SNN implemented on DYNAP-SE1 detects seizures from single-channel iEEG in real time with 100% sensitivity and a low false-alarm rate while using minimal hardware resources.","author":[{"family":"Davidhi","given":"Flavia"},{"family":"Costa","given":"Filippo"},{"family":"Ledergerber","given":"Debora"},{"family":"Indiveri","given":"Giacomo"},{"family":"Imbach","given":"Lukas"},{"family":"Sarnthein","given":"Johannes"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1101/2025.07.10.25331024","URL":"https://doi.org/10.1101/2025.07.10.25331024","source":"preprints"},{"id":"doi:10.1088/2634-4386/ae66b2","type":"article-journal","title":"CMOS implementation of field programmable spiking neural network for hardware reservoir computing","abstract":"Abstract The increasing complexity and energy demands of large-scale neural networks, such as deep neural networks and large language models, challenge their practical deployment in edge applications due to high power consumption, area requirements, and privacy concerns. Spiking neural networks, particularly in analog implementations, offer a promising low-power alternative but suffer from noise sensitivity and connectivity limitations. This work presents a novel CMOS-fabricated field-programmable neural network architecture for hardware reservoir computing (RC). We propose a leaky integrate-and-fire neuron circuit featuring integrated voltage-controlled oscillators and synaptic weights programmed via an on-chip field-programmable gate array framework. This framework enables direct connectivity between neurons, supporting the implementation of arbitrary reservoir configurations. The performance of the system is validated through simulation and chip measurements, demonstrating effective FORCE algorithm learning alongside competitive results in linear/non-linear memory capacity and NARMA10 benchmarks. The neuron design achieves compact area utilization (around 540 NAND2-equivalent units) and low energy consumption (21.7 pJ/pulse) without requiring ADCs for information readout, making it ideal for system-on-chip integration of RC. This architecture paves the way for scalable, energy-efficient neuromorphic systems capable of performing real-time learning and inference with high configurability and digital interfacing.","author":[{"family":"Duran","given":"Ckristian"},{"family":"Kimura","given":"Nanako"},{"family":"Byambadorj","given":"Zolboo"},{"family":"Iizuka","given":"Tetsuya"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/2634-4386/ae66b2","URL":"https://doi.org/10.1088/2634-4386/ae66b2","source":"crossref"},{"id":"doi:10.1038/s44335-025-00028-2","type":"article-journal","title":"Comparing quantum annealing and spiking neuromorphic computing for sampling binary sparse coding QUBO problems","abstract":"Abstract We consider the problem of computing a sparse binary representation of an image. Given an image and an overcomplete, non-orthonormal basis, we aim to find a sparse binary vector indicating the minimal set of basis vectors that when added together best reconstruct the given input. We formulate this problem with an L 2 loss on the reconstruction error, and an L 0 loss on the binary vector enforcing sparsity. First, we solve the sparse representation QUBOs by solving them both on a D-Wave quantum annealer with Pegasus chip connectivity, as well as on the Intel Loihi 2 spiking neuromorphic processor using a stochastic Non-equilibrium Boltzmann Machine (NEBM). Second, using Quantum Evolution Monte Carlo with Reverse Annealing and iterated warm starting on Loihi 2 to evolve the solution quality from the respective machines. We demonstrate that both quantum annealing and neuromorphic computing are suitable for solving binary sparse coding QUBOs.","author":[{"family":"Henke","given":"Kyle"},{"family":"Pelofske","given":"Elijah"},{"family":"Kenyon","given":"Garrett"},{"family":"Hahn","given":"Georg"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1038/s44335-025-00028-2","URL":"https://doi.org/10.1038/s44335-025-00028-2","source":"crossref"},{"id":"doi:10.1002/flm2.70012","type":"article-journal","title":"Advancements in flexible memristors for neuromorphic computing: Materials, mechanisms, and applications in synaptic emulation","abstract":"Abstract The brain orchestrates complex physiological processes through intricate neural networks, with synapses serving as the fundamental units for inter‐neuronal communication and ensuring the efficient functioning of these networks. Consequently, the development of devices capable of emulating synaptic functions represents a crucial avenue for advancing our understanding of neural networks. Among these devices, memristors have emerged as a promising candidate. Recognized as the fourth fundamental passive circuit element, memristors exhibit distinctive nonlinear memory characteristics. Their resistance values dynamically adjust in response to variations in the charge flowing through them and, importantly, retain these modified states even after power disconnection. These unique properties render memristors particularly suitable for emulating synaptic functions in neural systems. This paper provides a comprehensive overview of recent advancements in material selection and resistive switching mechanisms for flexible memristors, highlighting their applications in the construction of artificial neural networks. Furthermore, we discuss the feasibility of implementing neural networks using memristor‐based architectures, while also addressing the current challenges that need to be overcome. Finally, we outline the development prospects and ongoing challenges in this rapidly evolving field.","author":[{"family":"Li","given":"Weiwei"},{"family":"Duan","given":"Chunbo"},{"family":"Wei","given":"Ying"},{"family":"Xu","given":"Hui"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1002/flm2.70012","URL":"https://doi.org/10.1002/flm2.70012","source":"crossref"},{"id":"doi:10.1002/smll.202411596","type":"article-journal","title":"High‐Temperature Resilient Neuromorphic Device Based on Optically Configured Monolayer MoS\n                    <sub>2</sub>\n                    for Cognitive Computing","abstract":"Abstract High‐temperature neuromorphic devices are vital for space exploration and operations in harsh environments such as manufacturing units. To fulfil this need, researchers are developing technologies that imitate the human brain in structure and function. This need is further pushed by the growth of the Internet of Things (IoT), demanding massive computing power and processing of data. Herein, we present a scalable monolayer MoS 2 ‐based neuromorphic device that can operate at temperatures up to 100 °C. The device is fabricated using monolayer MoS 2 , a 2D semiconductor material known for its remarkable properties, such as mechanical flexibility and thermal stability. As a result, the device can operate at high temperatures and may be customized for different purposes. The obtained device is well characterized by excellent electrical properties, including low power consumption, fast switching rate, moderate resistance ratio of ≈102, low switching voltage, and good endurance up to ≈103 cycles. It also shows neuromorphic behavior as it mimics synaptic plasticity exhibited by biological neural networks. This study addresses high‐temperature requirements in electronics and lays the groundwork for connecting electronic systems with the environment to mutually adapt to demands.","author":[{"family":"Prajapat","given":"Pukhraj"},{"family":"Vashishtha","given":"Pargam"},{"family":"Gupta","given":"Govind"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1002/smll.202411596","URL":"https://doi.org/10.1002/smll.202411596","source":"crossref"},{"id":"doi:10.1088/2634-4386/add0d9","type":"article-journal","title":"Prospects of analog in-memory computing using ferroelectric tunnel junctions","abstract":"Abstract Artificial intelligence (AI) is set to disrupt the way businesses and civil society operate, but the large energy usage for training and running AI compute models remains a pressing concern in terms of its sustainability. Analog in-memory computing (AIMC) hardware based on memristor devices is a promising route to significant energy-savings. Among the various memristor technologies, ferroelectric tunnel junctions (FTJs) hold considerable promise for large scale AIMC. Their performance prospects remain to be fully assessed; therefore, we here evaluate the viability of FTJ memristors as compute units for AIMC. Based on the behavior of experimental TiN/HfZrO 4 /W FTJs we define three operating modes which we apply to standard AI tests and two real-world use cases: Image segmentation (YOLOv8) and natural language processing (BERT). We find that the inherent mechanism behind the analog state in FTJs limits the usable dynamic range and thus enforces strict control of noise. The best overall performance is obtained by constricting the dynamic range further and using ultrashort programming pulses (0.5 ns); on BERT matching the performance of digital hardware. We also present a more accessible approach, combining three FTJs, two of which are operated as binary memories, to achieve similar performance while also relaxing the requirements on data converter precision and level of noise. All in all, we here reveal benefits and intrinsic challenges of FTJ-based AIMC systems, providing a blueprint for future experimental implementations.","author":[{"family":"Borg","given":"Mattias"},{"family":"Papadopoulos","given":"Christos"},{"family":"Guerin","given":"Alec"},{"family":"Athle","given":"Robin"},{"family":"Bastani","given":"Saeed"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1088/2634-4386/add0d9","URL":"https://doi.org/10.1088/2634-4386/add0d9","source":"crossref"},{"id":"doi:10.1088/2634-4386/ae01d2","type":"article-journal","title":"Stochastic rounding for memory-efficient digital simulation of synaptic plasticity using 8-bit floating-point","abstract":"Abstract Simulating brain-scale networks digitally is often hindered by extensive memory access. In this context, using low-precision data types and more efficient models to represent state variables is a viable alternative to improve the scalability of the networks under consideration. However, understanding whether these approaches hurt the expected dynamics of the system is critical yet still poorly understood. This study aims to integrate 8-bit floating-point implementations with neural models optimised for efficient memory access to simulate spiking neural networks endowed with spike-timing-dependent plasticity. By doing this, we not only address scalability issues but also consider the biological plausibility of the network activity. We show that stochastic rounding (SR) is necessary to overcome the floating-point errors associated with weight updates and present our custom SR scheme, which was applied to all arithmetic operations. Under these conditions, we study the limitations and behaviour of plastic weight dynamics and large-scale balanced networks. Our results suggest that the models developed can be used to reproduce many prominent features previously described in studies of cortical stability and information processing. Such an approach offers a promising perspective on optimising spiking neural networks for real-time simulations and resource-constrained digital hardware. Such technology could be used to understand neurological disorders by providing insights into abnormal neural activity patterns. It could also enhance brain-computer interfaces, and aid in cognitive neuroscience research, contributing to novel therapeutic strategies.","author":[{"family":"Urbizagastegui","given":"Pablo"},{"family":"Schaik","given":"André"},{"family":"Wang","given":"Runchun"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1088/2634-4386/ae01d2","URL":"https://doi.org/10.1088/2634-4386/ae01d2","source":"crossref"},{"id":"doi:10.1063/5.0314289","type":"article-journal","title":"Memtransistor for bio-inspired neuromorphic computing: A perspective from device physics to neural and sensory systems","abstract":"The recent surge in data generated by emerging applications has exposed intrinsic bottlenecks in conventional von Neumann architectures, where physically separated processing and memory units limit bandwidth and energy efficiency. Neuromorphic computing offers a brain-inspired approach that unifies computation and memory within a single hardware framework, enabling massively parallel, low-power information processing. Among various device candidates, memtransistors incorporate a gate terminal that enables independent modulation of resistive switching and channel conductance, thereby allowing the realization of complex learning functions and effectively suppressing sneak-path currents in array architectures. This perspective outlines the operation principles of memtransistors based on diverse physical mechanisms, including ion migration, charge trapping, ferroelectric switching, and phase transitions, and then discusses recent materials and architectural engineering strategies, with particular emphasis on two-dimensional channels and scalable array integration. Beyond device-level behavior, bio-inspired functionalities, such as heterosynaptic and homeostatic plasticity, are highlighted as key ingredients for stable and self-regulated learning in neural networks. The integration of memtransistors with sensory modules is further examined to enable near-sensor and in-sensor computing, paving the way for multimodal signal processing that parallels biological perception. Finally, critical challenges and opportunities in variability control, CMOS-compatible processing, and three-dimensional, multisensory integration are identified, indicating that continued progress in material design and architecture optimization will be essential for positioning memtransistors as key enablers of autonomous, bio-inspired intelligence in future robotics, healthcare, and cognitive electronics.","author":[{"family":"Nam","given":"Minsu"},{"family":"Cho","given":"Hyun"},{"family":"Lee","given":"Seong"},{"family":"Yoon","given":"Jung"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1063/5.0314289","URL":"https://doi.org/10.1063/5.0314289","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-6567951/v1","type":"article-journal","title":"Neural networks for socio-labor regulation: a neuromorphic approach to human-centric AI in urban economies","abstract":"Abstract This study investigates the implementation and future potential of neural networks for socio-labor regulation within the urban economies of BRICS megacities, emphasizing a human-centric AI approach. Analysis reveals significant disparities in AI development across these urban centers, with Beijing and Shanghai leading in investment, while Moscow ranks third among all analyzed cities with an AI investment of $620 million, contributing to the growing global urban AI landscape where worldwide smart city spending is projected to reach hundreds of billions of dollars. The research examines key indicators of AI adoption, such as the number of startups and the percentage of companies utilizing AI solutions in these major cities. Specifically, Bangalore stands out with 320 AI startups and 58% of companies implementing AI, while in Russia, Moscow reports 230 startups and 48% company adoption, reflecting varying rates of AI integration within urban business ecosystems globally where the average AI adoption rate for enterprises is still below 30% according to some reports. Beyond general AI adoption, the study analyzes the deployment and effectiveness of neuromorphic AI approaches specifically for socio-labor regulation systems. Current data indicates uneven deployment of neuromorphic labor systems across BRICS megacities, with Shanghai showing a high deployment score of 9.5/10, significantly ahead of cities like Durban at 5.2/10, highlighting the uneven global progress in applying advanced AI for workforce management, including in Russian cities like Moscow with a 7.8/10 deployment score, as the worldwide market for AI in HR is rapidly expanding towards billions. Key metrics examined for these specialized systems include AI-driven job matching efficiency, the number of neural network workforce training programs, and labor market prediction accuracy. The study concludes that achieving effective and ethical socio-labor regulation through AI requires a human-centric approach that addresses disparities and integrates technological, social, and psychological considerations for inclusive urban development. Participation in the article: Irina Karabulatova - general editing, writing the \"introduction\" and \"discussion\" sections; Olga Ergunova - project idea, writing the \"results\" section, working on models (Fig.2-3), compiling tables; Andrey Somov - working on the project methodology and writing the code.","author":[{"family":"Karabulatova","given":"Irina"},{"family":"Ergunova","given":"Olga"},{"family":"Somov","given":"Andrey"}],"issued":{"date-parts":[[2025]]},"DOI":"10.21203/rs.3.rs-6567951/v1","URL":"https://doi.org/10.21203/rs.3.rs-6567951/v1","source":"crossref"},{"id":"doi:10.1088/1361-665x/ae8226","type":"article-journal","title":"Beyond biomimicry: a review of next-generation flexible tactile sensing from viscoelastic material design to neuromorphic computing","abstract":"Abstract Flexible tactile sensors have emerged as vital transduction interfaces for human–machine symbiosis and soft robotics. Despite significant advances, two persistent bottlenecks have hindered their practical viability. At the material level, the intrinsic viscoelasticity of polymeric matrices enforces stringent trade-offs among pressure sensitivity, output linearity, and loading–unloading hysteresis. At the system level, the conventional spatial separation of sensory acquisition from signal processing incurs considerable bandwidth and energy overheads when handling high-dimensional, unstructured analog data. To address these limitations, this review proposes a vertically integrated co-design framework that systematically couples material engineering, device architecture, and neuromorphic algorithm design. At the material-device level, we survey structural strategies, including dynamic network construction, rigid-phase anchoring, and topological interlocking, aimed at suppressing microscale viscoelastic relaxation and preventing interfacial delamination at mechanically heterogeneous junctions. Complementarily, alternating current excitation protocols have been evaluated to mitigate the electrochemical degradation intrinsic to iontronic architectures. We argue that the concurrent resolution of these mechanical and electrochemical instabilities provides the drift-free analog baseline essential for downstream neuromorphic computation, a cross-domain connection that has received limited systematic attention in the existing literature. Building on this stabilized platform, we further elucidate how intrinsic material dynamics, notably ferroelectric polarization switching and ion migration, can serve as native computational primitives, effectively linking material engineering and algorithm-design layers within the proposed framework. The coupling of these material-embedded dynamics with spiking neural networks enables in situ temporal encoding, whereby continuous mechanical stimuli are transduced into sparse event-driven spike trains, circumventing the von Neumann bottleneck. This co-design perspective offers a pathway for flexible tactile systems to evolve beyond passive structural biomimicry toward autonomous cognitive interfaces with real-time edge-localized adaptability.","author":[{"family":"Li","given":"Jun"},{"family":"Ji","given":"Yongcheng"},{"family":"Xin","given":"Lipan"},{"family":"Li","given":"Linan"},{"family":"Li","given":"Chuanwei"},{"family":"Wang","given":"Zhiyong"},{"family":"Wang","given":"Shibin"},{"family":"Zhou","given":"Lei"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/1361-665x/ae8226","URL":"https://doi.org/10.1088/1361-665x/ae8226","source":"crossref"},{"id":"doi:10.1002/smll.74180","type":"article-journal","title":"Self-Powered Neuromorphic Systems Based on Tribotronics Synaptic Devices.","abstract":"ABSTRACT Artificial synaptic devices based on field‐effect transistors (FETs) are essential building blocks for neuromorphic computing systems that emulate the signal processing and learning capabilities of biological neural networks. However, most FET‐based synaptic devices depend on external power sources, which increases energy consumption and limits their applications in self‐powered systems. Triboelectric nanogenerators (TENGs) have recently emerged as promising candidates for self‐powered neuromorphic systems that simultaneously harvest ambient mechanical energy and modulate synaptic signals, thereby enabling energy‐efficient neuromorphic operation. This review summarizes recent advances in TENG‐driven three‐terminal artificial synaptic devices, focusing on their device architectures, operating principles, and synaptic functionalities. It discusses the unique electrical characteristics of TENG outputs, including high internal impedance and pulsed voltage signals, and their correlation with synaptic behaviors such as excitatory and inhibitory postsynaptic currents, nonvolatile weight retention, and frequency‐ and time‐dependent plasticity. In addition, it explores the emerging applications of these devices in tactile sensing, human–machine interaction, auditory perception, and multimodal neuromorphic systems. Finally, the review highlights the current challenges, including signal stability, impedance matching, multisensory integration, and power consumption, and discusses the potential strategies to address these limitations, along with the future perspectives to advance the development of next‐generation self‐powered neuromorphic computing systems.","author":[{"family":"Shrestha","given":"Kumar"},{"family":"Akbari","given":"Mohammad"},{"family":"Anbari","given":"Alireza"},{"family":"Pandey","given":"Puran"},{"family":"Zhuiykov","given":"Serge"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1002/smll.74180","URL":"https://doi.org/10.1002/smll.74180","source":"europepmc"},{"id":"doi:10.1088/2634-4386/adce28","type":"article-journal","title":"Exploiting drain-erase scheme in ferroelectric FETs for logic-in-memory","abstract":"Abstract The conventional computing platforms based on von-Neumann architecture are highly space- and energy-intensive while handling the emerging applications such as AI, ML, and big data. To overcome the von Neumann bottleneck, compact and light-weight logic-in-memory (LiM) implementations of Boolean logic gates based on emerging non-volatile memory (e-NVM) such as RRAMs, PCM, STT-MRAMs, etc were proposed recently. However, these e-NVMs not only exhibit significant temporal and spatial variability, but their large-scale integration with CMOS process is also a technological challenge. To overcome these issues with the emerging non-volatile memories, ferroelectric FETs based on CMOS-compatible doped hafnium oxide with the capability of large-scale CMOS integration in the advanced logic nodes were proposed. Considering the high scalability and CMOS-compatibility of the FeFETs, in this work, for the first time, we propose a LiM implementation utilizing a single ferroelectric fully-depleted-silicon-on-insulator (Fe-FDSOI) FET exploiting the unique drain erase phenomenon. In our proposed LiM implementation, inputs are applied at the gate and drain terminals using a novel input-to-voltage mapping scheme, and output is obtained as the current flowing through the Fe-FDSOI FET. We utilize an experimentally calibrated compact model of the ferroelectric capacitor connected to the baseline industry standard BSIM-IMG compact model for the FDSOI transistor for proof of concept demonstration. We also perform a comprehensive analysis of the performance metrics of the proposed LiM implementation. Our results indicate that we can realize at least 10 Boolean logic gates with high energy and area-efficiency utilizing the proposed scheme.","author":[{"family":"Rafiq","given":"Musaib"},{"family":"Chauhan","given":"Yogesh"},{"family":"Sahay","given":"Shubham"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1088/2634-4386/adce28","URL":"https://doi.org/10.1088/2634-4386/adce28","source":"crossref"},{"id":"doi:10.1088/2634-4386/ae9982","type":"article-journal","title":"Moisture effects on diffusive memristors","abstract":"Abstract Biological nervous systems encode information through transient, event-driven spikes, motivating neuromorphic hardware that can reproduce such dynamics efficiently. Diffusive memristors are strong candidates because their volatile, threshold-driven responses generate spike-like signals and support temporal processing. However, their switching relies on ionic motion and transient metal clusters that are highly sensitive to interfacial chemistry, leaving the influence of environmental factors unresolved. Here, the chemical role of moisture in enabling volatile threshold switching is systematically established in symmetric Pt/Ag/SiO2/Ag/Pt diffusive memristors using a rigorously controlled fabrication and measurement framework. Systematic comparisons show that interfacial water facilitates Ag oxidation and hydrated transport pathways, whereas its depletion suppresses switching or increases variability. These findings demonstrate that moisture is not a minor perturbation but a fundamental chemical requirement for stable operation in such device systems. Controlling interfacial hydration emerges as a key design principle for achieving reliable and commercially scalable diffusive memristors for neuromorphic hardware.","author":[{"family":"Kim","given":"Seung"},{"family":"Zhao","given":"Ruoyu"},{"family":"Xu","given":"Yichun"},{"family":"Zhao","given":"Jian"},{"family":"Liao","given":"Han"},{"family":"Yang","given":"JJ"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/2634-4386/ae9982","URL":"https://doi.org/10.1088/2634-4386/ae9982","source":"crossref"},{"id":"doi:10.26599/nr.2026.94908419","type":"article-journal","title":"Green electronics based on biopolymer memristors toward sustainable neuromorphic devices","abstract":"Abstract The artificial intelligence-driven data deluge presents formidable challenges to conventional computing architectures, which are constrained by the von Neumann bottleneck and complementary metal-oxide-semiconductor (CMOS) scaling limits. Neuromorphic computing demonstrates breakthrough potential through its in-memory computing paradigm. Memristors, recognized as the most promising core devices in this field, excel at emulating synaptic plasticity while exhibiting nonlinear dynamic responses and conductance modulation capabilities. However, conventional inorganic memristors face critical limitations, including insufficient mechanical flexibility, biotoxicity, and non-degradability, which hinder their applications in wearable and implantable neuromorphic devices. In contrast, biomaterials and biological tissues emerge as viable platforms for overcoming these bottlenecks and realizing next-generation sustainable neuromorphic systems, owing to their exceptional biocompatibility, environmental benignity, and ultra-thin lightweight characteristics. This review focuses on cutting-edge developments in renewable biopolymer-based memristors (e.g., natural biomolecules like proteins and DNA, alongside bioengineered polymers such as poly(lactic acid) (PLA)). We elucidate how their molecular-level multifunctional groups and hierarchical structures drive high-performance memristive behaviors and bio-inspired synaptic functions. Special emphasis is placed on analyzing these devices’ superior biocompatibility and biodegradability, with in-depth discussions on how such properties enable implantable neuromorphic applications. Finally, we critically examine persisting challenges including environmental sensitivity/resistance state drift, stochastic ion migration pathways, and scalable integration hurdles. Potential countermeasures and feasible development pathways are systematically explored.","author":[{"family":"Zhang","given":"Xiaochao"},{"family":"Wang","given":"Haiting"},{"family":"Zhang","given":"Xuzhao"},{"family":"Wang","given":"Dongyue"}],"issued":{"date-parts":[[2026]]},"DOI":"10.26599/nr.2026.94908419","URL":"https://doi.org/10.26599/nr.2026.94908419","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-9477612/v1","type":"article-journal","title":"NHRN: A Neuromorphic Hierarchical Resonance Network for EEG-based Parkinson's Disease Classification","abstract":"Abstract Parkinson's disease (PD) is a progressive neurodegenerative disorder characterized by motor dysfunction and underlying neural rhythm disturbances, making early and objective diagnosis crucial for effective clinical intervention. Electroencephalogram (EEG) signals provide a non-invasive window into brain dynamics; however, their non-linear, multi-scale temporal behavior poses challenges for conventional approaches. This study presents a framework for PD detection using EEG signals, leveraging neural signatures such as reduced beta activity, altered gamma rhythms, disrupted cortico--basal ganglia interactions, and abnormal cross--frequency coupling. The proposed Neuromorphic Hierarchical Resonance Network (NHRN) model, which combines fractal multi-scale convolution, synaptic gating, oscillatory feature learning, adaptive receptive fields, cognitive load regulation, memory integration, and spectral--temporal fusion. Higher-level representations are refined through hierarchical abstraction, resonance-based attention, and adaptive channel weighting, enabling effective modeling of long-range dependencies and nonlinear brain activity. Clinical validity is established via UPDRS-III stratified analysis,revealing a near-perfect correlation between motor symptom severity andclassification performance across medication ON and OFF states, withmonotonically increasing accuracy from mild to severe disease stages,confirming that the NHRN captures genuine pathophysiological EEGbiomarkers and demonstrates potential for early detection, medicationmonitoring, and disease progression tracking.Rigorous evaluation spanning multiple independent benchmarks--UCSD,UNM, and Iowa under heterogeneous cohorts and varying recordingconditions achieves 98.1\\,%, 96.8\\,%, and 95.7\\,% accuracy,respectively, while multi-dataset training delivers 97.1\\,% accuracywith low variability and only 1.07M parameters, confirming stronggeneralization and suitability for online embedded clinical deployment.","author":[{"family":"Lalawat","given":"Rajveer"},{"family":"Kushwaha","given":"Nikhil"},{"family":"Yang","given":"Albert"}],"issued":{"date-parts":[[2026]]},"DOI":"10.21203/rs.3.rs-9477612/v1","URL":"https://doi.org/10.21203/rs.3.rs-9477612/v1","source":"crossref"},{"id":"doi:10.1126/sciadv.aed0971","type":"article-journal","title":"Neuromorphic tissues: Soft biomolecular networks for brain-inspired temporal computing.","abstract":"Brains achieve extraordinary efficiency in processing temporal information through dense interconnectivity and recurrent feedback among neurons. Inspired by this principle, we introduce neuromorphic tissues—soft biomolecular networks comprising cell-sized aqueous compartments interconnected by lipid membranes containing voltage-gated ion channels. When a compartment is electrically stimulated by current injection, the membranes separating it from neighboring compartments polarize until channel activation occurs, transiently transforming the interface into a conductive synapse that couples adjacent nodes. These dynamics generate intrinsic physical recurrence, enabling the network to encode, propagate, and reconstruct time-dependent signals without external feedback circuitry. Experiments and modeling demonstrate nonlinear, fading-memory, and recurrent dynamics characteristic of reservoir computing, enabling accurate prediction of nonlinear and chaotic sequences such as NARMA-10 and the Lorenz attractor. This work suggests that spatial interconnectivity can enhance the computational capabilities of physical reservoirs and highlights soft, self-assembled materials as a promising platform for implementing such interconnected systems.","author":[{"family":"Armendarez","given":"Nicholas"},{"family":"Mohamed","given":"Ahmed"},{"family":"Hasan","given":"Md"},{"family":"Najem","given":"Joseph"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1126/sciadv.aed0971","URL":"https://doi.org/10.1126/sciadv.aed0971","source":"europepmc"},{"id":"doi:10.3390/s26165311","type":"article-journal","title":"A Neuro-Inspired Rate-Encoded Descriptor for High-Speed Asynchronous Robotic Vision.","abstract":"This paper presents the Post-Stimulus Time-Dependent Event Descriptor (P-TED), a novel “pure event” feature descriptor designed for neuromorphic vision data. Unlike conventional frame-based approaches or hybrid methods that transform event data into intermediate representations, P-TED operates directly on asynchronous event streams, thereby preserving the intrinsic low-latency and high-temporal-resolution advantages of event-based sensors. The descriptor integrates two complementary feature sets: a motion feature vector, which aggregates spatial relationships within a Moore neighbourhood to quantify stimulus direction, and a pattern feature vector, which employs rate encoding to capture temporal excitation signatures. The efficacy of the P-TED framework is validated through three distinct experiments: object and character recognition (MNIST-DVS and CIFAR10-DVS), mobile robot movement analysis, and complex non-rigid robotic hand gesture recognition (RoShamBo). Experimental results demonstrate that the P-TED achieves a significant reduction in classification latency, requiring only 2.7 ms compared to the 10.3 ms recorded by the state-of-the-art Distribution-Aware Retinal Transform (DART) framework. Additionally, P-TED exhibits superior robustness in disambiguating symmetric and mirrored motions, as well as in maintaining stability under non-linear fluctuations in event density caused by changing scale. This work establishes P-TED as a high-speed, computationally efficient, and explainable solution for real-time neuromorphic robotic vision systems.","author":[{"family":"Harrigan","given":"Shane"},{"family":"Coleman","given":"Sonya"},{"family":"Kerr","given":"Dermot"},{"family":"Yogarajah","given":"Pratheepan"},{"family":"Wu","given":"Chengdong"},{"family":"Fang","given":"Zheng"}],"issued":{"date-parts":[[2026]]},"DOI":"10.3390/s26165311","URL":"https://doi.org/10.3390/s26165311","source":"europepmc"},{"id":"doi:10.21203/rs.3.rs-9546289/v1","type":"article-journal","title":"Spike-Based Neuromorphic Processing of Encrypted Data with BioEncryptSNN","abstract":"Abstract Deep learning is increasingly applied to everyday problems, particularly through neural networks. As these models grow in computational load and resource requirements, cloud-based solutions have become common. Spiking Neural Net-works (SNNs), the third generation of neural networks, offer improved efficiency and reduced power consumption by mimicking the brain’s sparse, dynamic, and event-driven computation. However, the use of cloud platforms raises concerns about preserving the privacy of confidential data processed by remote servers. Among the solutions proposed, asymmetric encryption is particularly promising because it uses a public–private key pair to secure communication without requiring shared secrets. This enables clients to send encrypted data to models owned by external parties while maintaining privacy. This article introduces BioEncryptSNN, an SNN-based framework for spike-based processing of encrypted data. Rather than proposing a new cryptographic primitive, BioEncryptSNN converts ciphertext into spike trains and exploits temporal neural dynamics to enable neuromorphic ciphertext processing, robust-ness analysis, and reliable plaintext recovery through the underlying encryption algorithms. The framework was benchmarked against widely used cryptographic standards, including Advanced Encryption Standard (AES)-128, Rivest-Shamir-Adleman (RSA)-2048, Data Encryption Standard (DES), ChaCha20, and Elliptic Curve Integrated Encryption Scheme (ECIES). Results show that BioEncryptSNN preserves plaintext integrity and, under the evaluated experimental conditions, achieves lower mean end-to-end latency than the selected AES soft-ware baseline while maintaining robustness under noisy conditions. Performance depends on encoding overhead, implementation details, and execution context, and may vary across hardware and deployment environments. BioEncryptSNN achieved a mean execution time of 0.016 ms, low variability, and a throughput of62,500 iterations per second, corresponding to a 1.19× speed improvement over the PyCryptodome AES baseline under the evaluated experimental conditions. The framework further demonstrates adaptability across symmetric and asymmetric ciphers, spanning block, stream, and public-key encryption settings, which supports its potential as a scalable approach to encrypted data processing for downstream applications such as secure analytics and vulnerability assessment. The cryptographic security of the framework is inherited from the underlying encryption algorithms, whereas the SNN enables ciphertext processing without providing formal mathematical guarantees of cryptographic strength. Accordingly, BioEncryptSNN is intended primarily as a proof-of-concept framework for neuromorphic ciphertext processing rather than as a replacement for production cryptographic systems.","author":[{"family":"Pulivathi","given":"Mahitha"},{"family":"Rodrigues","given":"Ana"},{"family":"Ihianle","given":"Isibor"},{"family":"Oikonomou","given":"Andreas"},{"family":"Boppu","given":"Srinivas"},{"family":"Machado","given":"Pedro"}],"issued":{"date-parts":[[2026]]},"DOI":"10.21203/rs.3.rs-9546289/v1","URL":"https://doi.org/10.21203/rs.3.rs-9546289/v1","source":"crossref"},{"id":"doi:10.2139/ssrn.6301398","type":"manuscript","title":"Neuromorphic Visual Odometry with Spiking Neural Networks: Evaluation and Benchmarking on the Akida Platform","abstract":"Event-driven vision sensors and neuromorphic processors represent two converging technologies enabling low-latency, energy-efficient perception for autonomous systems. This paper presents a spiking neural network (SNN)–based visual odometry (VO) framework that exploits the asynchronous output of an event camera to estimate 6-DoF motion. To enable compatibility with convolutional SNN architectures, incoming event streams are temporally aggregated into voxel-grid representations, trading strict event-level asynchrony for structured spatiotemporal encoding. These resulting representations are processed through a convolutional spiking architecture trained with LiDAR–IMU-fused ground truth poses. Evaluations on diverse indoor, outdoor, and hybrid environments demonstrate accurate motion estimation over long-duration trajectories of up to 1066 s (approximately 1.07 km of traversal), achieving a mean translation RMSE of 0.77 m and relative pose error of 0.18 m across test sequences. To assess neuromorphic deployment efficiency, a behaviorally comparable Akida-compatible model was benchmarked using BrainChip’s MetaTF runtime environment. The event-based implementation achieved an average inference latency of 63.15 ms with an estimated energy consumption of only 1.89 mJ per sample under MetaTF-based neuromorphic execution. Compared to conventional CPU and GPU baselines, this corresponds to energy reductions exceeding three orders of magnitude, highlighting the strong potential for event-based spiking VO for embedded and power-constrained robotic platforms.","author":[{"family":"Tenzin","given":"Sangay"},{"family":"Rassau","given":"Alexander"},{"family":"Chai","given":"Douglas"},{"family":"Moniruzzaman","given":"Md"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2139/ssrn.6301398","URL":"https://doi.org/10.2139/ssrn.6301398","source":"crossref"},{"id":"doi:10.1002/aisy.70508","type":"article-journal","title":"Low‐Voltage Operation of IGZO Memtransistors Enabled by Coupling Enhancement Through 2T Drain‐Driven Operation for Neuromorphic Computing","abstract":"Neuromorphic computing has emerged as a promising approach to overcome the limitations of the von Neumann architecture by enabling energy‐efficient data processing. Here, we demonstrate a two‐terminal (2T) flash‐type indium–gallium–zinc oxide memtransistor for voltage‐scalable and compact neuromorphic systems. By employing drain‐induced Fowler–Nordheim tunneling, the proposed device achieves an operating voltage below 10 V while maintaining a large memory window of 6 V and a high on/off ratio of 10 9 . The reduced operating voltage is enabled by coupling enhancement through 2T drain‐driven operation. The device represents linear weight modulation and stable retention with only 0.31% change after 10 4 s, and robust endurance over 10 000 cycles. As a system‐level validation of the 2T drain‐driven platform, a memtransistor array is fabricated and exhibits uniform device characteristics. Furthermore, spiking neural network simulations incorporating nonideal characteristics demonstrate that an intensity‐based input encoding improves classification accuracy with reduced time steps for both Fashion‐MNIST and CIFAR‐10 datasets.","author":[{"family":"Park","given":"Junhyeong"},{"family":"Bae","given":"Sunyeol"},{"family":"Yun","given":"Yumin"},{"family":"Park","given":"Chae‐hwan"},{"family":"Lee","given":"Donghyeon"},{"family":"Lee","given":"Soo‐yeon"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1002/aisy.70508","URL":"https://doi.org/10.1002/aisy.70508","source":"crossref"},{"id":"doi:10.1002/adom.71477","type":"article-journal","title":"LSPR‐Enhanced Reconfigurable NIR Organic Synapse for Neuromorphic Computing and Biomedical Sensing","abstract":"ABSTRACT Light‐controlled memristors are a key technology for overcoming the computational limits of von Neumann architectures, offering great potential for next‐generation neuromorphic computing. Organic optical synapses offer a promising route toward low‐power, multifunctional neuromorphic hardware, yet their weak response to near‐infrared (NIR) light limits applications in biomedical sensing and intelligent vision. Here, a reconfigurable NIR synaptic device based on an organic heterojunction (ITO/PEDOT:PSS:AgTNPs/PM6:L8‐BO/PFN‐Br/Ag) is demonstrated, in which silver triangular nanoplates (AgTNPs) embedded in the poly(3,4‐ethylenedioxythiophene):poly(styrenesulfonate)‌ (PEDOT:PSS) layer amplify NIR absorption and enhance synaptic performance. The resulting device exhibits reconfigurable synaptic plasticity—including short‐term memory (STM), long‐term memory (LTM), and paired‐pulse facilitation (PPF)—and achieves an ultralow operating energy of 2.7 fJ per event, among the lowest reported for NIR organic synapses. Beyond mimicking biological learning behaviors, the device performs optical encoding/decoding of Morse code and enables high‐accuracy neuromorphic tasks, achieving 98.9% accuracy in Modified National Institute of Standards and Technology database (MNIST) recognition and 94.2% accuracy in electrocardiogram (ECG) recognition using a convolutional neural network (CNN). Mechanistic analysis and finite element method (FEM) simulations confirm that localized surface plasmon resonance (LSPR)‐mediated photonic field enhancement increases the probabilities of exciton generation and dissociation, accounting for the improved NIR responsivity. This work establishes a practical strategy for developing NIR‐responsive neuromorphic and biomedical sensing platforms.","author":[{"family":"Xu","given":"Zhaoxin"},{"family":"Dong","given":"Shihui"},{"family":"Lian","given":"Hong"},{"family":"Wang","given":"Xianglin"},{"family":"Cao","given":"Jiangcheng"},{"family":"Ding","given":"Jiahui"},{"family":"Lan","given":"Guotao"},{"family":"Wang","given":"Shuanglong"},{"family":"Ding","given":"Xingdong"},{"family":"Gao","given":"Peng"},{"family":"Dong","given":"Qingchen"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1002/adom.71477","URL":"https://doi.org/10.1002/adom.71477","source":"crossref"},{"id":"doi:10.1007/s00521-026-11941-3","type":"article-journal","title":"Self-tuning neuromorphic controller for real-time UAS trajectory tracking based on prescribed error sensitivity","abstract":"Abstract Inspired by a learning mechanism encountered in the mammal brain, this study presents a Neuromorphic Self-Tuning Proportional-Integral-Derivative (PID) controller for Unmanned Aerial Systems (UAS). The controller is derived from a Spiking Neuronal Network (SNN) and enhanced with a biologically plausible supervised learning rule known as Prescribed Error Sensitivity (PES). This control framework enables the UAS to maintain stability near singular regions, improving robustness to input perturbations, and reducing trajectory tracking error. A series of experimental tests consisting of real-time UAS trajectory tracking demonstrate the applicability and effectiveness of the proposed approach.","author":[{"family":"Olivares","given":"Anel"},{"family":"Espinoza","given":"Eduardo"},{"family":"Ramos","given":"Luis"},{"family":"Carrillo","given":"Luis"},{"family":"Sornborger","given":"Andrew"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1007/s00521-026-11941-3","URL":"https://doi.org/10.1007/s00521-026-11941-3","source":"crossref"},{"id":"doi:10.35848/1347-4065/ae4be6","type":"article-journal","title":"A 319.3 TOPS W\n                    <sup>−1</sup>\n                    high-efficient computing-in-memory engine with 55 nm high-density multi-level NOR-flash array and event-driven neuromorphic architecture","abstract":"Abstract This paper presents a high-efficiency spiking neural network (SNN) accelerator based on a 1.5 T NOR flash device. By leveraging the non-volatile and tunable conductance characteristics of NOR flash, the proposed design enables precise synaptic weight storage and low-power updates in a computing-in-memory architecture. The accelerator adopts a 512 × 1024 array structure, supporting large-scale SNN parallel inference and event-driven computation. With optimized spike accumulation and neuron firing circuits, the system achieves high computational density with reduced power consumption. Experimental results demonstrate an energy efficiency of 319.3 TOPS W −1 , highlighting its superiority over SRAM- or RRAM-based implementations and its potential for edge artificial intelligence applications such as low-power sensing and on-device inference.","author":[{"family":"Cheng","given":"Yue"},{"family":"Guo","given":"Yaolei"},{"family":"Wang","given":"Hanfeng"},{"family":"He","given":"Zhaolong"},{"family":"Tang","given":"Chenhao"},{"family":"Wang","given":"Xinjian"},{"family":"Chi","given":"Jingyao"},{"family":"Gao","given":"Dawei"},{"family":"Qi","given":"Dianyu"},{"family":"Ma","given":"Yitao"}],"issued":{"date-parts":[[2026]]},"DOI":"10.35848/1347-4065/ae4be6","URL":"https://doi.org/10.35848/1347-4065/ae4be6","source":"crossref"},{"id":"doi:10.59717/j.xinn-inform.2026.100027","type":"article-journal","title":"Recent progress on integrated neuromorphic chips","abstract":"&lt;p&gt;In the post-Moore era, advancing integrated circuits necessitates the development of artificial synaptic devices capable of replicating the functions of human brain with neuromorphic computing, which consists of numerous biological synapses. The neuromorphic chips based on integrated synaptic devices are the promising candidates due to their unique properties, which enable the simulation of human brain and biological synapse plasticity. In recent years, significant progress has been made in the development of optoelectronic synaptic devices and integrated neuromorphic chips. This review summarizes the recent advancements in the integrated synaptic devices, focusing on the structures, integrated technologies and their applications of neuromorphic chips. It discusses the two-terminal and three-terminal configurations of synaptic devices, emphasizing their advantages in simulating biological synapse functions. These advantages include a simple structure, low power consumption, high stability, a wide switching ratio and fast programming speed. Furthermore, the discussion covers the key applications of synaptic devices and integrated neuromorphic chips, such as image recognition, brain-like computing, tactile and vision sensors. For the synaptic-device-based neural networks in image recognition, they have great potential in tactile technology and visual bionics. Other potential applications have also been reviewed and discussed, including intelligent robots, electric skin, human-computer interaction and brain-computer interface, &lt;i&gt;etc&lt;/i&gt;. Hence, neuromorphic chips have great promising application prospects, which can significantly promote the development of artificial intelligence and human civilization.&lt;/p&gt;","author":[{"family":"Gu","given":"Xiushuo"},{"family":"Bian","given":"Lifeng"},{"family":"Cheng","given":"Linrui"},{"family":"Song","given":"Chaoyun"},{"family":"Wang","given":"Yibin"},{"family":"Zhang","given":"Jianya"},{"family":"Zhao","given":"Yukun"}],"issued":{"date-parts":[[2026]]},"DOI":"10.59717/j.xinn-inform.2026.100027","URL":"https://doi.org/10.59717/j.xinn-inform.2026.100027","source":"crossref"},{"id":"doi:10.1063/5.0314841","type":"article-journal","title":"Superlattice-like Ge2Sb2Te5/Sb2S3 based phase-change memory enabling linear conductance modulation for neuromorphic computing","abstract":"Artificial synaptic devices that emulate biological synaptic behavior have garnered significant research interest for their potential in enabling efficient and low-power computing architectures. Phase-change memory (PCM) has emerged as a promising candidate for artificial synaptic devices, owing to its non-volatility, high speed, and low-power consumption. However, the inherent abrupt and hard-to-control resistance switching in PCM impedes linear and continuous conductance modulation, which substantially restricts the performance of PCM-based synaptic devices. This work demonstrates an electronic synaptic device fabricated from a Ge2Sb2Te5/Sb2S3 superlattice-like (SLL) phase-change thin film. The unique SLL structure effectively suppresses the rapid crystallization and resistance drift. The device operates at a driving voltage below 1 V with low-power consumption and exhibits eight stable resistance states. It achieves a notably low resistance drift coefficient of 0.0006, which remains stable for over 1000 s. Moreover, through a tailored programming strategy, the device successfully implements synaptic weight updates via long-term potentiation and long-term depression, thereby alleviating the nonlinearity and asymmetry commonly observed in PCM-based conductance modulation. In a handwritten digit recognition task, the device enabled a recognition accuracy of around 97.6%, highlighting its potential for enhancing the precision and reliability of neuromorphic computing systems.","author":[{"family":"Li","given":"Lele"},{"family":"Song","given":"Mengru"},{"family":"Gu","given":"Han"},{"family":"Hu","given":"Ziyang"},{"family":"Lu","given":"Yegang"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1063/5.0314841","URL":"https://doi.org/10.1063/5.0314841","source":"crossref"},{"id":"doi:10.1002/agt2.70319","type":"article-journal","title":"Plasmon‐Doped Organic Heterojunction Optoelectronic Synapses for Near‐Infrared Visual Memory and Neuromorphic Computing","abstract":"ABSTRACT The explosive growth of artificial intelligence has intensified demands for new computing paradigms beyond conventional von Neumann architectures. In response, brain‐inspired computing‐in‐memory technologies are emerging as a promising path forward. Here, we designed a two‐terminal optical synaptic device utilizing organic heterojunctions doped with gold nanorods (AuNRs), leveraging the electric field enhancement innate to the localized surface plasmon resonance (LSPR) effect. The device doped with 1 wt% AuNRs demonstrates a markedly enhanced light absorption capacity in the near‐infrared (NIR) region of 808 nm. The generation rate of photogenerated excitons increases by 16.8%, while the probability of exciton dissociation rises by 8.4%. The paired‐pulse facilitation (PPF) index reaches 114.6% (Δ t = 1 s), indicating heightened sensitivity to optical pulse parameters. Additionally, Hall effect measurements were performed to characterize the electrical properties of the PEDOT:PSS:AuNRs films. The carrier mobility of the doped films increased 20‐fold compared to pristine PEDOT:PSS due to electron injection from AuNRs. This enhanced mobility contributes to faster synaptic response and higher conductance tunability in the synapse device, further supporting its performance in neuromorphic computing tasks. Furthermore, we successfully simulated the dynamic “learning–forgetting–relearning” processes associated with human visual memory. By exploiting the tunable conductance of the optimized synaptic device, we implemented both convolutional neural networks (CNNs) and convolutional spiking neural networks (CSNNs) for weight updates. After 100 and 150 training epochs, the system achieved recognition accuracies up to 98.57% for handwritten digits and 92.01% for dynamic gestures. This work presents an effective plasmon‐doping approach to enhancing the performance of organic memristors and can be extended to other material systems.","author":[{"family":"Cao","given":"Jiangcheng"},{"family":"Lian","given":"Hong"},{"family":"Wang","given":"Xianglin"},{"family":"Huang","given":"Qishuai"},{"family":"Ding","given":"Jiahui"},{"family":"Xia","given":"Jiangnan"},{"family":"Wang","given":"Shuanglong"},{"family":"Hu","given":"Weijin"},{"family":"Wu","given":"Tom"},{"family":"Dong","given":"Qingchen"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1002/agt2.70319","URL":"https://doi.org/10.1002/agt2.70319","source":"crossref"},{"id":"doi:10.20517/iontronics.2026.04","type":"article-journal","title":"Nanofluidic neuromorphic iontronics: a nexus for biological signal transduction","abstract":"Neuromorphic iontronics is an emerging technology based on the sophisticated regulation of ions as signal carriers, demonstrating strong potential in human-machine interaction. Nanofluidic materials exhibit significant advantages in constructing iontronic neuromorphic devices due to their ability to precisely mimic the transport mechanisms of biological ion channels. By synergistically integrating multiple external stimuli, neuromorphic devices that combine perception fusion and memory behaviors are injecting new vitality into the development of artificial intelligence, including computation, neurorobotics, and healthcare. However, this field is still in its infancy and requires robust theoretical foundations, novel computational paradigms, and innovative approaches to functional material design and device integration to advance further. In this Perspective, we provide a comprehensive overview of recent advances in iontronic neuromorphic devices with respect to fabrication, integration, and applications, aiming to illustrate the rapid growth of this field. Finally, we discuss current challenges and future directions, including theoretical models based on ionic properties, neuromorphic signal simulation, and device integration.","author":[{"family":"Li","given":"Lin"},{"family":"Chen","given":"Weipeng"},{"family":"Kong","given":"Xiang"},{"family":"Wen","given":"Liping"}],"issued":{"date-parts":[[2026]]},"DOI":"10.20517/iontronics.2026.04","URL":"https://doi.org/10.20517/iontronics.2026.04","source":"crossref"},{"id":"doi:10.1002/adfm.76104","type":"article-journal","title":"Cavity‐Enhanced High Sensitivity InP Nanosheet Optoelectronic Synaptic Transistors for Neuromorphic Computing","abstract":"ABSTRACT Optoelectronic synaptic transistors that combine sensing and computing functionalities are emerging as key components for next‐generation neuromorphic systems, yet their scalability and performance remain limited by device architecture and fabrication constraints. Here, we report cavity‐enhanced optoelectronic synaptic transistors based on surface‐state modulation in selectively grown InP nanosheets. Benefiting from optical‐cavity‐induced high gain and effective dark‐current suppression, the device exhibits an ultrahigh responsivity of approximately 1.5 × 10 6 A W −1 , a specific detectivity of ≈5 × 10 14 Jones, and a noise‐equivalent power of ≈10 −16 W Hz −1/2 . Precise control of charge dynamics at surface states using electrical and optical stimuli allows the emulation of key synaptic functions, including short‐ and long‐term plasticity, paired‐pulse facilitation, and learning–forgetting–relearning behaviors. Furthermore, convolutional neural network simulations leveraging the device's continuously tunable conductance yield image recognition accuracies approaching 90%. These results establish InP nanosheet optoelectronic synaptic transistors as a scalable and high‐performance platform for integrated neuromorphic computing.","author":[{"family":"Zhang","given":"Xutao"},{"family":"Liu","given":"Liang"},{"family":"Pang","given":"Ningjie"},{"family":"Zhuang","given":"Yezhao"},{"family":"Ni","given":"Sheng"},{"family":"Zhan","given":"Wang"},{"family":"Liu","given":"Changlong"},{"family":"Huang","given":"Hai"},{"family":"Yuan","given":"Xiaoming"},{"family":"Gan","given":"Xuetao"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1002/adfm.76104","URL":"https://doi.org/10.1002/adfm.76104","source":"crossref"},{"id":"doi:10.1002/rar2.70406","type":"article-journal","title":"Efficient Spin‐Orbit Torques Enabled by Vanadium‐Induced Orbital Currents for Neuromorphic Computing","abstract":"ABSTRACT Spin‐orbit torque (SOT) provides an efficient electrical pathway for encoding spin states and underpins emerging, ultrafast, and nonvolatile spintronic memories and logic devices. Reducing the switching current density and power consumption remains a central challenge for practical applications. Here, we demonstrate that a light metal, vanadium (V), can serve as an efficient orbital‐current source to substantially enhance SOT efficiency. In V/Pt/CoFeB heterostructures, orbital currents generated in V are effectively converted into spin currents by the spin‐orbit‐active Pt layer, giving rise to enhanced torques acting on the perpendicularly magnetized CoFeB layer. As a result, the critical switching current density and power are reduced by 54% and 27%, respectively, compared with conventional spin current‐dominated Pt/CoFeB structures. Moreover, V/Pt/CoFeB‐based devices further enable neuromorphic computing functionalities, achieving a handwritten digit recognition accuracy of 89%. These results highlight orbital‐current engineering as a viable strategy for realizing low‐power SOT devices and advancing spintronic hardware for neuromorphic computing.","author":[{"family":"Yu","given":"Zhonghai"},{"family":"Du","given":"Yaohui"},{"family":"Yan","given":"Mengyang"},{"family":"Zhao","given":"Pengnan"},{"family":"Hou","given":"Rui"},{"family":"Li","given":"Keqin"},{"family":"Yang","given":"Lihuan"},{"family":"Guo","given":"Kaiwei"},{"family":"Bian","given":"Bingyue"},{"family":"Xiao","given":"Zhengyu"},{"family":"Cheng","given":"Lei"},{"family":"Wang","given":"Hongru"},{"family":"Lai","given":"Jia‐min"},{"family":"Quan","given":"Zhiyong"},{"family":"Yang","given":"Dongsheng"},{"family":"Liu","given":"Yakun"},{"family":"Wang","given":"Fei"},{"family":"Xu","given":"Xiaohong"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1002/rar2.70406","URL":"https://doi.org/10.1002/rar2.70406","source":"crossref"},{"id":"doi:10.55959/msu05799392.81.2610502","type":"article-journal","title":"Ti/HfOx/TiN-Based Memcapacitor for Neuromorphic Applications","abstract":"This paper presents a Ti/HfOx/TiN structure capable of switching between both resistive and capacitive states. Impedance spectroscopy data suggest an equivalent electrical circuit that increases the resistive switching of a hafnium oxide filament and suggests the presence of a TiON layer in the HfOx/TiN contact region, which may explain the ferroelectric potential switching. The capacitive switching window is 185 pF, resulting in a 75% increase in capacitance, while the resistance during resistive switching increases tenfold. These features represent an example of a versatile structure that can function as both a capacitor and a multi-state resistor. This enables encoding more information in a single cell and the creation of more complex logic elements. The problem of parasitic currents can also be solved, dramatically improving the energy efficiency and reliability of scalable memory arrays.","author":[{"family":"Kuchumov","given":"ID"},{"family":"Martyshov","given":"MN"},{"family":"Ilyin","given":"AS"},{"family":"Novoseltsev","given":"AI"},{"family":"Savchuk","given":"TP"},{"family":"Forsh","given":"PA"},{"family":"Kashkarov","given":"PK"}],"issued":{"date-parts":[[2026]]},"DOI":"10.55959/msu05799392.81.2610502","URL":"https://doi.org/10.55959/msu05799392.81.2610502","source":"crossref"},{"id":"doi:10.21863/jnis/2026.14.1.005","type":"article-journal","title":"Homeostatically Regulated Adaptive Threshold Spiking Neural Networks for Energy-Efficient Inference","abstract":"Spiking Neural Networks (SNNs) represent a biologically grounded computational paradigm in which information is carried by discrete spike events, conferring structural compatibility with low-power, event-driven neuromorphic hardware. A persistent impediment to realising the full energy benefit of SNNs in practice is the near-universal adoption of fixed firing thresholds, which prevent individual neurons from self-regulating their activity and permit task-irrelevant spike generation to accumulate unchecked. This paper introduces and evaluates a Homeostatically Regulated Adaptive Threshold (HAT) mechanism in which each neuron’s firing threshold is updated at every time step in proportion to the deviation between its exponentially smoothed firing rate and a designer-specified target rate. The rule is derived from the proportional control framework and mirrors the intrinsic excitability regulation observed in biological cortical circuits. A novel metric, the Target Achievement Error (TAE), is defined to quantify how faithfully the population reaches its intended operating point. Computational cost is assessed using a synaptic operations (SynOps) proxy that is hardware-agnostic and scales linearly with spike volume. A two-stage screening procedure selects the best configurations from 27 candidates by first enforcing an accuracy constraint and then ranking them by spike reduction, TAE, energy, and accuracy. Controlled experiments on MNIST show that the fixed-threshold baseline attains a mean validation accuracy of 92.52%±1.13% across three random seeds, with an average of 21.86 × 106 spikes and an estimated SynOps energy of 211.70 mJ. The top-ranked homeostatic configuration (α = 0.1, ftarget = 0.1, γ = 0.2) achieves 92.02% validation accuracy with a 15.54% reduction in total spike activity, incurring only a 0.50% absolute accuracy shortfall. These results indicate that homeostatic threshold regulation offers a structurally non-invasive route to improved inference energy in gradient-trained SNNs.","author":[{"family":"Mannickathan","given":"Gripsy"},{"family":"Varghese","given":"Yeldo"},{"family":"Sabu","given":"Noel"}],"issued":{"date-parts":[[2026]]},"DOI":"10.21863/jnis/2026.14.1.005","URL":"https://doi.org/10.21863/jnis/2026.14.1.005","source":"crossref"},{"id":"doi:10.1142/s0129065725500364","type":"article-journal","title":"Global–Local Feature Fusion Network Based on Nonlinear Spiking Neural Convolutional Model for MRI Brain Tumor Segmentation","abstract":"Due to the differences in size, shape, and location of brain tumors, brain tumor segmentation differs greatly from that of other organs. The purpose of brain tumor segmentation is to accurately locate and segment tumors from MRI images to assist doctors in diagnosis, treatment planning and surgical navigation. NSNP-like convolutional model is a new neural-like convolutional model inspired by nonlinear spiking mechanism of nonlinear spiking neural P (NSNP) systems. Therefore, this paper proposes a global–local feature fusion network based on NSNP-like convolutional model for MRI brain tumor segmentation. To this end, we have designed three characteristic modules that take full advantage of the NSNP-like convolution model: dilated SNP module (DSNP), multi-path dilated SNP pooling module (MDSP) and Poolformer module. The DSNP and MDSP modules are employed to construct the encoders. These modules help address the issue of feature loss and enable the fusion of more high-level features. On the other hand, the Poolformer module is used in the decoder. It processes features that contain global context information and facilitates the interaction between local and global features. In addition, channel spatial attention (CSA) module is designed at the skip connection between encoder and decoder to establish the long-range dependence between the same layers, thereby enhancing the relationship between channels and making the model have global modeling capabilities. In the experiments, our model achieves Dice coefficients of 85.71[Formula: see text], 92.32[Formula: see text], 87.75[Formula: see text] for ET, WT, and TC, respectively, on the N-BraTS2021 dataset. Moreover, our model achieves Dice coefficients of 83.91[Formula: see text], 91.96[Formula: see text], 90.14[Formula: see text] and 85.05[Formula: see text], 92.30[Formula: see text], 90.31[Formula: see text] on the BraTS2018 and BraTS2019 datasets respectively. Experimental results also indicate that our model not only achieves good brain tumor segmentation performance, but also has good generalization ability. The code is already available on GitHub: https://github.com/Li-JJ-1/NSNP-brain-tumor-segmentation .","author":[{"family":"Li","given":"Junjie"},{"family":"Peng","given":"Hong"},{"family":"Li","given":"Bing"},{"family":"Liu","given":"Zhicai"},{"family":"Lugu","given":"Rikong"},{"family":"He","given":"Bingyan"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1142/s0129065725500364","URL":"https://doi.org/10.1142/s0129065725500364","source":"crossref"},{"id":"doi:10.1142/s2301385026500044","type":"article-journal","title":"Development of a Centralized Conflict Free Greedy Assignment Learning Spiking Neural Network for Solving a Perimeter Defense Problem","abstract":"In this paper, a Centralized Conflict-free Assignment Learning Spiking neural network (C2ALS) is formulated for solving a Perimeter Defense Problem (PDP). Here, the region between the perimeter and the sensing range of the defender is divided into two layers. The outermost layer, in which the defender can only sense the intruders, is termed the sensing layer. The layer closest to the perimeter in which the defenders can sense and capture the intruders is termed the capture layer. Both layers are further divided into angular segments with respect to the center of the area to be protected. The spatiotemporal movements of both the defenders and intruders in these segments are converted into spikes and given as input to a Spiking Neural Network (SNN). The SNN is trained in a supervised manner to learn the intruder capture assignments of a defender in terms of these segments. Here, the desired assignments required for training a defender are referred to as greedy assignments because while generating them, priority is given to those segments that are relatively closer to the defender’s actual location. Inhibitory connections are used in the SNN architecture to obtain assignments without conflicts. Detailed performance study results show that the proposed C2ALS approach outperforms the other existing centralized optimization-based solutions for PDP by a 17% increase in the success rates of capturing the intruders. Also, C2ALS is the first centralized spike-based learning solution for PDP, capable of generating conflict-free assignments while ensuring robust intruder assignments with minimal defender movement.","author":[{"family":"Thousif","given":"Mohammed"},{"family":"Velhal","given":"Shridhar"},{"family":"Sundaram","given":"Suresh"},{"family":"Sundararajan","given":"Narasimhan"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1142/s2301385026500044","URL":"https://doi.org/10.1142/s2301385026500044","source":"crossref"},{"id":"doi:10.3389/fncel.2025.1534839","type":"article-journal","title":"Synapses mediate the effects of different types of stress on working memory: a brain-inspired spiking neural network study","abstract":"Acute stress results from sudden short-term events, and individuals need to quickly adjust their physiological and psychological to re-establish balance. Chronic stress, on the other hand, results in long-term physiological and psychological burdens due to the continued existence of stressors, making it difficult for individuals to recover and prone to pathological symptoms. Both types of stress can affect working memory and change cognitive function. In this study, we explored the impact of acute and chronic stress on synaptic modulation using a biologically inspired, data-driven rodent prefrontal neural network model. The model consists of a specific number of excitatory and inhibitory neurons that are connected through AMPA, NMDA, and GABA synapses. The study used a short-term recall to simulate working memory tasks and assess the ability of neuronal populations to maintain information over time. The results showed that acute stress can enhance working memory information retention by enhancing AMPA and NMDA synaptic currents. In contrast, chronic stress reduces dendritic spine density and weakens the regulatory effect of GABA currents on working memory tasks. In addition, this structural damage can be complemented by strong connections between excitatory neurons with the same selectivity. These findings provide a reference scheme for understanding the neural basis of working memory under different stress conditions.","author":[{"family":"Du","given":"Chengcheng"},{"family":"Sun","given":"Yinqian"},{"family":"Wang","given":"Jihang"},{"family":"Zhang","given":"Qian"},{"family":"Zeng","given":"Yi"}],"issued":{"date-parts":[[2025]]},"DOI":"10.3389/fncel.2025.1534839","URL":"https://doi.org/10.3389/fncel.2025.1534839","source":"europepmc"},{"id":"doi:10.36227/techrxiv.175037299.95344486/v1","type":"article-journal","title":"Reinforcement Learning with Spiking Neural Networks for Robotic Applications: A Survey","abstract":"Spiking Neural Networks (SNNs) have long been positioned as a biologically plausible and energy-efficient alternative to conventional deep learning models. Reinforcement Learning (RL), on the other hand, provides a framework for autonomous decision-making based on environmental interaction. While the convergence of SNNs and RL holds significant promise for robotics, this pathway remains in its infancy. This survey examines existing spiking RL approaches, categorizing them into bio-inspired models driven by reward-modulated local plasticity (e.g., R-STDP) and gradient-based models adapted from deep RL techniques. Additionally, we review neuron models, encoding schemes, and learning strategies, with emphasis on robotic applications. We further analyze evaluation metrics, emerging trends, limitations, and proposed future directions to bridge the gap between biological plausibility and scalable robotic intelligence. To the best of our knowledge, this work represents the first comprehensive survey specifically dedicated to RL with SNNs in robotic systems.","author":[{"family":"Oikonomou","given":"Katerina"},{"family":"Kansizoglou","given":"Ioannis"},{"family":"Gasteratos","given":"Antonios"}],"issued":{"date-parts":[[2025]]},"DOI":"10.36227/techrxiv.175037299.95344486/v1","URL":"https://doi.org/10.36227/techrxiv.175037299.95344486/v1","source":"crossref"},{"id":"doi:10.2139/ssrn.6291883","type":"manuscript","title":"Multimodal Representation Learning in a Spiking Neural Networks Framework","abstract":"Multimodal representation learning aims to integrate heterogeneous modalities—such as text, image, audio, and video—into a unified representation. However, modality imbalance and inter-modal interference often lead to modality collapse, limiting generalization, especially under noisy or incomplete inputs. To address this, we propose Spiking-based Multimodal Representation Learning(S-MRL), a spiking neural network (SNN) framework that reformulates multimodal learning as a process of spike sequence encoding and integration. S-MRL employs biologically inspired spike encoders and a Spiking Response Model(SRM) to extract temporally structured, modality-specific features. A Gaussian Kernel Spike Distance(GKSD) is introduced to construct a shared latent space across modalities, while Joint Entropy(JE) captures cross-modal information and preserves semantic diversity. A Recurrent Spiking Neural Networks(SNN) architecture is further designed to mitigate forgetting and stabilize training. During inference, an informative fusion mechanism aggregates spike-based representations, and Cross-Entropy(CE) supervises downstream classification. Extensive experiments on four multimodal benchmarks— under both complete and missing modality settings—demonstrate that S-MRL achieves 4%–19% accuracy improvements over existing methods, while enhancing energy efficiency, robustness, and generalization. These results establish S-MRL as a principled and biologically plausible solution for robust multimodal representation learning.","author":[{"family":"Zhang","given":"Yuping"},{"family":"Liu","given":"Yan"},{"family":"Yang","given":"Chunfang"},{"family":"Zhang","given":"Zilin"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2139/ssrn.6291883","URL":"https://doi.org/10.2139/ssrn.6291883","source":"crossref"},{"id":"doi:10.1088/1741-2552/adec1c","type":"article-journal","title":"Manipulation of neuronal activity by an artificial spiking neural network implemented on a closed-loop brain-computer interface in non-human primates","abstract":"Abstract Objective. Closed-loop brain-computer interfaces can be used to bridge, modulate, or repair damaged connections within the brain to restore functional deficits. Towards this goal, we demonstrate that small artificial spiking neural networks can be bidirectionally interfaced with single neurons (SNs) in the neocortex of non-human primates (NHPs) to create artificial connections between the SNs to manipulate their activity in predictable ways. Approach. Spikes from a small group of SNs were recorded from primary motor cortex of two awake NHPs during rest. The SNs were then interfaced with a small network of integrate-and-fire units (IFUs) that were programmed on a custom clBCI. Spikes from the SNs evoked excitatory and/or inhibitory postsynaptic potentials in the IFUs, which themselves spiked when their membrane potentials exceeded a predetermined threshold. Spikes from the IFUs triggered single pulses of intracortical microstimulation (ICMS) to modulate the activity of the cortical SNs. Main results. We show that the altered closed-loop dynamics within the cortex depends on several factors including the connectivity between the SNs and IFUs, as well as the precise timing of the ICMS. We additionally show that the closed-loop dynamics can reliably be modeled from open-loop measurements. Significance. Our results demonstrate a new type of hybrid biological-artificial neural system based on a clBCI that interfaces SNs in the brain with artificial IFUs to modulate biological activity in the brain. Our model of the closed-loop dynamics may be leveraged in the future to develop training algorithms that shape the closed-loop dynamics of networks in the brain to correct aberrant neural activity and rehabilitate damaged neural circuits.","author":[{"family":"Mishler","given":"Jonathan"},{"family":"Yun","given":"Richy"},{"family":"Perlmutter","given":"Steve"},{"family":"Rao","given":"Rajesh"},{"family":"Fetz","given":"Eberhard"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1088/1741-2552/adec1c","URL":"https://doi.org/10.1088/1741-2552/adec1c","source":"europepmc"},{"id":"doi:10.3390/machines14060603","type":"article-journal","title":"A Spiking Neural Network with Attention and Residual Mechanisms for Compound Fault Detection","abstract":"To address the challenges of severe multi-source coupling, easily masked spiking features, and limited selection of key responses in compound fault signals, this paper proposes a compound fault detection method based on a spiking attention residual network (SARN). This method uses the original time-domain vibration signal as input and constructs an end-to-end spiking neural network framework. A hierarchical spiking attention module is designed to enhance multi-level spiking features from both temporal response and feature channel perspectives, thereby highlighting fault-sensitive information and suppressing redundant responses. Furthermore, a cross-layer spiking residual gating mechanism is introduced to mitigate effective information attenuation in spiking neural networks and improve the representation capability of weak fault features. Simultaneously, a multi-label detection strategy is employed to jointly identify multiple fault attributes, thereby improving the recognition rate of coupled compound fault modes. Verification results show that the proposed method achieves high performance in compound fault detection tasks, and compared with other popular methods, it exhibits better feature separability and detection stability.","author":[{"family":"Xing","given":"Yulong"},{"family":"Li","given":"Kun"},{"family":"Li","given":"Xiaoshuai"},{"family":"Liu","given":"Congcong"},{"family":"Wang","given":"Qi"},{"family":"Peng","given":"Cong"},{"family":"Wang","given":"Zisheng"}],"issued":{"date-parts":[[2026]]},"DOI":"10.3390/machines14060603","URL":"https://doi.org/10.3390/machines14060603","source":"crossref"},{"id":"doi:10.3390/bdcc9070173","type":"article-journal","title":"Modeling the Effect of Prior Knowledge on Memory Efficiency for the Study of Transfer of Learning: A Spiking Neural Network Approach","abstract":"The transfer of learning (TL) is the process of applying knowledge and skills learned in one context to a new and different context. Efficient use of memory is essential in achieving successful TL and good learning outcomes. This study uses a cognitive computing approach to identify and explore brain activity patterns related to memory efficiency in the context of learning a new programming language. This study hypothesizes that prior programming knowledge reduces cognitive load, leading to improved memory efficiency. Spatio-temporal brain data (STBD) were collected from a sample of participants (n = 26) using an electroencephalogram (EEG) device and analyzed by applying a spiking neural network (SNN) approach and the SNN-based NeuCube architecture. The findings revealed the neural patterns demonstrating the effect of prior knowledge on memory efficiency. They showed that programming learning outcomes were aligned with specific theta and alpha waveband spike activities concerning prior knowledge and cognitive load, indicating that cognitive load was a feasible metric for measuring memory efficiency. Building on these findings, this study proposes that the methodology developed for examining the relationship between prior knowledge and TL in the context of learning a programming language can be extended to other educational domains.","author":[{"family":"Fard","given":"Mojgan"},{"family":"Petrova","given":"Krassie"},{"family":"Kasabov","given":"Nikola"},{"family":"Wang","given":"Grace"}],"issued":{"date-parts":[[2025]]},"DOI":"10.3390/bdcc9070173","URL":"https://doi.org/10.3390/bdcc9070173","source":"crossref"},{"id":"doi:10.37394/23208.2025.22.16","type":"article-journal","title":"A Spiking Neural Network Approach for Classifying Hand Movement and Relaxation from EEG Signal using Time Domain Features","abstract":"High-performance prosthetic and exoskeleton systems based on EEG signals can improve the quality of life of hand-impaired people. Effective controlling of these assistive devices requires accurate EEG signal classification. Although there have been advancements in the assistive Brain-Computer Interface (BCI) systems, still classifying the EEG signals with high accuracy is a great challenge. The objective of this research is to investigate the accuracy of the EEG signal classification of the Spiking Neural Network (SNN) classifier for factual and exact control of prosthetic and exoskeleton systems for individuals with hand impairment. The EEG dataset has been taken from the BNCI Horizon 2020 website, which is for hand movement-relax events of a patient with high spinal cord injury (SCI) to operate a neuro-prosthetic device attached to the paralyzed right upper limb. The fusion of Dispersion Entropy (DE), Fuzzy Entropy (FE), and Fluctuation based Dispersion Entropy (FDE) with mean and skewness features are extracted from the Motor Imagery (MI) EEG signals and applied to the Spiking Neural Network (SNN) classifier. To compare the performance of this algorithm, these same features have been used in Convolutional Neural Network (CNN), Random Forest (RF), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Logistic Regression (LR) classifiers. It has been found that SNN has given the highest classification accuracy of 80% with a precision of 80.95%, recall of 77.28%, and F1-score of 79.07%. This indicates that SNN with these five features has greater potential in BCI system-based applications.","author":[{"family":"Hossain","given":"Mohammad"},{"family":"Joy","given":"Md"},{"family":"Chowdhury","given":"Mohammed"}],"issued":{"date-parts":[[2025]]},"DOI":"10.37394/23208.2025.22.16","URL":"https://doi.org/10.37394/23208.2025.22.16","source":"crossref"},{"id":"doi:10.3389/fnins.2026.1875437","type":"article-journal","title":"A configurable streaming spiking neural network accelerator with decoupled pixel-level and output-channel parallelism for automatic modulation classification.","abstract":"Streaming spiking neural network (SNN) accelerators are widely adopted on edge platforms for their deterministic, low-latency inference. Realizing their full efficiency, however, requires three properties to hold simultaneously: a router-free streaming dataflow, joint exploitation of temporal and spatial sparsity within that dataflow, and configurable parallelism that can adapt to heterogeneous layers and hardware budgets. Existing streaming SNN accelerators typically achieve at most two of these properties: pixel-level and output-channel parallelism are bound to a single fixed operating point, limiting efficient mapping across heterogeneous layers in automatic modulation classification (AMC) workloads. This paper presents a configurable streaming SNN accelerator that satisfies all three properties by explicitly decoupling pixel-level multiple matrix–vector multiplication (MMV) parallelism from output-channel parallelism. The decoupling is realized within a weight-priority gated one-to-all product (GOAP) dataflow by deterministic offline scheduling, preserving router-free streaming execution while exploiting temporal and spatial sparsity. The proposed architecture is implemented on a Xilinx Virtex-7 field-programmable gate array (FPGA) and evaluated using the RadioML 2016 dataset and a compatible subset of RadioML 2018 under multiple sparsity and quantization settings. Experimental results show that, under comparable end-to-end latency, configurable parallelism enables effective layer-wise latency balancing and substantial hardware-resource savings while sustaining high throughput and classification accuracy. More broadly, the same accelerator description can be retargeted across operating points spanning more than an order of magnitude in hardware cost, providing a key enabler for deploying streaming SNNs at scale across diverse edge hardware platforms.","author":[{"family":"Yang","given":"Kuilian"},{"family":"Eltawil","given":"Ahmed"},{"family":"Salama","given":"Khaled"}],"issued":{"date-parts":[[2026]]},"DOI":"10.3389/fnins.2026.1875437","URL":"https://doi.org/10.3389/fnins.2026.1875437","source":"europepmc"},{"id":"doi:10.1007/s00422-025-01030-4","type":"article-journal","title":"Task success in trained spiking neural network models coincides with emergence of cross-stimulus-modulated inhibition.","abstract":"Abstract The neocortex is composed of spiking neurons interconnected in a sparse, recurrent network. Spiking activity within these networks underlies the computations that transform sensory inputs into appropriate behavioral responses. In this study, we train recurrent spiking neural network (SNN) models constrained by neocortical connectivity statistics and investigate the architectural changes that enable task-relevant, spike-based computations. We employ a binary state change detection task—an experimental paradigm used in animal behavioral studies. Our SNNs consist of interconnected excitatory and inhibitory units with connection probabilities and strengths modeled after the mouse neocortex and maintained throughout training and evaluation. Following training, we find that SNNs selectively modulate firing rates based on the binary input state, and that excitatory and inhibitory connectivity within and between input and recurrent layers adjusts accordingly. Notably, inhibitory neurons in the recurrent layer that positively modulate firing rates in response to one input state strengthen their connections to recurrent units with the opposite modulation. This push-pull connectivity—where excitation and inhibition are dynamically balanced in an opponent fashion—emerges as a key computational strategy and is reminiscent of connectivity observed in primary visual cortex. Using a one-hot output encoding yields identical firing rates to both input states, yet the push-pull inhibitory motif still arises. Importantly, this motif fails to emerge when Dale’s principle is not enforced during training, and task performance also declines.Furthermore, disrupting spike timing by a few milliseconds significantly impairs task performance, highlighting the importance of precise spike time coordination for computation in sparse networks like neocortex. The emergence of push-pull inhibition through task training in spiking models underscores the crucial role of interneurons and structured inhibition in shaping neural dynamics and spike-based information processing.","author":[{"family":"Zhu","given":"Yuqing"},{"family":"Smith","given":"Chadbourne"},{"family":"Jabri","given":"Tarek"},{"family":"Tang","given":"Mufeng"},{"family":"Scherr","given":"Franz"},{"family":"Maclean","given":"Jason"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1007/s00422-025-01030-4","URL":"https://doi.org/10.1007/s00422-025-01030-4","source":"europepmc"},{"id":"doi:10.3390/biomimetics11030208","type":"article-journal","title":"eXCube2: Explainable Brain-Inspired Spiking Neural Network Framework for Emotion Recognition from Audio, Visual and Multimodal Audio-Visual Data.","abstract":"This paper introduces a biomimetic framework and novel brain-inspired AI (BIAI) models based on spiking neural networks (SNNs) for emotional state recognition from audio (speech), visual (face), and integrated multimodal audio–visual data. The developed framework, named eXCube2, uses a three-dimensional SNN architecture NeuCube that is spatially structured according to a human brain template. The BIAI models developed in eXCube2 are trainable on spatio- and spectro-temporal data using brain-inspired learning rules. Such models are explainable in terms of revealing patterns in data and are adaptable to new data. The eXCube2 models are implemented as software systems and tested on speech and video data of subjects expressing emotional states. The use of a brain template for the SNN structure enables brain-inspired tonotopic and stereo mapping of audio inputs, topographic mapping of visual data, and the combined use of both modalities. This novel approach brings AI-based emotional state recognition closer to human perception, provides a better explainability and adaptability than existing AI systems. It also results in a higher or competitive accuracy, even though this was not the main goal here. This is demonstrated through experiments on benchmark datasets, achieving classification accuracy above 80% on single-modality data and 88.9% when multimodal audio–visual data are used, and a “don’t know” output is introduced. The paper further discusses possible applications of the proposed eXCube2 framework to other audio, visual, and audio–visual data for solving challenging problems, such as recognizing emotional states of people from different origins; brain state diagnosis (e.g., Parkinson’s disease, Alzheimer’s disease, ADHD, dementia); measuring response to treatment over time; evaluating satisfaction responses from online clients; cognitive robotics; human–robot interaction; chatbots; and interactive computer games. The SNN-based implementation of BIAI also enables the use of neuromorphic chips and platforms, leading to reduced power consumption, smaller device size, higher performance accuracy, and improved adaptability and explainability. This research shows a step toward building brain-inspired AI systems.","author":[{"family":"Kasabov","given":"NK"},{"family":"Yang","given":"A"},{"family":"Wang","given":"Z"},{"family":"Abouhassan","given":"I"},{"family":"Kassabova","given":"A"},{"family":"Lappas","given":"T"}],"issued":{"date-parts":[[2026]]},"DOI":"10.3390/biomimetics11030208","URL":"https://doi.org/10.3390/biomimetics11030208","source":"europepmc"},{"id":"doi:10.1177/20552076261466149","type":"article-journal","title":"NeurALLNet: An attention-based spiking neural network for energy-efficient multi-class classification of acute lymphoblastic leukemia.","abstract":"Objectives The classification of Acute Lymphoblastic Leukemia (ALL) from peripheral blood smear images using Convolutional Neural Networks (CNNs) has achieved expert-level accuracy. However, the computational and memory requirements of CNNs pose a barrier to their deployment in resource-constrained clinical settings and low-income countries. To bridge this gap, we propose NeurALLNet, a memory-efficient convolutional spiking neural network (SNN) augmented with Squeeze-and-Excitation channel attention for the multi-class classification of ALL subtypes. Methods NeurALLNet leverages sparse, event-driven temporal computation with an ultra-compact architecture of approximately 0.3M trainable parameters. The model was trained and evaluated on a primary dataset of ALL peripheral blood smear images, and its clinical generalizability was rigorously validated on an unseen external cohort of 3,242 images without retraining. We conducted hardware profiling on CPU and GPU platforms, alongside ablation studies and Grad-CAM visual explanations, to evaluate deployment viability and interpretability. Results NeurALLNet achieved a test accuracy of 98.16% on the primary dataset, with a bootstrapped 95% Confidence Interval (CI) of [0.9663, 0.9939]. On the external validation cohort, it yielded an accuracy of 96.02%, with a robust 95% CI of [0.9534, 0.9667]. The architecture requires a memory footprint of 1.35 MB, achieving single-image inference latencies of 454.67 ms on a standard CPU and 11.24 ms on a GPU. Ablation studies confirmed that the attention mechanism is critical to the network’s discriminative power, and Grad-CAM visualizations verified that predictions are grounded in clinically relevant morphological features. Conclusion Compared to recent state-of-the-art ensemble and hybrid CNNs that require millions of parameters, NeurALLNet delivers competitive diagnostic accuracy while reducing the computational footprint by orders of magnitude. By providing this precision within a 1.35 MB envelope, NeurALLNet offers a scalable, energy-efficient digital health intervention suitable for portable Lab-on-a-Chip devices and point-of-care diagnostics worldwide.","author":[{"family":"Hassan","given":"Md"},{"family":"Shanto","given":"Rejaul"},{"family":"Hasan","given":"Umar"},{"family":"Momen","given":"Sifat"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1177/20552076261466149","URL":"https://doi.org/10.1177/20552076261466149","source":"europepmc"},{"id":"doi:10.3390/biomimetics10100697","type":"article-journal","title":"SpiKon-E: Hybrid Soft Artificial Muscle Control Using Hardware Spiking Neural Network.","abstract":"Artificial muscles play a key role in the future of humanoid robotics and medical devices, with research on wire-driven joints leading the field. While electric servo motors were once at the forefront, the focus has shifted toward materials that react to changes in the environment (smart materials), including pneumatic silicone actuators and temperature-reactive metallic alloys, aiming to replicate human muscle actuation for improved performance. Initially designed for rigid actuators, control strategies were adapted to address the unique dynamics of artificial muscles. Although current controllers offer satisfactory performance, further optimization is necessary to mimic natural muscle control more rigorously. This study details the design and implementation of a novel system that mimics biological muscle. This system is designed to replicate the full range of motion and control functionalities, which can be utilized in various applications. This research has three significant contributions in the field of sustainable soft robotics. First, a novel shape memory alloy-based linear actuator is introduced, which achieves significantly higher displacements compared to traditional SMA wire-driven systems through a guiding mechanism. Second, this linear actuator is integrated into a hybrid soft actuation structure, which features a silicone PneuNet as the end effector and a force sensor for real-time pressure feedback. Lastly, a hardware Spiking Neural Network (HW-SNN) is utilized to control the exhibited force at the actuator’s endpoint. Experimental results showed that the displacement with the control system is significantly higher than that of the traditional control-based shape memory alloy systems. The system evaluation demonstrates good performance, thus advancing actuation and control in humanoid robotics.","author":[{"family":"Brașoveanu","given":"Florian"},{"family":"Hulea","given":"Mircea"},{"family":"Burlacu","given":"Adrian"}],"issued":{"date-parts":[[2025]]},"DOI":"10.3390/biomimetics10100697","URL":"https://doi.org/10.3390/biomimetics10100697","source":"europepmc"},{"id":"doi:10.3389/fnins.2025.1522788","type":"article-journal","title":"Optimizing event-driven spiking neural network with regularization and cutoff.","abstract":"Spiking neural networks (SNNs), which are the next generation of artificial neural networks (ANNs), offer a closer mimicry to natural neural networks and hold promise for significant improvements in computational efficiency. However, the current SNN model is trained to infer over a fixed duration, thereby overlooking the potential for dynamic inference in the SNN model. In this paper, we strengthen the relationship between SNN and event-driven processing by proposing the inclusion of a cutoff in SNN, that can terminate SNN at any time during inference to achieve efficient inference. Two novel optimization techniques are presented to achieve an inference-efficient SNN: a Top-K cutoff and regularization. The proposed regularization influences the training process by optimizing the SNN for the cutoff, whereas the Top-K cutoff technique optimizes the inference phase. We conducted an extensive set of experiments on multiple benchmark frame-based datasets, such as CIFAR10/100, Tiny-ImageNet, and event-based datasets, including CIFAR10-DVS, N-Caltech101, and DVS128 Gesture. The experimental results demonstrate the effectiveness of the proposed techniques in both the ANN-to-SNN conversion and direct training, enabling SNNs to require 1.76 to 2.76 × fewer timesteps for CIFAR-10, while achieving 1.64 to 1.95× fewer timesteps across all event-based datasets, with near-zero accuracy loss. These findings affirm the compatibility and potential benefits of the proposed techniques in terms of enhancing accuracy and reducing inference latency when integrated with existing methods. Code available: https://github.com/Dengyu-Wu/SNNCutoff .","author":[{"family":"Wu","given":"Dengyu"},{"family":"Jin","given":"Gaojie"},{"family":"Yu","given":"Han"},{"family":"Yi","given":"Xinping"},{"family":"Huang","given":"Xiaowei"}],"issued":{"date-parts":[[2025]]},"DOI":"10.3389/fnins.2025.1522788","URL":"https://doi.org/10.3389/fnins.2025.1522788","source":"europepmc"},{"id":"doi:10.1038/s41598-025-12611-5","type":"article-journal","title":"Short-term plasticity influences episodic memory recall: an interplay of synaptic traces in a spiking neural network model.","abstract":"Abstract We investigated the interaction of episodic memory processes with the short-term dynamics of recency effects. This work takes inspiration from a seminal experimental work involving an odor-in-context association task conducted on rats. In the experimental task, rats were presented with odor pairs in two arenas serving as old or new contexts for specific odor items. Rats were rewarded for selecting the odor that was new to the current context. These new-in-context odor items were deliberately presented with higher recency relative to old-in-context items, so that episodic memory was put in conflict with a short-term recency effect. To study our hypothesis about the major role of synaptic interplay of plasticity phenomena on different time-scales in explaining rats’ performance in such episodic memory tasks, we built a computational spiking neural network model consisting of two reciprocally connected networks that stored contextual and odor information as stable distributed memory patterns. We simulated the experimental task resulting in a dynamic context-item coupling between the two networks by means of Bayesian–Hebbian plasticity with eligibility traces to account for reward-based learning. We first reproduced quantitatively and explained mechanistically the findings of the experimental study, and then to further differentiate the impact of short-term plasticity we simulated an alternative task with old-in-context items presented with higher recency, thus synergistically confounding episodic memory with effects of recency. Our model predicted that higher recency of old-in-context items enhances episodic memory by boosting the activations of old-in-context items. We argue that the model offers a computational framework for studying behavioral implications of the synaptic underpinning of different memory effects in experimental episodic memory paradigms.","author":[{"family":"Chrysanthidis","given":"N"},{"family":"Fiebig","given":"F"},{"family":"Lansner","given":"A"},{"family":"Herman","given":"P"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1038/s41598-025-12611-5","URL":"https://doi.org/10.1038/s41598-025-12611-5","source":"europepmc"},{"id":"doi:10.3390/brainsci15030217","type":"article-journal","title":"Research on Anti-Interference Performance of Spiking Neural Network Under Network Connection Damage.","abstract":"Background: With the development of artificial intelligence, memristors have become an ideal choice to optimize new neural network architectures and improve computing efficiency and energy efficiency due to their combination of storage and computing power. In this context, spiking neural networks show the ability to resist Gaussian noise, spike interference, and AC electric field interference by adjusting synaptic plasticity. The anti-interference ability to spike neural networks has become an important direction of electromagnetic protection bionics research. Methods: Therefore, this research constructs two types of spiking neural network models with LIF model as nodes: VGG-SNN and FCNN-SNN, and combines pruning algorithm to simulate network connection damage during the training process. By comparing and analyzing the millimeter wave radar human motion dataset and MNIST dataset with traditional artificial neural networks, the anti-interference performance of spiking neural networks and traditional artificial neural networks under the same probability of edge loss was deeply explored. Results: The experimental results show that on the millimeter wave radar human motion dataset, the accuracy of the spiking neural network decreased by 5.83% at a sparsity of 30%, while the accuracy of the artificial neural network decreased by 18.71%. On the MNIST dataset, the accuracy of the spiking neural network decreased by 3.91% at a sparsity of 30%, while the artificial neural network decreased by 10.13%. Conclusions: Therefore, under the same network connection damage conditions, spiking neural networks exhibit unique anti-interference performance advantages. The performance of spiking neural networks in information processing and pattern recognition is relatively more stable and outstanding. Further analysis reveals that factors such as network structure, encoding method, and learning algorithm have a significant impact on the anti-interference performance of both.","author":[{"family":"Zhang","given":"Yongqiang"},{"family":"Pang","given":"Haijie"},{"family":"Ma","given":"Jinlong"},{"family":"Ma","given":"Guilei"},{"family":"Zhang","given":"Xiaoming"},{"family":"Man","given":"Menghua"}],"issued":{"date-parts":[[2025]]},"DOI":"10.3390/brainsci15030217","URL":"https://doi.org/10.3390/brainsci15030217","source":"europepmc"},{"id":"doi:10.3390/biomimetics10040240","type":"article-journal","title":"A Reinforced, Event-Driven, and Attention-Based Convolution Spiking Neural Network for Multivariate Time Series Prediction.","abstract":"Despite spiking neural networks (SNNs) inherently exceling at processing time series due to their rich spatio-temporal information and efficient event-driven computing, the challenge of extracting complex correlations between variables in multivariate time series (MTS) remains to be addressed. This paper proposes a reinforced, event-driven, and attention-based convolution SNN model (REAT-CSNN) with three novel features. First, a joint Gramian Angular Field and Rate (GAFR) coding scheme is proposed to convert MTS into spike images, preserving the inherent features in MTS, such as the temporal patterns and spatio-temporal correlations between time series. Second, an advanced LIF-pooling strategy is developed, which is then theoretically and empirically proved to be effective in preserving more features from the regions of interest in spike images than average-pooling strategies. Third, a convolutional block attention mechanism (CBAM) is redesigned to support spike-based input, enhancing event-driven characteristics in weighting operations while maintaining outstanding capability to capture the information encoded in spike images. Experiments on multiple MTS data sets, such as stocks and PM2.5 data sets, demonstrate that our model rivals, and even surpasses, some CNN- and RNN-based techniques, with up to 3% better performance, while consuming significantly less energy.","author":[{"family":"Li","given":"Ying"},{"family":"Guan","given":"Xikang"},{"family":"Yue","given":"Wenwei"},{"family":"Huang","given":"Yongsheng"},{"family":"Zhang","given":"Bin"},{"family":"Duan","given":"Peibo"}],"issued":{"date-parts":[[2025]]},"DOI":"10.3390/biomimetics10040240","URL":"https://doi.org/10.3390/biomimetics10040240","source":"europepmc"},{"id":"doi:10.5167/uzh-279845","type":"article-journal","title":"Temporally-Varying Stimulations for Cortical Visual Neuroprosthetic using Spiking Neural Networks","abstract":"Visual neuroprosthesis can help to restore a rudimentary form of sight in visually impaired subjects via electrical stimulation. These systems receive camera input images processed by a artificial neural network (ANN) to output stimulation patterns for driving a large set of electrodes. Previous optimization approaches for visual neuroprosthesis stimulation patterns provided a fixed stimulation amplitude for each electrode. In this work, we look at the feasibility of using a spiking neural network (SNN) instead of the ANN allowing the model to provide time-varying stimulation patterns. During training, we mapped pulses of the SNN through a phosphene simulator which models the perceptual response. Results on the MNIST dataset show that our SNNbased encoder demonstrates reasonable reconstruction quality compared to a state-of-the-art ANN while requiring 2× fewer operations (1.16 vs 2.44 GFLOPs). Preliminary results show that the network trained on N-MNIST, a dataset of spikes from the spiking retina event camera on MNIST images, shows successful reconstructed recognizable digit shapes, suggesting the possibility of using event cameras for visual prosthesis.","author":[{"family":"Moure","given":"Pehuen"},{"family":"Pak","given":"Tatyana"},{"family":"Hahn","given":"Niklas"},{"family":"De Ruyter Van Stevenick","given":"J"},{"family":"Liu","given":"Shih"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5167/uzh-279845","URL":"https://doi.org/10.5167/uzh-279845","source":"datacite"},{"id":"doi:10.5281/zenodo.18669202","type":"article-journal","title":"Data supporting: An Energy-Efficient Neuromorphic Front-End for Risk Pre-Screening Using Pulse-Encoded Biosensor Signals","abstract":"“An Energy-Efficient Neuromorphic Front-End for Risk Pre-Screening Using Pulse-Encoded Biosensor Signals”, IEEE Sensors Journal, 2024. The dataset includes: Synthetic pulse streams corresponding to free PSA (fPSA) and total PSA (tPSA), generated according to the empirical pulse-encoding model described in the manuscript Synaptic-filtered temporal representations used as inputs to the spiking neural network Risk category labels derived from clinically established free-to-total PSA ratio thresholds Python-based implementation of the spiking neural network architecture, including data generation, synaptic filtering, training, and evaluation modules The synthetic dataset consists of 500 samples balanced across four clinically defined prostate cancer risk categories. The neuromorphic front-end operates directly in the pulse domain without reconstructing continuous biomarker concentrations or explicitly computing biomarker ratios. This repository enables reproduction of the system-level evaluation results reported in the associated publication, including classification accuracy, confusion matrices, and layer-wise temporal activity analysis.","author":[{"family":"Chen","given":"Junrui"},{"family":"Pilehvar Meibody","given":"Ali"},{"family":"Carrara","given":"Sandro"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18669202","URL":"https://doi.org/10.5281/zenodo.18669202","source":"datacite"},{"id":"doi:10.5281/zenodo.18669203","type":"article-journal","title":"Data supporting: An Energy-Efficient Neuromorphic Front-End for Risk Pre-Screening Using Pulse-Encoded Biosensor Signals","abstract":"“An Energy-Efficient Neuromorphic Front-End for Risk Pre-Screening Using Pulse-Encoded Biosensor Signals”, IEEE Sensors Journal, 2024. The dataset includes: Synthetic pulse streams corresponding to free PSA (fPSA) and total PSA (tPSA), generated according to the empirical pulse-encoding model described in the manuscript Synaptic-filtered temporal representations used as inputs to the spiking neural network Risk category labels derived from clinically established free-to-total PSA ratio thresholds Python-based implementation of the spiking neural network architecture, including data generation, synaptic filtering, training, and evaluation modules The synthetic dataset consists of 500 samples balanced across four clinically defined prostate cancer risk categories. The neuromorphic front-end operates directly in the pulse domain without reconstructing continuous biomarker concentrations or explicitly computing biomarker ratios. This repository enables reproduction of the system-level evaluation results reported in the associated publication, including classification accuracy, confusion matrices, and layer-wise temporal activity analysis.","author":[{"family":"Chen","given":"Junrui"},{"family":"Pilehvar Meibody","given":"Ali"},{"family":"Carrara","given":"Sandro"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18669203","URL":"https://doi.org/10.5281/zenodo.18669203","source":"datacite"},{"id":"doi:10.48448/440w-ev06","type":"article-journal","title":"1234 - Spiking-Aided Neural Architecture for Efficient and Robust WiFi Sensing","abstract":"This paper introduces a spiking-aided wifi sensing network (SWS-Net), a novel hybrid neural architecture that integrates Spiking Neural Networks (SNNs) with conventional Artificial Neural Networks (ANNs) for robust WiFi-based indoor sensing. WiFi signals offer a low-cost and device-free solution for recognizing human activities, gestures, identities and etc. However, their susceptibility to multipath fading and environmental noise poses significant challenges. Inspired by the human brain’s capability to process noisy information, SWS-Net leverages the noise-resilient dynamics of spiking neurons alongside the feature extraction ability of ANNs. We present a theoretical analysis comparing the noise-handling capacities of SNNs and ANNs, and show how their combination yields both improved robustness and training efficiency. Experimental results across three WiFi sensing tasks demonstrate that SWS-Net consistently achieves higher accuracy and faster convergence compared to baseline models, validating its effectiveness in challenging indoor environments.","author":[{"family":"Jing","given":"Liwen"},{"family":"Lu","given":"Yisha"},{"family":"Zhang","given":"Bowen"},{"family":"Zheng","given":"Jiangmao"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48448/440w-ev06","URL":"https://doi.org/10.48448/440w-ev06","source":"datacite"},{"id":"doi:10.64898/2026.05.12.724100","type":"article-journal","title":"A biologically-grounded cerebellar spiking network model with realistic synaptic transmission captures complex circuit dynamics","abstract":"Abstract Cerebellar neural circuit dynamics rely on a rich repertoire of synaptic and excitability mechanisms, which are thought to determine network computation in physiological and pathological conditions. In this work, we develop and validate a biologically-grounded spiking neural network of the cerebellar cortex, embedding key mechanisms of cellular excitability and synaptic transmission, and assess their impact on signal processing. Neuronal input-output functions, short-term synaptic plasticity, receptor-specific kinetics, and NMDA channel voltage-dependent gating were calibrated against detailed multicompartmental models through automatic tuning procedures. Incorporating these realistic biological properties allowed the network model to simulate key features observed in recordings from acute cerebellar slices. The neuronal discharge and local field potentials elicited by mossy fiber stimulation faithfully reproduced the natural patterns with millisecond precision. Then, selective receptor switch-off revealed the contribution of NMDA, GABA, and AMPA receptors to the frequency-dependent input-output function of the granular layer and Purkinje cells, linking previous empirical findings to specific synaptic mechanisms. This model combines high computational performance with biological realism and offers a computationally efficient framework to investigate neurophysiological phenomena and the neural correlates of behavior in large-scale long-lasting simulations, such as those needed to address the neural underpinnings of learning and of cerebellar pathologies.","author":[{"family":"Grazia","given":"Marialaura"},{"family":"Benozzo","given":"Danilo"},{"family":"Rodarie","given":"Dimitri"},{"family":"Marchetti","given":"Filippo"},{"family":"Dangelo","given":"Egidio"},{"family":"Casellato","given":"Claudia"}],"issued":{"date-parts":[[2026]]},"DOI":"10.64898/2026.05.12.724100","URL":"https://doi.org/10.64898/2026.05.12.724100","source":"europepmc"},{"id":"doi:10.21203/rs.3.rs-8592450/v1","type":"article-journal","title":"A new recurrent spiking pi sigma artificial neural network for the forecasting problem","abstract":"Abstract Due to their flexible model structures and their success in nonlinear modelling, artificial neural networks provide good alternatives for solving the forecasting problem. It is seen that different artificial neuron models can positively affect the forecasting performance and pave the way for the creation of new artificial neural network models. In this study, a new artificial neural network with an architecture based on multiplicative and additive neuron models and using the feedback logic in exponential smoothing methods is presented. The training algorithm of the proposed neural network is based on particle swarm optimization using a dynamic fitness function that gives more weight to recent observations. The performance of the proposed new neural network is investigated with the help of statistical hypothesis tests by comparing it with other methods in the literature on Nasdaq stock exchange time series. As a result of the study, it is empirically observed that the proposed neural network has a successful forecasting performance.","author":[{"family":"Egrioglu","given":"Erol"},{"family":"Bas","given":"Eren"},{"family":"Albayrak","given":"Gulsen"}],"issued":{"date-parts":[[2026]]},"DOI":"10.21203/rs.3.rs-8592450/v1","URL":"https://doi.org/10.21203/rs.3.rs-8592450/v1","source":"europepmc"},{"id":"doi:10.20944/preprints202509.2072.v1","type":"manuscript","title":"A Practical Tutorial on Spiking Neural Networks: Comprehensive Review, Models, Experiments, Software Tools, and Implementation Guidelines","abstract":"Spiking Neural Networks (SNNs) provide a biologically inspired, event-driven alternative to Artificial Neural Networks (ANNs) with the potential to deliver competitive accuracy at substantially lower energy. This tutorial-study offers a unified, practice-oriented assessment combining critical reviews and standardized experiments. We benchmark a shallow Fully Connected Network (FCN) on MNIST and a deeper VGG7 architecture on CIFAR-10 across multiple neuron models (leaky Integrate-and-Fire (LIF), Sigma-Delta, etc.) and input encodings (direct, rate, temporal, etc.) using supervised surrogate-gradient training, implemented with Intel Lava/SLAYER, SpikingJelly, Norse and PyTorch. Empirically, we observe a consistent but tunable trade-off between accuracy and energy. On MNIST, Sigma-Delta neurons with rate or Sigma-Delta encodings reach 98.1% (ANN: 98.23%). On CIFAR-10, Sigma-Delta neurons with direct input achieve 83.0% at just 2 time steps (ANN: 83.6%). A GPU-based operation-count energy proxy indicates many SNN configurations operate below the ANN energy baseline; some frugal codes minimize energy at the cost of accuracy, whereas accuracy-leaning settings (e.g., Sigma-Delta with direct or rate coding ) narrow the performance gap while remaining energy-conscious, yielding up to 3-fold efficiency versus matched ANNs in our setup. Thresholds and the number of time steps are decisive: intermediate thresholds and the minimal time window that still meets accuracy targets typically maximize efficiency per joule. We distill actionable design rules: choose the neuron/encoding pair by application goal (accuracy-critical vs. energy-constrained) and co-tune thresholds and time steps. Finally, we outline how event-driven neuromorphic hardware can amplify these savings through sparse, local, asynchronous computation, providing a practical playbook for embedded, real-time, and sustainable AI deployments.","author":[{"family":"Ayasi","given":"Bahgat"},{"family":"Carmona","given":"Cristóbal"},{"family":"Saleh","given":"Mohammed"},{"family":"García-Vico","given":"Angel"}],"issued":{"date-parts":[[2025]]},"DOI":"10.20944/preprints202509.2072.v1","URL":"https://doi.org/10.20944/preprints202509.2072.v1","source":"europepmc"},{"id":"doi:10.21203/rs.3.rs-7944463/v1","type":"article-journal","title":"FPGA based Implementation of Neuromorphic Neural Networks for Early Detection of Alzheimer’s Disease","abstract":"Abstract Alzheimer’s Disease (AD) is a progressive, neurodegenerative condition causing cognitive decline with memory loss. To improve life quality and allow intervention in time, detection that is early and accurate is key, especially at Mild Cognitive Impairment (MCI). This paper proposes a Brain-Inspired Neuromorphic approach to classifying Alzheimer’s stages using advanced neural network models, that are Convolutional Neural Networks (CNN), Liquid Neural Networks (LNN), Spiking Neural Networks (SNN), and a hybrid CNN + LNN architecture. MRI datasets used within this study are obtained directly from Alzheimer’s Disease Neuroimaging Initiative (ADNI), which provides high-quality structural imaging data. Using the SPM toolbox, the raw DICOM images are converted into NIfTI format and preprocessed to segment and standardize the data. Dual-Tree Complex Wavelet Transform (DTCWT) is applied in feature extraction for preservation of both spatial with frequency information. These features are then used to train the neural network models in order to classify images into three stages. The trained models get translated into the RTL design as simulated using Xilinx AMD Vivado for exploring hardware - software co design deployment. This hardware-optimized Neuromorphic approach aims to provide a low-power, high-speed, and accurate Alzheimer’s detection solution, connecting research with practical, deployable medical tools.","author":[{"family":"Sujathakumari","given":"BA"},{"family":"Shalini","given":"MS"},{"family":"Sneha","given":"NS"}],"issued":{"date-parts":[[2025]]},"DOI":"10.21203/rs.3.rs-7944463/v1","URL":"https://doi.org/10.21203/rs.3.rs-7944463/v1","source":"europepmc"},{"id":"doi:10.21203/rs.3.rs-6767420/v1","type":"article-journal","title":"Hybrid Cross-Temporal Contrastive Model with Spiking Energy-Efficient Network Intrusion Detection in IOMT","abstract":"Abstract The Internet of Medical Things (IoMT), a key application of the Internet of Things (IoT), has played a key role, especially during the Covid 19 pandemic. Real-time patient monitoring and remote diagnostics help improve medical services, but this increases the mammoth size of network traffic, which impacts the security quite a bit. However, traditional intrusion detection systems lack synchronization between accuracy and energy efficiency in resource-constrained IoMT environments. To address this issue, we present a hybrid cross-temporal contrastive model coupled with a spiking energy-efficient network for intrusion detection. This approach uses contrastive learning to learn temporal dependencies in network traffic and spiking neural networks (SNNs) for energy-efficient computations. We evaluated the model on the WUSTL-EHMS-2020 dataset, which consists of 44 features (35 of them are network flow measurements, and 8 are biometric patient features), as well as the NSL-KDD dataset to perform a comparative validation. Furthermore, the experiment results prove that our proposed model achieves 99.95% accuracy on the WUSTL-EHMS-2020 dataset with an F1 score of 99.89%, precision of 98.23%, and recall of 99.55%, outperforming conventional models. The model attained 98.2% accuracy, 97.6% precision, 98.5% F1 score, and 97.3% recall on the NSL-KDD Dataset. Our approach shows that these results effectively secure IoMT networks at a low computational cost. Finally, the proposed hybrid model can achieve good performance and energy efficiency for intrusion detection in innovative healthcare systems. In future work, efforts will be made to improve the model's generalization property in diverse IoMT environments and minimize the energy consumption of spiking neural networks in real-time applications.","author":[{"family":"Alrayes","given":"Fatma"},{"family":"Zakariah","given":"Mohammed"},{"family":"Amin","given":"Syed"},{"family":"Khan","given":"Zafar"},{"family":"Helal","given":"Maha"}],"issued":{"date-parts":[[2025]]},"DOI":"10.21203/rs.3.rs-6767420/v1","URL":"https://doi.org/10.21203/rs.3.rs-6767420/v1","source":"preprints"},{"id":"doi:10.22541/au.174885691.16213858/v1","type":"article-journal","title":"Hybrid Cross-Temporal Contrastive Model with Spiking Energy-Efficient Network Intrusion Detection in IOMT","abstract":"The Internet of Medical Things (IoMT), a key application of the Internet of Things (IoT), has played a key role, especially during the Covid 19 pandemic. Real-time patient monitoring and remote diagnostics help improve medical services, but this increases the mammoth size of network traffic, which impacts the security quite a bit. However, traditional intrusion detection systems lack synchronization between accuracy and energy efficiency in resource-constrained IoMT environments. To address this issue, we present a hybrid cross-temporal contrastive model coupled with a spiking energy-efficient network for intrusion detection. This approach uses contrastive learning to learn temporal dependencies in network traffic and spiking neural networks (SNNs) for energy-efficient computations. We evaluated the model on the WUSTL-EHMS-2020 dataset, which consists of 44 features (35 of them are network flow measurements, and 8 are biometric patient features), as well as the NSL-KDD dataset to perform a comparative validation. Furthermore, the experiment results prove that our proposed model achieves 99.95% accuracy on the WUSTL-EHMS-2020 dataset with an F1 score of 99.89%, precision of 98.23%, and recall of 99.55%, outperforming conventional models. The model attained 98.2% accuracy, 97.6% precision, 98.5% F1 score, and 97.3% recall on the NSL-KDD Dataset. Our approach shows that these results effectively secure IoMT networks at a low computational cost. Finally, the proposed hybrid model can achieve good performance and energy efficiency for intrusion detection in innovative healthcare systems. In future work, efforts will be made to improve the model’s generalization property in diverse IoMT environments and minimize the energy consumption of spiking neural networks in real-time applications.","author":[{"family":"Alrayes","given":"Fatma"},{"family":"Zakariah","given":"Mohammed"},{"family":"Amin","given":"Syed"},{"family":"Khan","given":"Zafar"},{"family":"Helal","given":"Maha"}],"issued":{"date-parts":[[2025]]},"DOI":"10.22541/au.174885691.16213858/v1","URL":"https://doi.org/10.22541/au.174885691.16213858/v1","source":"preprints"},{"id":"doi:10.21203/rs.3.rs-6460749/v1","type":"article-journal","title":"Improving SpikeProp’s Training Efficiency in Spiking Neural Networks for Large Language Models Through Innovative Weight Initialization","abstract":"Abstract Spiking neural networks (SNNs) mimic the functions of biological neurons by leveraging individual temporal spikes for communication and computation. Since SNN was perceived to be complex and analytically challenging, it had long been overlooked. In this study, we explore the enhancement of SpikeProp, a supervised learning model customized for SNNs. Three distinct models are being investigated, including the Proposed Model 1, the Proposed Model 2, and the Proposed Model 3, each providing unique improvements to the SpikeProp algorithm. To accelerate convergence and adapt learning rates, momentum factors are integrated into Proposed Model 1. In Proposed Model 2, a rate dependency is introduced based on Angle Driven Learning. By incorporating particle swarm optimization (PSO), Model 3 combines the strengths of Model 1 and 2. SNNs can be trained and classified more efficiently and accurately using these models. Furthermore, we examine how large language models (LLMs) might inform the design and interpretability of neural architectures and learning methodologies while also enhancing SNN training. Through the use of LLMs, we seek to enhance model transparency and encourage more Responsible AI (RAI) principles. A thorough evaluation and comparison of Proposed Model 1, Proposed Model 2, and Proposed Model 3 with traditional methods confirms that these models consistently outperform them. Consequently, they have a high potential for practical applications in neural network training in real-world settings and LLM-informed development, contributing to the advancement of AI systems.","author":[{"family":"Ahmed","given":"Falah"},{"family":"Zakarya","given":"Muhammad"},{"family":"Khan","given":"Naveed"},{"family":"Zebari","given":"Dilovan"},{"family":"Al-Bahri","given":"Mahmood"},{"family":"Joseph","given":"Bwalya"}],"issued":{"date-parts":[[2025]]},"DOI":"10.21203/rs.3.rs-6460749/v1","URL":"https://doi.org/10.21203/rs.3.rs-6460749/v1","source":"preprints"},{"id":"doi:10.3929/ethz-b-000714674","type":"article-journal","title":"Recurrent models of orientation selectivity enable robust early-vision processing in mixed-signal neuromorphic hardware","abstract":"Mixed signal analog/digital neuromorphic circuits represent an ideal medium for reproducing bio-physically realistic dynamics of biological neural systems in real-time. However, similar to their biological counterparts, these circuits have limited resolution and are affected by a high degree of variability. By developing a recurrent spiking neural network model of the retinocortical visual pathway, we show how such noisy and heterogeneous computing substrate can produce linear receptive fields tuned to visual stimuli with specific orientations and spatial frequencies. Compared to strictly feed-forward schemes, the model generates highly structured Gabor-like receptive fields of any phase symmetry, making optimal use of the hardware resources available in terms of synaptic connections and neuron numbers. Experimental results validate the approach, demonstrating how principles of neural computation can lead to robust sensory processing electronic systems, even when they are affected by high degree of heterogeneity, e.g., due to the use of analog circuits or memristive devices.","author":[{"family":"Baruzzi","given":"Valentina"},{"family":"Indiveri","given":"Giacomo"},{"family":"Sabatini","given":"Silvio"}],"issued":{"date-parts":[[2025]]},"DOI":"10.3929/ethz-b-000714674","URL":"https://doi.org/10.3929/ethz-b-000714674","source":"datacite"},{"id":"doi:10.3929/ethz-b-000735596","type":"article-journal","title":"Event driven neural network on a mixed signal neuromorphic processor for EEG based epileptic seizure detection","abstract":"Long-term monitoring of biomedical signals is essential for the modern clinical management of neurological conditions such as epilepsy. However, developing wearable systems that are able to monitor, analyze, and detect epileptic seizures with long-lasting operation times using current technologies is still an open challenge. Brain-inspired spiking neural networks (SNNs) represent a promising signal processing and computing framework as they can be deployed on ultra-low power neuromorphic computing systems, for this purpose. Here, we introduce a novel SNN architecture, co-designed and validated on a mixed-signal neuromorphic chip, that shows potential for always-on monitoring of epileptic activity. We demonstrate how the hardware implementation of this SNN captures the phenomenon of partial synchronization within neural activity during seizure periods. We assess the network using a full-custom asynchronous mixed-signal neuromorphic platform, processing analog signals in real-time from an Electroencephalographic (EEG) seizure dataset. The neuromorphic chip comprises an analog front-end (AFE) signal conditioning stage and an asynchronous delta modulation (ADM) circuit directly integrated on the same die, which can produce the stream of spikes as input to the SNN, directly from the analog EEG signals. We show a linear classifier in a post processing stage that is sufficient to reliably classify and detect seizures, from the local features extracted by the SNN, indicating the feasibility of full on-chip seizure monitoring in the future. This research marks a significant advancement toward developing embedded intelligent “wear and forget” units for resource-constrained environments. These units could autonomously detect and log relevant EEG events of interest in out-of-hospital environments, offering new possibilities for patient care and management of neurological disorders.","author":[{"family":"Bartels","given":"Jim"},{"family":"Gallou","given":"Olympia"},{"family":"Ito","given":"Hiroyuki"},{"family":"Cook","given":"Matthew"},{"family":"Sarnthein","given":"Johannes"},{"family":"Indiveri","given":"Giacomo"},{"family":"Ghosh","given":"Saptarshi"}],"issued":{"date-parts":[[2025]]},"DOI":"10.3929/ethz-b-000735596","URL":"https://doi.org/10.3929/ethz-b-000735596","source":"datacite"},{"id":"doi:10.5281/zenodo.18286627","type":"article-journal","title":"Brain Inspired Computing Models and Architectures","abstract":"With an exponential increase in the amount of data collected per day, the fields of artificial intelligence and machinelearning continue to progress at a rapid pace with respect to algorithms, models, applications, and hardware. In particular,deep neural networks have revolutionized these fields by providing unprecedented human-like performance in solving manyreal-world problems such as image or speech recognition. There is also significant research aimed at unraveling the principles of computation in large biological neural networks and, in particular, biologically plausible spiking neural networks. This paperpresents an overview of the brain-inspired computing models starting with the development of the perceptron and multi-layerperceptron followed by convolutional neural networks (CNNs) and recurrent neural networks (RNNs). The paper brieflyreviews other neural network models such as Hopfield neural networks and Boltzmann machines. Other models such as spikingneural networks (SNNs) and hyperdimensional computing are then briefly reviewed. Recent advances in these neural networksand graph related neural networks are then described.Index Terms—Brain-inspired computing, spiking neural networks, neuromorphic architecture, graph neural networks, hyperdimensional computing, low-power inference, energy efficiency, surrogate gradients.","author":[{"family":"Katyal","given":"Diksha"},{"family":"Choudhary","given":"Kalpana"},{"family":"Singh","given":"Dimpy"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18286627","URL":"https://doi.org/10.5281/zenodo.18286627","source":"datacite"},{"id":"doi:10.5281/zenodo.18286628","type":"article-journal","title":"Brain Inspired Computing Models and Architectures","abstract":"With an exponential increase in the amount of data collected per day, the fields of artificial intelligence and machinelearning continue to progress at a rapid pace with respect to algorithms, models, applications, and hardware. In particular,deep neural networks have revolutionized these fields by providing unprecedented human-like performance in solving manyreal-world problems such as image or speech recognition. There is also significant research aimed at unraveling the principles of computation in large biological neural networks and, in particular, biologically plausible spiking neural networks. This paperpresents an overview of the brain-inspired computing models starting with the development of the perceptron and multi-layerperceptron followed by convolutional neural networks (CNNs) and recurrent neural networks (RNNs). The paper brieflyreviews other neural network models such as Hopfield neural networks and Boltzmann machines. Other models such as spikingneural networks (SNNs) and hyperdimensional computing are then briefly reviewed. Recent advances in these neural networksand graph related neural networks are then described.Index Terms—Brain-inspired computing, spiking neural networks, neuromorphic architecture, graph neural networks, hyperdimensional computing, low-power inference, energy efficiency, surrogate gradients.","author":[{"family":"Katyal","given":"Diksha"},{"family":"Choudhary","given":"Kalpana"},{"family":"Singh","given":"Dimpy"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18286628","URL":"https://doi.org/10.5281/zenodo.18286628","source":"datacite"},{"id":"doi:10.1186/s43074-026-00278-8","type":"article-journal","title":"Toward brain-inspired intelligence: a review of photonic neuromorphic computing systems","abstract":"Abstract Inspired by the key elements and principles in the brain, photonic neuromorphic computing shows great potential for building next-generation intelligent processing systems with high parallelism, low latency, low power consumption, and self-learning capabilities. While summarizing significant advances in this field, this review offers insights into how photonic neuromorphic computing systems support next-generation intelligent information processing. Specifically, we discuss emerging materials and devices, which support more compact integration of efficient physical architectures. Architectures such as photonic spiking neural networks and reservoir computing, together with associated learning paradigms, show a trend of collaborative development. Then we present promising applications, where neuromorphic computing systems provide broadband and multi-domain perception and processing. Finally, we discuss key challenges and future directions. This review aims to offer a clear and comprehensive overview for researchers across a broad community. We hope to present these insights in a “neuromorphic” manner, to reveal why learning from the brain has become increasingly important, especially for overcoming the efficiency bottleneck in traditional von Neumann architecture and data-intensive artificial neural networks.","author":[{"family":"Lan","given":"Dun"},{"family":"Ma","given":"Bowen"},{"family":"Ji","given":"Yuxiang"},{"family":"Zeng","given":"Yichen"},{"family":"Zou","given":"Zhihong"},{"family":"Li","given":"Zongsheng"},{"family":"Xu","given":"Yiyang"},{"family":"Chai","given":"Mengmeng"},{"family":"Zou","given":"Weiwen"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1186/s43074-026-00278-8","URL":"https://doi.org/10.1186/s43074-026-00278-8","source":"crossref"},{"id":"doi:10.1088/1361-6579/ae56cc","type":"article-journal","title":"NeuroSleep: neuromorphic event-driven single-channel EEG sleep staging for edge-efficient sensing","abstract":"Abstract Objective . Reliable, continuous neural sensing on wearable edge platforms is fundamental to long-term health monitoring; however, for electroencephalography (EEG)-based sleep monitoring, dense high-frequency processing is often computationally prohibitive under tight energy budgets. Approach . To address this bottleneck, this paper proposes NeuroSleep, an integrated event-driven sensing and inference system for energy-efficient sleep staging. NeuroSleep first converts raw EEG into complementary multi-scale bipolar event streams using Residual Adaptive Multi-Scale Delta Modulation, enabling an explicit fidelity–sparsity trade-off at the sensing front end. Furthermore, NeuroSleep adopts a hierarchical inference architecture that comprises an Event-based Adaptive Multi-scale Response module for local feature extraction, a Local Temporal-Attention Module for context aggregation, and an epoch-leaky integrate-and-fire module to capture long-term state persistence. Main results . Experimental results using subject-independent 5-fold cross-validation on the sleep-EDF expanded sleep-cassette subset (78 subjects, 153 overnight recordings) with single-channel EEG demonstrate that NeuroSleep achieves a mean accuracy of 74.2% with only 0.932 M parameters while reducing sparsity-adjusted effective operations by approximately 53.6% relative to dense processing. Compared to the representative dense Transformer baseline, NeuroSleep improves accuracy by 7.5% with a 45.8% reduction in computational load. Significance . By coupling neuromorphic event encoding with state-aware context modeling, NeuroSleep offers a deployment-oriented framework for single-channel sleep staging that reduces redundant high-rate processing and improves energy scalability for wearable and edge platforms.","author":[{"family":"Li","given":"Boyu"},{"family":"Zhu","given":"Xingchun"},{"family":"Wu","given":"Yonghui"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/1361-6579/ae56cc","URL":"https://doi.org/10.1088/1361-6579/ae56cc","source":"crossref"},{"id":"doi:10.1002/smll.202510540","type":"article-journal","title":"Beyond Sight: Neuromorphic Synapses Triggered by Invisible Light","abstract":"ABSTRACT Neuromorphic devices that emulate biological synaptic behavior are emerging as key enablers for in‐sensor intelligence. While visible‐light‐responsive systems have dominated the field, recent efforts have expanded toward invisible spectral regions: ultraviolet (UV), infrared (IR), and X‐ray, where unique photon‐matter interactions offer new avenues for optical plasticity. These invisible‐wavelength stimuli enable synaptic functions such as short‐term and long‐term potentiation through mechanisms like persistent photoconductivity, defect ionization, and interfacial charge trapping, often without the need for external programming circuitry. Although these devices are increasingly important for intelligent imaging, radiation‐tolerant electronics, and secure communication, related studies are still fragmented across different fields and lack an organized overview. In this review, we systematically categorize and analyze optoelectronic synapses that operate under UV, IR, and X‐ray illumination. We highlight representative material systems including Ga 2 O 3 , perovskites, wide‐bandgap oxides, and hybrid nanocomposites, and discuss their device architectures, synaptic behaviors, and operational metrics. Special emphasis is placed on the underlying physical mechanisms, spectral selectivity, and integration prospects for artificial retinas, neuromorphic vision systems, and multimodal sensing arrays. We also provide outlooks for scalable, multispectral, and energy‐efficient neuromorphic platforms beyond the visible.","author":[{"family":"Park","given":"Jisoo"},{"family":"Kim","given":"Kyounghoon"},{"family":"Lee","given":"Eun"},{"family":"Kim","given":"Young‐joon"},{"family":"Yoo","given":"Hocheon"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1002/smll.202510540","URL":"https://doi.org/10.1002/smll.202510540","source":"crossref"},{"id":"doi:10.36227/techrxiv.177040597.75784815/v1","type":"article-journal","title":"A Survey on Neuromorphic Navigation: Implementation Resources, Challenges and Perspectives","abstract":"Although traditional navigation technologies have seen continued progress, they remain constrained by persistent issues such as high power consumption, significant latency, and limited dynamic adaptability, especially in challenging environments like urban canyons and underground spaces. Neuromorphic technologies, leveraging eventdriven sensing, spike-based computation, and brain-inspired computein-memory, offer a promising new avenue to overcome these limitations. However, as an emerging field, neuromorphic navigation currently lacks dedicated survey efforts to systematically map the ecosystem and resources that support its development. Motivated by this, this survey provides a comprehensive overview of implementation resources across three key dimensions: data, hardware, and software. Building on this foundation, we identify main challenges, including the absence of evaluation benchmarks, fragmented ecosystem resources, the lack of a full-stack theoretical framework, and emerging concerns regarding security, trust, and data privacy. We also propose perspectives aimed at supporting community-driven growth and ecosystem maturation. A curated list of resources is available at: https://github.com/BINUCOE/Awesome-Neuromorphic-Navigation-Resources.","author":[{"family":"Zhang","given":"Youdong"},{"family":"He","given":"Xu"},{"family":"Meng","given":"Xiaolin"},{"family":"An","given":"Xiangdong"},{"family":"Mo","given":"Lingfei"},{"family":"Yin","given":"Wenxuan"},{"family":"Yu","given":"Fangwen"}],"issued":{"date-parts":[[2026]]},"DOI":"10.36227/techrxiv.177040597.75784815/v1","URL":"https://doi.org/10.36227/techrxiv.177040597.75784815/v1","source":"crossref"},{"id":"doi:10.1002/cnma.202500564","type":"article-journal","title":"Toward Neuromorphic Diagnostics: Memristors in Noncommunicable Disease Detection and Sensory Emulation","abstract":"Emerging materials‐based devices are revolutionizing healthcare by advancing diagnostics, monitoring, and therapeutic strategies. Among them, memristor devices capable of storing and processing information are attracting and receiving significant attention for their biomedical potential. Their tunable resistance enables sensitive, low‐power detection of major noncommunicable diseases such as diabetes and cancer. Devices fabricated with AlO x , SiO x , and GeO x have demonstrated effective glucose monitoring and biomarker detection, including sarcosine for prostate cancer and LOXL2 for breast cancer, offering pathways toward cost‐effective and early diagnosis. Beyond disease detection, memristor exhibit nociceptive properties such as threshold sensing, synaptic plasticity, and adaptive learning, enabling multifunctional applications in neurological and sensory systems. Artificial nociceptors emulate pain perception for prosthetics and robotics, photoreceptors integrate light sensing with memory for artificial vision, and memristors neuromodulatory platforms provide energy‐efficient seizure control and adaptive deep brain stimulation for Parkinson's disease. Collectively, these developments position memristors as versatile, compact, and biomimetic devices with the potential to unify diagnostics, therapy, and sensory replication in next‐generation biomedical technologies.","author":[{"family":"Panda","given":"Debashis"},{"family":"Acharya","given":"Arpan"},{"family":"Swain","given":"Subhrakali"},{"family":"Lo","given":"Cheng‐yao"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1002/cnma.202500564","URL":"https://doi.org/10.1002/cnma.202500564","source":"crossref"},{"id":"doi:10.1002/lpor.202502765","type":"article-journal","title":"VO\n                    <sub>2</sub>\n                    Nanoparticle‐Densely Packed Microwires for Flexible and Energy‐Efficient Photonic Synapses in Neuromorphic Computing","abstract":"ABSTRACT The growing demand for brain‐inspired computing in wearable electronics necessitates systems with high mechanical stability, biocompatibility, and low‐power processing. However, most existing neuromorphic technologies suffer from limited flexibility, reliance on ultraviolet light, and high energy consumption. Here, we report a flexible photonic synapse based on densely packed VO 2 microwires. The device achieves ultra‐low energy consumption (5.76 pJ per pulse) and robust performance in MNIST digit recognition with 92.5% accuracy. Even after 2000 bending cycles, it maintains 92.4% accuracy, demonstrating exceptional durability. In addition, the device retains synaptic memory responses under visible and near‐infrared stimulation, enabling RGB‐fusion reservoir computing.","author":[{"family":"Wang","given":"Yang"},{"family":"Zu","given":"Guang"},{"family":"Chen","given":"Xin"},{"family":"Li","given":"Shun‐xin"},{"family":"Zou","given":"Bo"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1002/lpor.202502765","URL":"https://doi.org/10.1002/lpor.202502765","source":"crossref"},{"id":"doi:10.1088/2752-5724/ae18e9","type":"article-journal","title":"Advances in flexible perovskite memristors for neuromorphic electronics","abstract":"Abstract The rapid growth toward digitization and artificial intelligence for high-performance storage technologies has spurred the development of brain-inspired neuromorphic electronics. In the post-Moore era, the conventional von Neumann computing architecture, characterized by physically separated processing and memory units, faces significant challenges including excessive power consumption and limited data processing capabilities. As a novel nano electronic device paradigm, memristor has emerged with integrated data storage and computing technology, which is promising for breaking through the von Neumann bottleneck. Perovskite semiconductors possess structural tunability, exceptional electronic and optical properties. When combined with mechanically robust substrates, perovskite-based memristors pave the way for the development of flexible and lightweight neuromorphic systems, suitable for wearable electronics and the internet of things. In this review, we provide a comprehensive overview of recent progress in flexible perovskite memristor technologies, with special focus on the optimization engineering for improving the resistive switching characteristics of memory devices. We then review the role of perovskite-based flexible memories in neuromorphic applications from artificial synapses to image recognition. We outline the importance of clarifying the resistance-switching mechanism in perovskite memristors. Finally, we highlight the challenges that researchers face upon fabricating operationally stable perovskite-based flexible memristors and what is still missing to unlock their full potential in next-generation neuromorphic electronics.","author":[{"family":"Wang","given":"Shuanglong"},{"family":"Liu","given":"Aqiang"},{"family":"Wu","given":"Hao"},{"family":"Xia","given":"Jiangnan"},{"family":"Yang","given":"Yongge"},{"family":"Lian","given":"Hong"},{"family":"Bì","given":"Huān"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1088/2752-5724/ae18e9","URL":"https://doi.org/10.1088/2752-5724/ae18e9","source":"crossref"},{"id":"doi:10.2139/ssrn.7090704","type":"manuscript","title":"Near-Infrared Circular-Polarization-Sensitive Chiral Au Metasurface/MoS2 Neuromorphic Device for Polarization Encryption","abstract":"Near-infrared circularly polarized light, with strong penetration capability, low background interference, and helicity freedom, holds great promise for polarization imaging, information encryption, near-infrared communication, and neuromorphic perception. Herein, we construct a 980 nm near-infrared circular-polarization-sensitive neuromorphic device based on a C3-symmetric chiral Au metasurface/monolayer MoS2 heterojunction. The device exhibits pronounced field-effect modulation, delivering a heterojunction mobility of 1.264 cm2 V–1 s–1, a photoresponsivity of approximately 825.9 A W–1. and a circular polarization response ratio of 1.534. Electromagnetic simulations reveal helicity-dependent field localization in the chiral resonator, explaining the circular-polarization-sensitive photoresponse. Moreover, the device emulates typical optoelectronic synaptic functions, including excitatory postsynaptic current, paired-pulse facilitation, and the transition from short-term plasticity to long-term plasticity. Benefiting from its near-infrared circular polarization discrimination capability, a simple information encryption and decryption system is demonstrated, highlighting its potential for polarization imaging. This work provides a promising strategy for developing two-dimensional material heterojunction devices integrating near-infrared polarization sensing, information storage, and neuromorphic processing.","author":[{"family":"Huang","given":"Ming"},{"family":"Ali","given":"Wajid"},{"family":"Li","given":"Wanying"},{"family":"Wang","given":"Lei"},{"family":"Jia","given":"Yonglei"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2139/ssrn.7090704","URL":"https://doi.org/10.2139/ssrn.7090704","source":"crossref"},{"id":"doi:10.3389/fnins.2025.1735027","type":"article-journal","title":"Sequential analysis and its applications to neuromorphic engineering","abstract":"Introduction: Neuromorphic circuits operate by comparing fluctuating signals to thresholds. This operation underpins sensing and computation in both neuromorphic architectures and biological nervous systems. Rigorous analysis of such systems is rarely attempted because the statistical tools to study them are both inaccessible and largely unknown to the neuromorphic community. Methods We offer a gentle introduction to one such tool, sequential analysis, a classical framework that addresses a particular class of threshold-crossing problems. We define the formal problem analyzed in sequential analysis and present Abraham Wald's elegant methodology for solving it. Results We then apply this framework to three examples in neuromorphic engineering, demonstrating how it can serve as a benchmark, proxy model, and design tool. Our introduction is understandable without prior training in probability or statistics. Discussion Sequential analysis provides the statistical limits of circuit performance, tractable abstractions of complex circuit behavior, and constructive rules for circuit design. It establishes rigorous statistical baselines for evaluating hardware. It links low-level circuit parameters to observable dynamics, clarifying the computational role of neuromorphic architectures. By translating performance goals into optimal thresholds and design parameters, it offers principled prescriptions that go beyond empirical tuning.","author":[{"family":"Mani","given":"Shivaram"},{"family":"Afshar","given":"Saeed"},{"family":"Monk","given":"Travis"}],"issued":{"date-parts":[[2026]]},"DOI":"10.3389/fnins.2025.1735027","URL":"https://doi.org/10.3389/fnins.2025.1735027","source":"crossref"},{"id":"doi:10.64898/2026.03.25.714179","type":"article-journal","title":"A Bidirectional Neural Interface With Direct On-Device Neuromorphic Decoding for Closed-Loop Optogenetics","abstract":"Abstract Bidirectional interfaces combined with neural de-coding algorithms are essential for closed-loop (CL) neuromodulation, enabling simultaneous neural monitoring and responsive optogenetic stimulation. However, implementing these capabilities in compact wireless headstages for freely moving animals remains challenging, as most existing platforms rely on tethered setups and external processors to execute computationally intensive decoders. This work presents the design and optimization of a neural decoder integrated into a bidirectional wireless system for CL optogenetic experiments in rodents. The proposed platform combines 32-channel electrophysiological recording with neuromorphic feature extraction, dimensionality reduction, and a nonlinear support vector machine (NL-SVM) decoder implemented on a resource-constrained Spartan-6 FPGA. Temporal dynamics are captured using spike-count features and leaky integrators, while principal component analysis (PCA) reduces the feature space to six components, enabling sub-millisecond inference with minimal memory and power requirements. Model size is further reduced using k-means clustering during training to limit the number of support vectors. Decoder performance was validated using datasets from non-human primate and rat motor cortex recordings. The proposed decoder achieved accuracy comparable to convolutional neural networks ( R 2 =0.85 vs. 0.87) and outperformed Wiener filters ( R 2 = 0.81) while requiring significantly fewer computational resources. The full system was further demonstrated in vivo through wireless closed-loop optogenetic stimulation in rats, achieving a variance accounted for (VAF) of 0.9148. Overall, this work introduces a versatile, fully self-contained, and resource-efficient platform for real-time untethered closed-loop neuroscience experiments.","author":[{"family":"Bilodeau","given":"G"},{"family":"Miao","given":"A"},{"family":"Gagnon-Turcotte","given":"G"},{"family":"Ethier","given":"C"},{"family":"Gosselin","given":"B"}],"issued":{"date-parts":[[2026]]},"DOI":"10.64898/2026.03.25.714179","URL":"https://doi.org/10.64898/2026.03.25.714179","source":"crossref"},{"id":"doi:10.1149/ma2026-01221244mtgabs","type":"article-journal","title":"Redox-Controlled Charge Localization and Hopping Transport in BBL for Neuromorphic Electronics","abstract":"Neuromorphic electronic materials seek to emulate synaptic functionality by exhibiting conductivity that depends on redox history, ionic environment, and charge localization, rather than solely on band-like transport. Such behavior is naturally realized in electrochemically gated systems, where coupled ion-electron dynamics continuously reorganize electronic states. Poly(benzimidazobenzophenanthroline) (BBL) is a particularly attractive platform in this context due to its intrinsic n-type character, electrochemical robustness, and experimentally observed non-linear conductivity. In this work, density functional theory is used to investigate how key components of electrochemically operated and antiambipolar devices such as redox state, ionic environment, and hydration govern charge distribution and electronic structure in BBL. We find that odd and even redox states exhibit qualitatively different electronic behavior, reflecting a transition from localized single-charge states to paired charge configurations. This alternation provides a microscopic explanation for the non-monotonic evolution of the bandgap and indicates the presence of distinct transport regimes under successive redox steps. The ionic environment further modulates structural and electrostatic response, while hydration stabilizes charge distributions and partially restores electronic gaps, showing the importance of realistic electrochemical conditions in predictive simulations. Complementary molecular calculations reveal that reduced states favor localized, LUMO-like frontier orbitals, whereas oxidized states exhibit more delocalized, HOMO-like character. This consistent behavior across molecular and solid-state models points to a redox-controlled electronic response. Finally, this redox-dependent localization behavior points toward a transport mechanism dominated by thermally activated charge hopping rather than band-like conduction. Comparison of electron-transfer rates across redox states reveals a non-monotonic dependence, with transport optimized at a specific redox condition. This rate maximum naturally explains the antiambipolar conductivity observed experimentally and reinforces a Marcus-type hopping mechanism as the basis for neuromorphic switching behavior in BBL. More broadly, this study provides a pathway for linking redox chemistry, ion coupling, and charge-transfer kinetics to the design of neuromorphic polymer materials.","author":[{"family":"Ghotbi","given":"Maryam"},{"family":"Sanchez","given":"Alejandro"},{"family":"Balbuena","given":"Perla"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1149/ma2026-01221244mtgabs","URL":"https://doi.org/10.1149/ma2026-01221244mtgabs","source":"crossref"},{"id":"doi:10.1063/5.0325815","type":"article-journal","title":"Flexible electrolyte-gated oxide transistors for synaptic memory and neuromorphic computing","abstract":"Flexible neuromorphic hardware integrating learning, memory, and reliable information processing is crucial for next-generation wearable electronics and artificial intelligent systems. Here, we developed flexible electrolyte-gated oxide transistors (EGOTs) for neuromorphic computing and memory applications. The optimized devices demonstrate key synaptic functionalities under electrical stimulation. The cognitive processes, such as repetitive learning, forgetting, and memory efficiency via a gate-voltage-controlled phenomenological weighting model, were also emulated. In addition, a convolutional neural network trained using EGOT characteristics achieves high classification accuracy on the Fashion-MNIST dataset for both raw and noise-perturbed inputs. These results highlight the promise of flexible devices for neuromorphic computing, wearable intelligence, and bioinspired artificial intelligence hardware.","author":[{"family":"Touqeer","given":"Ayesha"},{"family":"Sadiq","given":"Muhammad"},{"family":"Chen","given":"Zhenhao"},{"family":"Zahid","given":"Muhammad"},{"family":"Sun","given":"Jia"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1063/5.0325815","URL":"https://doi.org/10.1063/5.0325815","source":"crossref"},{"id":"doi:10.5281/zenodo.17974726","type":"article-journal","title":"SYNtzulA: Open-Source Hardware for Energy-Efficient Spiking Neural Network Inference","abstract":"Spiking Neural Networks (SNNs) represent a promising class of neural networks that emulate the behavior of biological neurons,offering significant advantages in terms of energy efficiency and computational power. These networks achieve optimal performance on neuromorphic processors, however, such hardware remains constrained by prohibitive costs and limited accessibility. To address this gap, open source Process Design Kits (PDKs) and ElectronicDesign Automation (EDA) tools can be used to facilitate broader access to hardware development. In this work, we present SYNtzulA (SYNtzulu on ASIC), a system-on-chip that integrates a RISC-V softcore with a dedicated accelerator for SNNs. The chip layout was developed using the open-source IHP-SG13G2 PDK and the OpenROAD flow, reducing development costs and promoting wider accessibility to neuromorphic hardware solutions. SYNtzulA occupies an area of approximately 5.2 mm2 and operates at a maximum frequency of 125 MHz. The system is capable of processing 109 synapses per second at its maximum frequency, while maintaining a power consumption during inference that scales linearly with the operating frequency, dissipating approximately 632 𝜇W/MHz.","author":[{"family":"Martis","given":"Luca"},{"family":"Leone","given":"Gianluca"},{"family":"Raffo","given":"Luigi"},{"family":"Meloni","given":"Paolo"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.17974726","URL":"https://doi.org/10.5281/zenodo.17974726","source":"datacite"},{"id":"doi:10.5281/zenodo.17974727","type":"article-journal","title":"SYNtzulA: Open-Source Hardware for Energy-Efficient Spiking Neural Network Inference","abstract":"Spiking Neural Networks (SNNs) represent a promising class of neural networks that emulate the behavior of biological neurons,offering significant advantages in terms of energy efficiency and computational power. These networks achieve optimal performance on neuromorphic processors, however, such hardware remains constrained by prohibitive costs and limited accessibility. To address this gap, open source Process Design Kits (PDKs) and ElectronicDesign Automation (EDA) tools can be used to facilitate broader access to hardware development. In this work, we present SYNtzulA (SYNtzulu on ASIC), a system-on-chip that integrates a RISC-V softcore with a dedicated accelerator for SNNs. The chip layout was developed using the open-source IHP-SG13G2 PDK and the OpenROAD flow, reducing development costs and promoting wider accessibility to neuromorphic hardware solutions. SYNtzulA occupies an area of approximately 5.2 mm2 and operates at a maximum frequency of 125 MHz. The system is capable of processing 109 synapses per second at its maximum frequency, while maintaining a power consumption during inference that scales linearly with the operating frequency, dissipating approximately 632 𝜇W/MHz.","author":[{"family":"Martis","given":"Luca"},{"family":"Leone","given":"Gianluca"},{"family":"Raffo","given":"Luigi"},{"family":"Meloni","given":"Paolo"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.17974727","URL":"https://doi.org/10.5281/zenodo.17974727","source":"datacite"},{"id":"doi:10.5281/zenodo.17924141","type":"article-journal","title":"Rosetta Stone of Neural Mass Models","abstract":"Brain dynamics dominate every level of neural organization—from single-neuron spiking to the macroscopic waves captured by fMRI, MEG, and EEG—yet the mathematical tools used to interrogate those dynamics remain scattered across a patchwork of traditions. Neural mass models (NMMs) (aggregate neural models) provide one of the most popular gateways into this landscape, but their sheer variety---spanning lumped parameter models, firing‐rate equations, and multi‐layer generators--- demands a unifying framework that situates diverse architectures along a continuum of abstraction and biological detail. Here, we start from the idea that oscillations originate from a simple push-pull interaction between two or more neural populations. We build from the undamped harmonic oscillator and, guided by a simple push–pull motif between excitatory and inhibitory populations, climb a systematic ladder of detail. Each rung is presented first in isolation, next under forcing, and then within a coupled network, reflecting the progression from single‐node to whole‐brain modeling. By transforming a repertoire of disparate formalisms into a navigable ladder, we hope to turn NMM choice from a subjective act into a principled design decision, helping both theorists and experimentalists translate between scales, modalities, and interventions. In doing so, we offer a \\emph{Rosetta Stone} for brain oscillation models—one that lets the field speak a common dynamical language while preserving the dialectical richness that fuels discovery.","author":[{"family":"Castaldo","given":"Francesca"},{"family":"De Palma Aristides","given":"Raul"},{"family":"Clusella","given":"Pau"},{"family":"Garcia-Ojalvo","given":"Jordi"},{"family":"Ruffini","given":"Giulio"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.17924141","URL":"https://doi.org/10.5281/zenodo.17924141","source":"datacite"},{"id":"doi:10.5281/zenodo.17924142","type":"article-journal","title":"Rosetta Stone of Neural Mass Models","abstract":"Brain dynamics dominate every level of neural organization—from single-neuron spiking to the macroscopic waves captured by fMRI, MEG, and EEG—yet the mathematical tools used to interrogate those dynamics remain scattered across a patchwork of traditions. Neural mass models (NMMs) (aggregate neural models) provide one of the most popular gateways into this landscape, but their sheer variety---spanning lumped parameter models, firing‐rate equations, and multi‐layer generators--- demands a unifying framework that situates diverse architectures along a continuum of abstraction and biological detail. Here, we start from the idea that oscillations originate from a simple push-pull interaction between two or more neural populations. We build from the undamped harmonic oscillator and, guided by a simple push–pull motif between excitatory and inhibitory populations, climb a systematic ladder of detail. Each rung is presented first in isolation, next under forcing, and then within a coupled network, reflecting the progression from single‐node to whole‐brain modeling. By transforming a repertoire of disparate formalisms into a navigable ladder, we hope to turn NMM choice from a subjective act into a principled design decision, helping both theorists and experimentalists translate between scales, modalities, and interventions. In doing so, we offer a \\emph{Rosetta Stone} for brain oscillation models—one that lets the field speak a common dynamical language while preserving the dialectical richness that fuels discovery.","author":[{"family":"Castaldo","given":"Francesca"},{"family":"De Palma Aristides","given":"Raul"},{"family":"Clusella","given":"Pau"},{"family":"Garcia-Ojalvo","given":"Jordi"},{"family":"Ruffini","given":"Giulio"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.17924142","URL":"https://doi.org/10.5281/zenodo.17924142","source":"datacite"},{"id":"doi:10.5282/ubm/epub.127336","type":"article-journal","title":"Time to Spike? Understanding the Representational Power of Spiking Neural Networks in Discrete Time","abstract":"Recent years have seen significant progress in developing spiking neural networks (SNNs) as a potential solution to the energy challenges posed by conventional artificial neural networks (ANNs). However, our theoretical understanding of SNNs remains relatively limited compared to the ever-growing body of literature on ANNs. In this paper, we study a discrete-time model of SNNs based on leaky integrate-and-fire (LIF) neurons, referred to as discrete-time LIF-SNNs, a widely used framework that still lacks solid theoretical foundations. We demonstrate that discrete-time LIF-SNNs with static inputs and outputs realize piecewise constant functions defined on polyhedral regions, and more importantly, we quantify the network size required to approximate continuous functions. Moreover, we investigate the impact of latency (number of time steps) and depth (number of layers) on the complexity of the input space partitioning induced by discrete-time LIF-SNNs. Our analysis highlights the importance of latency and contrasts these networks with ANNs employing piecewise linear activation functions. Finally, we present numerical experiments to support our theoretical findings.","author":[{"family":"Nguyen","given":"Duc"},{"family":"Araya","given":"Ernesto"},{"family":"Fono","given":"Adalbert"},{"family":"Kutyniok","given":"Gitta"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5282/ubm/epub.127336","URL":"https://doi.org/10.5282/ubm/epub.127336","source":"datacite"},{"id":"doi:10.5281/zenodo.17860614","type":"article-journal","title":"Object Segmentation: From Neuromorphic Sensing to Neuromorphic Machine Learning","abstract":"Autonomous robotic systems are increasingly deployed in a variety of applications, ranging from industrial automation to search and rescue missions. A fundamental challenge for these systems lies in effective perception and navigation, especially in environments with complex dynamics or limited visibility. Neuromorphic vision systems address these challenges, featuring attributes like high dynamic range, low latency, and microsecond-level temporal resolution. However, utilizing neuromorphic vision for advanced autonomous perception in robotics remains a largely untapped area. The objective of this dissertation is to advance the field of autonomous perception in robotics through the application of neuromorphic vision systems. In achieving this aim, three major contributions are introduced:(i) First, this work introduces Bimodal SegNet, a novel encoder-decoder architecture equipped with crossattention mechanisms. This architecture is tailored to enhance multi-modal signal processing,particularly for the task of segmentation. (ii) Second, the Graph Mixer Neural Network (GMNN) is developed, which features a unique Collaborative Contextual Mixing (CCM) layer. This innovation enhances panoptic segmentation capabilities in event-based vision systems by effectively leveraging spatiotemporal correlations. (iii) Lastly, a dynamic thresholding technique is proposed for spiking neural networks. This approach incorporates bio-plausible elements such as Spike Frequency Adaptation and Burst Suppression, aiming to boost both the efficiency and performance of the system. Extensive tests on both publicly available and customcollected ESD datasets demonstrate the efficacy of these contributions. Quantitative evaluations based on commonly accepted metrics show that the approaches introduced in this dissertationoutperform current state-of-the-art methods. The research findings from this dissertation present a step forward in overcoming the limitations of current autonomous robotic perception systems, particularly in challenging and visually constrained environments. These contributions set the stage for more robust, efficient, and intelligent autonomous systems, unlocking new possibilities for their deployment in real-world scenarios.","author":[{"family":"Kachole","given":"Sanket"}],"issued":{"date-parts":[[2024]]},"DOI":"10.5281/zenodo.17860614","URL":"https://doi.org/10.5281/zenodo.17860614","source":"datacite"},{"id":"doi:10.18452/35648","type":"article-journal","title":"How neuronal morphology impacts the synchronisation state of neuronal networks","abstract":"The biophysical properties of neurons not only affect how information is processed within cells, they can also impact the dynamical states of the network. Specifically, the cellular dynamics of action-potential generation have shown relevance for setting the (de)synchronisation state of the network. The dynamics of tonically spiking neurons typically fall into one of three qualitatively distinct types that arise from distinct mathematical bifurcations of voltage dynamics at the onset of spiking. Accordingly, changes in ion channel composition or even external factors, like temperature, have been demonstrated to switch network behaviour via changes in the spike onset bifurcation and hence its associated dynamical type. A thus far less addressed modulator of neuronal dynamics is cellular morphology. Based on simplified and anatomically realistic mathematical neuron models, we show here that the extent of dendritic arborisation has an influence on the neuronal dynamical spiking type and therefore on the (de)synchronisation state of the network. Specifically, larger dendritic trees prime neuronal dynamics for in-phase-synchronised or splayed-out activity in weakly coupled networks, in contrast to cells with otherwise identical properties yet smaller dendrites. Our biophysical insights hold for generic multicompartmental classes of spiking neuron models (from ball-and-stick-type to anatomically reconstructed models) and establish a connection between neuronal morphology and the susceptibility of neural tissue to synchronisation in health and disease.","author":[{"family":"Gowers","given":"Robert"},{"family":"Schreiber","given":"Susanne"}],"issued":{"date-parts":[[2024]]},"DOI":"10.18452/35648","URL":"https://doi.org/10.18452/35648","source":"datacite"},{"id":"doi:10.7302/7436","type":"article-journal","title":"Cholinergic Modulation of Network Activity and Applications in Sleep, Memory and Anesthesia","abstract":"In the nervous system, neurons work in tandem with a full range of complex signaling chemicals known as neuromodulators which tune neuron function to fit different behavioral tasks by varying temporal firing of the neurons. The aim of this dissertation is to use biophysically-based in-silico modeling to study how acetylcholine (ACh), one of the major neuromodulatory molecules in the brain, through its effect on cellular firing behavior, can affect brain function. As ACh modulates, among others, m-type voltage gated potassium currents through muscarinic receptors, neurons change their firing behavior in response to extracellular input. These changes are exhibited in both the average neuron firing rate as well as differences in phase relationships between coupled neurons. Our modeling results focus on elucidating how these cellular-level changes lead to modulation of network dynamics that can influence brain network functions. First, we investigated the influence of ACh on neuron firing behavior and its network-wide implications in the transition between rate and phase coding of information. We used direct current input as a proxy for the effects of external stimuli on the network and found that for high ACh conditions, increased neural gain causes a dispersion of firing rates in response to the different magnitudes of these inputs. Additionally, ACh-induced increased neural responsiveness to input allowed neurons to persist in firing to maintain a representation in frequency space (rate coding). Alternatively, in low ACh conditions, phase coding was promoted through reduced frequency spread, increased neural resonance, and augmented propensity for synchronization. Next, we analyzed how ACh-induced changes in firing behavior can contribute to the formation and consolidation of memories during non-rapid eye movement (NREM) sleep. Combining reduced neuronal network models and analysis of in vivo recordings, we tested the hypothesis that ACh-induced neuromodulatory changes during non-rapid eye movement (NREM) sleep mediate stabilization of network-wide temporal firing patterns, with the temporal order of neuronal firing dependent on their intrinsic mean firing rate during wake. We found, in both reduced models and in vivo recordings from mouse hippocampus, that the temporal order of firing among neurons during NREM sleep initially reflects their relative firing rates during prior wake. We also showed that learning-dependent reordering of sequential firing in the hippocampus during NREM sleep, together with spike timing-dependent plasticity (STDP), reconfigures neuronal firing rates across the network, similarly as has been reported in multiple brain circuits across periods of sleep. Finally, we investigated changes in electrophysiological activity associated with anesthesia and showed that differences in synaptic transmission properties can emulate the observed alteration of neural firing patterns observed during states of anesthesia. We then proposed how these effects can be ameliorated by ACh-induced changes to the muscarinic receptor-based potassium currents. Specifically, we showed that increasing the influence of the muscarinic-mediated ACh effects under simulated anesthesia leads to an increase in firing rate and neural interaction measures, showing a population level reversal of anesthesia-induced changes in activity. We found that the simulated ACh reversal restored neurons’ spiking activity, functional connectivity, as well as other measures of pairwise and population interactions.","author":[{"family":"Eniwaye","given":"Bolaji"}],"issued":{"date-parts":[[2023]]},"DOI":"10.7302/7436","URL":"https://doi.org/10.7302/7436","source":"datacite"},{"id":"doi:10.7302/8473","type":"article-journal","title":"RRAM-Based In-Memory Computing Architecture Designs","abstract":"New computing applications, e.g., deep neural network (DNN) training and inference, have been a driving force that changed the semiconductor industry landscape. The data-intensive nature of DNN applications usually leads to high computation costs and complexity, and different hardware accelerators have thus been developed to improve the efficiency of running these models. For example, NVIDIA GPU, google TPU and near-memory computing architectures can enhance DNN performance and energy efficiency compared to conventional CPUs. Past that, IMC methods can circumvent the fundamental von Neumann bottleneck and enable highly parallel computing, leading to even higher hardware efficiency and performance. This thesis examines several aspects of IMC accelerators based on emerging memory devices such as RRAM, which can offer high computation density, throughput and energy efficiency. We present a reconfigurable IMC design that can accelerate general arithmetic and logic functions. The system consists of small look-up tables (LUTs), a memory block, and search auxiliary blocks integrated into an RRAM crossbar array. External data access and data conversions are eliminated to allow operations fully in-memory. Details of logic and arithmetic functions such as addition, AND and multiplication are discussed based on search and writeback steps. A compact instruction set is demonstrated in this architecture through circuit-level simulations. Performance evaluations show that the proposed IMC architecture suits data-intensive tasks with low power consumption. Next, we discuss DNN accelerator designs using a tiled IMC architecture. Popular models including VGG-16 and MobileNet are successfully mapped and tested on the RRAM-based tiled IMC architecture. Effects of finite RRAM array size and quantized partial sums (Psums) due to ADC precision constraints are analyzed. Methods are developed to address these challenges and preserve DNN accuracy and IMC performance gains. For practical IMC implementations and to support larger models, we develop a Tiled Architecture for In-memory Computing and Heterogeneous Integration (TAICHI), a general IMC DNN accelerator design. TAICHI is based on tiled RRAM crossbar arrays heterogeneously integrated with local arithmetic units and global co-processors to allow the same chip to efficiently map different models while maintaining high energy efficiency and throughput. A hierarchical mesh network-on-chip is implemented to facilitate communication among clusters in TAICHI to balance reconfigurability and efficiency. Detailed implementation of the different circuit components is presented, and the system performance is benchmarked at several technology nodes. The heterogeneous design also allows the system to accommodate models larger than the on-chip storage capability to make the hardware system future-proof. Large-scale implementations of IMC accelerators face two technological challenges – high ADC overhead and device variability. We note these challenges can be addressed by restricting neuron activations to single-bit values, i.e., spikes, and by employing binary weights. Based on these principles, we propose efficient hardware implementation of binary-weight SNNs (BSNNs) that can be achieved using current RRAM devices and simple circuits. Binary activations also provide opportunities for intra and inter-layer data routing, and neuron circuit design optimizations. Through high-precision backpropagation-through-time (HP-BPTT) and a proper neuron design, we show BSNN can achieve accuracies comparable to floating-point models. With these co-designs, the proposed architecture can achieve high energy efficiency and accuracy for common SNN datasets. The robustness of the BSNN model against device non-idealities is further verified through experimental chip measurements. Finally, we discuss other opportunities to further enhance IMC architecture performance, including possible pipelining optimization, mapping strategy an","author":[{"family":"Wang","given":"Xinxin"}],"issued":{"date-parts":[[2023]]},"DOI":"10.7302/8473","URL":"https://doi.org/10.7302/8473","source":"datacite"},{"id":"doi:10.3929/ethz-c-000786445","type":"article-journal","title":"Neural networks for singular perturbations","abstract":"We prove deep neural network (DNN for short) expressivity rate bounds for solution sets of a model class of singularly perturbed, elliptic two-point boundary value problems, in Sobolev norms, on the bounded interval (-1,1). We assume that the given source term and reaction coefficient are analytic in [-1,1]. The expression rate bounds in Sobolev norms in terms of the NN size are robust, i.e. uniform with respect to the singular perturbation parameter ε∈(0,1] for several classes of DNN architectures. In particular, ReLU NNs, spiking NNs, and tanh- and sigmoid-activated NNs. The latter activations can represent “exponential boundary layer solution features” explicitly, in the last hidden layer of the DNN, i.e. in a shallow subnetwork, and afford improved robust expression rate bounds in terms of the NN size. All DNN architectures allow robust exponential solution expression in so-called ‘energy’ as well as in ‘balanced’ Sobolev norms, for analytic input data.","author":[{"family":"Opschoor","given":"Joost"},{"family":"Schwab","given":"Christopn"},{"family":"Xenophontos","given":"Christos"}],"issued":{"date-parts":[[2025]]},"DOI":"10.3929/ethz-c-000786445","URL":"https://doi.org/10.3929/ethz-c-000786445","source":"datacite"},{"id":"doi:10.5075/epfl-thesis-10637","type":"article-journal","title":"Supervised learning and inference of spiking neural networks with temporal coding","abstract":"The way biological brains carry out advanced yet extremely energy efficient signal processing remains both fascinating and unintelligible. It is known however that at least some areas of the brain perform fast and low-cost processing relying only on a small number of temporally encoded spikes. This thesis investigates supervised learning and inference of spiking neural networks (SNNs) with sparse temporally encoded communication. We explore different setups and compare the performance of our SNNs with that of the standard artificial neural networks (ANNs) on data classification tasks. In the first setup we consider: A family of exact mappings between a single-spike network and a ReLU network. We dismiss training for a moment and analyse deep SNNs with time-to-first-spike (TTFS) encoding. There exist a neural dynamics and a set of parameter constraints which guarantee an approximation-free mapping (conversion) from a ReLU network to an SNN with TTFS encoding. We find that a pretrained deep ReLU network can be replaced with our deep SNN without any performance loss on large-scale image classification tasks (CIFAR100 and PLACES365). However, we hypothesise that in many cases there is a need for training or fine-tuning deep spiking neural network for the specific problem at hand. In the second setup we consider: Training a deep single-spike network using a family of exact mappings from a ReLU network. We thoroughly investigate the reasons for unsuccessful training of deep SNNs with TTFS encoding and uncover an instance of the vanishing-and-exploding gradient problem. We find that a particular exact mapping solves this problem and yields an SNN with learning trajectories equivalent to those of ReLU network on large image classification tasks (CIFAR100 and PLACES365). Training is crucial for fine-tuning SNNs for the specific device properties such as low latency, the amount of noise or quantization. We hope that this study will eventually lead to an SNN hardware implementation offering a low-power inference with ANN performance on data classification tasks.","author":[{"family":"Stanojevic","given":"Ana"}],"issued":{"date-parts":[[2023]]},"DOI":"10.5075/epfl-thesis-10637","URL":"https://doi.org/10.5075/epfl-thesis-10637","source":"datacite"},{"id":"doi:10.5281/zenodo.17638161","type":"article-journal","title":"Memristor-Based Digital Twin of Mycelium for Unconventional Computing","abstract":"Fungal mycelium is widespread in nature and stands out among Engineered Living Materials (ELMs). When stimulated, it shows intricate electrical signaling activity that with the help of digital twins can be used for sustainable computing. For its further study, its emulation with novel hardware is important. Because of their special state-transition properties and non-volatility, memristors show promise for creating an effective, low-power replica of mycelium. This paper focuses on how RRAMs can be utilized to form a flexible spiking reservoir network as an alternative of common spiking neural networks (SNNs). It serves as a novel computational primitive that mimic the adaptive, self-organizing characteristics of mycelium, capable for unconventional computing.","author":[{"family":"Chatzipaschalis","given":"Ioannis"},{"family":"Tompris","given":"Ioannis"},{"family":"Kleitsiotis","given":"Georgios"},{"family":"Chatzinikolaou","given":"Theodoros"},{"family":"Fyrigos","given":"Iosif"},{"family":"Calomarde","given":"Antonio"},{"family":"Sirakoulis","given":"Georgios"},{"family":"Rubio","given":"Antonio"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.17638161","URL":"https://doi.org/10.5281/zenodo.17638161","source":"datacite"},{"id":"doi:10.5281/zenodo.17638162","type":"article-journal","title":"Memristor-Based Digital Twin of Mycelium for Unconventional Computing","abstract":"Fungal mycelium is widespread in nature and stands out among Engineered Living Materials (ELMs). When stimulated, it shows intricate electrical signaling activity that with the help of digital twins can be used for sustainable computing. For its further study, its emulation with novel hardware is important. Because of their special state-transition properties and non-volatility, memristors show promise for creating an effective, low-power replica of mycelium. This paper focuses on how RRAMs can be utilized to form a flexible spiking reservoir network as an alternative of common spiking neural networks (SNNs). It serves as a novel computational primitive that mimic the adaptive, self-organizing characteristics of mycelium, capable for unconventional computing.","author":[{"family":"Chatzipaschalis","given":"Ioannis"},{"family":"Tompris","given":"Ioannis"},{"family":"Kleitsiotis","given":"Georgios"},{"family":"Chatzinikolaou","given":"Theodoros"},{"family":"Fyrigos","given":"Iosif"},{"family":"Calomarde","given":"Antonio"},{"family":"Sirakoulis","given":"Georgios"},{"family":"Rubio","given":"Antonio"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.17638162","URL":"https://doi.org/10.5281/zenodo.17638162","source":"datacite"},{"id":"doi:10.34734/fzj-2025-04366","type":"article-journal","title":"Learning sequence timing and controlling recall speed in networks of spiking neurons","abstract":"Processing sequential inputs is a fundamental aspect of brain function, underlying tasks such as sensory perception, reading, and mathematical reasoning. At the core of the cortical algorithm, sequence processing involves learning the order and timing of elements, predicting future events, detecting unexpected deviations, and recalling learned sequences. The spiking Temporal Memory (sTM) model (Bouhadjar, 2022), a biologically inspired spiking neuronal network, provides a framework for key aspects of sequence processing. In its original version, however, it can not learn the timing of sequence elements. Further, it remains an open question how the speed of sequential recall can be flexibly modulated. We propose a mechanism in which the duration of sequence elements is represented by repeated activations of element specific neuronal populations. The sTM model can thereby represent even long time intervals, providing a biologically plausible basis for learning and recalling not only the order of sequence elements, but also complex rhythms. Additionally, we demonstrate that oscillatory background inputs can serve as a clock signal and thereby provide a robust mechanism for controlling the speed of sequence recall. Modulation of oscillation frequency and amplitude enable a stable recall across a wide range of speeds,offering a biologically relevant strategy for flexible temporal adaptation. Our findings suggest that time is encoded by unique and sparse spatio-temporal patterns of neural activity, and that the speed of sequence recall during wakefulness and sleep is correlated to the characteristics of global oscillatory activity, as observed in EEG or LFP recordings. In summary, our results contribute to the understanding of sequence processing and time representation in the brain.","author":[{"family":"Lober","given":"Melissa"},{"family":"Bouhadjar","given":"Younes"},{"family":"Diesmann","given":"Markus"},{"family":"Tetzlaff","given":"Tom"}],"issued":{"date-parts":[[2025]]},"DOI":"10.34734/fzj-2025-04366","URL":"https://doi.org/10.34734/fzj-2025-04366","source":"datacite"},{"id":"doi:10.1609/aaai.v40i10.37769","type":"article-journal","title":"SpikingIR: A Novel Converted Spiking Neural Network for Efficient Image Restoration","abstract":"Image restoration has made great progress with the rise of deep learning, but its energy consumption limits its real-world applications. Spiking Neural Networks (SNNs) are seen as energy-efficient alternatives to Artificial Neural Networks (ANNs). Applying SNNs to image restoration (IR) remains challenging, primarily due to the limited information capacity of spike-based signals. This limitation leads to quantization errors and information loss, while IR tasks are highly sensitive to output precision and error. Thus, the restoration performance suffers significantly. To address this challenge, we propose SpikingIR, an ANN-to-SNN conversion framework for IR that reduces information loss and quantization error. SpikingIR mainly consists of two components: Convolutional Pixel Mapping (CPM) and Membrane Potential Reuse Neuron (MPRN), which are designed to alleviate quantization errors and information loss in the output and intermediate layers, respectively. Specifically, CPM maps discrete outputs into a continuous space, better aligning with pixel-level details. From the perspective of information entropy, we show that outputs of CPM contain more information than the original outputs. MPRN introduces a post-processing step with relaxed firing conditions to extract residual membrane potential, reducing information waste. Furthermore, we fine-tune the converted model to jointly optimize both accuracy and energy efficiency. Experimental results demonstrate that SpikingIR achieves performance comparable to ANN counterparts across various IR benchmarks while reducing energy consumption by up to 50%.","author":[{"family":"Ouyang","given":"Yang"},{"family":"Cheng","given":"Zihan"},{"family":"Luo","given":"Xiaotong"},{"family":"Li","given":"Guoqi"},{"family":"Qu","given":"Yanyun"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1609/aaai.v40i10.37769","URL":"https://doi.org/10.1609/aaai.v40i10.37769","source":"crossref"},{"id":"doi:10.7717/peerj-cs.3554","type":"article-journal","title":"Parameter efficient hybrid spiking-quantum convolutional neural network with surrogate gradient and quantum data-reupload","abstract":"The rapid advancement of artificial intelligence (AI) and deep learning (DL) has catalyzed the emergence of several optimization-driven subfields, notably neuromorphic computing and quantum machine learning. Leveraging the differentiable nature of hybrid models, researchers have explored their potential to address complex problems through unified optimization strategies. One such development is the Spiking Quantum Neural Network (SQNN), which combines principles from spiking neural networks (SNNs) and quantum computing. However, existing SQNN implementations often depend on pretrained SNNs due to the non-differentiable nature of spiking activity and the limited scalability of current SNN encoders. In this work, we propose a novel architecture—Spiking-Quantum Data Re-upload Convolutional Neural Network (SQDR-CNN)—that enables joint training of convolutional SNNs and quantum circuits within a single backpropagation framework. Unlike its predecessor, SQDR-CNN allow convergence to reasonable performance without the reliance of pretrained spiking encoder and subsetting datasets. We also clarified some theoretical foundations, testing new design using quantum data-reupload with different training algorithm-initialization and evaluate the performance of the proposed model under noisy simulated quantum environments. As a result, we were able to achieve 86% of the mean top-performing accuracy of the state of the art (SOTA) SNN baselines, yet uses only 0.5% of the smallest spiking model’s parameters. Through this integration of neuromorphic and quantum paradigms, we aim to open new research directions and foster technological progress in multi-modal, learnable systems.","author":[{"family":"Luu","given":"Nhan"},{"family":"Luu","given":"Duong"},{"family":"Pham","given":"Nam"},{"family":"Truong","given":"Thang"}],"issued":{"date-parts":[[2026]]},"DOI":"10.7717/peerj-cs.3554","URL":"https://doi.org/10.7717/peerj-cs.3554","source":"crossref"},{"id":"doi:10.1609/aaai.v40i3.37175","type":"article-journal","title":"A Closer Look at Knowledge Distillation in Spiking Neural Network Training","abstract":"Spiking Neural Networks (SNNs) become popular due to excellent energy efficiency, yet facing challenges for effective model training. Recent works improve this by introducing knowledge distillation (KD) techniques, with the pre-trained artificial neural networks (ANNs) used as teachers and the target SNNs as students. This is commonly accomplished through a straightforward element-wise alignment of intermediate features and prediction logits from ANNs and SNNs, often neglecting the intrinsic differences between their architectures. Specifically, ANN's outputs exhibit a continuous distribution, whereas SNN's outputs are characterized by sparsity and discreteness. To mitigate this issue, we introduce two innovative KD strategies. Firstly, we propose the Saliency-scaled Activation Map Distillation (SAMD), which aligns the spike activation map of the student SNN with the class-aware activation map of the teacher ANN. Rather than performing KD directly on the raw features of ANN and SNN, our SAMD directs the student to learn from saliency activation maps that exhibit greater semantic and distribution consistency. Additionally, we propose a Noise-smoothed Logits Distillation (NLD), which utilizes Gaussian noise to smooth the sparse logits of student SNN, facilitating the alignment with continuous logits from teacher ANN. Extensive experiments on multiple datasets demonstrate the effectiveness of our methods.","author":[{"family":"Liu","given":"Xu"},{"family":"Xia","given":"Na"},{"family":"Zhou","given":"Jinxing"},{"family":"Xu","given":"Jingyuan"},{"family":"Guo","given":"Dan"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1609/aaai.v40i3.37175","URL":"https://doi.org/10.1609/aaai.v40i3.37175","source":"crossref"},{"id":"doi:10.3389/fnins.2025.1593580","type":"article-journal","title":"SpiNeRF: direct-trained spiking neural networks for efficient neural radiance field rendering","abstract":"Spiking neural networks (SNNs) have recently demonstrated significant progress across various computational tasks, due to their potential for energy efficiency. Neural radiance fields (NeRFs) excel at rendering high-quality 3D scenes but require substantial energy consumption, with limited exploration of energy-saving solutions from a neuromorphic approach. In this paper, we present SpiNeRF, a novel method that integrates the sequential processing capabilities of SNNs with the ray-casting mechanism of NeRFs, aiming to enhance compatibility and unlock new prospects for energy-efficient 3D scene synthesis. Unlike conventional SNN encoding schemes, our method considers the spatial continuity inherent in NeRF, achieving superior rendering quality. To further improve training and inference efficiency, we adopt a hybrid volumetric representation that allows the predefinition and masking of invalid sampled points along pixel-rendering rays. However, this masking introduces irregular temporal lengths, making it intractable for hardware processors, such as graphics processing units (GPUs), to conduct effective parallel training. To address this issue, we present two methods: Temporal padding (TP) and temporal condensing-and-padding (TCP). Experiments on multiple datasets demonstrate that our method outperforms previous SNN encoding schemes and artificial neural network (ANN) quantization methods in both rendering quality and energy efficiency. Compared to the full-precision ANN baseline, our method reduces energy consumption by up to 72.95% while maintaining comparable synthesis quality. Further verification using a neuromorphic hardware simulator shows that TCP-based SpiNeRF achieves additional energy efficiency gains over the ANN-based approaches by leveraging the advantages of neuromorphic computing. Codes are in https://github.com/Ikarosy/SpikingNeRF-of-CASIA .","author":[{"family":"Yao","given":"Xingting"},{"family":"Hu","given":"Qinghao"},{"family":"Zhou","given":"Fei"},{"family":"Liu","given":"Tielong"},{"family":"Mo","given":"Zitao"},{"family":"Zhu","given":"Zeyu"},{"family":"Zhuge","given":"Zhengyang"},{"family":"Cheng","given":"Jian"}],"issued":{"date-parts":[[2025]]},"DOI":"10.3389/fnins.2025.1593580","URL":"https://doi.org/10.3389/fnins.2025.1593580","source":"crossref"},{"id":"doi:10.1609/aaai.v40i45.41237","type":"article-journal","title":"HypoxSpike: Ternary Spiking Neural Network for Opioid Overdose Detection","abstract":"Opioid overdose is a growing global health crisis that claims more than 120,000 lives annually, of which more than half use opioids alone, without access to bystander intervention. Fatal overdose events are marked by motionlessness, respiratory depression, and hypoxemia, yet current wearable systems often rely on a single biomarker, limiting detection speed and accuracy. We present HypoxSpike, a novel ternary spiking neural network designed for real-time, multi-biomarker overdose detection for low-power neuromorphic hardware, optimized for integration into shoulder-based wearables. HypoxSpike combines motion, respiration, and oxygen saturation signals, while accounting for skin tone and body physiology, thus addressing known racial bias in pulse oximetry. Our research leverages an open-source shoulder-worn dataset from 19 patients experiencing sleep apnea, exploiting the shared physiological mechanisms underlying apnea and opioid overdose. This allows a direct comparison of our model with existing overdose detection approaches. HypoxSpike classifies three stages of hypoxemia with an average accuracy of 94%, outperforming state-of-the-art shoulder-based hypoxemia estimation while reducing false positive alert rates by 23.5%. By minimizing false positives, HypoxSpike supports accurate and power-efficient overdose detection, improving trust and usability for high-risk populations often overlooked by conventional systems.","author":[{"family":"Lingamoorthy","given":"Anush"},{"family":"Mishra","given":"Abhishek"},{"family":"Oni","given":"Olumuyiwa"},{"family":"Brenner","given":"Jacob"},{"family":"Kandasamy","given":"Nagarajan"},{"family":"Watson","given":"Amanda"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1609/aaai.v40i45.41237","URL":"https://doi.org/10.1609/aaai.v40i45.41237","source":"crossref"},{"id":"doi:10.1002/dac.70195","type":"article-journal","title":"Optimized Resource Management Using Binarized Spiking Neural Network With Pyramid Attention and Smart Contract Blockchain for Sustainable Spectrum Allocation in 6G","abstract":"ABSTRACT The rapid expansion of connected devices and the need for high‐speed communication make sustainable spectrum allocation and resource management crucial issues in 6G networks. Traditional spectrum management methods often rely on static or centralized techniques, limiting adaptability and efficiency in dynamic network environments. Additionally, these methods struggle to balance energy efficiency, latency, and spectrum utilization. To address limitations, this work proposes a novel Binarized Spiking Neural Network with Pyramid Attention Network (BSNN‐PAN)‐based spectrum allocation approach that integrates Pyramid Attention Networks (PANs) with binarized spiking neural networks (BSNNs). The approach employs Time‐to‐First‐Spike (TTFS) coding to encode input spectrum data into spike trains, which are processed by binary Integrate‐and‐Fire (IF) neurons. PAN introduces attention weights to capture multiscale spatial and temporal dependencies, enhancing feature extraction and decision‐making. The prairie dog optimization algorithm (PDOA) is applied to optimize hyperparameters such as weight, error, and loss, improving adaptability, efficiency, and resource allocation. The collaboration of the three significantly improves decision‐making efficiency and security in dynamic environments, demonstrating the added value of their technological integration. Experimental results demonstrate significant improvements, including 1.8‐kJ energy consumption, 412‐ms transaction latency, and 15‐ms average channel allocation time. These outcomes validate the proposed BSNN‐PAN framework as an effective and sustainable solution for next‐generation 6G spectrum management.","author":[{"family":"Basha","given":"SKJ"},{"family":"Krishnapriya","given":"Singamaneni"},{"family":"Kirubasri","given":"G"},{"family":"Varshney","given":"Gunjan"},{"family":"Mohan","given":"CR"},{"family":"Chouhan","given":"Kuldeep"},{"family":"Thokala","given":"Mohan"},{"family":"Koppala","given":"Neelima"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1002/dac.70195","URL":"https://doi.org/10.1002/dac.70195","source":"crossref"},{"id":"doi:10.1371/journal.pone.0344997","type":"article-journal","title":"Topology-aware design of spiking neural networks via modular graph architectures.","abstract":"Spiking Neural Networks (SNNs) offer a biologically plausible and energy-efficient alternative to traditional artificial neural networks (ANNs), yet their design remains constrained by limited architectural flexibility and slow training dynamics. In this work, we introduce a novel SNN framework that leverages modular graph-based topologies and explicit synaptic delays to significantly enhance both training efficiency and classification performance. Our architecture, TANet-Tiny, incorporates structured graph stages with up to 32 nodes and diverse community-driven connectivity patterns derived from KMeans clustering, Louvain modularity, and Watts–Strogatz small-world models. We integrate these topologies into a topology-aware search space and explore them via a Spatio-Temporal Topology Sampling (STTS) approach, enabling the discovery of high-performing networks without exhaustive search. Experimental results on MNIST, CIFAR-10, and CIFAR-100 demonstrate that our modular designs achieve state-of-the-art accuracy while requiring 6–10 × fewer training epochs, with top-1 accuracy reaching 99.57% on MNIST and over 92% on CIFAR-10, all with reduced parameter counts. We introduce an accuracy-per-epoch metric to quantify training efficiency and show that modularity, rather than network size, is the critical driver of performance. This work lays the groundwork for scalable, interpretable, and low-latency SNN architectures suitable for deployment in neuromorphic and edge computing environments.","author":[{"family":"Motaghian","given":"Farideh"},{"family":"Nazari","given":"Soheila"},{"family":"Dominguez-Morales","given":"Juan"},{"family":"Jafari","given":"Reza"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1371/journal.pone.0344997","URL":"https://doi.org/10.1371/journal.pone.0344997","source":"europepmc"},{"id":"doi:10.1038/s41598-025-18004-y","type":"article-journal","title":"A mean-field approach to criticality in spiking neural networks for reservoir computing.","abstract":"Abstract Spiking Neural Networks (SNNs) exhibit their optimal information-processing capability at the edge of chaos, but tuning them to this critical regime in reservoir-computing architectures usually relies on costly trial-and-error or plasticity-driven adaptation. This work presents an analytical framework for configuring in the critical regime a SNN-based reservoir with a highly general topology. Specifically, we derive and solve a mean-field equation that governs the evolution of the average membrane potential in leaky integrate-and-fire neurons, and provide an approximation for the critical point. This framework reduces the need for an extensive online fine-tuning, offering a streamlined path to near-optimal network performance from the outset. Through extensive numerical experiments, we validate the theoretical predictions by analyzing the network’s spiking dynamics and quantifying its computational capacity using the information-based Lempel-Ziv-Welch complexity near criticality. Finally, we explore self-organized quasi-criticality by implementing a local homeostatic learning rule for synaptic weights, demonstrating that the network’s dynamics remain close to the theoretical critical point. Beyond AI, our approach and findings also have significant implications for computational neuroscience, providing a principled framework for quantitatively understanding how (neuro)biological networks exploit criticality for efficient information processing. The paper is accompanied by Python code, enabling the reproducibility of the findings.","author":[{"family":"Freddi","given":"Ruggero"},{"family":"Cicala","given":"Francesco"},{"family":"Marzetti","given":"Laura"},{"family":"Basti","given":"Alessio"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1038/s41598-025-18004-y","URL":"https://doi.org/10.1038/s41598-025-18004-y","source":"europepmc"},{"id":"doi:10.7717/peerj-cs.3077","type":"article-journal","title":"Hardware implementation of FPGA-based spiking attention neural network accelerator.","abstract":"Spiking neural networks (SNNs) are recognized as third-generation neural networks and have garnered significant attention due to their biological plausibility and energy efficiency. To address the resource constraints associated with using field programmable gate arrays (FPGAs) for numerical recognition in SNNs, we proposed a lightweight spiking efficient attention neural network (SeaSNN) accelerator. We designed a simple, four-layer structured network, achieving a recognition accuracy of 93.73% through software testing on the MNIST dataset. To further enhance the model’s accuracy, we developed a highly spiking efficient channel attention mechanism (SECA), resulting in a significant performance improvement and an increase in test accuracy to 94.28%. For higher recognition speed, we optimized circuit parallelism by applying techniques such as loop unrolling, loop pipelining, and array partitioning. Finally, SeaSNN was implemented and verified on an FPGA board, achieving an inference speed of 0.000401 seconds per frame and a power efficiency of 0.42 TOPS/W at a frequency of 200 MHz. These results demonstrate that the proposed low-power, high-precision, and fast handwritten digit recognition system is well-suited for handwritten digit recognition tasks.","author":[{"family":"Geng","given":"Shiyong"},{"family":"Wang","given":"Zhida"},{"family":"Liu","given":"Zhipeng"},{"family":"Zhang","given":"Mengzhao"},{"family":"Zhu","given":"Xuelong"},{"family":"Dan","given":"Yongping"}],"issued":{"date-parts":[[2025]]},"DOI":"10.7717/peerj-cs.3077","URL":"https://doi.org/10.7717/peerj-cs.3077","source":"europepmc"},{"id":"doi:10.1101/2025.02.20.636945","type":"article-journal","title":"Hybrid Neural Network Models Explain Cortical Neuronal Activity During Volitional Movement","abstract":"Abstract Massive interconnectivity in large-scale neural networks is the key feature underlying their powerful and complex functionality. We have developed hybrid neural network (HNN) models that allow us to find statistical structure in this connectivity. Describing this structure is critical for understanding biological and artificial neural networks. The HNNs are composed of artificial neurons, a subset of which are trained to reproduce the responses of individual neurons recorded experimentally. The experimentally observed firing rates came from populations of neurons recorded in the motor cortices of monkeys performing a reaching task. After training, these networks (recurrent and spiking) underwent the same state transitions as those observed in the empirical data, a result that helps resolve a long-standing question of prescribed vs ongoing control of volitional movement. Because all aspects of the models are exposed, we were able to analyze the dynamic statistics of the connections between neurons. Our results show that the dynamics of extrinsic input to the network changed this connectivity to cause the state transitions. Two processes at the synaptic level were recognized: one in which many different neurons contributed to a buildup of membrane potential and another in which more specific neurons triggered an action potential. HNNs facilitate modeling of realistic neuron-neuron connectivity and provide foundational descriptions of large-scale network functionality.","author":[{"family":"Mao","given":"Hongwei"},{"family":"Hasse","given":"Brady"},{"family":"Schwartz","given":"Andrew"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1101/2025.02.20.636945","URL":"https://doi.org/10.1101/2025.02.20.636945","source":"preprints"},{"id":"doi:10.3390/biomimetics11010075","type":"article-journal","title":"Hybrid Spike-Encoded Spiking Neural Networks for Real-Time EEG Seizure Detection: A Comparative Benchmark","abstract":"Reliable and low-latency seizure detection from electroencephalography (EEG) is critical for continuous clinical monitoring and emerging wearable health technologies. Spiking neural networks (SNNs) provide an event-driven computational paradigm that is well suited to real-time signal processing, yet achieving competitive seizure detection performance with constrained model complexity remains challenging. This work introduces a hybrid spike encoding scheme that combines Delta–Sigma (change-based) and stochastic rate representations, together with two spiking architectures designed for real-time EEG analysis: a compact feed-forward HybridSNN and a convolution-enhanced ConvSNN incorporating depthwise-separable convolutions and temporal self-attention. The architectures are intentionally designed to operate on short EEG segments and to balance detection performance with computational practicality for continuous inference. Experiments on the CHB–MIT dataset show that the HybridSNN attains 91.8% accuracy with an F1-score of 0.834 for seizure detection, while the ConvSNN further improves detection performance to 94.7% accuracy and an F1-score of 0.893. Event-level evaluation on continuous EEG recordings yields false-alarm rates of 0.82 and 0.62 per day for the HybridSNN and ConvSNN, respectively. Both models exhibit inference latencies of approximately 1.2 ms per 0.5 s window on standard CPU hardware, supporting continuous real-time operation. These results demonstrate that hybrid spike encoding enables spiking architectures with controlled complexity to achieve seizure detection performance comparable to larger deep learning models reported in the literature, while maintaining low latency and suitability for real-time clinical and wearable EEG monitoring.","author":[{"family":"Mehrabi","given":"Ali"},{"family":"Sreenivasan","given":"Neethu"},{"family":"Gunawardana","given":"Upul"},{"family":"Gargiulo","given":"Gaetano"}],"issued":{"date-parts":[[2026]]},"DOI":"10.3390/biomimetics11010075","URL":"https://doi.org/10.3390/biomimetics11010075","source":"europepmc"},{"id":"doi:10.37965/jait.2025.0848","type":"article-journal","title":"Feature-Optimized Intrusion Detection Based on a Hybrid Spiking Neural Network for the Internet of Things","abstract":"The intrusion detection system (IDS) has gained significant attention due to its ability to enhance network utilization. However, different types of IDS approaches have been developed in traditional research that concentrate on recognizing intrusions from datasets with the help of classification. This research proposes a Lyrebird Optimization Algorithm (LOA) with Beta Hebbian Learning-based Elite Spike Neural Network (BHLESNN) for IDS classification. The LOA selects optimal features and reduces redundancy because it can explore and exploit the search space, thereby enhancing classifier performance. The usage of the beta function for spike encoding enhances temporal precision and allows a better presentation of dynamic features in network traffic. Furthermore, the network’s capability to learn temporal and spatial patterns makes it efficient in detecting IDS. The metrics, including precision, accuracy, F1-score, and recall, are assessed to show the efficiency of LOA-BHLESNN. The proposed LOA-BHLESNN achieves accuracy of 99.96%, 99.94% and 99.81% for ToN-IoT, BoT-IoT, and IoT-23 datasets, respectively, which is better than Dual Phase Feature Extraction-Conditional Tabular Generative Adversarial Networks(DPFEN-CTGAN).","author":[{"family":"Vishwanath","given":"Manu"},{"family":"Jayachandra","given":"Ananda"},{"family":"Chen","given":"Chin"},{"family":"Liu","given":"Ling"}],"issued":{"date-parts":[[2025]]},"DOI":"10.37965/jait.2025.0848","URL":"https://doi.org/10.37965/jait.2025.0848","source":"crossref"},{"id":"doi:10.1142/s2010324726500049","type":"article-journal","title":"Prediction of Electronic and Structural Properties in Two-Dimensional Materials Employing Spiking Deep Residual Network","abstract":"Two-dimensional (2D) materials exhibit unique electronic and structural properties, but accurately predicting these characteristics remains challenging due to complex feature interactions and computational constraints. This paper proposes a Prediction of Electronic and Structural Properties in 2D Materials employing Spiking Deep Residual Network (PESP-2D-SDRN). Initially, Input data is sourced from the 2DMatPedia dataset and preprocessed using the Multivariate Fast Iterative Filter (M-FIF), which effectively removes noise. Relevant features from preprocessed data are then selected using the Oscillating Spider Monkey Optimization Algorithm (OSMO), enhancing the model’s focus on informative inputs. The selected features are given to the Spiking Deep Residual Network (SDRN) for precise property prediction across diverse material classes, including semiconducting, metallic, magnetic, insulating and topological systems. Implemented in MATLAB, the proposed method demonstrates superior performance, achieving 98.5% precision and 99.4% accuracy. These results confirm that PESP-2D-SDRN offers a robust and computationally efficient solution for high-dimensional material property prediction.","author":[{"family":"Chaudhary","given":"Megha"},{"family":"Kumar","given":"Sushil"},{"family":"Kumar","given":"Sarvendra"},{"family":"Kumar","given":"Nitin"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1142/s2010324726500049","URL":"https://doi.org/10.1142/s2010324726500049","source":"crossref"},{"id":"doi:10.33425/3066-1226.1202","type":"article-journal","title":"Analyzing The Efficiency of Spiking Neural Networks in Real -Time Edge Computing Applications","abstract":"This study investigates the efficiency of Spiking Neural Networks (SNNs) compared to traditional neural network architectures—Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and Fully Connected Neural Networks (FCNNs)—in real-time edge computing applications. Through experimental evaluation, we examine key performance metrics including latency, energy consumption, accuracy, and computational complexity. Our results indicate that SNNs exhibit superior performance in terms of latency and energy efficiency, with an average latency of 15 milliseconds and energy consumption of 2.5 millijoules, significantly outperforming CNNs, RNNs, and FCNNs. While SNNs show slightly lower accuracy (85%) compared to CNNs (90%), they require fewer computational resources, with a total of 1.2 × 10^9 floating-point operations (FLOPs), making them particularly suitable for power-constrained edge devices. This study highlights the potential of SNNs to address the stringent requirements of real-time edge computing while offering insights into the trade-offs between efficiency and accuracy.","author":[{"family":"Ragipani","given":"Sowmya"},{"family":"Bushra","given":"Muneeb"},{"family":"Yerraginnela","given":"Shravani"}],"issued":{"date-parts":[[2026]]},"DOI":"10.33425/3066-1226.1202","URL":"https://doi.org/10.33425/3066-1226.1202","source":"crossref"},{"id":"doi:10.3389/fninf.2026.1854811","type":"article-journal","title":"The role of inhibition in modeling decision making with spiking neural networks.","abstract":"Decision making (DM) requires coordination of elementary information processes subserved by a distributed network of brain areas. Computational models help to understand these processes, but most of the existing models focus on simulating only one of the many parallel operations. An existing spiking neural network (SNN) model attempts to simulate DM holistically, however it does not take advantage of the significant role of inhibition at the neural level as a possible mechanism underlying DM. To address this limitation, we propose to examine the impact of neural inhibition on decision strategy selection in value-based DM using the mentioned model. In this study we outline the methodology and perform successful in-silico validation of the inhibition hypothesis with the SNN model of DM. To perform the simulation, we use a well-studied multi-attribute choice task and we validate simulation results against human behavioral data. The inhibition model achieved approximately 17% lower mean prediction error than the no-inhibition model (0.55 vs. 0.67) when evaluated on held-out, compensatory-condition data not used for fitting (Wilcoxon signed-rank test, p = 0.009, r = 0.55), with no significant difference observed in the condition used for fitting. These findings indicate that the advantage conferred by inhibition is not attributable to model complexity alone, and support neural inhibition as a plausible, biologically grounded mechanism for adaptive, context-sensitive decision strategy selection.","author":[{"family":"Król-Józaga","given":"Bartłomiej"},{"family":"Duggins","given":"Peter"},{"family":"Broniec-Wójcik","given":"Anna"},{"family":"Wichary","given":"Szymon"}],"issued":{"date-parts":[[2026]]},"DOI":"10.3389/fninf.2026.1854811","URL":"https://doi.org/10.3389/fninf.2026.1854811","source":"europepmc"},{"id":"doi:10.1063/5.0338509","type":"article-journal","title":"General aspects of internal noise in spiking neural networks.","abstract":"This study examines the impact of additive and multiplicative noise on both a single leaky integrate-and-fire neuron and a trained spiking neural network (SNN). Noise was introduced at different stages of neural processing, including the input current, membrane potential, and output spike generation. The results show that multiplicative noise applied to the membrane potential has the most detrimental effect on network performance, leading to significant degradation in accuracy. This is primarily due to its tendency to suppress membrane potentials toward large negative values, effectively silencing neuronal activity. To address this issue, input pre-filtering strategies were evaluated, with a sigmoid-based filter demonstrating the best performance by shifting inputs to a strictly positive range. Under these conditions, additive noise in the input current becomes the dominant source of performance degradation, while other noise configurations reduce accuracy by no more than 1%, even at high noise intensity. Additionally, the study compares the effects of common and uncommon noise across neuron populations in the hidden layer, revealing that SNNs exhibit greater robustness to common noise. Overall, the findings identify the most critical noise mechanisms affecting SNNs and provide practical approaches for improving their robustness.","author":[{"family":"Kolesnikov","given":"ID"},{"family":"Maksimov","given":"DA"},{"family":"Moskvitin","given":"VM"},{"family":"Semenova","given":"N"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1063/5.0338509","URL":"https://doi.org/10.1063/5.0338509","source":"europepmc"},{"id":"doi:10.1016/j.neunet.2026.109325","type":"article-journal","title":"Modeling multiple classical conditioning mechanisms in a Memristor-Based learning circuit.","abstract":"To achieve circuit-level modeling of multiple complex associative learning mechanisms in brain-inspired systems, this paper proposes an analog neural circuit based on threshold memristors, capable of emulating multiple classical conditioning phenomena in a biologically inspired manner. Specifically, the circuit incorporates multimodal sensory stimuli, including gustatory, visual, and auditory signals, to construct biologically inspired associative pathways. Through dynamical synaptic regulation by memristors across multiple neural pathways, the proposed rate-based analog circuit successfully reproduces several classical conditioning processes, such as acquisition, second-order conditioning, overshadowing, blocking, extinction, and reacquisition. It is noted that the circuit adopts a rate-based analog design paradigm, in which neuronal activation is governed by threshold comparison of weighted voltage summations, rather than spike-timing dynamics as in biologically accurate spiking neural networks. Moreover, the second-order conditioning pathway and the synaptic competition underlying overshadowing and blocking are validated through PSPICE simulations, demonstrating the synaptic plasticity of threshold memristors in modeling higher-order associative learning. Unlike prior single-mechanism approaches, this circuit realizes multiple associative processes within a unified memristive framework, extending memristor use in brain-inspired computing and cognitive hardware modeling.","author":[{"family":"Song","given":"Yueqi"},{"family":"Gao","given":"Suo"},{"family":"Iu","given":"Herbert"},{"family":"Banerjee","given":"Santo"},{"family":"Cao","given":"Yinghong"},{"family":"Chen","given":"Junxin"},{"family":"Zhang","given":"Yushu"},{"family":"Mou","given":"Jun"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1016/j.neunet.2026.109325","URL":"https://doi.org/10.1016/j.neunet.2026.109325","source":"europepmc"},{"id":"doi:10.1609/aaai.v40i24.39109","type":"article-journal","title":"Spatial-Frequency Spiking Neural Network for Underwater Object Detection","abstract":"Underwater object detection presents significant challenges due to the unique visual degradations in underwater environments, such as low contrast, poor visibility, and blurry object boundaries. While ANNs have achieved impressive detection accuracy, their high computational cost and power consumption limit their deployment in resource-constrained underwater platforms. In this work, we propose a Spatial-Frequency Spiking Neural Network (SFSNN) that combines the energy-efficient and event-driven nature of Spiking Neural Networks (SNNs) with the discriminative power of spatial-frequency analysis. SFSNN introduces a novel spatial-frequency spiking module that integrates spatial and frequency-domain representations, enhancing edge and texture features crucial for object detection in murky waters. Furthermore, we adapt the YOLOX architecture into a spike-based detector via ANN-to-SNN conversion using signed spiking neurons. Extensive experiments on the RUOD dataset demonstrate that SFSNN achieves superior performance over both SNN- and ANN-based detection models, offering a compelling solution for low-power underwater object detection.","author":[{"family":"Chen","given":"Long"},{"family":"Miao","given":"Wei"},{"family":"Gao","given":"Xin"},{"family":"Zhuge","given":"Yunzhi"},{"family":"Xu","given":"Hongming"},{"family":"Li","given":"Yaxin"},{"family":"Xu","given":"Qi"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1609/aaai.v40i24.39109","URL":"https://doi.org/10.1609/aaai.v40i24.39109","source":"crossref"},{"id":"doi:10.1142/s0129065726500139","type":"article-journal","title":"Spiking Neural Membrane Systems with Multiplexed Neurons for Enhanced Parallel Computing","abstract":"Spiking neural membrane systems (SNP systems) are distributed parallel computing models inspired by neuronal spike mechanisms. Traditional SNP systems execute rules serially within each neuron, limiting their efficiency. This paper introduces MNSNP systems, a novel variant where neurons can distinguish spike sources and execute multiple rules in parallel at one time step. MNSNP systems maintain global distributed parallelism while integrating local parallelism, significantly enhancing information processing capabilities. Computational completeness is demonstrated, proving MNSNP systems as Turing universal devices for number generation, acceptance, and function computation. Compared to existing models, MNSNP systems require fewer neurons (only 60 for universal computation), showcasing resource efficiency. An application in smoke detection achieves an AUC value of 0.9840, demonstrating practical utility. This work advances SNP systems by introducing multiplexing, paving the way for applications in robotics, feature recognition, and real-time processing.","author":[{"family":"Wang","given":"Liping"},{"family":"Liu","given":"Xiyu"},{"family":"Zhao","given":"Yuzhen"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1142/s0129065726500139","URL":"https://doi.org/10.1142/s0129065726500139","source":"crossref"},{"id":"doi:10.1162/neco.a.1501","type":"article-journal","title":"ReBaCCA-ss: Relevance-Balanced Continuum Correlation Analysis With Smoothing and Surrogating for Quantifying Similarity Between Population Spiking Activities","abstract":"Abstract Quantifying similarity between population spike patterns is essential for understanding how neural dynamics encode information. Traditional approaches, which combine kernel smoothing, principal component analysis, and canonical correlation analysis (CCA), have limitations: smoothing kernel bandwidths are often empirically chosen, CCA maximizes alignment between patterns without considering the variance explained within patterns, and baseline correlations from stochastic spiking are rarely corrected. We introduce ReBaCCA-ss (relevance-balanced continuum correlation analysis with smoothing and surrogating), a novel framework that addresses these challenges through three innovations: (1) balancing alignment and variance explanation via continuum canonical correlation, (2) correcting for noise using surrogate spike trains, and (3) selecting the optimal kernel bandwidth by maximizing the difference between true and surrogate correlations. ReBaCCA-ss is validated on both simulated data and hippocampal recordings from rats performing a delayed nonmatch-to-sample task. It reliably identifies spatiotemporal similarities between spike patterns. Combined with multidimensional scaling, ReBaCCA-ss reveals structured neural representations across trials, events, sessions, and animals, offering a powerful tool for neural population analysis.","author":[{"family":"Zhang","given":"Xiang"},{"family":"Xu","given":"Chenlin"},{"family":"Lu","given":"Zhouxiao"},{"family":"Wang","given":"Haonan"},{"family":"Song","given":"Dong"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1162/neco.a.1501","URL":"https://doi.org/10.1162/neco.a.1501","source":"crossref"},{"id":"doi:10.31224/4781","type":"article-journal","title":"Explainability Techniques and Training Strategies for Spiking Neural Networks","abstract":"Spiking Neural Networks (SNNs) have emerged as a promising class of biologically inspired models that process information through discrete spike events and temporal dynamics, closely emulating neural computation in the brain. Their inherent advantages include energy-efficient processing, event-driven operation, and natural suitability for neuromorphic hardware, positioning them as strong candidates for next-generation artificial intelligence systems. However, despite significant advancements, challenges remain in training SNNs effectively, understanding their internal decision-making processes, and deploying them in practical applications. This survey provides a comprehensive and detailed overview of the state-of-the-art in explainable and effective Spiking Neural Networks, addressing fundamental principles, training algorithms, neuromorphic hardware integration, and interpretability methodologies. We systematically explore how recent developments in surrogate gradient techniques, biologically plausible learning rules, and hybrid architectures have improved the accuracy and efficiency of SNNs, while maintaining or enhancing model transparency. Furthermore, we review a broad spectrum of explainability approaches specifically tailored to the temporal and event-driven nature of SNNs, including spike visualization, saliency mapping, rule extraction, and biologically grounded interpretability frameworks. Key application domains are surveyed, highlighting successes and challenges in neuromorphic vision, brain-machine interfaces, robotics, and biomedical signal processing, where the combination of explainability and effectiveness is critical for trustworthiness and deployment. Finally, we discuss open challenges related to training complexity, standardization of interpretability metrics, hardware-software co-design, and the balance between biological realism and engineering practicality, proposing future research directions to overcome these barriers. By synthesizing current knowledge and identifying promising avenues, this survey aims to guide researchers and practitioners in advancing the development of Spiking Neural Networks that are not only high-performing but also transparent, interpretable, and suitable for real-world applications.","author":[{"family":"Zhang","given":"Wei"},{"family":"Chen","given":"Lina"},{"family":"Huang","given":"Tao"},{"family":"Xue","given":"Heng"}],"issued":{"date-parts":[[2025]]},"DOI":"10.31224/4781","URL":"https://doi.org/10.31224/4781","source":"crossref"},{"id":"doi:10.1145/3773283","type":"article-journal","title":"RCNNshift: Moving Object Tracking with Kernel and Training-Free Spiking Neural Network","abstract":"Moving object tracking is essential in computer vision applications such as autonomous navigation and surveillance. Traditional kernel-based methods like Meanshift and Camshift are computationally efficient but often falter with complex motions and occlusions. Deep learning approaches improve performance but require significant computational resources and extensive pretraining. In this paper, we present RCNNshift, a novel tracking algorithm that integrates Three-Dimensional Random-Coupled Neural Networks (3DRCNN) with the kernel-based window update process, eliminating the need for pretraining. RCNNshift constructs a robust feature space by leveraging spatiotemporal information from 3DRCNN, enabling effective tracking with reduced computational overhead. Additionally, it employs a one-dimensional feature space for the kernel-based window update step, which can improve efficiency compared to conventional multi-dimensional approaches. We evaluated RCNNshift on multiple datasets, where it consistently outperformed Meanshift and Camshift in accuracy and reliability, particularly in high-resolution and stable remote sensing environments. RCNNshift offers a promising balance between performance and computational efficiency, making it suitable for real-world applications where resources are limited.","author":[{"family":"Liu","given":"Haoran"},{"family":"Li","given":"Peng"},{"family":"Liu","given":"Mingzhe"},{"family":"Chen","given":"Yiran"},{"family":"Yao","given":"Rui"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1145/3773283","URL":"https://doi.org/10.1145/3773283","source":"crossref"},{"id":"doi:10.1049/cim2.70043","type":"article-journal","title":"DLC‐NddMode: A Spiking Neural Network Tactile Object Recognition Model With Adaptive Optimisation and Regularisation","abstract":"ABSTRACT Empowering robots with tactile perception is crucial for the future development of intelligent robots. Tactile perception can expand the application scenarios of robots to perform more complex tasks. Unfortunately, existing approaches are flawed in their use of data collected by robotic tactile sensors because they either do not consider that tactile sensation is event‐driven, which means that tactile data are spatiotemporal, or they ignore that too few samples of tactile data would cause overfitting problems in the network model. We introduce DLC‐NddModel, a method based on spiking neural networks (SNNs) that incorporates Adam optimisation, regularisation and cosine annealing method. DLC‐NddModel aims to fully interpret the spatiotemporal nature of the tactile data using the spatiotemporal dynamics of SNNs and to alleviate the overfitting problem caused by the few samples. Furthermore, unlike previous work using SNNs, we use a different approximation function to surmount the nondifferentiable spiking activity of the spiking neurons, thus making the gradient descent method usable and effective. To effectively alleviate the overfitting problem caused by too few tactile data samples, we explore solutions through regularisation strategies that add training noise or regularisation terms to the loss function. We compare DLC‐NddModel against four prior state‐of‐the‐art approaches on the EvTouch‐Objects tactile spike dataset. Our experimental results demonstrate that DLC‐NddModel has higher recognition accuracy than the comparison method when recognising household object data with an ACC value improvement of at least 2.362%.","author":[{"family":"Liu","given":"Lin"},{"family":"Li","given":"Shaobo"},{"family":"Ji","given":"Xiaoyang"},{"family":"Yang","given":"Jing"},{"family":"Yu","given":"Zukun"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1049/cim2.70043","URL":"https://doi.org/10.1049/cim2.70043","source":"crossref"},{"id":"doi:10.1038/s41598-025-97913-4","type":"article-journal","title":"Research on target detection for autonomous driving based on ECS-spiking neural networks","abstract":"Abstract In response to the increasing demands for improved model performance and reduced energy consumption in object detection tasks relevant to autonomous driving, this research presents an advanced YOLO model, designated as ECSLIF-YOLO, which is based on the Leaky Integrate-and-Fire with Extracellular Space (ECS-LIF) framework. The primary aim of this model is to tackle the issues associated with the high energy consumption of traditional artificial neural networks (ANNs) and the suboptimal performance of existing spiking neural networks (SNNs). Empirical findings demonstrate that ECSLIF-YOLO achieves a peak mean Average Precision (mAP) of 0.917 on the BDD100K and KITTI datasets, thereby aligning with the accuracy levels of conventional ANNs while exceeding the performance of current direct-training SNN approaches without incurring additional energy costs. These findings suggest that ECSLIF-YOLO is particularly well-suited to assist the development of efficient and reliable systems for autonomous driving.","author":[{"family":"Jin","given":"Miao"},{"family":"Wang","given":"Xiaohong"},{"family":"Guo","given":"Ce"},{"family":"Yang","given":"Shufan"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1038/s41598-025-97913-4","URL":"https://doi.org/10.1038/s41598-025-97913-4","source":"crossref"},{"id":"doi:10.1088/1402-4896/add660","type":"article-journal","title":"First demonstration of leaky-integrate and fire neuron based GAA nanosheet FET with ultra-low energy consumption down to 0.8fJ/spike for spiking neural network applications","abstract":"Abstract The fundamental component of an artificial spiking neural network (SNN) is an electronic device designed to emulate a biological neuron effectively. However, a key concern is the high energy consumption and large area associated with these artificial neurons, rendering them highly inefficient. This article presents a CMOS-compatible Gate-All-Around Nanosheet Field-Effect Transistor (GAA NSFET)-based Leaky Integrate-and-Fire (LIF) neuron with gate length of 50 nm. This design achieves an exceptionally low energy consumption of 0.8 fJ, thereby establishing a new benchmark in the field. Utilizing well-calibrated 3D TCAD simulation and the SRH recombination model, along with the Unibo2 model to represent impact ionization phenomena, the proposed GAA NSFET LIF neuron effectively captures integration and recombination aspects within it. Moreover, it operates with a supply voltage of merely 1.8 V, significantly lower than conventional bulk FinFET and PD-SOI MOSFET-based LIF neurons. Also, it demonstrates spiking frequency of ~100 MHz, approximately 10 5 times higher than that of a biological neuron, controlled by the input voltage of the device, allowing the neuron adapt dynamically. Nanosheet FET technology offers superior electrostatic control, minimizing parasitic capacitances and facilitating rapid switching capabilities, thereby optimizing both power and performance simultaneously. Furthermore, the effective area of the NSFET amounts to 0.004 μ m 2 , making it an appealing choice for low-power, area-efficient, large-scale hardware implementations of SNN applications.","author":[{"family":"Kanche","given":"Madhu"},{"family":"Adwaith","given":"Dannayak"},{"family":"Kotha","given":"Venkata"},{"family":"Valasa","given":"Sresta"},{"family":"Bhukya","given":"Sunitha"},{"family":"Tayal","given":"Shubham"},{"family":"Malishetty","given":"Narender"},{"family":"Vadthiya","given":"Narendar"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1088/1402-4896/add660","URL":"https://doi.org/10.1088/1402-4896/add660","source":"crossref"},{"id":"doi:10.2139/ssrn.6402372","type":"manuscript","title":"Spiking Neural P Systems with Microglia and Delay on Synapses","abstract":"Spiking neural P systems with microglia (MSNP systems) belong to the third generating of artificial neural networks. It is a computational model inspired by biological neurons that transmit and process information through spikes, and by microglia that inhibit excessive neuronal excitation. Spiking neural P systems with microglia and delay on synapses (MSNP-DS systems) are proposed in this work, with the consideration of the phenomenon that spike transmission across synapses requires time, a factor associated with synaptic length and other relevant elements. In MSNP-DS systems, synaptic delay determines that the time required for spikes sent from the same presynaptic neuron to reach different postsynaptic neurons via different synapses varies. It has been demonstrated by research that Turing universality is exhibited by MSNP-DS systems in both number generating and number accepting models, with its computational capabilities thus proven. Additionally, we constructed a small universal MSNP-DS system comprising only 37 neurons and 2 microglia for computational functions. Compared to other universal SNP systems, the advantage of consuming fewer computational resources is demonstrated by MSNP-DS systems.","author":[{"family":"Yang","given":"Zhen"},{"family":"Wang","given":"Lin"},{"family":"Zhao","given":"Yuzhen"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2139/ssrn.6402372","URL":"https://doi.org/10.2139/ssrn.6402372","source":"crossref"},{"id":"doi:10.3233/faia250137","type":"article-journal","title":"Data Encryption Method for Financial Approval Workflow in Public Institutions Based on Spiking Neural Network Technology","abstract":"With the advent of the digital and intelligent finance era, in recent years, criminals have frequently utilized artificial intelligence technology to forge and tamper with financial data. Additionally, incidents of financial data and personal privacy breaches have occurred frequently during cross-organizational data sharing through financial sharing systems. To address this issue, this paper proposes a data encryption method for financial approval workflows based on spiking neural networks (SNN). This method establishes a new data encryption system by integrating a data exchange center, hierarchical permission management, differentiated encryption management, and SNN encryption. Leveraging the high complexity and randomness of SNN, it converts financial approval workflow data into difficult-to-decipher pulse sequences, thereby achieving financial data encryption.","author":[{"family":"Ye","given":"Aiying"},{"family":"Liu","given":"Qi"},{"family":"Zhou","given":"Lili"}],"issued":{"date-parts":[[2025]]},"DOI":"10.3233/faia250137","URL":"https://doi.org/10.3233/faia250137","source":"crossref"},{"id":"doi:10.3390/app16042073","type":"article-journal","title":"AEFSNN: Adaptive Filtering Spiking Neural Network for Event-Based Sensors","abstract":"Dynamic Vision Sensor (DVS) is an event-based imaging technology inspired by biological photoreceptors, which holds great promise for edge computing. The event streams produced by DVS are often contaminated by Background Activity (BA) noise and hot-pixel noise, which degrade downstream processing. Existing filters typically use fixed parameters, resulting in poor adaptability to changing illumination. In this paper, we propose a lightweight Adaptive Event-based Filtering Spiking Neural Network (AEFSNN) to address these limitations. Inspired by homeostatic plasticity, AEFSNN dynamically adjusts neuronal thresholds by monitoring the input-to-output spike ratio, allowing the network to autonomously converge to an optimal operating point across different lighting conditions. Furthermore, we introduce a novel neuronal wake-up mechanism that inhibits processing neurons until triggered by valid input, which effectively suppresses redundant events generated by neighboring activity. Experiments show that AEFSNN is more robust under varying illumination. Compared with current filters, our method increases the Signal-to-Noise Ratio (SNR) of the output data by 1.42–2.33 dB. Additionally, the filtered data improves classification accuracy on downstream tasks, validating its practical value for neuromorphic vision systems.","author":[{"family":"Xu","given":"Yue"},{"family":"Zhao","given":"Ye"},{"family":"Ren","given":"Yumeng"},{"family":"Chen","given":"Long"},{"family":"Chen","given":"Liang"},{"family":"Zhang","given":"Yulin"},{"family":"Qiao","given":"Shushan"}],"issued":{"date-parts":[[2026]]},"DOI":"10.3390/app16042073","URL":"https://doi.org/10.3390/app16042073","source":"crossref"},{"id":"doi:10.1002/cpe.70404","type":"article-journal","title":"Brain‐Inspired Efficient Pruning: Exploiting Criticality in Spiking Neural Networks","abstract":"ABSTRACT Spiking neural networks (SNNs) have gained significant attention due to their energy‐efficient and multiplication‐free characteristics. Despite these advantages, deploying large‐scale SNNs on edge hardware is challenging due to limited resource availability. Network pruning offers a viable approach to compress the network scale and reduce hardware resource requirements for model deployment. However, existing SNN pruning methods cause high pruning costs and performance loss because they lack efficiency in processing the sparse spike representation of SNNs. In this paper, inspired by the critical brain hypothesis in neuroscience and the high biological plausibility of SNNs, we explore and leverage criticality to facilitate efficient pruning in deep SNNs. We first explain criticality in SNNs from the perspective of maximizing feature information entropy. Second, we propose a low‐cost metric to assess neuron criticality in feature transmission and design a pruning‐regeneration method that incorporates this criticality into the pruning process. Experimental results demonstrate that our method achieves higher performance than the current state‐of‐the‐art (SOTA) method with up to 95.26% reduction in pruning cost. The criticality‐based regeneration process efficiently selects potential structures and facilitates consistent feature representation. Our code is available at https://github.com/MmPple/pruning‐criticality .","author":[{"family":"Chen","given":"Shuo"},{"family":"Liu","given":"Zeshi"},{"family":"You","given":"Haihang"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1002/cpe.70404","URL":"https://doi.org/10.1002/cpe.70404","source":"crossref"},{"id":"doi:10.1121/10.0037581","type":"article-journal","title":"Spiking neural networks for sound localization: A new perspective on illuminating auditory spatial perception","abstract":"Humans estimate sound source direction using information from their auditory neural system. Traditional methods use auditory cues [e.g., interaural time differences (ITDs) and interaural level differences (ILDs), etc] to perform sound localization. These cues are extracted from binaural signals or decoded from neuronal firing rates. In contrast, we proposed a new computational model that directly localizes sound sources using the firing rates of auditory neurons, eliminating the need for physical cue extraction and the template-matching process. This model incorporates spiking neural networks (SNNs) and artificial neural networks (ANNs) to emulate auditory spatial perception. To get firing rates, the SNN uses auditory peripheral processing and physiological models of the cochlear nucleus and medial superior olive (MSO). The SNN calculates a database of firing rates from sine tones across varying positions and frequencies to train the ANN. The ANN performs nonlinear regression to predict azimuth and elevation angles, accommodating both narrowband and broadband signals. The integration of dynamic cues resolves front–back confusion by aligning with human auditory perception. We conducted a localization listening test with 10 participants with normal hearing, enabling the refinement of network parameters to closely mimic human behavior. In the future, hearing loss can be simulated by adjusting parameters related to inner hair cell dysfunction, thereby providing a robust framework for real-world spatial hearing.","author":[{"family":"Liu","given":"Qin"},{"family":"Simon","given":"Laurent"},{"family":"Lissek","given":"Hervé"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1121/10.0037581","URL":"https://doi.org/10.1121/10.0037581","source":"crossref"},{"id":"doi:10.1007/s10791-026-10423-3","type":"article-journal","title":"Dygenn dynamic Gaussian weighted evolving Spiking Neural Network model for enhanced neurological disease detection","abstract":"Abstract Spiking Neural Network (SNN) is a promising field for modelling neuronal activity in the brain, with applications in healthcare, agriculture, finance, manufacturing and others. However, traditional SNNs often suffer from static synaptic weights, limiting their adaptability to the changes in data. Hence, to enhance performance and more accurately model the biological process of the brain, a novel SNN model, Dynamic Gaussian-Weighted Evolving Spiking Neural Network (DyGENN) is developed and applied for the detection of Parkinson’s disease and epileptic seizure. DyGENN incorporates a time-varying synaptic weight function modeled by a Gaussian distribution. The output layer evolves dynamically by either adding new output neurons, updating parameters, or skipping sample learning based on the least error margin calculated from postsynaptic firing time and the desired class labels. The suggested model is evaluated on two different imbalanced datasets: the UCI Oxford Parkinson’s Disease Detection and the Bonn Epileptic Seizure dataset using percentage splitting and stratified K -fold cross validation. Compared to popular machine learning (Support Vector Machine (SVM) and Random Forest (RF)), deep learning (Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM)) and state-of-the-art SNN frameworks (Norse, SpikingJelly, and a custom PyTorch-based model), DyGENN achieves superior performance: 97.44% accuracy, 87.5% sensitivity, 100% specificity, and geometric mean (G-Mean) of 0.9354 for Parkinson’s disease detection; 97.98% accuracy, 95.83% sensitivity, 98.67% specificity, and G-Mean of 0.972 for epileptic seizure detection. The proposed model outperforms existing models in both scenarios, suggesting its potential for broader application to other neurological conditions characterized by structured biomarker data.","author":[{"family":"Das","given":"Priya"},{"family":"Nanda","given":"Sarita"},{"family":"Sahoo","given":"Prabodh"},{"family":"Samantaray","given":"Aswini"},{"family":"Panda","given":"Ganapati"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1007/s10791-026-10423-3","URL":"https://doi.org/10.1007/s10791-026-10423-3","source":"crossref"},{"id":"doi:10.3390/electronics14081602","type":"article-journal","title":"Batchnorm-Free Binarized Deep Spiking Neural Network for a Lightweight Machine Learning Model","abstract":"The development of deep neural networks, although demonstrating astounding capabilities, leads to more complex models, high energy consumption, and expensive hardware costs. While network quantization is a widely used method to address this problem, the typical binary neural networks often require the batch normalization (batchnorm) layer to preserve their classification performances. The batchnorm layer contains full-precision multiplication and the addition operation that requires extra hardware and memory access. To address this issue, we present a batch normalization-free binarized deep spiking neural network (B-SNN). We combine spike-based backpropagation in a spiking neural network with weight binarization to further reduce the memory and computation overhead while maintaining comparable accuracy. Weight binarization reduces the huge amount of memory storage for a large number of parameters by replacing the full-precision weights (32 bit) with binary weights (1 bit). Moreover, the proposed B-SNN employs the stochastic input encoding scheme together with a spiking neuron model, thereby enabling networks to perform efficient bitwise computations without the necessity of using a batchnorm layer. As a result, our experimental results demonstrate that the efficacy of the proposed binarization scheme on deep SNNs outperforms the conventional binarized convolutional neural network.","author":[{"family":"Karimah","given":"Hasna"},{"family":"Lee","given":"Chankyu"},{"family":"Seo","given":"Yeongkyo"}],"issued":{"date-parts":[[2025]]},"DOI":"10.3390/electronics14081602","URL":"https://doi.org/10.3390/electronics14081602","source":"crossref"},{"id":"doi:10.9734/jerr/2026/v28i61935","type":"article-journal","title":"Enhancing Sequential Learning with a Hybrid EWC- Integrated Spiking Neural Network","abstract":"When neural networks are trained on tasks one after another, they often forget previously learned information, a problem known as catastrophic forgetting. While Elastic Weight Consolidation (EWC) has been effective in reducing this issue in conventional neural networks, its application to Spiking Neural Networks (SNNs) has received limited attention. In this study, EWC is integrated with SNNs to develop a continual learning model capable of retaining earlier knowledge while learning new tasks sequentially. The model is evaluated using multiple task variants of the MNIST dataset, including rotated and permuted versions, which introduce distribution shifts across tasks in a controlled manner. Experimental results show that incorporating EWC significantly reduces forgetting compared to standard SNN training, while preserving the biologically inspired and energy- efficient properties of spiking models, as demonstrated through task-wise accuracy and forgetting metrics.","author":[{"family":"Jyothi","given":"BS"},{"family":"Sireesha","given":"M"},{"family":"Chowdary","given":"KSN"},{"family":"Vandana","given":"K"},{"family":"Pranitha","given":"PS"}],"issued":{"date-parts":[[2026]]},"DOI":"10.9734/jerr/2026/v28i61935","URL":"https://doi.org/10.9734/jerr/2026/v28i61935","source":"crossref"},{"id":"doi:10.1038/s42005-025-02420-7","type":"article-journal","title":"Spiking neural networks for radio frequency interference detection in radio astronomy","abstract":"Abstract Automated systems capable of real-time operation with minimal energy consumption are increasingly important in modern radio telescopes. Spiking Neural Networks (SNNs) promise efficient and dynamic spatio-temporal data processing. This paper reformulates a significant challenge in radio astronomy, Radio Frequency Interference (RFI) detection, as a time-series segmentation task suited for SNN execution. We explore several spectrogram encoding methods and network parameters, applying first and second-order leaky integrate and fire SNNs to tackle RFI detection. We introduce a divisive normalisation-inspired pre-processing step, improving detection performance across multiple encodings strategies. Our approach achieves competitive performance on a synthetic dataset and compelling initial results on real data from the Low-Frequency Array (LOFAR) establishing a baseline for future work. We position SNNs as a viable path towards real-time RFI detection, with many possibilities for follow-up studies. These findings highlight the potential for SNNs performing complex time-series tasks, paving the way towards efficient, real-time processing in radio astronomy and other data-intensive fields.","author":[{"family":"Pritchard","given":"Nicholas"},{"family":"Wicenec","given":"Andreas"},{"family":"Bennamoun","given":"Mohammed"},{"family":"Dodson","given":"Richard"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1038/s42005-025-02420-7","URL":"https://doi.org/10.1038/s42005-025-02420-7","source":"crossref"},{"id":"doi:10.3390/electronics14030578","type":"article-journal","title":"Designing Spiking Neural Network-Based Reinforcement Learning for 3D Robotic Arm Applications","abstract":"This study investigates a novel approach to robotic arm control through integrating spiking neural networks with the twin delayed deep deterministic policy gradient reinforcement learning algorithm. Specifically, it presents the first application of spiking neural networks-based twin delayed deep deterministic policy gradient in 3D robotic manipulation, demonstrating its extension from traditional 2D tasks to complex 3D target-reaching scenarios with improved energy efficiency and stability. Additionally, with the inertial measurement unit data the system successfully mimics human arm movements, achieving a success rate of 0.95 among 50 trials and enabling an intuitive and accurate human–robot interaction system. This pioneering attempt highlights the feasibility of combining the biologically inspired spiking neural networks with the reinforcement learning algorithm to address the real-time challenges in high-dimensional robotic environments and advance the field of human–robot interaction systems.","author":[{"family":"Park","given":"Yuntae"},{"family":"Lee","given":"Jiwoon"},{"family":"Sim","given":"Donggyu"},{"family":"Cho","given":"Youngho"},{"family":"Park","given":"Cheolsoo"}],"issued":{"date-parts":[[2025]]},"DOI":"10.3390/electronics14030578","URL":"https://doi.org/10.3390/electronics14030578","source":"crossref"},{"id":"doi:10.1038/s41467-025-60818-x","type":"article-journal","title":"Experimental demonstration of third-order memristor-based artificial sensory nervous system for neuro-inspired robotics.","abstract":"Abstract The sensory nervous system in animals enables the perception of external stimuli. Developing an artificial sensory nervous system has been widely conducted to realize neuro-inspired robots capable of effectively responding to external stimuli. However, it remains challenging to develop artificial sensory nervous systems that possess sophisticated biological functions, such as habituation and sensitization, enabling efficient responses without bulky peripheral circuitry. Here, we introduce a memristor device with third-order switching complexity, emulating an artificial synapse that inherently possesses habituation and sensitization properties. Incorporating an additional resistive switching TiO x layer into the HfO 2 memristor exhibits third-order switching complexity and non-volatile habituation characteristics. Based on the third-order memristor, we propose a robotic system equipped with a memristor-based artificial sensory nervous system for optimizing the robot arm’s response to external stimuli without the aid of processors. It is experimentally demonstrated that the robot arm with the developed memristor-based artificial sensory nervous system ignores approximately 71% of safe and familiar stimuli while sensitively responding to threatening and significant stimuli, similar to the habituation and sensitization of biological sensory nervous systems. Our findings can be a stepping stone for energy-efficient and intelligent robotic systems with reduced hardware burden.","author":[{"family":"Park","given":"See"},{"family":"Jeong","given":"Hakcheon"},{"family":"Seo","given":"Seokho"},{"family":"Kwon","given":"Youna"},{"family":"Lee","given":"Jongwon"},{"family":"Choi","given":"Shinhyun"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1038/s41467-025-60818-x","URL":"https://doi.org/10.1038/s41467-025-60818-x","source":"europepmc"},{"id":"doi:10.1109/icicr65456.2025.00153","type":"article-journal","title":"BioRWKV: RWKV Large Model Inspired by Biological Brain Inspired","abstract":"BioRWKV aims to enhance the biological plausibility and efficiency of artificial intelligence systems by developing a novel large-scale Spiking Neural Network (SNN) architecture inspired by principles from biological neural systems. This research integrates Leaky Integrate-and-Fire (LIF) neuron activation layers, event-driven temporal encoding mechanisms, and Spike-Timing-Dependent Plasticity (STDP) learning rules into the model, effectively simulating synaptic plasticity in the brain. Additionally, the extended BioViRWKV model incorporates patch embedding and positional encoding to improve visual data processing capabilities. The results demonstrate that BioRWKV significantly enhances the processing of temporal sequence data and offers new perspectives for multimodal learning tasks. These advancements mark a significant step toward developing more efficient and biologically inspired AI systems, with potential applications in intelligent sensing, robotic vision, and biomedical signal processing.","author":[{"family":"Liu","given":"Xin"},{"family":"Zhang","given":"Yiwen"},{"family":"Liu","given":"Lei"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1109/icicr65456.2025.00153","URL":"https://doi.org/10.1109/icicr65456.2025.00153","source":"crossref"},{"id":"doi:10.3390/e27080811","type":"article-journal","title":"Deep Reinforcement Learning-Based Resource Allocation for UAV-GAP Downlink Cooperative NOMA in IIoT Systems","abstract":"This paper studies deep reinforcement learning (DRL)-based joint resource allocation and three-dimensional (3D) trajectory optimization for unmanned aerial vehicle (UAV)–ground access point (GAP) cooperative non-orthogonal multiple access (NOMA) communication in Industrial Internet of Things (IIoT) systems. Cooperative and non-cooperative users adopt different signal transmission strategies to meet diverse, task-oriented, quality-of-service requirements. Specifically, the DRL framework based on the Soft Actor–Critic algorithm is proposed to jointly optimize user scheduling, power allocation, and UAV trajectory in continuous action spaces. Closed-form power allocation and maximum weight bipartite matching are integrated to enable efficient user pairing and resource management. Simulation results show that the proposed scheme significantly enhances system performance in terms of throughput, spectral efficiency, and interference management, while enabling robustness against channel uncertainties in dynamic IIoT environments. The findings indicate that combining model-free reinforcement learning with conventional optimization provides a viable solution for adaptive resource management in dynamic UAV-GAP cooperative communication scenarios.","author":[{"family":"Huang","given":"Yuanyan"},{"family":"Su","given":"Jingjing"},{"family":"Lu","given":"Xuan"},{"family":"Huang","given":"Shoulin"},{"family":"Zhu","given":"Hongyan"},{"family":"Zeng","given":"Haiyong"}],"issued":{"date-parts":[[2025]]},"DOI":"10.3390/e27080811","URL":"https://doi.org/10.3390/e27080811","source":"crossref"},{"id":"doi:10.1007/s10791-026-10234-6","type":"article-journal","title":"Domesticated animal-inspired metaheuristic algorithms for static and dynamic optimization problems","abstract":"Abstract Metaheuristic optimization algorithms inspired by animal behaviors have predominantly focused on wild and predatory species, while the structured and cooperative behaviors of domesticated animals remain largely unexplored. Motivated by this gap, this paper proposes a unified family of four domesticated animal–inspired metaheuristic algorithms: Canine–Human Trust Optimization (CHTO), Feline Curiosity Optimization (FCO), Equine–Human Harmony Optimization (EHHO), and Lagomorph Bond Optimization (LBO), where lagomorph refers to rabbit-inspired bonding behavior. Each algorithm models companionship-oriented traits such as trust, curiosity, harmony, and social bonding to achieve an effective balance between exploration and exploitation. The proposed algorithms are evaluated on standard unimodal and multimodal benchmark functions and dynamic optimization problems using the Generalized Moving Peaks Benchmark (GMPB). Experimental results show that EHHO consistently outperforms classical and recent metaheuristics, including Particle Swarm Optimization (PSO), Harris Hawks Optimization (HHO), Slime Mould Algorithm (SMA), and Aquila Optimizer (AO). On highly multimodal benchmark functions such as Rastrigin (f₃), Griewank (f₄), and Levy (f₇), EHHO achieves up to 35% improvement in mean fitness values compared to Harris Hawks Optimization (HHO), which is identified as the strongest competing baseline algorithm. In dynamic optimization scenarios using the GMPB, EHHO reduces the average offline error by approximately 28% compared to Harris Hawks Optimization (HHO), while achieving a success rate of 92% across changing environments. Statistical validation using Friedman and Wilcoxon signed-rank tests confirms the robustness and statistical significance of the proposed algorithms. The findings establish domesticated animal companionship behaviors as a competitive and effective paradigm for solving both static and dynamic optimization problems.","author":[{"family":"Tantubay","given":"Neeraj"},{"family":"Solanki","given":"Surendra"},{"family":"Gupta","given":"Shabya"},{"family":"Kumar","given":"Lalit"},{"family":"Jhariya","given":"Mahendra"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1007/s10791-026-10234-6","URL":"https://doi.org/10.1007/s10791-026-10234-6","source":"crossref"},{"id":"doi:10.3390/eng6110304","type":"article-journal","title":"A Practical Tutorial on Spiking Neural Networks: Comprehensive Review, Models, Experiments, Software Tools, and Implementation Guidelines","abstract":"Spiking neural networks (SNNs) provide a biologically inspired, event-driven alternative to artificial neural networks (ANNs), potentially delivering competitive accuracy at substantially lower energy. This tutorial-study offers a unified, practice-oriented assessment that combines critical review and standardized experiments. We benchmark a shallow fully connected network (FCN) on MNIST and a deeper VGG7 architecture on CIFAR-10 across multiple neuron models (leaky integrate-and-fire (LIF), sigma–delta, etc.) and input encodings (direct, rate, temporal, etc.), using supervised surrogate-gradient training implemented in Intel Lava, SLAYER, SpikingJelly, Norse, and PyTorch. Empirically, we observe a consistent but tunable trade-off between accuracy and energy. On MNIST, sigma–delta neurons with rate or sigma–delta encodings achieve 98.1% accuracy (ANN baseline: 98.23%). On CIFAR-10, sigma–delta neurons with direct input reach 83.0% accuracy at just two time steps (ANN baseline: 83.6%). A GPU-based operation-count energy proxy indicates that many SNN configurations operate below the ANN energy baseline; some frugal codes minimize energy at the cost of accuracy, whereas accuracy-oriented settings (e.g., sigma–delta with direct or rate coding) narrow the performance gap while remaining energy-conscious—yielding up to threefold efficiency compared with matched ANNs in our setup. Thresholds and the number of time steps are decisive factors: intermediate thresholds and the minimal time window that still meets accuracy targets typically maximize efficiency per joule. We distill actionable design rules—choose the neuron–encoding pair according to the application goal (accuracy-critical vs. energy-constrained) and co-tune thresholds and time steps. Finally, we outline how event-driven neuromorphic hardware can amplify these savings through sparse, local, asynchronous computation, providing a practical playbook for embedded, real-time, and sustainable AI deployments.","author":[{"family":"Ayasi","given":"Bahgat"},{"family":"Carmona","given":"Cristóbal"},{"family":"Saleh","given":"Mohammed"},{"family":"García-Vico","given":"Angel"}],"issued":{"date-parts":[[2025]]},"DOI":"10.3390/eng6110304","URL":"https://doi.org/10.3390/eng6110304","source":"crossref"},{"id":"doi:10.1515/jisys-2024-0235","type":"article-journal","title":"Hyperparameters optimization of evolving spiking neural network using artificial bee colony for unsupervised anomaly detection","abstract":"Abstract Nowadays, anomaly detection in streaming data has gained considerable attention due to the exponential growth in the data gathered by Internet of Things applications. Analyzing and processing vast data volumes requires a system capable of working in real-time. Moreover, obtaining labeled data for supervised learning is challenging, as it requires human involvement, is time-consuming, and costly. A promising direction is implementing evolving spiking neural networks (eSNN), which can be updated whenever new data becomes available without re-training previous samples. However, eSNN encounters significant challenges when it comes to manually tuning its hyperparameter values. As such, this work covers the current research gap by suggesting a novel method to optimize the hyperparameters of eSNN called online evolving spiking neural networks-artificial bee colony (OeSNN-ABC). Multiple scenarios have been utilized to evaluate the proposed method using two benchmark datasets: the Numenta anomaly benchmark (NAB) and the Yahoo Webscope using different criteria. Further validation was provided by comparing the proposed OeSNN-ABC against five well-known optimization algorithms: particle swarm optimization, grey wolf optimization, flower pollination algorithm, whale optimization algorithm, and grid search, alongside other classifiers such as random forest, support vector machine, and k-nearest neighbor. The findings revealed that OeSNN-ABC had the best performance among all compared optimization algorithms and classifiers, outperformed prior anomaly detection techniques for the NAB dataset, and achieved competitive results for the Yahoo Webscope dataset.","author":[{"family":"Rehan","given":"Rabie"},{"family":"Sahran","given":"Shahnorbanun"},{"family":"Alyasseri","given":"Zaid"},{"family":"Sani","given":"Nor"},{"family":"Al-Betar","given":"Mohammed"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1515/jisys-2024-0235","URL":"https://doi.org/10.1515/jisys-2024-0235","source":"crossref"},{"id":"doi:10.2139/ssrn.6741860","type":"manuscript","title":"Training Spiking Neural Networks with Real-Time Propagation Through Time","abstract":"Online learning for spiking neural networks (SNNs) offers a memory-efficient alternative to Backpropagation Through Time (BPTT). However, the frequent parameter perturbations inherent in online learning may exacerbate membranepotential distribution drift previously observed in offline learning. Drift-mitigation methods developed for offline learning often assume parameters remain stable across time steps, but this assumption is violated in online algorithms with single-step parameter updates. To mitigate potential drift, we propose Real-Time Propagation Through Time (RPTT), which preserves singlestep updates while incorporating two regularizers: Spatio-Temporal Gradient Regularization (STGR) for stabilizing weight updates via moving averages, and Membrane Potential Distribution Regularization (MPDR) for constraining layerwise membrane-potential statistics. We further show that the extra gradients from STGR/MPDR are bounded and do not derail the optimization direction, ensuring RPTT converges to stationary points of the empirical risk. Experiments on CIFAR-10/100, ImageNet-1k, and DVS-CIFAR10 show that RPTT outperforms prior online methods while reducing memory by over 50% relative to BPTT. Ablations verify the roles of STGR and MPDR in drift mitigation, and streaming experiments with controlled drift demonstrate that RPTT adapts better to distribution shifts than methods using delayed parameter updates.","author":[{"family":"Wang","given":"Yaokun"},{"family":"Chen","given":"Wanze"},{"family":"Xiao","given":"Tiantian"},{"family":"Long","given":"Zhiying"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2139/ssrn.6741860","URL":"https://doi.org/10.2139/ssrn.6741860","source":"crossref"},{"id":"doi:10.5194/egusphere-egu26-9219","type":"article-journal","title":"Quantum Computing and Bayesian Optimization-Inspired Multilayer Perceptron Approach for Suspended Sediment Concentration Estimates at Rivers","abstract":"Suspended Sediment Concentration (SSC) is an indicator of a river system's quality and has several ecological impacts on aquatic life. Increased levels of SSC typically reduce the transparency of the water, thereby reducing photosynthetic activity. Moreover, it has an adverse effect on aquatic organisms due to impediments to respiration and altered habitat conditions. Additionally, contaminants, including heavy metals and organic pollutants, can be carried by suspended sediments as they attach to them. Thus, understanding and prediction of SSC became an active research topic in hydrologic science for better analyzing the sources and variations of SSC . The current research aims to estimate and predict SSC levels using measured data from the flow station 07373420 located on the Mississippi River in West Feliciana Parish in the Gulf of Mexico operated by the US Geological Survey. Over 29 years (1992 to 2021) of monthly flow and SSC data were obtained and used in this research. The SSC values were estimated using three machine learning-based modeling frameworks: Multilayer Perceptron (MLP), hybrid Quantum-Inspired Algorithms - Multilayer Perceptron (QIS-MLP), and Bayesian Optimization-Multilayer Perceptron (BO-MLP). The MLP models the complex, nonlinear relationship between independent and dependent variables using an artificial neural network. Each neuron receives the weighted inputs from the previous layer and processes them through the activation function, allowing the MLP to detect nonlinear relationships among the various input variables. With BO, the optimal hyperparameters of the MLP model (number of hidden layers, number of neurons, and learning rate) were tuned automatically. BO-MLP selects promising configurations through iterative evaluations of the acquisition function, optimally exploring the hyperparameter search space for the MLP model. The QIS-MLP utilizes an improved global search strategy combined with enhanced ability to conduct local searches by virtue of its advanced quantum optimization algorithm. To evaluate the performance of each model, the standard statistical measures of model performance were utilized, including Root Mean Squared Error (RMSE), Nash-Sutcliffe Efficiency (NSE), Kling-Gupta Efficiency (KGE), and Correlation Coefficient (CC). Among all tested input combinations, the QIS-MLP with three lagged SSC values and three lagged discharge values represents the best performance with the highest degree of accuracy and greatest potential for generalization of SSC values (RMSE = 37.34, NSE = 0.418, KGE = 0.537, and CC = 0.7). As a result, the QIS-MLP achieved superior output through optimization of functions and reduction in error rates. Based on the results, QIS-MLP were found to demonstrate superior accuracy and reliability of monthly SSC estimates and, therefore, considered an excellent candidate for intelligent modeling of complex hydrological systems.Keywords: Hydrological Modeling, Quantum-Inspired Algorithms, Suspended Sediment Concentration","author":[{"family":"Beirami","given":"Neda"},{"family":"Samadianfard","given":"Saeed"},{"family":"Gündüz","given":"Orhan"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5194/egusphere-egu26-9219","URL":"https://doi.org/10.5194/egusphere-egu26-9219","source":"crossref"},{"id":"doi:10.1002/aisy.202500526","type":"article-journal","title":"Speech Recognition with Cochlea‐Inspired In‐Sensor Computing","abstract":"Traditional speech recognition methods rely on software‐based feature extraction that introduces latency and high energy costs, making them unsuitable for low‐power devices. A proof‐of‐concept demonstration is provided of a bioinspired tonotopic sensor for speech recognition that mimics the human cochlea, using a spiral‐shaped elastic metamaterial. The measured modal response of the structure at different frequencies generates a spatially distributed signal, providing a spatiotemporal map of the input named “tonogram”. The device acts as an in‐sensor physical reservoir computing system, working simultaneously as a sensor and as a computing unit, capable of extracting features of spoken words relevant to speech recognition. Results indicate that this can serve as a valid alternative to traditional software‐based digital preprocessing, ensuring high accuracy in terms of classification, while reducing computational requirements. This work demonstrates the potential of bioinspired metamaterials for energy‐efficient auditory sensing and, beyond speech recognition, for applications such as IoT devices and edge computing artificial intelligence systems.","author":[{"family":"Beoletto","given":"Paolo"},{"family":"Milano","given":"Gianluca"},{"family":"Ricciardi","given":"Carlo"},{"family":"Bosia","given":"Federico"},{"family":"Gliozzi","given":"Antonio"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1002/aisy.202500526","URL":"https://doi.org/10.1002/aisy.202500526","source":"crossref"},{"id":"doi:10.1140/epjqt/s40507-025-00443-1","type":"article-journal","title":"A hybrid spiking neural network - quantum framework for spatio-temporal data classification: a case study on EEG data","abstract":"Abstract The study introduces a hybrid computational framework that combines neuro-inspired information processing using spiking neural networks (SNNs) and quantum information processing using quantum kernels to develop quantum-enhanced machine learning models for spatio-temporal data, demonstrated through the classification of EEG data as a case study. In the proposed SNN-quantum computation (SNN-QC) framework, SNN with spike time information representation is employed to learn spatio-temporal interactions (EEG recorded from multiple channels over time). Frequency-based (rate-based) information as spike frequency state vectors are extracted from the SNN and classified using a quantum classifier. In the latter part, we use the quantum kernel approach utilising feature maps for classification tasks. The proposed SNN-QC is demonstrated on a benchmark EEG dataset to classify three distinct wrist movement tasks in six binary classification setups as a proof of concept. We introduce a novel high-order nonlinear feature map that demonstrates improved performance over state-of-the-art feature maps and several machine learning methods across most of the tasks studied. Furthermore, the role of hyperparameters for enhanced feature maps is also highlighted. The performance of SNN-QC is evaluated using statistical metrics and cross-validation techniques, demonstrating its efficacy across multiple binary classifiers. Quantum hardware validation is conducted using both a superconducting IBM-QPU and a high-fidelity noisy simulation that replicates a real QPU. Furthermore, the results demonstrate that the SNN-QC outperforms models that use statistical features rather than features extracted from the SNN, as the SNN accounts for the temporal interaction between the spatio-temporal input variables. Finally, we conclude that the SNN-QC offers a potential pathway for developing more accurate neuromorphic-quantum enhanced systems that are both energy-efficient and biologically-inspired, well-suited for dealing with spatio-temporal data.","author":[{"family":"Jha","given":"Ravi"},{"family":"Kasabov","given":"Nikola"},{"family":"Bhattacharyya","given":"Saugat"},{"family":"Coyle","given":"Damien"},{"family":"Prasad","given":"Girijesh"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1140/epjqt/s40507-025-00443-1","URL":"https://doi.org/10.1140/epjqt/s40507-025-00443-1","source":"crossref"},{"id":"doi:10.1002/htj.23294","type":"article-journal","title":"Sheaf Attention–Based Osprey Spiking Neural Network for Effective Thermal Management and Self‐Heating Mitigation in GaAs and GaN HEMTs","abstract":"ABSTRACT This research introduces the sheaf attention–based osprey spiking neural network (SA‐OSNN) to optimize the thermal performance of GaAs and GaN high electron mobility transistors (HEMTs), which are critical for radio frequency and microwave circuits due to their excellent electron characteristics. By integrating modified osprey optimization, the SA‐OSNN approach enhances thermal management by dynamically adjusting model parameters in response to changing environmental conditions, ensuring efficient and effective thermal control. This method is used as an optimization tool that works in conjunction with established thermal management solutions, such as GaN, SiC, and AlN materials, which provide the physical properties necessary for effective heat dissipation. This analysis covers a temperature between −100°C and 200°C, examining frequencies up to 50 GHz validating the accuracy and reliability for GaAs and GaN HEMT thermal optimization. Overall, this research achieves a minimum error of 5.97834e−01 and 6.01251e−05. Also, SA‐OSNN achieves an accuracy of 97% with better performances than existing methods.","author":[{"family":"Iype","given":"Preethi"},{"family":"Babu","given":"VS"},{"family":"Paul","given":"Geenu"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1002/htj.23294","URL":"https://doi.org/10.1002/htj.23294","source":"crossref"},{"id":"doi:10.1002/aisy.202500806","type":"article-journal","title":"Review of Memristors for In‐Memory Computing and Spiking Neural Networks","abstract":"The convergence of in‐memory computing (IMC) and neuromorphic architectures offers a promising path toward energy‐efficient, scalable artificial intelligence, particularly for edge and real‐time applications. Memristors, resistive devices with nonvolatile, analog switching, uniquely enable this convergence by serving both as computational memory units for matrix‐vector multiplication and as synaptic elements for spike‐based learning. This review comprehensively explores the physical mechanisms, material classes, and integration strategies of memristors tailored for IMC and spiking neural networks, with emphasis on their implementation in crossbar arrays, synapse‐neuron emulation, and hybrid CMOS circuits. It discusses how memristors facilitate key biological learning rules like STDP and LTP/LTD and examine their deployment in edge artificial intelligence, adaptive robotics, and neuromorphic sensors. Despite their potential, device variability, noise, relaxation, scalability limits, and standardization remain pressing challenges. By synthesizing device‐level insights with architectural innovation and emerging applications, this work outlines a roadmap toward fully integrated, low‐power, and brain‐inspired computing systems.","author":[{"family":"Shooshtari","given":"Mostafa"},{"family":"Serranogotarredona","given":"Teresa"},{"family":"Linaresbarranco","given":"Bernabé"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1002/aisy.202500806","URL":"https://doi.org/10.1002/aisy.202500806","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-7019575/v1","type":"article-journal","title":"Cell Attention Networks Integrated with Spiking Neural Networks and Binary Light Spectrum Optimization for Efficient Task Scheduling in Cloud Computing: Blockchain-Enhanced","abstract":"Abstract Task scheduling in cloud computing remains a persistent and complex challenge due to inherently dynamic resource requirements and fluctuating workload patterns. Traditional methodologies—whether heuristic or AI-driven—often fall short in delivering real-time adaptability, robust scalability, and verifiable security. To address these limitations, this study introduces a composite framework integrating Cell Attention Networks (CANs), Spiking Neural Networks (SNNs), and Binary Light Spectrum Optimization (BLSO), underpinned by a Blockchain infrastructure for secure and trustworthy task management. CANs facilitate dynamic task prioritization by attending in real-time to resource-sensitive attributes. SNNs, inspired by the spike-based signaling of biological neurons, enable efficient temporal encoding and representation of task data. BLSO, as a discrete binary metaheuristics technique, ensures lightweight yet effective optimization in the task-to-resource allocation process. Security and trust are reinforced through blockchain deployment, guaranteeing immutability, auditability, and decentralized validation via smart contracts. The proposed architecture was empirically validated using simulated cloud workload environments and standard benchmark datasets within a Python–TensorFlow–BindsNET framework, integrated with Ethereum Testnet for decentralized verification. Experimental outcomes demonstrated significant enhancements in execution latency, energy efficiency, SLA compliance, and trustworthiness when compared to conventional scheduling algorithms.","author":[{"family":"Abhale","given":"Babasaheb"},{"family":"Shelake","given":"Nitin"},{"family":"Rokade","given":"Prakash"},{"family":"Gade","given":"Somnath"}],"issued":{"date-parts":[[2025]]},"DOI":"10.21203/rs.3.rs-7019575/v1","URL":"https://doi.org/10.21203/rs.3.rs-7019575/v1","source":"crossref"},{"id":"doi:10.2139/ssrn.6343052","type":"manuscript","title":"LogSNN: Spiking Neural Networks for System Log Anomaly Detection","abstract":"Log anomaly detection is the most important method for identifying potential system failure. However, current Artificial Neural Networks (ANNs) based methods significantly increase the energy consumption of the system while improving the accuracy of log classification. In this paper, we present a novel SNN framework specifically designed for system log anomaly detection, offering distinct advantages in energy efficiency and temporal information processing. First, we introduce a new LLMs-based Spike Feature Encoding (LSFE) method, which encodes word embeddings from a pre-trained T5-base model into spatiotemporal spike sequences. This approach bridge the gap between high accuracy and low energy consumption in SNNs. Second, we propose an SNN Fine-Tuning (SFT) method that integrates SNN with the T5-base model, allowing supervised fine-tuning to enhance the learning capabilities of the network. This fine-tuning process addresses the accuracy limitations typically encountered in traditional SNNs. Experimental results demonstrate that 1) LogSNN consumes only 23\\% of the energy required by UniLog and LogBERT; and 2) LogSNN achieves the highest performance with an F1-score of 0.9951 on the HDFS dataset and 0.9948 on the BGL dataset. This work presents a breakthrough in the application of SNNs to log anomaly detection, positioning SNNs as a competitive solution in resource-constrained environments.","author":[{"family":"Chen","given":"Song"},{"family":"Liao","given":"Hai"},{"family":"Liang","given":"Yan"},{"family":"Zhou","given":"Hang"},{"family":"Min","given":"Fan"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2139/ssrn.6343052","URL":"https://doi.org/10.2139/ssrn.6343052","source":"crossref"},{"id":"doi:10.1142/s0129065727500213","type":"article-journal","title":"Compensation-Balanced Recurrent Neural Architecture Based on Nonlinear Spiking Neural Membrane Systems","abstract":"Nonlinear spiking neural P (NSNP) systems offer a biologically inspired framework for modeling nonlinear temporal dynamics via spike consumption and generation. Existing NSNP-based recurrent architectures, such as the long short-term memory model inspired from spiking neural P systems (LSTM-SNP), primarily rely on implicit gating mechanisms and lack explicit state correction during recurrent propagation, which can lead to unstable state evolution and excessive information decay in long-term temporal modeling. To address this limitation, a compensation-balanced LSTM-SNP (CBLSTM-SNP) architecture is proposed. The model introduces a compensation-driven regulation mechanism, comprising a regulation branch and an adaptive compensation branch, to dynamically adjust hidden state evolution. From a nonlinear dynamical systems perspective, CBLSTM-SNP establishes a balanced recurrent state transition integrating memory preservation, nonlinear spike processing, and adaptive state restoration. Theoretical analysis confirms the boundedness and Lipschitz continuity of the proposed recurrent dynamics. Empirical evaluations on five benchmark time series datasets demonstrate that CBLSTM-SNP produces smoother hidden state trajectories and achieves comparable forecasting performance compared with several representative recurrent and hybrid models.","author":[{"family":"Fu","given":"Jun"},{"family":"Peng","given":"Hong"},{"family":"Li","given":"Bing"},{"family":"Zhou","given":"Ziyin"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1142/s0129065727500213","URL":"https://doi.org/10.1142/s0129065727500213","source":"crossref"},{"id":"doi:10.59461/ijdiic.v5i2.281","type":"article-journal","title":"Hybrid Quantum-Inspired Intelligent Computing Model for Real-Time Optimization of Massive Heterogeneous Climate Datasets","abstract":"The volume of climate observations is now growing faster than classical optimizers can process it. With petabyte-scale reanalysis archives and multi-source satellite streams in routine operation, the training budget of large climate models has become a genuine bottleneck for real-time forecasting. In this paper, we propose a hybrid quantum-inspired intelligent computing model (HQI-Opt) for fast and energy-efficient optimization of deep forecasting networks trained on heterogeneous climate tensors. The method combines a classical gradient branch, driven by a parameter-shift-style analytic estimator over a latitude-weighted loss, with a quantum-inspired branch that encodes the parameter state into an Ising/QUBO representation and applies a quantum rotation-gate mixing step modulated by an adiabatic schedule. A training loop with three concurrent branches is developed and evaluated on an ERA5 subset following the WeatherBench 2 protocol. The method is compared against six widely used optimizers: SGD with momentum, Adam, AdamW, LAMB, Lion, and Sophia. Across 69 variables and four lead times, HQI-Opt reaches the target latitude-weighted loss in approximately 43% fewer epochs; the training energy per run falls from 29.8 to 17.6 kWh, the associated carbon footprint is reduced by 41%, and the RMSE at the 3-day lead on geopotential height at 500 hPa improves from 140.8 to 137.3 m²/s². A convergence analysis is provided showing that the iterates approach a stationary point of the latitude-weighted loss as the annealing schedule decays. The results indicate that a classically simulated quantum-inspired update step, with no quantum hardware in the loop, is already a viable building block for real-time climate informatics pipelines.","author":[{"family":"Gaikwad","given":"Anirudha"},{"family":"Gaikwad","given":"Atit"},{"family":"Chauhan","given":"Shardul"}],"issued":{"date-parts":[[2026]]},"DOI":"10.59461/ijdiic.v5i2.281","URL":"https://doi.org/10.59461/ijdiic.v5i2.281","source":"crossref"},{"id":"doi:10.3233/atde260221","type":"article-journal","title":"Quantum-Behaved Pigeon-Inspired Optimization Algorithm for Multi-UAV Mission Assignment","abstract":"In order to address the challenge of large-scale reconnaissance mission assignment, the K-means cluster algorithm is applied to divide the numerous targets into several clusters. Every cluster of targets is reconnoitred by a single unmanned aerial vehicle (UAV). The quantum-behaved pigeon-inspired optimization (QPIO) algorithm with a discretization algorithm is developed to determine the reconnaissance sequence of every cluster of targets. The discretization algorithm consists of three steps: rounding the elements of the position vector, eliminating the repeated elements, and supplementing the missing elements. Two simulation experiments are conducted for multi-UAV mission assignment. The QPIO algorithm with the discretization algorithm is applied to the first experiment. The conventional PIO algorithm with the discretization algorithm is applied to the second one. The comparison of experimental results demonstrates that the QPIO algorithm with the discretization algorithm has superior performance and effectiveness.","author":[{"family":"Zhao","given":"Hongchao"},{"family":"Zhao","given":"Jianzhong"},{"family":"Ge","given":"Tingting"}],"issued":{"date-parts":[[2026]]},"DOI":"10.3233/atde260221","URL":"https://doi.org/10.3233/atde260221","source":"crossref"},{"id":"doi:10.64751/ajaccm.2026.v6.n2(2).691","type":"article-journal","title":"Neuro Fusion-CART: A Hybrid Intelligence Framework for Anomaly Detection in VANETs","abstract":"Vehicular Ad Hoc Network (VANET) generate large volumes of real-time data such as packet size, latency, vehicle speed, and signal strength, which can be vulnerable to malicious attacks. Ensuring secure and reliable communication is therefore a critical challenge. Traditionally, anomaly detection in such networks relied on rule-based systems and statistical threshold techniques. These methods depended heavily on predefined rules and manual monitoring, making them less effective in handling complex, dynamic traffic patterns. They often failed to detect unknown or evolving attack behaviors and lacked adaptability to real-time scenarios. In the proposed system, a Machine Learning (ML) and Deep Learning (DL) – Classification and Regression Trees (CART) based hybrid approach is implemented to improve detection accuracy and adaptability. The system utilizes multiple algorithms including Linear Regression (LR), Decision Tree (DT), Passive Aggressive (PA) algorithms, and a novel proposed model called KNeuroFusion CART. This hybrid model integrates Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) for feature extraction, combined with K-Nearest Neighbors (KNN) for final classification and regression. The models are trained on VANET dataset features such as packet rate, vehicle speed, and anomaly score, achieving high accuracy in detecting normal and attack patterns. The research is developed using Flask for web deployment, SQLite for database management, and libraries like Scikit-learn, TensorFlow, Pandas, and Matplotlib. The system provides both single and batch prediction capabilities, along with visualization and analysis tools","author":[{"family":"Kumar","given":"SS"},{"family":"Bhavani","given":"Penti"},{"family":"Mudassir","given":"Mohammed"},{"family":"Navya","given":"Goshika"},{"family":"Kumar","given":"Kurmeti"}],"issued":{"date-parts":[[2026]]},"DOI":"10.64751/ajaccm.2026.v6.n2(2).691","URL":"https://doi.org/10.64751/ajaccm.2026.v6.n2(2).691","source":"crossref"},{"id":"doi:10.1002/aelm.202500794","type":"article-journal","title":"Millisecond‐Scale Relaxation in Metastable HZO Ferroelectric Capacitors for Bio‐Inspired Temporal Computing","abstract":"ABSTRACT Depolarization effects in ‐based ferroelectric devices have gained significant interest for research on non‐volatile memory applications. Understanding retention mechanisms enables device optimization for temporal and brain‐inspired computing in hardware. This work presents a ferroelectric capacitor stack that shows a metastable state that relaxes on a millisecond scale through a unique interface configuration. Electrical characterization demonstrates the device's capability for implementing tunable time constants in hardware, followed by investigation of the electronic mechanisms underlying observed retention times to facilitate modeling of retention processes in ‐based ferroelectric capacitors. Internal electric fields stabilize one polarization state, enabling unipolar operation with millisecond retention for the unstable polarization. The observed retention loss depends on both polarization state and programming conditions, allowing exploitation of multiple memory time‐scales in a single device. The defective interface in the material stack provides insight into retention loss mechanisms and internal bias field origins in ‐based ferroelectric devices. These internal bias fields are related to interface composition, with oxygen vacancies identified as a possible source. The results demonstrate that ‐based ferroelectric devices offer an elegant solution for realizing tunable time constants in scaled memory devices, providing memory elements for brain‐inspired temporal computing hardware.","author":[{"family":"Fehlings","given":"Luca"},{"family":"Mikolajick","given":"Thomas"},{"family":"Noheda","given":"Beatriz"},{"family":"Covi","given":"Erika"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1002/aelm.202500794","URL":"https://doi.org/10.1002/aelm.202500794","source":"crossref"},{"id":"doi:10.62762/tmi.2025.403059","type":"article-journal","title":"Neuro-Inspired Alert System for Air Quality Prediction Using Ensemble Preprocessing and SNN Classification","abstract":"Air pollution has emerged as a critical challenge, directly affecting human health, urban sustainability, and climate systems. Traditional air-quality index (AQI) prediction models often struggle to provide timely alerts because they are not very sensitive to changes over time and are hard to understand. This paper proposes a Neuro-Inspired Alert System for Air Quality Prediction (NAS--AQP) that incorporates an ensemble learning approach using voting regression to enhance input quality, followed by classification through a Spiking Neural Network (SNN). The system is designed such that it captures the temporal and nonlinear relationships between air pollutants such as Nitrogen Dioxide ($NO_2$), Sulphur Dioxide ($SO_2$), Respirable Suspended Particulate Matter (RSPM) and Suspended Particulate Matter (SPM). The proposed method starts with preprocessing of the data and normalizing the features. After that, models like Linear Regression (LR), Random Forest (RF), and Decision Tree (DT) are trained and evaluated using Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Coefficient of Determination ($R^2$) metrics. After the training of above models, a voting based ensemble approach is used to improve AQI regression accuracy. A threshold based rule is then used to convert received, AQI predictions into binary alerts. Finally, SNN is trained to classify these alerts to achieve energy, efficient, real time, alerting by using its temporal coding and sparse activation. The ensemble voting regression model achieved an RMSE of 8.43 and MAE of 6.21, while the SNN classifier attained a classification accuracy of 92.4%.","author":[{"family":"Sharma","given":"Sneh"},{"family":"Devgan","given":"Kashish"},{"family":"Jangra","given":"Devanshi"},{"family":"Bhardwaj","given":"Aanshi"},{"family":"Aggarwal","given":"Shubhani"}],"issued":{"date-parts":[[2026]]},"DOI":"10.62762/tmi.2025.403059","URL":"https://doi.org/10.62762/tmi.2025.403059","source":"crossref"},{"id":"doi:10.2139/ssrn.5136545","type":"manuscript","title":"Optimizing Load Balancing and Task Scheduling in Cloud Computing Based on Nature-Inspired Optimization Algorithms","abstract":"A paradigm for high-performance computing services, cloud computing integrates the latest developments in distributed computing, virtualization, load balancing, parallel processing, network storage, and hot backup redundancy. In light of the fact that there is currently no reliable method for DAG task scheduling that guarantees a balanced distribution of resources across nodes, this study suggests an approach for LB algorithms in cloud computing. This research presents a new paradigm for resource selection and job scheduling, which is abstracted from swarm intelligence such as PSO, SA, Cuckoo search, etc. The LBA tackles critical issues with the system under investigation, such as system overhead or resource limits, with the goal of achieving lowest makespan and execution time while concurrently improving resource usage. This component is incorporated in the CloudSim simulation environment where the framework measures performance under different cloud models, task intensity and resource provisioning. Experimental outcomes prove that an LBA is better than an existing algorithm, where it gives an average makespan of 894.85ms, execution time of 614.88ms, and resource utilization was 69%. Comparative analysis with PSO and CSSA confirms the superior efficiency of LBA in maximizing resource allocation. These results highlight the possibility of optimization methods derived from nature to improve cloud performance by means of efficient scheduling and load balancing.","author":[{"family":"Chippagiri","given":"Srinivas"},{"family":"Ravula","given":"Preethi"},{"family":"Gangwani","given":"Divya"}],"issued":{"date-parts":[[2025]]},"DOI":"10.2139/ssrn.5136545","URL":"https://doi.org/10.2139/ssrn.5136545","source":"crossref"},{"id":"doi:10.2478/jaiscr-2026-0001","type":"article-journal","title":"A Hybridizing-Enhanced Quantum-Inspired Differential Evolution Algorithm with Multi-Strategy for Complicated Optimization","abstract":"Abstract Differential Evolution (DE) has been found to be inefficient and inaccurate for high-dimensional complex problems. Quantum-inspired Differential Evolution (QDE) possesses quantum computational properties, enabling effective handling of high-dimensional problems. However, QDE is plagued by issues of excessive mutation and poor convergence. Therefore, a hybrid enhanced Quantum-inspired Differential Evolution algorithm, termed QAHQDE, is proposed. Within QAHQDE, an improved chaotic strategy is designed. Non-repeating distributed quantum positions are generated, enhancing the diversity of initialized individuals. A quantum-adaptive mutation strategy is adopted to address the over-mutation problem inherent in QDE. The mutation degree is adaptively reduced, and convergence performance is thereby improved. A novel hybrid mutation strategy is constructed. Weighted mutation operators are combined with standard differential evolution. Local and global search capabilities are balanced, and convergence accuracy is enhanced. The performance of QAHQDE was evaluated against 38 algorithms using 48 benchmark functions from CEC2005, CEC2010, and CEC2013, across dimensions D=100, 500, 1000, and 3000. Experimental results demonstrate that QAHQDE outperforms QDE by at least three orders of magnitude. Superior convergence performance, higher convergence accuracy, and excellent stability are exhibited by QAHQDE on most functions.","author":[{"family":"Chen","given":"Yu"},{"family":"Xu","given":"Haotian"},{"family":"Liu","given":"Jie"},{"family":"Hou","given":"Ming"},{"family":"Li","given":"Yang"},{"family":"Qiu","given":"Shaopeng"},{"family":"Sun","given":"Maohua"},{"family":"Zhao","given":"Huimin"},{"family":"Deng","given":"Wu"}],"issued":{"date-parts":[[2025]]},"DOI":"10.2478/jaiscr-2026-0001","URL":"https://doi.org/10.2478/jaiscr-2026-0001","source":"crossref"},{"id":"doi:10.1101/2025.07.06.663394","type":"article-journal","title":"Large-scale classification of metagenomic samples: a comparative analysis of classical machine learning techniques vs a novel brain-inspired hyperdimensional computing approach","abstract":"Abstract Classical machine learning techniques have revolutionized bioinformatics, enabling researchers to extract knowledge from complex biological data. However, these techniques often struggle with high-dimensional data, where the increasing number of features leads to decreased performance, also affecting models accuracy. To address this problem, we explore hyperdimensional computing (HDC), an emerging brain-inspired computational paradigm that leverages high-dimensional vectors and simple arithmetic operations to represent and manipulate complex patterns, as an alternative approach in the context of supervised machine learning. In this work, we present a comprehensive comparative analysis of HDC against established machine learning techniques across a range of classification tasks. As a representative use case, we focus on classifying heterogeneous metagenomic samples based on their quantitative microbial profiles, using publicly available microbiome datasets. Our results demonstrate that HDC achieves comparable, and in some cases, superior classification accuracy to classical methods. Furthermore, our findings highlight the potential of HDC for improved computational efficiency, particularly when dealing with large-scale datasets, suggesting the HDC-based classifier as a promising tool for bioinformatics research, particularly in areas characterized by high-dimensional data. We also offer a Galaxy powered toolset to analyze your own datasets and generate reproducible workflows and adopt these methods in your own research with ease. Our investigation into the application of a HDC-based supervised machine learning technique for classifying microbial profiles in metagenomic samples yielded promising results, demonstrating the potential of this novel computational paradigm to complement and, in some cases, surpass the performances of well established machine learning techniques. Importance The growing complexity and dimensionality of biological data require more efficient and scalable machine learning approaches. HDC offers a novel alternative to conventional methods, showing resilience to high-dimensionality while maintaining competitive accuracy. This study demonstrates the effectiveness of HDC in classifying metagenomic samples based on their microbial composition. Our results suggest that HDC not only matches, but sometimes exceeds the performance of well-established methods. We make this approach accessible to the broader bioinformatics community with an open-source tool fully integrated into the Galaxy platform, facilitating its adoption and reproducibility, with the aim of integrating HDC into mainstream biological data analysis pipelines, especially for complex, high-dimensional tasks in microbiome research.","author":[{"family":"Joshi","given":"Jayadev"},{"family":"Cumbo","given":"Fabio"},{"family":"Blankenberg","given":"Daniel"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1101/2025.07.06.663394","URL":"https://doi.org/10.1101/2025.07.06.663394","source":"crossref"},{"id":"doi:10.1093/neuonc/noaf201.0308","type":"article-journal","title":"CNSC-101. CHARACTERIZATION OF GBM INDUCED NEURO-IMMUNE AXIS DYSREGULATION","abstract":"Abstract Glioblastoma (GBM) is the most common and aggressive form of malignant brain tumor in adults. Although GBM is believed to remain localized within the central nervous system (CNS), it exerts systemic effects; most notably, a profound immune suppression. Patients with GBM frequently exhibit lymphopenia, diminished T cell responsiveness, and elevated levels of immunosuppressive myeloid-derived suppressor cells (MDSCs). Through murine studies, our group has previously shown that intracranial gliomas induce changes in the bone marrow, skewing hematopoiesis toward immunosuppressive myeloid lineages at the expense of lymphoid progenitors. This is a striking observation of altered hematopoietic stem cell dynamics, given the anatomical separation between the brain and bone marrow and in lieu of studies of metastatic spread, suggests the presence of a signaling conduit—potentially neural in nature—that enables a novel form of CNS tumor-host communication. Spinal cord and peripheral nerves—particularly the sciatic nerve—represent major neuroimmune interfaces. These regions contain resident immune cells, including microglia and macrophages, that respond dynamically to neuronal activity and injury. These immune cells are capable of producing cytokines, expressing neurotransmitter receptors, and relaying signals that could influence distal organs, including the bone marrow. Considering this, we hypothesize that GBM induces disruption of neuroimmune signaling pathways that alters the phenotype and function of peripheral immune cells and hematopoietic progenitors. To investigate this, we have identified and characterized genotypic and phenotypic GBM-associated changes in spinal and peripheral nerve tissues. We have also probed the spinal and peripheral nerve tissues to assess variances in population, spatial distribution and expression of key neurotransmitter receptors of the immune cells within them.","author":[{"family":"Reid","given":"Alexandra"},{"family":"Jin","given":"Dan"},{"family":"Flores","given":"Catherine"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1093/neuonc/noaf201.0308","URL":"https://doi.org/10.1093/neuonc/noaf201.0308","source":"crossref"},{"id":"doi:10.1088/2634-4386/adef76","type":"article-journal","title":"A retina-inspired pathway to real-time motion prediction inside image sensors for extreme-edge intelligence","abstract":"Abstract The ability to predict motion in real time is fundamental to many maneuvering activities in animals, particularly those critical for survival, such as attack and escape responses. Given its significance, it is no surprise that motion prediction (MP) in animals begins in the retina. Similarly, autonomous systems utilizing computer vision could greatly benefit from the capability to predict motion in real time. Therefore, for computer vision applications, MP should be integrated directly at the camera pixel level. Towards that end, we present a retina-inspired neuromorphic framework capable of performing real-time, energy-efficient MP directly within camera pixels. Our hardware-algorithm framework, implemented using GlobalFoundries’ 22 nm FDSOI technology, integrates key retinal MP compute blocks, including a biphasic filter, spike adder, non-linear circuit, and a 2D array for multi-directional MP. Additionally, integrating the sensor and MP compute die using a 3D Cu–Cu hybrid bonding approach improves design compactness by minimizing area usage and simplifying routing complexity. Validated on real-world object stimuli, the model delivers efficient, low-latency MP for decision-making scenarios reliant on predictive visual computation, while consuming only 18.56 pJ/MP in our mixed-signal hardware implementation.","author":[{"family":"Chakraborty","given":"Subhradip"},{"family":"Snyder","given":"Shay"},{"family":"Kaiser","given":"Md"},{"family":"Parsa","given":"Maryam"},{"family":"Schwartz","given":"Gregory"},{"family":"Jaiswal","given":"Akhilesh"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1088/2634-4386/adef76","URL":"https://doi.org/10.1088/2634-4386/adef76","source":"crossref"},{"id":"doi:10.1088/2631-8695/adadc7","type":"article-journal","title":"Synaptic neural circuit inspired by side-gated graphene synaptic transistors for neuromorphic computing","abstract":"Abstract With the rapid advancement of artificial intelligence, the energy consumption bottleneck inherent in the von Neumann computing architecture poses a significant obstacle to the future development of edge computing, artificial intelligence, and information technology. Consequently, it is crucial to develop synaptic neural circuits that exhibit memory and learning properties through synaptic plasticity. Drawing inspiration from the side-gated graphene synaptic transistor, we have designed a synaptic neural circuit comprising four key components: pre-voltage input, synaptic weight modulation, electric double-layer effect, and post-membrane current response. Through comprehensive simulations, we have successfully mimicked various synaptic behaviours, including long-term and short-term synaptic plasticity, paired-pulse facilitation, spiking-rate-dependent plasticity, spiking time-dependent plasticity, and Pavlovian associative learning. This approach establishes a robust framework for designing synaptic neural network circuits with advanced learning capabilities, thereby enhancing the practical applications of neural networks and machine learning.","author":[{"family":"Wen","given":"Huifeng"},{"family":"Yong","given":"Haoran"},{"family":"He","given":"Xiaoying"},{"family":"Rao","given":"Lan"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1088/2631-8695/adadc7","URL":"https://doi.org/10.1088/2631-8695/adadc7","source":"crossref"},{"id":"doi:10.1145/3807784","type":"article-journal","title":"Cross-Layer Design of Vector-Symbolic Computing: Bridging Cognition and Brain-Inspired Hardware Acceleration","abstract":"Vector Symbolic Architectures (VSAs), also known as hyperdimensional (HD) computing, are increasingly deployed in cognitive applications due to their simple and efficient operations. The widespread adoption has, in turn, spurred the development of a diverse set of hardware solutions that optimize VSA performance for embedded and edge AI systems. Despite these advances, there remains a lack of comprehensive, unified discussion on the co-design and co-evolution of VSA algorithms and hardware. This survey aims at bridging that gap by linking theoretical, software-level explorations with efficient hardware architectures and emerging technology fabrics for VSAs, providing co-design insights that are accessible to both algorithm and hardware communities. First, we introduce the principles of vector-symbolic computing, including its core mathematical operations and learning paradigms. Second, we provide an in-depth discussion on hardware technologies for VSAs, analyzing analog, mixed-signal, and digital circuit design styles. We compare hardware implementations of VSAs by carrying out detailed analysis of their performance characteristics and tradeoffs, from which we distill design guidelines that are applicable across arbitrary VSA formulations. Third, we discuss a methodology for cross-layer design of VSAs that identifies synergies across layers and explores key ingredients for hardware/software co-design of VSAs. Finally, as a concrete case study of this methodology, we present an in-memory computing hardware design for VSA-based hierarchical cognition, illustrating how the proposed co-design principles translate into efficient architectures. The article concludes with a discussion of open research challenges and opportunities for future explorations.","author":[{"family":"Du","given":"Shuting"},{"family":"Ibrahim","given":"Mohamed"},{"family":"Wan","given":"Zishen"},{"family":"Zheng","given":"Luqi"},{"family":"Zhao","given":"Boheng"},{"family":"Fan","given":"Zhenkun"},{"family":"Liu","given":"Che"},{"family":"Krishna","given":"Tushar"},{"family":"Raychowdhury","given":"Arijit"},{"family":"Li","given":"Haitong"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1145/3807784","URL":"https://doi.org/10.1145/3807784","source":"crossref"},{"id":"doi:10.1093/neuonc/noag172.022","type":"article-journal","title":"54 Access to Prehabilitation in UK Neuro-Oncology: Insights from a UK-wide review of adult brain tumour services (2024-2026)","abstract":"Abstract Background Prehabilitation aims to optimise physical, psychological and functional health prior to and during cancer treatment. While an increasing focus in neuro-oncology, little is known about its availability or structure across the United Kingdom (UK). Methods As part of the Tessa Jowell Centre of Excellence programme, workforce and service data were collected from 25 UK adult neuro-oncology centres (2024–2026), covering 90+% of the UK population. Responses were extracted from application forms and thematically analysed. Data were self-reported and validated through expert review. Results 17/25 centres (68%) reported involving allied health professionals (AHPs) to prepare patients for surgical/oncological treatment. Only 7 centres (28%) reported providing components of prehabilitation to all patients; a further 10 offered restricted access to specific subgroups (e.g. awake craniotomy, high-risk patients, or tumour subtypes). Service scope varied: 7 centres offered pre-surgical input only, 3 pre-oncology only, and 7 both. Of the 17 centres providing input, 9 delivered active intervention, while the remainder offered assessment alone. Prehabilitation was delivered by AHPs from multiple disciplines, most commonly occupational therapists (12 centres), physiotherapists (10), speech and language therapists (8), neuropsychologists (7) and dietitians (4). Assessment approaches ranged from comprehensive evaluation to clinic-based screening. Barriers were common: 9 centres reported funding and workforce constraints, 2 cited limited time before treatment initiation, and 2 highlighted insufficient data to support further service development. Despite variation, two-thirds of centres (15/25) reported active plans to establish or expand prehabilitation. Conclusions Access to neuro-oncology prehabilitation across the UK is inconsistent, with marked variation in eligibility, scope and workforce. Many services remain assessment-based or limited to specific patient groups. Data collection, workforce investment and clearer definitions may be required to ensure equitable and sustainable prehabilitation access.","author":[{"family":"Melhuish","given":"Sara"},{"family":"Peat","given":"Nicola"},{"family":"Wright","given":"Andrew"},{"family":"Goetz","given":"Camille"},{"family":"Huskens","given":"Nicky"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1093/neuonc/noag172.022","URL":"https://doi.org/10.1093/neuonc/noag172.022","source":"crossref"},{"id":"doi:10.1177/29498732261469489","type":"article-journal","title":"Neuro-LENS: A Neuro-Symbolic Framework Integrating Incomplete Background Knowledge and Deep Learning","abstract":"In this study, we propose Neuro-LENS, a neuro-symbolic evidence-based logic and symbolic reasoning framework that combines incomplete symbolic knowledge with neural learning to address ambiguity and improve the accuracy and interpretability of the results. We explore three strategies for integrating symbolic reasoning with deep learning and evaluate their effectiveness in practical settings: (i) applying the symbolic component to the neural output (neural-to-symbolic chaining); (ii) generating additional neural input features through symbolic rules (symbolic-to-neural chaining); (iii) creating an ensemble reasoning model (parallel neural-symbolic integration). The potential of the proposed Neuro-LENS framework is demonstrated on two real-world use cases: scene classification with abandoned object detection and prognostic health monitoring in vehicle failure prediction.","author":[{"family":"Murtas","given":"Giulia"},{"family":"Boeva","given":"Veselka"},{"family":"Tsiporkova","given":"Elena"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1177/29498732261469489","URL":"https://doi.org/10.1177/29498732261469489","source":"crossref"},{"id":"doi:10.1063/5.0306671","type":"article-journal","title":"Imitating neuro-plasticity in SrTiO3 based synaptic optoelectronic memristor for in-memory computing applications","abstract":"Recently, perovskite oxide-based synaptic optoelectronic memristors (SOMs) present a favorable path toward the advancement of neuromorphic (or in-memory) computing. In this study, we propose a SrTiO3 (STO) functional layer-based simple and transparent two-terminal memristor device stacked with ITO/STO/ITO/glass, which shows improved resistive switching behavior after post-oxide nitrogen annealing treatment at 300 °C. The annealed device shows more stable resistive switching characteristics compared to the unannealed device, with a low set voltage (+0.9 V) and an enhanced memory window (∼35). The presence of nitrogen in the STO layer assists in confining the conductive filament path, which helps to enhance the device's uniformity and stability. The device successfully imitates key biological synaptic behaviors, such as long-term potentiation/depression, paired pulse facilitaion (PPF), and spike timing-dependent plasticity. In addition, artificial neural network models with convolutional layers and vision transformer architectures are simulated for image classification tasks on the extended- and fashion-modified National Institute of Standards and Technology datasets with ∼88% and ∼83% accuracy, respectively. The same device was illuminated with violet light (wavelength: 405 nm) at 40 mW/cm2 and produced the excitatory postsynaptic current response that gradually decayed under a dark environment. The exposure conditions are adjusted to simulate short- to long-term memory transition, optical PPF, and image sharpening functions. Along with the learning phase, the impact of illumination is also analyzed on the forgetting (or memory) phase of the current response, which is further utilized for simulating the image memory function. These findings emphasize the remarkable potential of the STO-based SOM for use in in-memory computing applications.","author":[{"family":"Uong","given":"Phan"},{"family":"Limantoro","given":"Stephen"},{"family":"Shrivastava","given":"Saransh"},{"family":"Juliano","given":"Hans"},{"family":"Tseng","given":"Tseung"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1063/5.0306671","URL":"https://doi.org/10.1063/5.0306671","source":"crossref"},{"id":"doi:10.2118/224045-ms","type":"article-journal","title":"Advanced Wireline Conveyance Modeling Powered by Physics-Informed Neural Network","abstract":"Abstract Accurately estimating the interaction between tool string, cable, and wellbore is essential for wireline conveyance design and operational risk evaluation. Traditional physics-based models suffer from inaccuracies due to simplifications and uncertain parameters, such as friction coefficients. We propose an advanced wireline conveyance model that overcomes these limitations by integrating a neural network with a physics-based model to enhance prediction accuracy of various operational parameters, such as tensions. The solution optimizes wireline conveyance in both planning and real-time phases. The proposed model integrates a multilayer perceptron neural network with a physics-based model, forming a physics-informed neural network (PINN). This fusion of physical simulation and historical data improves prediction accuracy and tolerates input uncertainties. Inputs include cable, tool string, wellbore properties and geometry, along with time series data like cable speed, length, tractor force, and pumping rate. The model is currently tuned to predict surface tension, as it is the dominant indicator and consistently available in historical data. However, with minimal adjustments and additional data, it can be adapted to predict other parameters, such as head tension, open-hole sticking probability, conveyance time, and the effects of temperature and pressure gradients. We collected datasets of wireline conveyance jobs from different geological locations, spanning a wide range of configurations, covering various cables, tool strings, well geometries, winch speeds, and conveyance methods. Representative jobs were selected as validation sets to test the model's prediction accuracy, while the remaining jobs were used as references for manual parameter tuning and training the physics-based and PINN models. After the corresponding parameter tuning or training, both models were used to simulate jobs in the validation set, with errors quantified using root-mean-square error. Surface tension predictions by both models, along with their errors, were compared and analyzed. Results show that the PINN model outperformed the physics-based model in prediction accuracy, even when trained with data from a single trip. The accuracy further improved as more data were incorporated into the training process. These findings highlight the PINN model's robust capability to accurately predict surface tension, where the physics-based model may struggle, indicating its strong potential for broader applications in challenging environments. The proposed PINN wireline conveyance model innovatively simulates complex jobs by fusing data-driven and physical approaches, enhancing prediction accuracy. By relying on data, it reduces the dependency on uncertain inputs for the physics-based model. The improved accuracy offers a competitive advantage by enhancing design and execution, potentially leading to faster conveyance, less human interaction, and reduced anomalies with accurate risk estimation. Furthermore, the model's adaptability to various job configurations and its ability to simulate a range of conveyance scenarios enable broader application through precise tuning across diverse basins.","author":[{"family":"Wang","given":"J"},{"family":"Baklanov","given":"N"},{"family":"Durand","given":"B"}],"issued":{"date-parts":[[2025]]},"DOI":"10.2118/224045-ms","URL":"https://doi.org/10.2118/224045-ms","source":"crossref"},{"id":"doi:10.31705/eru.2025.18","type":"article-journal","title":"Convolutional neural network–based automated quality grading of cinnamon peel","abstract":"Ceylon cinnamon, endemic to Sri Lanka, holds significant culinary, medicinal, and economic value, reinforcing the country’s position as the leading global exporter of authentic cinnamon. However, current processing techniques, particularly manual peeling and quill preparation, are highly labor dependent and introduce quality inconsistencies, limiting scalability. While national efforts increasingly promote value addition and product diversification, maintaining consistent export-grade quality remains challenging. Recent research highlights the potential of technological advancement within the cinnamon value chain, with image processing applied for maturity estimation, and diameter-based grading [1], [2]. Additionally, deep learning models, particularly Convolutional Neural Networks (CNNs), have demonstrated strong performance in agricultural applications such as disease diagnosis and adulteration detection [3], [4]. Visual grading of cinnamon based on peel texture, surface defects and residual bark remains subjective and inconsistent. This study addresses this gap by proposing an objective, scalable deep learning–based automated visual inspection framework.","author":[{"family":"Darshani","given":"AA"},{"family":"Hewavitharana","given":"DC"},{"family":"Amarasinghe","given":"ADUS"},{"family":"Gamage","given":"JR"}],"issued":{"date-parts":[[2026]]},"DOI":"10.31705/eru.2025.18","URL":"https://doi.org/10.31705/eru.2025.18","source":"crossref"},{"id":"doi:10.5121/csit.2025.150907","type":"article-journal","title":"Improved Fire Recognition in VTOL UAVs through Convolutional Neural Network Algorithmss","abstract":"South Korea, with approximately 63% of its land covered by forests, is highly susceptible to wildfires. Traditional fire detection methods—such as satellite imagery and ground-based observation—face significant limitations, including high operational costs, delayed response times, and vulnerability to weather conditions. This paper presents an efficient fire detection system for Vertical Take-Off and Landing (VTOL) Unmanned Aerial Vehicles (UAVs), utilizing Convolutional Neural Networks (CNNs). The integration of CNNs significantly improves detection accuracy, even in complex environments that challenge conventional approaches. In simulations designed to closely mimic real-world scenarios, the optimized algorithm achieved a 93% detection rate with 20% false positives and a frame latency of just 1.2 seconds. Additionally, deploying the model on a Raspberry Pi onboard a VTOL drone demonstrated its practical viability for real-time forest fire surveillance and rapid response. This study highlights the potential of drone-based, AI-powered fire detection systems as a powerful supplement to existing wildfire monitoring and prevention strategies.","author":[{"family":"Lee","given":"Seo"},{"family":"Oh","given":"Jisoo"},{"family":"Choi","given":"Jennifer"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5121/csit.2025.150907","URL":"https://doi.org/10.5121/csit.2025.150907","source":"crossref"},{"id":"doi:10.4018/979-8-3373-0523-3.ch004","type":"article-journal","title":"Evolutionary Metaheuristics for Neural Network Optimization","abstract":"Neural networks have numerous hyperparameters, network designs, and weight combinations, making optimization challenging. Traditional methods like grid search and gradient descent often produce suboptimal solutions due to limited exploration. Genetic Algorithms (GAs), inspired by natural selection, provide an alternative by evolving populations across generations. Key GA operators include selection (e.g., Roulette Wheel, Tournament), crossover (e.g., Single-Point, Multi-Point), and mutation (e.g., Bit-Flip, Gaussian), which maintain diversity and avoid premature convergence. This chapter explores the use of GAs in neural network optimization, focusing on weight evolution, architectural search, and hyperparameter tuning. It also discusses GAs in medical image processing, improving tasks like disease detection, feature selection, and segmentation. GAs enhance diagnostic accuracy and treatment planning, especially in medical image analysis, offering effective optimization solutions in neural network training.","author":[{"family":"Tharanidharan","given":"Sridevi"},{"family":"Balaji","given":"Prasanalakshmi"},{"family":"Yue","given":"Gabriel"},{"family":"Devi","given":"Renuka"}],"issued":{"date-parts":[[2025]]},"DOI":"10.4018/979-8-3373-0523-3.ch004","URL":"https://doi.org/10.4018/979-8-3373-0523-3.ch004","source":"crossref"},{"id":"doi:10.1007/s11063-025-11740-2","type":"article-journal","title":"WACSO: Wolf Crow Search Optimizer for Convolutional Neural Network Hyperparameter Optimization","abstract":"Abstract Convolutional Neural Networks (CNNs) experience performance and training efficiency changes according to the selection of correct hyperparameters. The research presents WACSO which combines Crow Search Optimization with Grey Wolf Optimizer to improve Convolutional Neural Networks hyperparameter selection through a hybrid metaheuristic algorithm. The hybrid algorithm WACSO uses exploration parts from CSO together with GWO exploitation mechanics to obtain optimized performance. WACSO reaches higher classification accuracy than traditional optimization algorithms when performing tests on the MNIST and CIFAR-10 datasets along with Random Search and particle swarm optimization and genetic algorithms and standalone CSO and standalone GWO. The best classification results reached 98.9% accuracy levels on MNIST along with 91.5% accuracy levels on CIFAR-10. The final outcomes of this system depend on the combination of model structure along with dataset challenges and available computational power. The investigation demonstrates that mixing algorithms drawn from nature can lead to successful CNN hyperparameter optimization. The promising outcomes of WACSO depend on multiple variables including computation expenses and sensitive parameter adjustments and universal result adaptability between different datasets and network setups. Research into WACSO should expand to involve longer evaluations across multiple datasets and various models to confirm widespread usage.","author":[{"family":"Papalkar","given":"Rahul"},{"family":"Jadhav","given":"Jayendra"},{"family":"Pattewar","given":"Tareek"},{"family":"Thorat","given":"Vivek"},{"family":"Morey","given":"Pallavi"},{"family":"Deshmukh","given":"Mayur"},{"family":"Jagdale","given":"Rajkumar"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1007/s11063-025-11740-2","URL":"https://doi.org/10.1007/s11063-025-11740-2","source":"crossref"},{"id":"doi:10.29303/ijasds.v2i1.5855","type":"article-journal","title":"Peramalan Nilai Tukar Petani Kalimantan Timur Menggunakan Metode Neural Network","abstract":"Nilai tukar petani (NTP) merupakan indikator yang sangat penting untuk mengukur daya beli para petani Indonesia, yang dimana mereka merupakan pelaku utama dalam sektor pertanian. Hal tersebut dikarenakan sektor pertanian adalah salah satu sektor utama di Indonesia, salah satunya di Provinsi Kalimantan Timur. Penelitian ini bertujuan untuk memprediksi dan meramalkan NTP Provinsi Kalimantan Timur menggunakan metode Neural Network (NN) dengan algoritma backpropagation. Data yang digunakan adalah data NTP Provinsi Kalimantan Timur periode Januari 2020 sampai dengan September 2024 yang diperoleh dari BPS Provinsi Kalimantan Timur. Penelitian ini menguji 5 model arsitektur NN yang berbeda jumlah lapisan pada hidden layer, yaitu 1, 2, 3, 4, dan 5 lapisan pada hidden layer. Penelitian dilakukan dengan menggunakan 1 variabel input, learning rate sebesar 0,01, maksimal 10.000 iterasi, dan threshold sebesar 0,5, Berdasarkan proses pelatihan yang telah dilakukan, diperoleh kesimpulan bahwa arsitektur NN terbaik yang dapat digunakan untuk meramalkan NTP Provinsi Kalimantan Timur adalah NN dengan 5 lapisan pada hidden layer dengan MAPE sebesar 2,087%.","author":[{"family":"Rahmah","given":"Putri"},{"family":"Hayati","given":"Memi"},{"family":"Cahyaningsih","given":"Ariyanti"}],"issued":{"date-parts":[[2025]]},"DOI":"10.29303/ijasds.v2i1.5855","URL":"https://doi.org/10.29303/ijasds.v2i1.5855","source":"crossref"},{"id":"doi:10.4018/979-8-3693-7250-0.ch014","type":"article-journal","title":"Artificial Neural Network Predictions of Heat Transfer of Flat Plate Collector With Hybrid Nanofluids","abstract":"The Nusselt number, heat transfer, and friction factor of a flat plate collector operating with hybrid nanofluids of Al2O3-CuO and water are utilized in the Support Vector Regression method of artificial neural networks. The volume loadings used in the experiments were 0.048%, 0.096%, 0.144%, 0.192%, and 0.24%. The experiments ran from 09:00 to 16:30 hours. Time zone 1 (09:00 to 13:00 hr) and time zone 2 (13:00 to 16:30) are applied. Nusselt number is increased by 20.43% at time zone-1, at 13:00 hours, at 0.24% vol. and at a Reynolds number of 364.66, and by 14.08% at time zone-2, at 16:30 hours, at 0.24% vol. and at a Reynolds number of 211.23, respectively, above the base fluid. Similarly, the friction factor is increased by 15.34% and 11.50% above the base fluid for time zones 1 and 2, respectively, at 13:00 and 16:30, at 0.24% vol. and Reynolds numbers 364.66 and 211.23. The support vector regression algorithm in use makes accurate predictions about the values. Nusselt number, heat transfer, and friction factor have correlation values of 0.99497, 0.9947, and 0.9995.","author":[{"family":"Syam","given":"Lingala"},{"family":"Mesfin","given":"Solomon"},{"family":"Rao","given":"Veeredhi"}],"issued":{"date-parts":[[2025]]},"DOI":"10.4018/979-8-3693-7250-0.ch014","URL":"https://doi.org/10.4018/979-8-3693-7250-0.ch014","source":"crossref"},{"id":"doi:10.54914/jtt.v11i2.1900","type":"article-journal","title":"Deteksi Penyakit Kulit dengan Metode Convolutional Neural Network Menggunakan Arsitektur VGG19","abstract":"Deteksi dini penyakit kulit merupakan tantangan signifikan, khususnya di daerah dengan keterbatasan layanan medis dermatologis. Masalah ini diperparah oleh kurangnya tenaga medis spesialis serta maraknya informasi kesehatan yang tidak akurat di internet. Tujuan dari penelitian ini adalah mengembangkan sistem klasifikasi otomatis berbasis citra digital guna mengidentifikasi lima jenis penyakit kulit: Eczema, Melanocytic Nevus, Melanoma, Benign Keratosis, dan Basal Cell Carcinoma. Metode yang digunakan adalah Convolutional Neural Network (CNN) dengan arsitektur VGG19 melalui pendekatan transfer learning dan fine-tuning parsial pada layer block4_conv1. Dataset terdiri dari 10.000 gambar berformat .jpg yang telah melalui tahap normalisasi, augmentasi, deteksi tepi, dan penyeimbangan kelas. Evaluasi dilakukan menggunakan metrik akurasi, presisi, recall, F1-score, dan confusion matrix. Hasil eksperimen menunjukkan bahwa model mencapai akurasi hingga 84% pada skenario terbaik, dengan keseimbangan metrik lainnya yang menunjukkan kinerja klasifikasi multi-kelas yang andal. Kesimpulan dari penelitian ini menunjukkan bahwa arsitektur VGG19 efektif untuk mendeteksi berbagai penyakit kulit berbasis citra. Implikasi dari hasil ini membuka peluang pengembangan sistem deteksi awal berbasis aplikasi mobile, terutama untuk membantu masyarakat di daerah dengan keterbatasan layanan medis.","author":[{"family":"Rizqya","given":"Ainunnisa"},{"family":"Anggadimas","given":"Nanda"},{"family":"Misdram","given":"Muhammad"}],"issued":{"date-parts":[[2025]]},"DOI":"10.54914/jtt.v11i2.1900","URL":"https://doi.org/10.54914/jtt.v11i2.1900","source":"crossref"},{"id":"doi:10.51903/jgn2sd35","type":"article-journal","title":"IDENTIFICATION OF ROAD DAMAGE USING THE CONVOLUTIONAL NEURAL NETWORK (CNN) METHOD","abstract":"Kerusakan jalan merupakan permasalahan umum yang berdampak langsung terhadap keselamatan dan kenyamanan pengguna jalan, serta terhadap efisiensi transportasi. Selama ini, proses inspeksi jalan masih dilakukan secara manual, yang memerlukan waktu, biaya, dan sumber daya yang tidak sedikit. Penelitian ini bertujuan untuk mengembangkan sistem klasifikasi otomatis guna mendeteksi kerusakan jalan berdasarkan citra permukaan, dengan memanfaatkan metode Convolutional Neural Network (CNN). Dataset yang digunakan dalam penelitian ini terdiri dari 400 gambar dengan distribusi seimbang, yaitu 200 gambar kategori Cracks (jalan retak) dan 200 gambar kategori non-Cracks (jalan tidak retak), yang diambil dari sumber dataset terbuka di platform Mendeley Data. Arsitektur CNN dirancang secara khusus dengan empat lapisan konvolusi yang dilengkapi fungsi aktivasi ReLU, pooling layer, dropout layer untuk mengurangi overfitting, serta fully connected layer pada tahap akhir klasifikasi. Proses pelatihan dilakukan menggunakan TensorFlow dan Keras di platform Google Colab, dengan pembagian data sebesar 80% untuk data pelatihan dan 20% untuk data validasi. Hasil evaluasi menunjukkan bahwa model memiliki performa klasifikasi yang sangat baik. Berdasarkan rata-rata dari seluruh skenario pelatihan (epoch 30, 40, dan 50), model CNN yang dikembangkan mampu mencapai akurasi keseluruhan sebesar 97,50%, presisi rata-rata 96,77%, recall rata-rata 99,37%, dan F1-score rata-rata 97,72%. Dengan kinerja yang konsisten dan tingkat kesalahan yang rendah, model CNN ini memiliki potensi besar untuk diterapkan sebagai alat bantu dalam proses identifikasi kerusakan jalan berbasis citra secara otomatis dan efisien, sehingga dapat mempercepat inspeksi, mengurangi beban kerja manual, dan membantu instansi terkait dalam pengambilan keputusan pemeliharaan infrastruktur jalan secara tepat waktu.","author":[{"family":"Zuhri","given":"Ahmad"},{"family":"Nafiiyah","given":"Nur"},{"family":"Budi","given":"Agus"}],"issued":{"date-parts":[[2025]]},"DOI":"10.51903/jgn2sd35","URL":"https://doi.org/10.51903/jgn2sd35","source":"crossref"},{"id":"doi:10.1063/5.0312026","type":"article-journal","title":"Dual-function Sb2S3/HfO2 memristor for reservoir computing and neural network learning via decoupled short- and long-term memory","abstract":"Conventional computing architectures typically rely on separate devices to achieve dynamic sensing and long-term storage, leading to low integration density, high energy consumption, and significant data movement bottlenecks. Here, a biomimetic dual-function memristor based on an Sb2S3/HfO2 heterostructure is proposed, in which synergistic regulation of ion migration and electronic transport enables the materials-assisted decoupling and coordinated integration of short-term memory (STM) and long-term memory (LTM) functions within a single device. The device successfully emulates various biological synaptic behaviors, including paired-pulse facilitation/depression, tunable excitatory postsynaptic currents (EPSCs), and highly linear long-term potentiation/depression. Subsequently, utilizing the LTM characteristics of the device, a nonvolatile synaptic array is built to implement a fully connected neural network, achieving 94.5% accuracy. Meanwhile, a physical reservoir computing system is constructed using the STM dynamics to directly encode and recognize spatiotemporal features in iris image sequences, achieving 98% accuracy. Through coordinated innovation in materials, devices, and architecture, this work advances memristors from single-function memory elements toward multifunctional, all-electrical intelligent processing units.","author":[{"family":"Song","given":"Mengru"},{"family":"Li","given":"Lele"},{"family":"Gu","given":"Han"},{"family":"Hu","given":"Ziyang"},{"family":"Lu","given":"Yegang"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1063/5.0312026","URL":"https://doi.org/10.1063/5.0312026","source":"crossref"},{"id":"doi:10.4103/abr.abr_407_23","type":"article-journal","title":"Predicting Pregnant Women’s Abortion: Artificial Neural Network, Wavelet Neural Network and Adaptive Neural Fuzzy Inference System","abstract":"Background: Abortion is an important and controversial issue and one of the important reasons for the mortality of pregnant women worldwide. This study aimed to predict the risk factors of abortion in pregnant women using artificial neural network, wavelet neural network, and adaptive neural fuzzy inference system. Materials and Methods: The study is an analytical-comparative modeling and data of 4437 pregnant women from the Ravansar Non-Communicable Disease (RaNCD) cohort study from 2014 to 2016 was used. First, six variables were chosen through the genetic algorithm approach, then artificial neural network (ANN), wavelet neural network (WNN), and adaptive neural fuzzy inference system (ANFIS) were run. Finally, the performance of the models was compared based on the evaluation criteria. All analyses were done in MATLAB R2019b software. Results: ANN with RMSE of 0.019 showed better performance than ANFIS and WNN with 0.42 and 1.445, respectively. Further, the accuracy, sensitivity, and specificity in ANN were 100%, 99%, and 100%, while in WNN, they were 76.2%, 76.4%, and 66.7%. However, when the researchers used three selected variables, the accuracy, sensitivity, and specificity as well as RMSE in ANFIS were 100%, 100% 100%, and 0; 100%, 99%, 100%, and 0.021 in ANN; and finally 76.2%, 76.4%, 38.5%, and 1.553 in WNN. Conclusion: The models with six input variables indicated that the artificial neural network has a better performance than the other two models, but based on the three variables, the fuzzy neural inference system performed better than the other two models.","author":[{"family":"Hashemian","given":"Amir"},{"family":"Mahaki","given":"Behzad"},{"family":"Rezaei","given":"Mansour"},{"family":"Solouki","given":"Leila"},{"family":"Cheqabaleki","given":"Mohammad"},{"family":"Cheqabaleki","given":"Somayeh"}],"issued":{"date-parts":[[2025]]},"DOI":"10.4103/abr.abr_407_23","URL":"https://doi.org/10.4103/abr.abr_407_23","source":"crossref"},{"id":"doi:10.36499/psnst.v14i1.12035","type":"article-journal","title":"KLASIFIKASI GENDER BERDASARKAN CITRA MATA MANUSIA MENGGUNAKAN ALGORITMA CONVOLUTION NEURAL NETWORK","abstract":"Teknologi otentikasi biometrik yang memanfaatkan karakteristik manusia seperti wajah, sidik jari, suara, dan iris mata semakin banyak digunakan untuk seperti wajah, sidik jari, suara, dan iris mata semakin banyak digunakan untuk identifikasi individu. Meskipun efektif dalam memastikan keaslian, sistem-sistem ini umumnya tidak memberikan informasi tambahan seperti jenis kelamin atau etnis dari individu yang diverifikasi. Penelitian sebelumnya telah meneliti klasifikasi jenis kelamin berdasarkan citra wajah, sedangkan penggunaan citra iris mata masih terbatas. Penelitian ini bertujuan untuk mengklasifikasikan jenis kelamin manusia berdasarkan citra iris mata dengan menerapkan pendekatan deep learning, khususnya menggunakan algoritma Convolutional Neural Network (CNN). Pengujian dilakukan pada sejumlah besar data untuk mengukur kinerja. Hasil penelitian menunjukkan bahwa model CNN mampu mencapai akurasi hingga 92% dalam mengklasifikasikan jenis kelamin berdasarkan citra iris mata. Penelitian ini membuka peluang baru untuk pengembangan lebih lanjut dalam biometrik berbasis iris mata.","author":[{"family":"Wicaksono","given":"Rizky"},{"family":"Pratama","given":"Fandy"},{"family":"Budianita","given":"Avira"}],"issued":{"date-parts":[[2026]]},"DOI":"10.36499/psnst.v14i1.12035","URL":"https://doi.org/10.36499/psnst.v14i1.12035","source":"crossref"},{"id":"doi:10.15388/namc.2026.31.45269","type":"article-journal","title":"Finite-time matrix projective synchronization of fractional-order memristor-based delayed neural networks with parameter uncertainty","abstract":"This research examines the phenomenon of finite-time matrix projective synchronization (FTMPS) in the context of two distinct fractional-order memristor-based delayed neural networks (FOMDNNs). For the FOMDNNs with indeterminate parameters, some suitable controllers are structured, and sufficient conditions for implementing the FTMPS are demonstrated through some inequality techniques and relevant lemmas pertaining to fractional calculus. The synchronization issues under two different norm cases are fully considered. Subsequently, two numerical examples of the FTMPS are revealed and the discrepancies in their synchronization gradually tend to zero, which shows the validity and accuracy of the obtained synchronization results.","author":[{"family":"Xu","given":"Shangbin"},{"family":"Hu","given":"Bo"},{"family":"Zhang","given":"Hai"},{"family":"Chen","given":"Xinbin"},{"family":"Cao","given":"Jinde"}],"issued":{"date-parts":[[2026]]},"DOI":"10.15388/namc.2026.31.45269","URL":"https://doi.org/10.15388/namc.2026.31.45269","source":"crossref"},{"id":"doi:10.1002/asjc.70135","type":"article-journal","title":"Predefined‐time synchronization of memristor‐based bidirectional associative memory neural networks with time‐varying delays","abstract":"Abstract In this paper, two novel and general predefined‐time stability (PTSt) lemmas are introduced and applied to address the predefined‐time synchronization (PTSy) problem in memristor‐based bidirectional associative memory (BAM) neural networks. Unlike traditional finite‐time synchronization (FinTS) or fixed‐time synchronization (FixTS) approaches, the synchronization time in this study is independent of the initial states (InSts) and system parameters. It can be predetermined according to task requirements without any estimation. Two effective and innovative controllers are designed to achieve synchronization in time‐delayed memristor‐based BAM neural networks (MBAMNNs) based on PTSt. The proposed controllers employ general PTSy analysis to ensure robust synchronization performance. These diverse predefined‐time controllers (PTCos) design facilitates the practical implementation of synchronization control in neural networks. By employing the fraction‐order power function and exponential function, the PTCos offer flexibility and adaptability in controller design. The results are validated through two numerical examples using MATLAB simulations.","author":[{"family":"Chen","given":"Yan"},{"family":"Muniyandi","given":"Ravie"},{"family":"Sahran","given":"Shahnorbanun"},{"family":"Cai","given":"Zuowei"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1002/asjc.70135","URL":"https://doi.org/10.1002/asjc.70135","source":"crossref"},{"id":"doi:10.7554/elife.106871.1","type":"article-journal","title":"Neural signatures of motor memories emerge in neural network models","abstract":"Abstract Animals can learn and seamlessly perform a great number of behaviors. However, it is unclear how neural activity can accommodate new behaviors without interfering with those an animal has already acquired. Recent studies in monkeys performing motor and brain-computer interface (BCI) learning tasks have identified neural signatures—so-called “memory traces” and “uniform shifts”—that appear in the neural activity of a familiar task after learning a new task. Here we asked when these signatures arise and how they are related to continual learning. By modeling a BCI learning paradigm, we show that both signatures emerge naturally as a consequence of learning, without requiring a specific mechanism. In general, memory traces and uniform shifts reflected savings by capturing how information from different tasks coexisted in the same neural activity patterns. Yet, although the properties of these two different signatures were both indicative of savings, they were uncorrelated with each other. When we added contextual inputs that separated the activity for the different tasks, these signatures decreased even when savings were maintained, demonstrating the challenges of defining a clear relationship between neural activity changes and continual learning.","author":[{"family":"Chang","given":"Joanna"},{"family":"Clopath","given":"Claudia"},{"family":"Gallego","given":"Juan"}],"issued":{"date-parts":[[2025]]},"DOI":"10.7554/elife.106871.1","URL":"https://doi.org/10.7554/elife.106871.1","source":"crossref"},{"id":"doi:10.1101/2025.06.20.660661","type":"article-journal","title":"Learning, sleep replay and consolidation of contextual fear memories: A neural network model","abstract":"Abstract Contextual fear conditioning is an experimental framework widely used to investigate how aversive experiences affect the valence an animal associates with an environment. While the initial formation of associative context-fear memories is well studied – dependent on plasticity in hippocampus and amygdala – the neural mechanisms underlying their subsequent consolidation remain less understood. Recent evidence suggests that the recall of contextual fear memories shifts from hippocampal-amygdalar to amygdalo-cortical networks as they age. This transition is thought to rely on sleep. In particular, neural replay during hippocampal sharp-wave ripple events seems crucial, though open questions regarding the involved neural interactions remain. Here, we propose a biologically informed neural network model of context-fear learning. It expands the scope of previous models through the addition of a sleep phase. Hippocampal representations of context, formed during wakefulness, are replayed in conjunction with cortical and amygdalar activity patterns to establish long-term encodings of learned fear associations. Additionally, valence-coding synapses within the amygdala undergo overnight adjustments consistent with the synaptic homeostasis hypothesis of sleep. The model reproduces experimentally observed phenomena, including context-dependent fear renewal and time-dependent increases in fear generalisation. Few neural network models have addressed fear memory consolidation and to our knowledge, ours is the first to incorporate a neural mechanism enabling it. Our framework yields testable predictions about how disruptions in synaptic homeostasis may lead to pathological fear sensitization and generalisation, thus potentially bridging computational models of fear learning and mechanisms underlying anxiety symptoms in disorders such as PTSD. Author Summary How do we learn to fear certain environments? Why do some fear memories fade while others persist or even grow stronger over time? Scientists have long used laboratory experiments to study how animals associate danger with a particular context. These studies have helped identify brain regions involved in fear learning, including the amygdala, hippocampus, and cortex, and have inspired many computational models of how fear is acquired in the brain. However, most models focus only on what happens when fear is first learned, overlooking how these memories evolve in the days that follow and the role of sleep in this process. In this work, we present a neural network model that captures how fear memories are strengthened or reshaped during sleep. It builds on earlier models by incorporating memory replay and synaptic homeostasis, two brain processes believed to support emotional memory consolidation. Our model identifies neural processes that help make fear memories persistent, suggests that sleep is necessary to maintain adaptive behaviour after threatening experiences, and proposes that sleep disruptions mediate the harmful impact of stress on emotional regulation. By extending amygdala-based models of fear learning to include post-learning dynamics, our work offers new insight into how emotional memories are stabilised.","author":[{"family":"Werne","given":"Lars"},{"family":"Chadwick","given":"Angus"},{"family":"Seriès","given":"Peggy"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1101/2025.06.20.660661","URL":"https://doi.org/10.1101/2025.06.20.660661","source":"crossref"},{"id":"doi:10.20944/preprints202512.0715.v1","type":"manuscript","title":"Wind Power Forecast Using Multilevel Adaptive Graph Convolution Neural Network","abstract":"Accurate forecasting of wind power is essential for maintaining the stability and efficiency of power networks as renewable energy sources become more integrated. This study proposes a multilevel spatial-temporal graph convolution network (MLAGCN) for wind power forecasting. The framework combines a multilevel adaptive graph convolution (MLAGC) and a lightweight temporal transformer (LWTT) to jointly model complex spatial-temporal relationships in wind power data. MLAGC is constructed using three adaptive graphs: a local-aware graph, a global-aware graph, and a structure-aware graph. These components form a flexible graph structure that effectively represents dynamic spatial interactions while LWTT learns short- and long-term sequential patterns. Experiments on real wind farm datasets demonstrated that the proposed model outperforms existing baselines. The model achieved an improved prediction accuracy and generalization, as indicated by a lower score of 43.44, mean absolute error (38.83), root mean square error (48.05) and a forecast loss of 0.22. These results demonstrates the effectiveness of temporal modeling and multilevel attention-based adaptive graph learning for high-resolution wind power forecasting.","author":[{"family":"Duntoye","given":"Oluwaseun"},{"family":"Alowonou","given":"Kowovi"},{"family":"Kwon","given":"Do"}],"issued":{"date-parts":[[2025]]},"DOI":"10.20944/preprints202512.0715.v1","URL":"https://doi.org/10.20944/preprints202512.0715.v1","source":"crossref"},{"id":"doi:10.36227/techrxiv.174114593.34611655/v1","type":"article-journal","title":"Fast Acquisition of Sensor Array Geometry of Whole-head Magnetoencephalograph Systems Using Neural Network","abstract":"Acquiring position, orientation, and sensitivity of magnetometers in a helmet-shaped sensor array is crucial for accurate current source reconstruction in magnetoencephalography. To determine these parameters for each magnetometer, we utilize a spherical calibration coil array. In our previous study, the position and orientation of each magnetometer were determined as the solution of an inverse problem through a numerical search that minimized the difference between the theoretical magnetic field signals from each coil and the measured signals detected by the magnetometer. In this study, we applied a deep neural network to estimate the position and orientation of each magnetometer in the helmet-shaped sensor array without solving inverse problem. A total of 223 million pairs of a given magnetometer's five parameters (x, y, z, θ, and φ) and the corresponding theoretical magnetic field signals from the coils were used to train the neural network. The training process required approximately 53 hours using a commercially available GPU-equipped computer. The trained neural network was then applied to acquire the sensor geometry from magnetic field data obtained during the conventional calibration procedure for a 160-channel whole-head magnetoencephalograph system using a spherical calibration coil array. The position and orientation of each magnetometer estimated by this method deviated by an average of 0.65 mm and 0.51 degree, respectively, from those obtained via the conventional inverse problem approach. The acquisition of the geometry for all 160 magnetometers required less than 8 ms. With such high-speed acquisition, this approach opens possibilities for future applications in acquiring positional information of wearable sensor arrays whose structures change in real-time.","author":[{"family":"Adachi","given":"Yoshiaki"},{"family":"Oyama","given":"Daisuke"},{"family":"Uehara","given":"Gen"}],"issued":{"date-parts":[[2025]]},"DOI":"10.36227/techrxiv.174114593.34611655/v1","URL":"https://doi.org/10.36227/techrxiv.174114593.34611655/v1","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-6581715/v1","type":"article-journal","title":"DiffNet-A Diffusion Convolutional Neural Network for Classification of Epileptic Seizure","abstract":"Abstract Epileptic seizure detection using electroencephalogram (EEG) signals remains a challenging problem in neuroscience and biomedical engineering. In this study, we propose DiffNet, a novel Diffusion Convolutional Neural Network (DCNN) designed for accurate and automated classification of epileptic seizures. DiffNet combines spatial and temporal feature extraction capabilities, leveraging graph-based diffusion processes and convolutional layers to enhance classification performance. The model was evaluated on multiple benchmark EEG datasets, including Bonn, ResearchGate, and Mendeley Data, achieving an average accuracy of 99.7% during training and 98.3% during validation. Additionally, DiffNet outperformed existing state-of-theart algorithms across various statistical metrics such as precision (97.2%), sensitivity (96.8%), and F1-score (97.5%). These results highlight the robustness and reliability of DiffNet in addressing variability in EEG signal patterns. The proposed architecture also demonstrates reduced computational complexity, making it suitable for real-time seizure detection applications.","author":[{"family":"Shaw","given":"Laxmi"},{"family":"Ajitha","given":"D"},{"family":"Induvasi","given":"Sai"}],"issued":{"date-parts":[[2025]]},"DOI":"10.21203/rs.3.rs-6581715/v1","URL":"https://doi.org/10.21203/rs.3.rs-6581715/v1","source":"crossref"},{"id":"doi:10.1364/opticaopen.28902587","type":"article-journal","title":"Q-factor Analysis in Free Space Optical Communication and Neural Network-Based Prediction","abstract":"Free Space Optics (FSO) provides a promising alternative where fiber-optic deployment is impractical due to cost or fragility. However, FSO performance is highly vulnerable to atmospheric disturbances such as fog, rain, and dust, which can significantly degrade signal quality. To optimize system performance under varying conditions, it is crucial to understand how the Q-factor responds to changes in system parameters. This study investigates the effects of bit rate, filter type, transmitter and receiver aperture diameters, and transmission range on the Q-factor in FSO systems. We developed a detailed simulation model using OptiSystem to generate data, which was then used to train a feedforward neural network via MATLAB’s Neural Network Tool (NN-Tool) using the Levenberg–Marquardt algorithm. This model effectively captures complex, nonlinear relationships between input parameters and Q-factor outcomes, allowing accurate predictions without further simulations. The hybrid approach of combining simulation data with neural network-based modeling offers a practical and user-friendly tool for performance prediction and system planning. This research contributes to the design and optimization of high-data-rate FSO systems by addressing existing limitations in modeling and parameter tuning.","author":[{"family":"Sikder","given":"Mohammad"},{"family":"Sakib","given":"Fahim"},{"family":"Hossen","given":"Md"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1364/opticaopen.28902587","URL":"https://doi.org/10.1364/opticaopen.28902587","source":"crossref"},{"id":"doi:10.2139/ssrn.5048252","type":"manuscript","title":"Advancing Autonomous Vehicle Navigation through Hybrid Fuzzy-Neural Network Training Systems","abstract":"In the evolving realm of autonomous vehicle navigation, the integration of fuzzy logic and neural networks presents a formidable challenge, particularly in the context of real-time, on-the-fly neural network training. This paper addresses the gap in dynamic and adaptable training methods necessary for navigating unpredictable environments with limited computational resources. The primary objective of our study is to empirically validate a hybrid training approach that combines&amp;nbsp;fuzzy logic with back-propagation learning algorithms, aiming to optimize neural network performance under hardware constraints. Our methodology leverages a fuzzy logic trainer to provide initial training sets dynamically, which guide the neural network in adjusting its weights in real time, thus facilitating adaptive learning during navigation tasks. The findings reveal that this integrated approach not only enhances the learning efficiency of neural networks but also significantly improves navigation accuracy in real-time scenarios. These advancements contribute to the field by demonstrating the feasibility of deploying more adaptable and robust autonomous navigation systems, potentially expanding their application in more diverse and challenging environments.","author":[{"family":"Alzaydi","given":"Ammar"},{"family":"Abedalrhman","given":"Kahtan"},{"family":"Alotaibi","given":"Ibrahim"},{"family":"Alessa","given":"Fahad"}],"issued":{"date-parts":[[2025]]},"DOI":"10.2139/ssrn.5048252","URL":"https://doi.org/10.2139/ssrn.5048252","source":"crossref"},{"id":"doi:10.1101/2025.02.03.25321462","type":"article-journal","title":"Miniaturization of Epileptic Abnormal Electrocorticogram Detector Using 3D Convolutional Neural Network","abstract":"Abstract Epilepsy is a neurological disorder characterized by sudden and recurrent seizures caused by abnormal electrical activity in the brain. Responsive Neurostimulation (RNS) offers a promising treatment option for patients with drug-resistant epilepsy. Responsive Neurostimulation (RNS) is an implantable device that employs a closed-loop system. It continuously monitors brain activity through electro-corticogram (ECoG) recordings. When the system detects seizure activity, it delivers direct electrical stimulation to the brain to suppress the seizure. Seizure detection algorithms require patient-specific optimization, leading to increased interest in deep learning approaches in recent years. While deeper network architectures generally improve detection accuracy, their implementation in implantable devices is constrained by limited hardware resources and the restricted number of electrode channels available for ECoG monitoring. To ensure the practical feasibility of RNS, it is crucial to systematically minimize both the computational costs of patient-specific deep learning models and the number of connected ECoG electrodes. This study systematically reduced the number of electrode channels and computational costs in seizure detection models by analyzing the spatiotemporal kernels learned by the first convolutional layer of a 3D convolutional neural network (3D CNN) trained on 3D ECoG data. This approach capitalizes on the network’s ability to learn spatial relationships between grid electrodes and the temporal dynamics of ECoG signals. The performance comparison between the downsized seizure detection CNN model and the original CNN model revealed that, for at least some patients, it is possible to maintain inference performance while reducing the model size.","author":[{"family":"Yamaji","given":"Moemi"},{"family":"Yamamasu","given":"Shinjiro"},{"family":"Hirano","given":"Yuto"},{"family":"Hayashida","given":"Yuki"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1101/2025.02.03.25321462","URL":"https://doi.org/10.1101/2025.02.03.25321462","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-6787728/v1","type":"article-journal","title":"Forecasting the residual stress components in wires using an artificial neural network","abstract":"Abstract In this paper, a computer simulation of round wire drawing processes with different equations of state for steel A12 has been carried out. In addition, the methods of improving the configuration of neural networks based on multilayer perceptron (MLP) for estimating the distributions of residual stress tensor components have been investigated. The study demonstrated that the strain rate exerted a significant influence on the character of the processes, particularly within the central region (0r–0.4r) of the investigated specimens. In addition, the employment of software tools for the purpose of tuning the hyperparameters of trained machine learning models, including Optuna, BayesianOpt, and Skopt, has been demonstrated to enhance the predictive capability of the models. Consequently, this results in an improvement in the accuracy of the obtained distributions of the required characteristics.","author":[{"family":"Demin","given":"Dmitriy"},{"family":"Grebenkin","given":"Ilya"},{"family":"Barinov","given":"Alexey"}],"issued":{"date-parts":[[2025]]},"DOI":"10.21203/rs.3.rs-6787728/v1","URL":"https://doi.org/10.21203/rs.3.rs-6787728/v1","source":"crossref"},{"id":"doi:10.7554/elife.106871","type":"article-journal","title":"Neural signatures of motor memories emerge in neural network models","abstract":"Abstract Animals can learn and seamlessly perform a great number of behaviors. However, it is unclear how neural activity can accommodate new behaviors without interfering with those an animal has already acquired. Recent studies in monkeys performing motor and brain-computer interface (BCI) learning tasks have identified neural signatures—so-called “memory traces” and “uniform shifts”—that appear in the neural activity of a familiar task after learning a new task. Here we asked when these signatures arise and how they are related to continual learning. By modeling a BCI learning paradigm, we show that both signatures emerge naturally as a consequence of learning, without requiring a specific mechanism. In general, memory traces and uniform shifts reflected savings by capturing how information from different tasks coexisted in the same neural activity patterns. Yet, although the properties of these two different signatures were both indicative of savings, they were uncorrelated with each other. When we added contextual inputs that separated the activity for the different tasks, these signatures decreased even when savings were maintained, demonstrating the challenges of defining a clear relationship between neural activity changes and continual learning.","author":[{"family":"Chang","given":"Joanna"},{"family":"Clopath","given":"Claudia"},{"family":"Gallego","given":"Juan"}],"issued":{"date-parts":[[2025]]},"DOI":"10.7554/elife.106871","URL":"https://doi.org/10.7554/elife.106871","source":"crossref"},{"id":"doi:10.48185/jfcns.v7i1.1816","type":"article-journal","title":"Analysis of Stability in Rulkov Neural Networks with Fractional Orders and Asymmetric Memristor Synapses","abstract":"Fractional-order models effectively capture memory and hereditary effects in neural and nonlinear dy namical systems. Memristors are ideal components for modeling synaptic connections due to their ability to emulate plasticity and memory effects. Discrete models of memristor-coupled neurons simplify computa tions and enable efficient analysis of large-scale networks. Despite their potential, discrete fractional-order memristor-coupled models have been less explored. To address this, we propose two novel discrete fractional order neural systems. The first system is a two-neuron motif coupled via dual memristors, while the second extends this configuration to a ring-shaped network of similar subnetworks. A new theorem on the stabil ity of discrete fractional-order systems is established, defining stability regions for both models. Numerical simulations illustrate the theoretical results and investigate how the fractional order, asymmetric memristive coupling, and other model and network parameters jointly influence the dynamics and stability of discrete time neurons.","author":[{"family":"Eftekhari","given":"Leila"},{"family":"Khalighi","given":"Moein"},{"family":"Abbasbandy","given":"Saeid"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48185/jfcns.v7i1.1816","URL":"https://doi.org/10.48185/jfcns.v7i1.1816","source":"crossref"},{"id":"doi:10.1088/1674-1056/adb8bb","type":"article-journal","title":"Resonant tunneling diode cellular neural network with memristor coupling and its application in police forensic digital image protection","abstract":"Abstract Due to their biological interpretability, memristors are widely used to simulate synapses between artificial neural networks. As a type of neural network whose dynamic behavior can be explained, the coupling of resonant tunneling diode-based cellular neural networks (RTD-CNNs) with memristors has rarely been reported in the literature. Therefore, this paper designs a coupled RTD-CNN model with memristors (RTD-MCNN), investigating and analyzing the dynamic behavior of the RTD-MCNN. Based on this model, a simple encryption scheme for the protection of digital images in police forensic applications is proposed. The results show that the RTD-MCNN can have two positive Lyapunov exponents, and its output is influenced by the initial values, exhibiting multistability. Furthermore, a set of amplitudes in its output sequence is affected by the internal parameters of the memristor, leading to nonlinear variations. Undoubtedly, the rich dynamic behaviors described above make the RTD-MCNN highly suitable for the design of chaos-based encryption schemes in the field of privacy protection. Encryption tests and security analyses validate the effectiveness of this scheme.","author":[],"issued":{"date-parts":[[2025]]},"DOI":"10.1088/1674-1056/adb8bb","URL":"https://doi.org/10.1088/1674-1056/adb8bb","source":"crossref"},{"id":"doi:10.26434/chemrxiv-2025-hpcws","type":"manuscript","title":"Architecture independent absolute solvation free energy calculations with neural network potentials","abstract":"Allowing atoms or molecules to disappear is a critical step in alchemical free energy simulations (FES). The necessary tricks are well understood when using force fields. Over the past few years, neural network potentials (NNPs) have seen rapid development. Their potentially higher accuracy compared to force fields makes them attractive for use in FES. Here, we outline a method for gradually decoupling atoms and molecules in systems that are fully described by NNPs. Specifically, we show that manipulating the neighbor list is equivalent to using soft-core potentials in force-field-based FES. Since constructing the neighbor list is a central step, regardless of the NNP's inner workings, our approach is agnostic to NNP architecture. We validate the correctness of our methodology by demonstrating cycle closure for a model problem and report solvation free energies obtained with the MACE-OFF23(S/M) NNP.","author":[{"family":"Picha","given":"Anna"},{"family":"Tkaczyk","given":"Sara"},{"family":"Wieder","given":"Marcus"},{"family":"Boresch","given":"Stefan"}],"issued":{"date-parts":[[2025]]},"DOI":"10.26434/chemrxiv-2025-hpcws","URL":"https://doi.org/10.26434/chemrxiv-2025-hpcws","source":"crossref"},{"id":"doi:10.1101/2025.10.21.25338451","type":"article-journal","title":"ALTARN: A Tabular Residual Neural Network for Alzheimer’s Disease Classification and Prediction","abstract":"Abstract Early and accurate prediction of Alzheimer’s disease (AD) from accessible clinical data remains a significant challenge in healthcare. This study proposes ALTARN, a tabular attention residual neural network architecture for robust classification of AD through heterogeneous patient data from a publicly available dataset of 2149 subjects, with medical, demographic, and lifestyle variables organized as structured tabular data. With sigmoid attention mechanisms for the dynamic reweighing of input variables for each patient, deep residual connections capturing complex, and non-linear relationships in tabular features, we propose ALTARN-an early Alzheimer prediction tool. ALTARN achieved an average cross-validated training accuracy of roughly 92.73%, alongside robust validation metrics, including an average accuracy of 85.06%, average precision of 80.73%, mean recall of 76.18%, and a mean validation F1-score of 78.32% when evaluated through five-fold stratified validation. When tested against other approaches, ALTARN meets and also exceeds the performance of supervised deep learning models for non-imaging AD classification. This further illustrates that deep neural network based approaches with tabular attention offer a promising direction for interpretable diagnosis of AD via non-imaging medical data.","author":[{"family":"Balakrishna","given":"Keshav"},{"family":"Hammond","given":"Alessandro"},{"family":"Idrissi","given":"Abdeslem"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1101/2025.10.21.25338451","URL":"https://doi.org/10.1101/2025.10.21.25338451","source":"crossref"},{"id":"doi:10.20944/preprints202506.2227.v1","type":"manuscript","title":"Neural Network-Informed Lotka-Volterra Dynamics for Cryptocurrency Market Analysis","abstract":"Mathematical modeling plays a crucial role in supporting decision-making across a wide range of scientific disciplines. These models often involve multiple parameters, the estimation of which is critical to assessing their reliability and predictive power. Recent advancements in artificial intelligence have made it possible to efficiently estimate such parameters with high accuracy. In this study, we focus on modeling the dynamics of cryptocurrency market shares by employing a Lotka-Volterra system. We introduce a methodology based on a deep neural network (DNN) to estimate the parameters of the Lotka-Volterra model, which are subsequently used to numerically solve the system using a fourth-order Runge-Kutta method. The proposed approach, when applied to real-world market share data for Bitcoin, Ethereum, and alternative cryptocurrencies, demonstrates excellent alignment with empirical observations. Moreover, our method outperforms ARIMA models in terms of accuracy, showcasing its effectiveness for crypto market forecasting. The entire framework, including neural network training and Runge-Kutta integration, was implemented in MATLAB.","author":[{"family":"Kastoris","given":"Dimitris"},{"family":"Papadopoulos","given":"Dimitris"},{"family":"Giotopoulos","given":"Konstantinos"}],"issued":{"date-parts":[[2025]]},"DOI":"10.20944/preprints202506.2227.v1","URL":"https://doi.org/10.20944/preprints202506.2227.v1","source":"crossref"},{"id":"doi:10.36227/techrxiv.176291893.35097559/v1","type":"article-journal","title":"Concurrent Generation of RSMTs for Multiple Nets using Graph Neural Network","abstract":"A Rectilinear Steiner Minimum Tree (RSMT) is the shortest (in Manhattan distance) tree to interconnect a given set of 2D points called pins of a net, formed by joining only vertical and horizontal line (wire) segments through the pins and a few extra points called Steiner points. It is used as an initial skeleton route in many state-of-the-art global routers. However, finding an optimal RSMT is NP-hard. Existing techniques either generate optimal RSMTs in exponential/sub-exponential time or near-optimal RSMTs using heuristics or machine learning in polynomial time. However, each of these methods considers the construction of an RSMT of a single net at a time. The concurrent construction of RSMTs for multiple nets has the potential to reduce the overall construction time and edge-overlap mitigation of multiple nets. In this work, we propose a method for concurrent construction of RSMTs of multiple nets based on Graph Neural Network (GNN) with the aim of minimizing the overall construction time, group-wise edge-overlap and wirelength. Experimental results show that our model can generate RSMTs of K nets concurrently with significant reduction in construction time, group-wise edge-overlapping and wirelength as compared with multiple baseline methods.","author":[{"family":"Saha","given":"Kritanta"},{"family":"Banik","given":"Mrinmoy"},{"family":"Banerjee","given":"Pritha"},{"family":"Sur-Kolay","given":"Susmita"}],"issued":{"date-parts":[[2025]]},"DOI":"10.36227/techrxiv.176291893.35097559/v1","URL":"https://doi.org/10.36227/techrxiv.176291893.35097559/v1","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-7088954/v1","type":"article-journal","title":"A Novel RBF Neural Network-based Hybrid Technique and Its Applications","abstract":"Abstract Recent studies have demonstrated that Radial Basis Function Neural Networks (RBFNNs), based on traditional methods like gradient descent, frequently produce local approximates after training. This motivation led us to propose an RBFNN-based novel hybrid algorithm that maintains the local and global approximation properties throughout the training process and maintains the balance between these approximation properties. This study introduces the RBFNN-based hybrid particle swarm optimization and cuckoo search with biogeography-based optimization (PSO-CS-BBO) algorithm for global exploration and optimal regions. Several experiments have taken place through six distinct methods for the initialization of RBFNN fitting to perceive the best alignment. Initially, the proposed hybrid was applied to a simple trigonometric function to find the ability of our approach to approximate the function, convergence graph, MSE, and stability analysis. In the final method, the Broyden-Fletcher-Goldfarb-Shanno (BFGS) method has been incorporated with our proposed technique to refine the outcome further and improve the convergence to the local optimum. This new hybrid scheme is capable of enhancing the training speed and convergence accuracy. Also, this hybrid-based network has been utilized in different real-world applications to evaluate its problem-solving capabilities, efficacy, and accuracy. The obtained results are rigorously compared with three existing RBFNN-based algorithms. The comparative analysis focused on approximate solutions, the algorithm's convergence, and mean square error (MSE). The proposed results reveal that the RBFNN-based hybrid technique is capable of solving complex, high-dimensional, stochastic, and nonlinear problems.","author":[{"family":"Ali","given":"Sabir"},{"family":"Kurz","given":"Jason"},{"family":"Oughton","given":"Sean"}],"issued":{"date-parts":[[2025]]},"DOI":"10.21203/rs.3.rs-7088954/v1","URL":"https://doi.org/10.21203/rs.3.rs-7088954/v1","source":"crossref"},{"id":"doi:10.2139/ssrn.5426897","type":"manuscript","title":"What Hinders Electric Vehicle Diffusion? Insights from a Neural Network Approach","abstract":"The transition to a zero-emission vehicle fleet represents a pivotal element of Europe's decarbonization strategy, with Italy's participation being particularly significant given the size of its automotive market. This study investigates the potential for battery electric cars (BEVs) to drive decarbonization of Italy's passenger vehicle fleet, focusing on the feasibility of targets set in the National Integrated Plan for Energy and Climate (PNIEC). Leveraging artificial neural networks, we integrate macroeconomic indicators, market-specific variables, and policy instruments to predict fleet dynamics and identify key factors influencing BEV adoption. We forecast that while BEV registrations will continue growing through 2030, the growth rate is projected to decelerate, presenting challenges for meeting ambitious policy targets. Our feature importance analysis demonstrates that BEV adoption is driven by an interconnected set of economic, infrastructural, and behavioral factors. Specifically, our model highlights that hybrid vehicle registrations and the vehicle purchase index exert the strongest influence on BEV registrations, suggesting that policy interventions should prioritize these areas to maximize impact. By offering data-driven insights and methodological innovations, our findings contribute to more effective policy design for accelerating sustainable mobility adoption while accounting for market realities and consumer behavior.","author":[{"family":"Bonacina","given":"Monica"},{"family":"Demir","given":"Mert"},{"family":"Sileo","given":"Antonio"},{"family":"Zanoni","given":"Angela"}],"issued":{"date-parts":[[2025]]},"DOI":"10.2139/ssrn.5426897","URL":"https://doi.org/10.2139/ssrn.5426897","source":"crossref"},{"id":"doi:10.1007/s00521-024-10963-z","type":"article-journal","title":"FLAME: fire detection in videos combining a deep neural network with a model-based motion analysis","abstract":"Abstract Among the catastrophic natural events posing hazards to human lives and infrastructures, fire is the phenomenon causing more frequent damages. Thanks to the spread of smart cameras, video fire detection is gaining more attention as a solution to monitor wide outdoor areas where no specific sensors for smoke detection are available. However, state-of-the-art fire detectors assure a satisfactory Recall but exhibit a high false-positive rate that renders the application practically unusable. In this paper, we propose FLAME, an efficient and adaptive classification framework to address fire detection from videos. The framework integrates a state-of-the-art deep neural network for frame-wise object detection, in an automatic video analysis tool. The advantages of our approach are twofold. On the one side, we exploit advances in image detector technology to ensure a high Recall. On the other side, we design a model-based motion analysis that improves the system’s Precision by filtering out fire candidates occurring in the scene’s background or whose movements differ from those of the fire. The proposed technique, able to be executed in real-time on embedded systems, has proven to surpass the methods considered for comparison on a recent literature dataset representing several scenarios. The code and the dataset used for designing the system have been made publicly available by the authors at ( https://mivia.unisa.it/large-fire-dataset-with-negative-samples-lfdn/ ).","author":[{"family":"Gragnaniello","given":"Diego"},{"family":"Greco","given":"Antonio"},{"family":"Sansone","given":"Carlo"},{"family":"Vento","given":"Bruno"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1007/s00521-024-10963-z","URL":"https://doi.org/10.1007/s00521-024-10963-z","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-6479514/v1","type":"article-journal","title":"BP Neural Network Model for Late-Night Effects Prediction in Postgraduates' Hypertension","abstract":"Abstract Graduate students exhibit a higher propensity for nocturnal hypertension episodes compared to the general working population. This study investigated the correlation between blood pressure and various physiological parameters, developing a continuous blood pressure prediction model based on highly correlated characteristics. Using a convenience sampling method involving physical examinations and structured questionnaires, 119 master's and doctoral students were selected. A BP neural network prediction model was constructed to analyze the impact of late-night academic work and dietary habits on blood pressure, identifying key influencing factors. Analysis demonstrated the model's high accuracy in blood pressure prediction, facilitating personalized health behavior recommendations. Evaluation metrics such as mean absolute error (MAE), standard deviation of error (SDE), root mean square error (RMSE), and R 2 were employed, achieving MAE = 3.56, SDE = 5.16, RMSE = 8.09, and R 2 = 0.99866 for systolic blood pressure (SBP), and MAE = 3.34, SDE = 4.46, RMSE = 7.26, and R 2 = 0.99284 for diastolic blood pressure (DBP) across test sets. These results met the standards set by the Association for the Advancement of Medical Instrumentation (AAMI). Factor weight analysis identified sleep duration (19.39%) and body weight as key hypertension drivers, followed by exercise duration, dietary habits, and emotional regulation. Prioritizing these factors is critical for developing targeted interventions. Systematic analysis of health data in research cohorts, particularly quantifying their pathogenic contributions to hypertension pathogenesis, advances predictive modeling and precision treatment frameworks.","author":[{"family":"Che","given":"Yanrui"},{"family":"Li","given":"Guangqing"},{"family":"Li","given":"Jingyao"}],"issued":{"date-parts":[[2025]]},"DOI":"10.21203/rs.3.rs-6479514/v1","URL":"https://doi.org/10.21203/rs.3.rs-6479514/v1","source":"crossref"},{"id":"doi:10.36227/techrxiv.176592058.84380101/v1","type":"article-journal","title":"An Energy-Efficient RFET-Based Stochastic Computing Neural Network Accelerator","abstract":"Stochastic computing (SC) offers significant reductions in hardware complexity for traditional convolutional neural networks (CNNs). However, despite its advantages, stochastic computing neural networks (SCNNs) often suffer from high resource consumption due to components such as stochastic number generators (SNGs) and accumulative parallel counters (APCs), which limit overall performance. This paper proposes a novel SCNN architecture leveraging reconfigurable field-effect transistors (RFETs). The inherent reconfigurability at the device level enables the design of highly efficient and compact SNGs, APCs, and other essential components. Furthermore, a dedicated SCNN accelerator architecture is developed to facilitate system-level simulation. Based on accessible open-source standard cell libraries, experimental results demonstrate that the proposed RFET-based SCNN accelerator achieves significant reductions in area, latency, and energy consumption compared to its FinFET-based counterpart at the same technology node.","author":[{"family":"Lu","given":"Sheng"},{"family":"Qu","given":"Qainhou"},{"family":"Jung","given":"Sungyong"},{"family":"Liang","given":"Qilian"},{"family":"Pan","given":"Chenyun"}],"issued":{"date-parts":[[2025]]},"DOI":"10.36227/techrxiv.176592058.84380101/v1","URL":"https://doi.org/10.36227/techrxiv.176592058.84380101/v1","source":"crossref"},{"id":"doi:10.33407/lib.naes.id/748283","type":"article-journal","title":"Chapter ХVІ. Neural network modeling and optimization of technological modes in polyvinyl chloride production","abstract":"This study is dedicated to enhancing the efficiency of industrial polyvinyl chloride synthesis via the suspension polymerization of vinyl chloride. Research has been conducted on constructing the \"temperature trajectory\" of the suspension polymerization process based on the \"thermal shock\" strategy, according to which the final reaction temperature is increased to accelerate the reaction rate (reduce its duration). In accordance with the research objectives, a comprehensive analysis was carried out to assess the impact of technological parameters (the timing of additional vinyl chloride monomer introduction and the moment of thermal shock) on the quality of the obtained polyvinyl chloride. As a result of neural network approximation of active experimental data and multi-criteria optimization, the optimal values of technological factors were determined, ensuring the best quality indicators of polyvinyl chloride compared to operations conducted under the standard (isothermal) technological mode.","author":[{"family":"Shakhnovsky","given":"Arcady"},{"family":"Kvitka","given":"Oleksandr"},{"family":"Koliushko","given":"Oleg"}],"issued":{"date-parts":[[2026]]},"DOI":"10.33407/lib.naes.id/748283","URL":"https://doi.org/10.33407/lib.naes.id/748283","source":"crossref"},{"id":"doi:10.1364/opticaopen.28902587.v1","type":"article-journal","title":"Q-factor Analysis in Free Space Optical Communication and Neural Network-Based Prediction","abstract":"Free Space Optics (FSO) provides a promising alternative where fiber-optic deployment is impractical due to cost or fragility. However, FSO performance is highly vulnerable to atmospheric disturbances such as fog, rain, and dust, which can significantly degrade signal quality. To optimize system performance under varying conditions, it is crucial to understand how the Q-factor responds to changes in system parameters. This study investigates the effects of bit rate, filter type, transmitter and receiver aperture diameters, and transmission range on the Q-factor in FSO systems. We developed a detailed simulation model using OptiSystem to generate data, which was then used to train a feedforward neural network via MATLAB’s Neural Network Tool (NN-Tool) using the Levenberg–Marquardt algorithm. This model effectively captures complex, nonlinear relationships between input parameters and Q-factor outcomes, allowing accurate predictions without further simulations. The hybrid approach of combining simulation data with neural network-based modeling offers a practical and user-friendly tool for performance prediction and system planning. This research contributes to the design and optimization of high-data-rate FSO systems by addressing existing limitations in modeling and parameter tuning.","author":[{"family":"Sikder","given":"Mohammad"},{"family":"Sakib","given":"Fahim"},{"family":"Hossen","given":"Md"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1364/opticaopen.28902587.v1","URL":"https://doi.org/10.1364/opticaopen.28902587.v1","source":"crossref"},{"id":"doi:10.31224/4738","type":"article-journal","title":"Physics-informed neural network framework for solving forward and inverse flexoelectric problems","abstract":"Flexoelectricity, the coupling between strain gradients and electric polarization, poses significant computational challenges due to its governing fourth-order partial differential equations that require C¹-continuous solutions. To address these issues, we propose a physics-informed neural network (PINN) framework grounded in an energy-based formulation that treats both forward and inverse problems within a unified architecture. The forward problem is recast as a saddle-point optimization of the total potential energy, solved via the deep energy method (DEM), which circumvents the direct computation of high-order derivatives. For the inverse problem of identifying unknown flexoelectric coefficients from sparse measurements, we introduce an additional variational loss that enforces stationarity with respect to the electric potential, ensuring robust and stable parameter inference. The framework integrates finite element-based numerical quadrature for stable energy evaluation and employs hard constraints to rigorously enforce boundary conditions. Numerical results for both direct and converse flexoelectric effects show excellent agreement with mixed-FEM solutions, and the inverse model accurately recovers material parameters from limited data. This study establishes a unified, mesh-compatible, and scalable PINN approach for high-order electromechanical problems, offering a promising alternative to traditional simulation techniques.","author":[{"family":"Moon","given":"Hyeonbin"},{"family":"Park","given":"Donggeun"},{"family":"Yeo","given":"Jinwook"},{"family":"Ryu","given":"Seunghwa"}],"issued":{"date-parts":[[2025]]},"DOI":"10.31224/4738","URL":"https://doi.org/10.31224/4738","source":"crossref"},{"id":"doi:10.53656/adpe-2025.09","type":"article-journal","title":"Surface shaping mechatronic neural network","abstract":"Neural networks are fundamental concept in artificial intelligence and machine learning, inspired by the structure and function of the human brain. A neural network is a computational model composed of interconnected nodes (called neurons or units) organized in layers. These networks are designed to recognize patterns, learn from data, and make predictions or decisions. As a part of the neural networks, mechanical neural networks are physical systems designed to mimic the behavior of artificial neural networks using mechanical components. These systems leverage the properties of materials and structures to process information, adapt to external stimuli, or perform tasks such as pattern recognition, optimization, or control. The developed surface-shaping mechanical neural network is based on systems with parallel kinematics but in addition contains sensing elements part of the electronic system controlled by developed software. The system can adapt its surface geometry depend on the external stimuli, in this case the force applied on it.","author":[{"family":"Komarski","given":"Dobri"},{"family":"Vassilev","given":"Velizar"},{"family":"Nikolova","given":"Hristiana"}],"issued":{"date-parts":[[2025]]},"DOI":"10.53656/adpe-2025.09","URL":"https://doi.org/10.53656/adpe-2025.09","source":"crossref"},{"id":"doi:10.36227/techrxiv.176617687.72969471/v1","type":"article-journal","title":"Neural Network Model Tree Prediction for Passive Microwave Sensor Quality Control","abstract":"Earth remote sensing algorithms for spaceborne passive microwave sensors depend on high quality observations of top-of-atmosphere radiances. When these observations are adversely affected by non-geophysical processes, such as radio frequency interference (RFI), calibration drifts, or sensor anomalies, they can have detrimental impacts on retrievals or their assimilation into models. The rise of machine learning applications has exacerbated this problem because non-linear schemes can reveal even small issues with the data. A simple and computationally inexpensive method is presented for assessing the quality of observations both instantaneously and over time using a fully connected neural network model tree. The basic premise of this scheme involves using n-1 channels to predict the nth channel, and then identifying when the prediction error is larger than normal in order to identify potentially anomalous observations. Prediction error thresholds are chosen based on the application, and well-trained models with high correlations between channels can have prediction errors on the order of the noise-equivalent differential temperature (NEDT) of the sensor (less than 1 K in most cases). We find that this method is effective for identifying various types of observational problems and present case studies to this effect that demonstrate radio frequency interference, sensor electronic and physical anomalies, sun glint, and channel performance degradation. We also find that its sensitivity is ideal for quality control, even when the observational data has previously passed operational quality checks, and that this tool can be used to effectively prescreen problematic observations that may not be initially identifiable in the data application. This tool is readily available to the remote sensing community through freely distributed code with examples on training and application via a link provided in the Data Availability statement of this article.","author":[{"family":"Jones","given":"Spencer"},{"family":"Brown","given":"Paula"},{"family":"Kummerow","given":"Christian"}],"issued":{"date-parts":[[2025]]},"DOI":"10.36227/techrxiv.176617687.72969471/v1","URL":"https://doi.org/10.36227/techrxiv.176617687.72969471/v1","source":"crossref"},{"id":"doi:10.1007/s00521-025-11325-z","type":"article-journal","title":"Stock price forecasting through symbolic dynamics and state transition graphs with a convolutional recurrent neural network architecture","abstract":"Abstract Accurate stock price forecasting remains a critical challenge in financial analytics due to volatile market conditions, non-stationary dynamics, and abrupt regime shifts that often defy traditional modeling techniques. This study proposes a comprehensive framework for stock price forecasting that integrates symbolic dynamics, graph-based state representations, and deep learning. By converting continuous-valued stock prices into discrete symbolic states representing amplitude and trend information, the method constructs transition matrices capturing probabilistic relationships within financial time series. These transition matrices are then processed by a convolutional recurrent neural network (CRNN), in which convolutional layers isolate local spatial dependencies in the symbolic-state domain, while recurrent LSTM layers capture multi-scale temporal dynamics extending across multiple time horizons. Experimental evaluations are conducted over prediction horizons of 1 day, 10 days, and 100 days, spanning pre-COVID, COVID, and post-COVID market regimes. The results indicate that while longer prediction horizons naturally incur greater forecasting uncertainty due to compounding variability, the integration of symbolic-state preprocessing with deep temporal modeling demonstrates significant robustness in handling non-stationary financial environments. During the stable pre-COVID period, the proposed methodology achieves reductions in mean squared error (MSE) of up to 98% relative to the volatile COVID phase, highlighting its capability to effectively leverage well-defined market patterns in stable economic conditions. Furthermore, the model consistently delivers competitive forecasting performance across all prediction horizons and market regimes. Collectively, these findings emphasize the potential of symbolic-state-based deep learning architectures as a viable pathway to address the complexity and volatility characteristic of modern financial markets.","author":[{"family":"Mirza","given":"Fuat"},{"family":"Pekcan","given":"Önder"},{"family":"Hekimoğlu","given":"Mustafa"},{"family":"Baykaş","given":"Tunçer"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1007/s00521-025-11325-z","URL":"https://doi.org/10.1007/s00521-025-11325-z","source":"crossref"},{"id":"doi:10.31219/osf.io/gzjfy_v1","type":"article-journal","title":"Exploring Quantum Convolutional Neural Network for Fault Detection in Semiconductor Wafers","abstract":"This research work on the well-known WM811K wafer map dataset intends to integrate the insights provided by wafer manufacturing techniques and the computing power of quantum convolutional neural networks (QCNN) approach. The work begins with demonstrating high accuracy (𝑹 𝟐 =99.32%) of the classical convolutional neural network (CCNN) consisting of three consecutive convolution layers on the WM-811K dataset. In later part of the paper, optimized for multi-class classification, the applied QCNN model consisting of four qubits used for convolutional and pooling layers exhibit 99.38% accuracy. The QCNN model establishes supremacy over CCNN model by reducing time requirement to train the model for classifying defects on the wafer images.","author":[{"family":"Biswas","given":"Debajyoti"},{"family":"Marwaha","given":"Shikha"},{"family":"Atmakuru","given":"Sriya"}],"issued":{"date-parts":[[2025]]},"DOI":"10.31219/osf.io/gzjfy_v1","URL":"https://doi.org/10.31219/osf.io/gzjfy_v1","source":"crossref"},{"id":"doi:10.2139/ssrn.6391418","type":"manuscript","title":"AI-based Computational Engineering using Hierarchical Deep Learning Neural Network","abstract":"Computational engineering is undergoing a rapid transformation due to artificial intelligence (AI), which combines data-driven learning with physics-based modeling. An important advancement is the development of Hierarchical Deep Learning Neural Networks (HDLNNs), which employ layered architectures to capture multi-scale and multi-physics features. Operator learning models like DeepONet, Fourier Neural Operator (FNO), and Physics-Informed Neural Operator (PINO) built upon the ideas of early frameworks like Physics-Informed Neural Networks (PINNs), which incorporated governing equations into training, to enable resolution-invariant PDE surrogates. Hierarchical extensions like HiDeNN and Convolutional HiDeNN (C-HiDeNN) improve structural and material simulations, while Graph Neural Networks (GNNs) provide mesh-based multigrid-like learning. Despite the rapid progress, challenges remain in handling stiffness, generalization to complex geometries, and multi- fidelity integration. Through comparison with traditional solvers and a description of possible future directions for scalable, physics-consistent AI systems, this study investigates the role of HDLNNs in computational engineering.","author":[{"family":"Singhal","given":"Kanika"},{"family":"Kaur","given":"Inderpreet"},{"family":"Bansal","given":"Madhav"},{"family":"Kushwah","given":"Kirti"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2139/ssrn.6391418","URL":"https://doi.org/10.2139/ssrn.6391418","source":"crossref"},{"id":"doi:10.2139/ssrn.6760256","type":"manuscript","title":"A Crack Growth-Guided Symbolic Regression-Neural Network Framework for Fatigue Life Prediction","abstract":"To reduce the restrictions of empirical fatigue models and improve the physical consistency of purely data-driven approaches, this study proposes a fatigue life prediction framework integrating symbolic regression with physics-informed neural networks. In this framework, the local crack growth relation is connected with total fatigue life through two steps. Symbolic regression is first performed using fracture-mechanics variables, and explicit phenomenological expressions are obtained directly from crack growth rate data rather than from a prescribed empirical form. The equivalent initial flaw size is then used as the lower limit in the life integration. In this way, the identified crack growth law can be used as a constraint for total fatigue life prediction. A stress-ratio-dependent local correction is incorporated in the equivalent initial flaw size calculation to improve the physical consistency of life inference under different loading ratios. The identified crack-growth relationship, life inference constraint, and experimental fatigue life data are jointly embedded into the physics-informed neural network, enabling coordinated optimization between data supervision and physical constraints. Validation on public fatigue datasets of TC4, A36, and AA 2024-T3 shows that the proposed framework captures fatigue life evolution under different material conditions and provides better prediction accuracy, stability, and applicability than classical crack-growth-constrained models and purely data-driven models. Under limited target-material samples, the model also shows small-sample cross-material transfer potential. This study provides a physically interpretable and prediction-oriented modeling approach for fatigue life prediction under complex conditions.","author":[{"family":"Zhang","given":"Dongxu"},{"family":"Shi","given":"Junjie"},{"family":"Wen","given":"Zhixun"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2139/ssrn.6760256","URL":"https://doi.org/10.2139/ssrn.6760256","source":"crossref"},{"id":"doi:10.26434/chemrxiv.15003976/v2","type":"manuscript","title":"Beyond Multiscale: A Neural Network-Unified Molecular Model for Asymmetry-Free Dynamics Simulation","abstract":"Adaptive multiscale simulations, such as adaptive QM/MM, are indispensable for investigating complex solution dynamics but historically suffer from a fundamental dilemma: the dynamic exchange of identical solvent molecules across different model resolutions introduces a physical asymmetry. This asymmetry forces a prohibitive trade-off between severe structural distortions and the violation of Hamiltonian conservation (continuous energy drift). To definitively resolve this issue, we propose the ”Integrated Molecular Model,” a paradigm-shifting framework that mathematically unifies multiple resolutions into a single, dynamically consistent potential. By employing a neural network that takes only the multiscale weight functions as input descriptors, our model learns to systematically cancel the non-physical transition forces originating from spatial discontinuities. As a proof-of-concept, we validated this unified framework on bulk water, aqueous ionic solutions, and reactive hydronium ions. Our approach not only faithfully restores artifact-free solvation structures but also strictly conserves the Hamiltonian, enabling unprecedentedly stable, long-term simulations under the microcanonical (NVE) ensemble. Consequently, the model successfully eliminates artificial thermostat-induced breathing modes, allowing for the rigorous evaluation of intrinsic dynamical properties such as diffusion coefficients. Furthermore, owing to its physically informed network design, the correction potential introduces less than 2\\% additional computational overhead and demonstrates remarkable zero-shot transferability across chemically similar ionic species. By fundamentally eradicating the inherent asymmetry of adaptive multiscale modeling, this highly scalable framework establishes a robust foundation for high-fidelity simulations of complex condensed-phase phenomena, including reactive dynamics at solid-liquid interfaces and ion transport in battery electrolytes.","author":[{"family":"Ikeda","given":"Takuma"},{"family":"Watanabe","given":"Hiroshi"},{"family":"Nakano","given":"Haruyuki"}],"issued":{"date-parts":[[2026]]},"DOI":"10.26434/chemrxiv.15003976/v2","URL":"https://doi.org/10.26434/chemrxiv.15003976/v2","source":"crossref"},{"id":"doi:10.2139/ssrn.6988684","type":"manuscript","title":"CABORT: A deep neural network inference accelerator based on on-chip optical interconnect","abstract":"Deep neural networks (DNNs) possess strong feature extraction abilities and have been widely used across various applications, serving as a core component of current artificial intelligence systems. However, the complex data communication within DNNs greatly limits model inference speed on local hardware devices. The development of silicon-based photonics has introduced on-chip optical interconnect technologies that offer fresh solutions for complex intra-chip communication. This study presents an innovative DNN accelerator architecture named CABORT, which employs optical interconnects organized in a ring topology. The CABORT framework incorporates wavelength division multiplexing (WDM) to enhance communication performance among convolutional layers. Its design includes an optical processing array composed of micro-ring resonators, power splitters, and photodetectors to facilitate efficient data transmission during convolutional computations. To further increase network throughput and alleviate the influence of optical signal crosstalk on the optical signal-to-noise ratio (OSNR), a neural task allocation algorithm based on the genetic algorithm is developed. Experimental findings reveal that CABORT, through its specialized task mapping strategy and distinctive network structure, achieves up to 68.0% lower latency and 55.9% less power consumption compared with existing DNN accelerators. These results verify the architecture’s effectiveness and highlight its potential as a highly promising direction for advanced deep neural network processing.","author":[{"family":"Chen","given":"Zhenjiao"},{"family":"Wang","given":"Zehao"},{"family":"Xu","given":"Wentao"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2139/ssrn.6988684","URL":"https://doi.org/10.2139/ssrn.6988684","source":"crossref"},{"id":"doi:10.36341/rabit.v11i2.8061","type":"article-journal","title":"PENGEMBANGAN MODEL DEEP LEARNING UNTUK DETEKSI SUARA  MENGGUNAKAN CONVOLUTIONAL NEURAL NETWORK","abstract":"Deteksi suara dan pengenalan kata kunci dalam sinyal audio telah menjadi bidang penelitian yang berkembang pesat karena aplikasinya yang luas, mulai dari pengawasan audio cerdas hingga sistem interaksi manusia-komputer. Penelitian ini bertujuan untuk mengembangkan model deep learning berbasis Convolutional Neural Network (CNN) untuk mendeteksi dan mengklasifikasikan kata-kata spesifik dalam rekaman suara secara otomatis. Fokus penelitian ini adalah pada pendeteksian kata kunci \"anjing\" dan \"anak-anak\" yang terdapat dalam data ucapan. Metodologi penelitian mencakup pra-pemrosesan data melalui pembersihan gangguan (noise) dan normalisasi, serta teknik augmentasi data seperti pitch shifting untuk meningkatkan performa model. Fitur suara diekstraksi menggunakan Short-Time Fourier Transform (STFT) untuk menghasilkan representasi visual berupa spektrogram sebagai input utama bagi arsitektur CNN. Hasil pengujian menunjukkan bahwa model yang dikembangkan berhasil mengklasifikasikan kata kunci tersebut dengan tingkat akurasi sebesar 90,00%. Model ini terbukti efektif dalam mengenali pola spektral dan temporal dari ucapan kata kunci dan dapat diimplementasikan untuk sistem deteksi suara waktu nyata","author":[{"family":"Sinaga","given":"Norita"},{"family":"Riadi","given":"Imam"},{"family":"Herman"}],"issued":{"date-parts":[[2026]]},"DOI":"10.36341/rabit.v11i2.8061","URL":"https://doi.org/10.36341/rabit.v11i2.8061","source":"crossref"},{"id":"doi:10.2139/ssrn.7019278","type":"manuscript","title":"Virtual boundary integral neural network for three-dimensional exterior acoustic problems​","abstract":"This paper presents a virtual boundary integral neural network (VBINN) for three-dimensional exterior acoustic problems. The method introduces a virtual boundary within the scatterer or vibrating body and represents the associated source density using a neural network. Based on the acoustic Green’s function, this representation satisfies the Sommerfeld radiation condition and enables direct evaluation of the acoustic pressure and its normal derivative at arbitrary field points. Because the integration surface is separated from the physical boundary, the formulation avoids the singular and near singular kernel evaluations in conventional boundary integral learning methods. To reduce sensitivity to boundary placement, the geometric parameters of the virtual boundary are optimized jointly with the source density during training. Numerical examples for acoustic scattering, multiple body interaction, and underwater acoustic propagation show close agreement with analytical solutions and COMSOL results, and the Burton-Miller extension further improves stability near characteristic frequencies. These results demonstrate the potential of VBINN for three-dimensional exterior acoustic analysis.","author":[{"family":"Li","given":"Jiahao"},{"family":"Xi","given":"Qiang"},{"family":"Marchevsky","given":"Ilia"},{"family":"Fu","given":"Zhuojia"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2139/ssrn.7019278","URL":"https://doi.org/10.2139/ssrn.7019278","source":"crossref"},{"id":"doi:10.31645/jisrc.26.24.1.12","type":"article-journal","title":"An Enhanced Variant of Graph-Constrained Neural Multi-Objective Evolutionary Algorithm for Sparse Neural Network Training","abstract":"Sparse neural networks are essential for deploying deep learning models on resource-limited and latency-sensitive platforms, where efficiency must be improved without compromising predictive performance. While the Graph-Constrained Neural Multi-Objective Evolutionary Algorithm (GCNMOEA) has demonstrated advantages in sparsity promotion, runtime efficiency, and population diversity, its performance is hindered by reduced accuracy, unstable early-stage convergence, and lower hypervolume under moderate sparsity. To address these limitations, this study proposes Enhanced GCNMOEA, an improved variant that integrates accuracy-aware graph mutation, a two- phase dominance–decomposition selection mechanism, multifidelity evaluation, and adaptive diversity scheduling. These enhancements collectively strengthen structural preservation, reduce fitness noise, and improve convergence behaviour. Experimental evaluations on MNIST, Fashion-MNIST, CIFAR-10, and CIFAR-100 across LeNet-5, ResNet-18, and MobileNetV2 architectures demonstrate that the proposed variant achieves consistently higher accuracy, improved hypervolume, and reduced IGD variance, while maintaining the sparsity and runtime benefits of the baseline algorithm. The results confirm that Enhanced GCNMOEA deliverreal-worldsuperior Pareto-front quality and improved robustness, making it a strong candidate for real world edge and embedded neural network applications.","author":[{"family":"Channer","given":"Collen"},{"family":"Rizvi","given":"Syed"},{"family":"Ndajah","given":"Peter"},{"family":"Osbourne","given":"Otis"}],"issued":{"date-parts":[[2026]]},"DOI":"10.31645/jisrc.26.24.1.12","URL":"https://doi.org/10.31645/jisrc.26.24.1.12","source":"crossref"},{"id":"doi:10.1007/s00521-026-11898-3","type":"article-journal","title":"PolyNeXt: a novel semantic segmentation network for polyp detection in colonoscopy images","abstract":"Abstract Colorectal polyps are precancerous lesions with a high risk of developing into cancer if left untreated. Colonoscopy is the gold standard for detecting and removing these polyps, but it has high miss rates, especially for small and flat polyps. Deep learning methods have been increasingly used to aid in polyp detection, but their performance remains limited when applied to samples captured under unconstrained conditions or when processing images of small and flat polyps. To address this challenge, we propose PolyNeXt, a novel polyp segmentation network that leverages ConvNeXtV2 and global response normalization (GRN) layers. We train PolyNeXt on a combined dataset of samples from Kvasir-SEG and CVC-ClinicDB and evaluate it on four public distinct datasets: ETIS-LaribPolypDB, CVC-ColonDB, CVC-300, and BKAI-IGH NeoPolyp-Small. PolyNeXt achieves state-of-the-art performance on polyps captured in suboptimal conditions, outperforming other methods in terms of Intersection over Union (IoU) and Dice coefficient. Our work demonstrates that PolyNeXt is an effective polyp segmentation network that can improve the accuracy and reliability of detecting polyps captured under challenging circumstances.","author":[{"family":"Elkarazle","given":"Khaled"},{"family":"Raman","given":"Valliappan"},{"family":"Chua","given":"Caslon"},{"family":"Then","given":"Patrick"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1007/s00521-026-11898-3","URL":"https://doi.org/10.1007/s00521-026-11898-3","source":"crossref"},{"id":"doi:10.63367/199115992026063703017","type":"article-journal","title":"Biomedical Event Extraction via Multi-Grained Graph Neural Network","abstract":"Biomedical event extraction (BEE) is a significant task which plays an important role for some downstream biomedical applications. The recent works have applied joint methods using Graph Convolutional Network (GCN) based on the dependency tree to capture non-local syntactic features for the BEE task. However, in the graph of GCN, some edges are redundant which need to be pruned, and some dependency type labels are neglected which only can make the model capture coarse-grained features. In this paper, we propose a novel BEE method based on a Multi-Grained Graph Neural Network (MGGNN) structure. Specifically, the coarse-grained module using a standard GCN in MGGNN can first capture coarse-grained features in the biological text. Then, based on the proposed relation-attention guided GCN (RAGGCN), the fine-grained module intro-duces dependency type labels to prune edges and generates full-connection weight adjacent matrices during the node aggregation, which can capture fine-grained features by refining the coarse-grained ones. The experiment results on the MLEE dataset show that the proposed model achieves state-of-the-art performance with 1.06% higher F1 score for the BEE task. The extended experiment on the ACE2005 dataset also confirms the great generalization in the general domain.","author":[{"family":"Tan","given":"Fangyong"},{"family":"Zhang","given":"Jing"},{"family":"Zhao","given":"Ruifeng"}],"issued":{"date-parts":[[2026]]},"DOI":"10.63367/199115992026063703017","URL":"https://doi.org/10.63367/199115992026063703017","source":"crossref"},{"id":"doi:10.1007/s00702-026-03128-w","type":"article-journal","title":"Language network disruption in patients with Lewy body diseases","abstract":"Abstract Lewy body diseases, including Parkinson’s disease and dementia with Lewy bodies, are marked by neuronal α‑synuclein aggregation, motor parkinsonism, cognitive impairment and diverse non‑motor symptoms including communication impairments. Compared to other symptoms, non‑motor communication impairments remain under-explored, especially outside English‑speaking cohorts. The aim of this study was to elucidate the neural underpinnings of linguistic deficits as assessed by fMRI in Czech-speaking patients diagnosed with neuronal Lewy body disease and mild cognitive impairment (LBD-MCI). Scores from the Short Neuropsychological Battery and resting‑state functional MRI data were analyzed in 26 Czech‑speaking patients with LBD‑MCI and 24 healthy controls. Analyses targeted regions of interest within the dorsal and ventral language networks. We applied graph theory metrics, within‑network connectivity and seed‑based functional connectivity analysis. Graph analysis revealed dorsal‑stream disruption in LBD‑MCI: reduced clustering coefficient, increased path length, and diminished node strength, each correlating with language functions; ventral‑stream topology remained intact. Within‑network analyses showed impaired connectivity across both pathways, but weaker coupling between the opercular inferior frontal gyrus and posterior superior temporal gyrus predicted behavioral scores in language functioning, further highlighting dorsal vulnerability. Seed‑based analysis identified reduced frontotemporal connectivity in the dorsal stream and decreased fronto‑occipital and temporo-cerebellar connections in the ventral stream. LBD‑MCI is characterized particularly by network‑specific reductions in dorsal language stream efficiency and functional connectivity, underpinning syntactic and phonological processing deficits. These findings offer novel insights in the neural basis of language impairment in LBD-MCI.","author":[{"family":"Carbol","given":"Daniel"},{"family":"Novakova","given":"Lubomira"},{"family":"Klobusiakova","given":"Patricia"},{"family":"Rektorova","given":"Irena"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1007/s00702-026-03128-w","URL":"https://doi.org/10.1007/s00702-026-03128-w","source":"crossref"},{"id":"doi:10.3389/fphar.2026.1686243","type":"article-journal","title":"Analysis of the interaction network relationship between drugs using a graph neural network","abstract":"Introduction The ever-increasing complexity of biochemical systems, alongside the rapid growth of pharmaceutical and biomedical data, underscores the urgent need for intelligent, scalable, and interpretable computational models. These models must be capable of supporting next-generation decision-support systems and driving knowledge discovery in the realm of computational science. Traditional approaches to relational biomedical modeling, however, often struggle to accurately capture intricate multi-relational dependencies and typically lack robustness in sparse or incomplete interaction domains. To address these pressing limitations, we present a novel, biologically grounded graph-based learning framework designed to overcome such challenges. Methods Our approach comprises a two-tiered system: PHARMNet, a multi-relational graph neural network (GNN) equipped with memory-augmented attention mechanisms, and INTERACT-SCOPE, an advanced, context-aware optimization strategy that leverages structured biomedical ontologies and domain knowledge. PHARMNet employs relation-specific graph convolutions and semantic embedding alignment to effectively model latent relational dependencies in biochemical and pharmacological datasets. In parallel, INTERACT-SCOPE improves predictive generalization and stability by incorporating ontology-guided constraints, estimating epistemic uncertainty, and applying adaptive graph regularization techniques tailored to biomedical structures. Results and Discussion Through rigorous experimental evaluations across a variety of pharmacological interaction categories, our framework consistently achieves state-of-the-art (SOTA) predictive performance, enhanced model interpretability, and notable robustness—especially in low-data or high-noise scenarios. These outcomes strongly align with the journal’s mission to promote innovative and knowledge-driven advances in software engineering, artificial intelligence, and biomedical informatics. Ultimately, our article illustrates the synergistic integration of computational intelligence, domain-informed graph representation learning, and scalable modeling, contributing a powerful and interpretable solution to real-world challenges in healthcare informatics and biomedical discovery. Experimental results demonstrate that MGTNSyn outperforms existing methods, achieving an AUC of 0.873 and an F1-score of 0.831 on drug–drug interaction (DDI) benchmark datasets.","author":[{"family":"Chai","given":"Zhongyi"},{"family":"Wang","given":"Jing"},{"family":"Du","given":"Huili"}],"issued":{"date-parts":[[2026]]},"DOI":"10.3389/fphar.2026.1686243","URL":"https://doi.org/10.3389/fphar.2026.1686243","source":"crossref"},{"id":"doi:10.1088/1674-1056/ae1dec","type":"article-journal","title":"Dynamic analysis and DNA coding-based image encryption of memristor synapse-coupled hyperchaotic IN-HNN network","abstract":"Abstract The rapid development of brain-like neural networks and secure data transmission technologies has placed greater demands on highly complex neural network systems and highly secure encryption methods. To this end, the paper proposes a novel high-dimensional memristor synapse-coupled hyperchaotic neural network by using the designed memristor as the synapse to connect an inertial neuron (IN) and a Hopfield neural network (HNN). By using numerical tools including bifurcation plots, phase plots, and basins of attraction, it is found that the dynamics of this system are closely related to the memristor coupling strength, self-connection synaptic weights, and inter-connection synaptic weights, and it can exhibit excellent hyperchaotic behaviors and coexisting multi-stable patterns. Through PSIM circuit simulations, the complex dynamics of the coupled IN-HNN system are verified. Furthermore, a DNA-encoded encryption algorithm is given, which utilizes generated hyperchaotic sequences to achieve encoding, operation, and decoding of DNA. The results show that this algorithm possesses strong robustness against statistical attacks, differential attacks, and noise interference, and can effectively resist known/selected plaintext attacks. This work will provide new ideas for the modeling of large-scale brain-like neural networks and high-security image encryption.","author":[],"issued":{"date-parts":[[2025]]},"DOI":"10.1088/1674-1056/ae1dec","URL":"https://doi.org/10.1088/1674-1056/ae1dec","source":"crossref"},{"id":"doi:10.2139/ssrn.6642465","type":"manuscript","title":"Spectral Analysis Reveals Fundamental Differences between Human and Deep Neural Network Shape Representations","abstract":"While the human visual system is known to be highly sensitive to global and configural shape information, deep neural networks models (DNNs) trained on ImageNet seem to favour local shape features.&amp;nbsp; However, a more exact understanding of these differences has remained elusive, in part due to a lack of systematic methods for exploring the nature of high-dimensional shape representations.&lt;br&gt;&lt;br&gt;Here we argue that a novel shape frequency analysis can provide important insights into these representations.&amp;nbsp; We explore this hypothesis through a series of experiments in which we measure human and DNN sensitivity across the shape frequency spectrum.&amp;nbsp; These experiments reveal systematic differences between human and DNN models in spectral tuning and sensitivity to the amplitude and phase of shape frequency components.&amp;nbsp; Using this frequency analysis approach, we also show that recent curriculum modifications claimed to create more human-like AI systems primarily act to attenuate sensitivity to higher shape frequencies rather than sharpening discriminative tuning to lower frequencies.&amp;nbsp;&amp;nbsp; Overall, we find that humans are tuned to much lower shape frequencies than DNN models, and as a consequence these models are unable to predict the majority of variance in human judgements of shape stimuli.","author":[{"family":"Baker","given":"Nicholas"},{"family":"Wilder","given":"John"},{"family":"Elder","given":"James"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2139/ssrn.6642465","URL":"https://doi.org/10.2139/ssrn.6642465","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-10282772/v1","type":"article-journal","title":"A Physics-Informed Neural Network Frameworkfor Elastodynamic Wave Propagation inBimaterial Systems","abstract":"Abstract Physics-informed neural networks (PINNs) provide a promising framework for solving partial differential equations while embedding the underlying physical laws directly into the learning process. This study presents a PINN-based framework for modeling transient elastodynamic wave propagation in bimaterial systems governed by the axisymmetric equations of linear elasticity. A steel-aluminum specimen representative of a Split Hopkinson Pressure Bar configuration is considered, and the governing elastodynamic equations, together with the corresponding initial, boundary, and interface conditions, are incorporated directly into the network through a physics-informed loss function. High-fidelity finite-element simulations performed using ANSYS Workbench Explicit Dynamics are used for validation and as supplementary data constraints during training. The proposed framework accurately predicts wave transmission and reflection across the bimaterial interface and reproduces axial and radial displacement histories, face-averaged responses, and the dominant stress and strain evolution with close agreement to the finite-element solutions. The trained network further demonstrates the ability to predict wave responses at previously unseen time instants and for modified material properties without requiring additional finite-element simulations, providing a continuous surrogate model for elastodynamic analysis. Mesh-sensitivity studies confirm numerical robustness, while additional material combinations demonstrate the generality of the proposed methodology. The results show that integrating physics-informed neural networks with explicit finite-element analysis provides an accurate and computationally efficient framework for elastodynamic wave propagation in heterogeneous solids, offering an effective surrogate modeling approach for high-rate solid mechanics and impact engineering applications.","author":[{"family":"Chibire","given":"Sonal"},{"family":"Gau","given":"Jenn"},{"family":"Zhang","given":"Bo"}],"issued":{"date-parts":[[2026]]},"DOI":"10.21203/rs.3.rs-10282772/v1","URL":"https://doi.org/10.21203/rs.3.rs-10282772/v1","source":"crossref"},{"id":"doi:10.1002/cem.70088","type":"article-journal","title":"In‐Situ Detection of Microplastic Particles on Food Using Hyperspectral Imaging With One‐Dimensional Convolutional Neural Network and Artificial Neural Network","abstract":"ABSTRACT Hyperspectral imaging (HSI) has emerged as a promising technique for microplastic detection through analysis of reflectance variations across multiple wavelengths. Traditional approaches have focused primarily on isolated microplastic particles, requiring labor‐intensive separation procedures impractical for routine monitoring. The challenge of detecting microplastics directly on food surfaces stems from spectral similarities between microplastics and food matrices, making differentiation difficult using conventional methods. Leveraging recent advances in machine learning, this study explores how artificial neural networks (ANN) and one‐dimensional convolutional neural networks (1D‐CNN) can identify subtle spectral differences to detect microplastic particles on seafood without isolation. We systematically evaluated model architectures, preprocessing techniques, and hyperparameter configurations to optimize detection performance using hyperspectral data from tilapia samples contaminated with polyethylene microspheres. Our findings demonstrate that 1D‐CNN models trained on hyperspectral data without dimensionality reduction significantly outperform other approaches, achieving object‐level detection F1 scores of 0.963 for 600‐μm particles and 0.950 for 300‐μm particles. This detection strategy represents a substantial improvement over traditional methods and highlights the potential of deep learning–based approaches for non‐destructive, efficient microplastic detection in food safety applications.","author":[{"family":"Nayani","given":"Nikhita"},{"family":"Yang","given":"Ran"},{"family":"Sun","given":"Yue"},{"family":"Yang","given":"Lihong"},{"family":"Zhou","given":"Lifeng"},{"family":"Feng","given":"Yiming"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1002/cem.70088","URL":"https://doi.org/10.1002/cem.70088","source":"crossref"},{"id":"doi:10.1088/1402-4896/adca65","type":"article-journal","title":"A four-dimensional memristor-coupled neural network chaotic dynamical system based on multi-level logic pulse stimulation","abstract":"Abstract In this paper, we propose a chaotic dynamics system based on a three-neuron cosine magnetically controlled memristor synapse-coupled Hopfield neural network. The system consists of a line equilibrium set composed of an infinite number of points, these points in the equilibrium set are dictated by the coupling strength, and when the coupling strength reaches 3, the line equilibrium set is composed of an infinite number of stability points and the index-2 saddle-foci. Numerical analysis of the bifurcation diagram, Lyapunov exponents and phase plots, reveals that there are attractors with different orbits in the system with the change in coupling strength. Moreover, after multilevel-logic pulse is added, the neural network has the capability to generate and control intricate multi-scroll attractors according to different system parameter values. The analog circuit of the four-dimensional magnetically controlled memristor synapse-coupled Hopfield neural network was designed, and the correctness of the numerical simulation was verified via PSIM circuit simulation software.","author":[{"family":"Fan","given":"Manhong"},{"family":"Xu","given":"Shiqi"},{"family":"Liu","given":"Qingsong"},{"family":"Xiao","given":"Qian"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1088/1402-4896/adca65","URL":"https://doi.org/10.1088/1402-4896/adca65","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-6035148/v1","type":"article-journal","title":"Forecasting Gold Price using Hybrid Deep Neural Network LSTM-Autoencoder","abstract":"Abstract Gold prices hold a significant place in the global economy as they reflect the economic health, which influences markets, investments, and currency values. Industries that rely on commodities, investors and decision-makers, need accurate forecasting gold prices. Existing models for gold price forecasting, struggle with overfitting, poor adaptability and difficulty in handling volatile or long term trend changes. For this research, various deep learning models were evaluated, which includes Long Short-Term Memory, Convolutional Neural Networks, hybrid LSTM-CNN. The proposed hybrid model of LSTM-Autoencoders, to predict the gold prices for the data taken from September 2000 to January 2024. We also examines the impact of external parameters such as US dollar price, silver price, and crude oil price on forecasting gold prices. Furthermore, a comparative analysis of these external parameters shows that, only gold prices as a prediction parameter yields the highest accuracy across various evaluation metrics. While the silver prices showed some association with the gold price prediction, crude oil had a comparatively low predictive value. Additionally the proposed LSTM-Autoencoders hybrid model has shown the highest accuracy outperforming the other models, while addressing the challenge of overfitting effectively. The results and findings from this study, aid in exploring the role of deep learning in financial time series domain, offering insights which contributes to financial analysts market strategist and economic forecasters.","author":[{"family":"Saini","given":"Agampreet"},{"family":"Singh","given":"Rahul"},{"family":"Sinha","given":"Puneet"}],"issued":{"date-parts":[[2025]]},"DOI":"10.21203/rs.3.rs-6035148/v1","URL":"https://doi.org/10.21203/rs.3.rs-6035148/v1","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-7624030/v1","type":"article-journal","title":"Neural network smoothing of American options payoff with grid refinement strategies","abstract":"Abstract For an American call or put option, at the expiry time, the value function is a ramp function, the Delta sensitivity is a Heaviside function and Gamma is a Dirac delta measure. In addition, the first derivative with respect to the early exercise boundary is singular and relates to Theta sensitivity at the free boundary. This research explores the possibility of whether deep learning can efficiently account for the payoff condition and learn these irregularities. To this end, we introduce a smoothed modular learning approximant (SMLA) that combines key expressions, and a regulator representing a smoothed terminal condition for the value function. The SMLA's derivatives at the terminal time correspond to smoothed Heaviside and Dirac delta functions. With appropriate grid refinement strategies, our neural network solver can predict the early exercise boundary and its derivative, value function, and Greeks. The predictive performance of SMLA is illustrated by examples, and accurate results are achieved even for extreme maturity time and volatility values","author":[{"family":"Nwankwo","given":"Chinonso"},{"family":"Ware","given":"Tony"},{"family":"Dai","given":"Weizhong"}],"issued":{"date-parts":[[2025]]},"DOI":"10.21203/rs.3.rs-7624030/v1","URL":"https://doi.org/10.21203/rs.3.rs-7624030/v1","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-7693532/v1","type":"article-journal","title":"Physics-informed Fourier Basis Neural Network for Fluid Mechanics","abstract":"Abstract Conventional machine learning approaches struggle to capture periodic patterns and solve quasi-periodic boundary problems in fluid mechanics. This study proposes a physics-informed Fourier basis neural network (PIFBNN) that integrates adaptive Fourier series with physical constraints to address canonical partial differential equations. The architecture preserves Fourier series' natural mathematical compatibility with periodic phenomena while incorporating trainable parameters (angular frequencies and weight coefficients) that enhance basis function flexibility and nonlinear learning capacity. We evaluate the framework on six fundamental fluid dynamics benchmarks: two-dimensional cylinder wake flow, lid-driven cavity flow, Kovasznay flow, Helmholtz equation, Burgers equation, and Allen-Cahn equation. Results demonstrate PIFBNN's consistent superiority over standard physics-informed neural network (PINN) in accuracy. Through sparse data reconstruction experiments and adjusting the activation functions of neural networks and comparing , we further validate the dual advantages of Fourier basis neural network (FBNN) over conventional artificial neural network (ANN): inherent periodicity handling and reduced sensitivity to activation function selection. The FBNN architecture maintains robust performance across different activation functions, as verified through systematic comparisons with ANN and PINN baselines. These findings position PIFBNN as a promising computational framework for complex fluid dynamics problems.","author":[{"family":"Wang","given":"Chao"},{"family":"Li","given":"Shilong"},{"family":"Yuan","given":"Zelong"},{"family":"Guo","given":"Chunyu"}],"issued":{"date-parts":[[2025]]},"DOI":"10.21203/rs.3.rs-7693532/v1","URL":"https://doi.org/10.21203/rs.3.rs-7693532/v1","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-9189154/v1","type":"article-journal","title":"Performance of Artificial Neural Network and Physics-Informed Neural Networks for Flexural Strength Using Sugarcane Bagasse Ash and Distilled Sewage Water ","abstract":"Abstract An increasing number of eco-conscious construction projects are looking to agricultural waste products, such as sugarcane bagasse ash (SCBA) and Distilled Sewage Water, to partially replace cement and Pure Water in their mixes. This study evaluates the predictive power of Artificial Neural Networks (ANNs) and Physics-Informed Neural Networks (PINNs) by examining the increase in flexural strength of SCBA-based concrete at 7, 14, and 28 days. The results of the experiment showed a steady increase in flexural strength with curing age, indicating constant hydration and good matrix densification. The nonlinear correlation between mix design factors and strength was anticipated using ANN and PINN methods and tested and validated through training, cross-validation (OOF), testing and full-dataset appraisals. The ANN demonstrated relatively high training accuracy, but it showed evidence of overtraining and poor generalisation, as indicated by negative R2 values on the validation and test datasets. Conversely, the PINN, where constraints are included in the loss function based on the physics, demonstrated a relatively higher stability and reduced error values (MSE, MAE, RMSE) on every data split. Reduced variance dispersion in PINN was demonstrated using residual analysis, and SHAP-based interpretability showed that cement content, percentage of SCBA and ratio of water-binder were the strongest predictors of flexural strength. Despite the fact that both models still need additional improvement to be more generalisable, the findings indicate that physics-informed learning can improve the robustness and interpretability of strength prediction. The suggested integrated experimental-PINN system presents a potential solution to the sustainable concrete performance modelling and the intelligent mix design optimisation.","author":[{"family":"Ahirrao","given":"Mayuri"},{"family":"Patel","given":"Rakesh"},{"family":"Mishra","given":"Chaitanya"}],"issued":{"date-parts":[[2026]]},"DOI":"10.21203/rs.3.rs-9189154/v1","URL":"https://doi.org/10.21203/rs.3.rs-9189154/v1","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-8576637/v1","type":"article-journal","title":"NB-Net: A Biologically-Inspired Framework for Neural Network Width Expansion","abstract":"Abstract Deep learning has achieved remarkable success across various fields, however, most research in this area has primarily focused on increasing network depth, leading to a relative lack of systematic exploration of network width. Inspired by the structure of biological neural systems, widening a neural network can enhance the resolution of the feature space, enabling the capture of richer and more fine-grained visual features. To this end, we propose a novel architecture called Neuron Bundle Network (NB-Net), featuring a groundbreaking dual 1×1 convolution fusion mechanism that revolutionizes multi-branch feature integration. NB-Net is inspired by the morphological structure of dendrites and axons in biological neurons. It aggregates multiple parallel processing paths using diverse convolutional layers and attention modules through our innovative dual 1×1 convolution design. Our dual 1×1 convolution fusion mechanism serves as the core innovation, providing superior feature integration compared to traditional single-stage merge operations by enabling gradual dimensionality reduction and enhanced gradient stability. The framework introduces several key innovations, including the dual 1×1 convolution two-stage merge mechanism for stable gradient flow, adaptive learning rate scaling for attention integration, and a systematic approach to kernel diversity orchestrated through our fusion strategy. Ablation studies demonstrate that the dual 1×1 convolution mechanism drives performance gains, enabling competitive results with efficient structured sparsity and feature fusion.","author":[{"family":"Tan","given":"Longfei"},{"family":"Huang","given":"Zhaohui"},{"family":"Meng","given":"Wei"}],"issued":{"date-parts":[[2026]]},"DOI":"10.21203/rs.3.rs-8576637/v1","URL":"https://doi.org/10.21203/rs.3.rs-8576637/v1","source":"crossref"},{"id":"doi:10.2139/ssrn.7112170","type":"manuscript","title":"Engineering structural synthetic reliability estimation using enhanced moving neural network-based dimensional modeling method","abstract":"To perform synthetic reliability estimation of engineering structures, an enhanced moving neural network-based dimensional modeling (DMEMNN) method is proposed based on artificial neural network (ANN) model, heuristic algorithm, moving regression (MR) method and dimensional modeling concept, in which the ANN model is used to reflect relationships between each response and these related input parameters, the heuristic algorithm is applied to improve the ability of searching compact support region (CSR) radius and determining effective modeling samples, the MR technique is adopted to find these unknown coefficients with the determined effective modeling samples, and the dimensional modeling concept is employed to derive synchronous modeling of multiple responses. In this case, four models are developed by introducing four chaotic mapping (CM) strategies and sled dog optimizer (SDO). For illustrating the applicability of the proposed DMEMNN method and the effectiveness of four developed models, a multi-objective nonlinear function (MONF) fitting and probabilistic analyses, an aircraft hydraulic system (HS) pressure and temperature prediction and reliability evaluation and an aeroengine high-pressure turbine blisk (HPTB) multi-failure response and reliability analyses are implemented, the analytical results show that the proposed framework holds excellent advantages in modeling and simulation capabilities.","author":[{"family":"Wang","given":"Junyao"},{"family":"Lu","given":"Cheng"},{"family":"Teng","given":"Da"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2139/ssrn.7112170","URL":"https://doi.org/10.2139/ssrn.7112170","source":"crossref"},{"id":"doi:10.26434/chemrxiv.10001743/v2","type":"manuscript","title":"A Multitask Graph Neural Network Framework for AMES Mutagenicity Prediction","abstract":"The process of drug screening and development is long, complex, and costly, making early and reliable assessment of toxicity a central challenge in pharmaceutical research. In this context, the ability to accurately predict drug-induced toxicity in general, and mutagenicity in particular, is crucial for guiding decision-making and reducing late-stage attrition. In particular, deep learning approaches can be utilized to predict the results of the Ames mutagenicity test, which is commonly employed for toxicity screening in drug development. Recent work has shown that deep learning multitask approaches that account for the contributions of individual bacterial strains improve mutagenicity prediction. However, most existing models rely on molecular descriptors, which are limited in their representation of molecular structure and properties and often lack interpretability. Graph neural networks are chemically intuitive and overcome many of these limitations. Here, we present a deep learning, multitask graph neural network model for predicting Ames mutagenicity. The proposed approach outperforms existing models, including those from the Ames/QSAR International Challenge, and demonstrates high sensitivity for identifying mutagenic compounds. In addition, our model is highly interpretable. GNNExplainer analysis of atom-level contributions revealed that the model learned known structural alerts associated with mutagenicity, as well as additional molecular patterns and contextual features, including three-dimensional structural properties. Importantly, these insights were captured in a strain-specific manner, highlighting the value of a multitask approach for modeling mutagenic mechanisms. Finally, the model identified previously unreported molecular substructures potentially associated with mutagenicity, supporting its utility in drug design and development.","author":[{"family":"Teitgen","given":"Abigail"},{"family":"Ulzurrun","given":"Eugenia"},{"family":"Campillo","given":"Nuria"},{"family":"Hernández","given":"Eduardo"}],"issued":{"date-parts":[[2026]]},"DOI":"10.26434/chemrxiv.10001743/v2","URL":"https://doi.org/10.26434/chemrxiv.10001743/v2","source":"crossref"},{"id":"doi:10.5194/egusphere-egu26-9179","type":"article-journal","title":"Detecting hail-prone environments using a W-Net convolutional neural network","abstract":"Hailstorms are known to cause great damage to agriculture and infrastructure. However, understanding and identifying the conditions favorable to hail remains a major challenge due to the complex, multi-scale nature of deep convective processes. In this study, we investigate the use of a W-Net convolutional neural network (CNN), which proved successful in image segmentation tasks, to identify atmospheric environments susceptible to hail from high-resolution numerical weather prediction data. The NOAA High-resolution Rapid Refresh (HRRR) model explicitly resolves convective processes owing to its fine spatial (3 km) and temporal (hourly) resolution. We consider meteorological variables from HRRR outputs relevant to deep convection and hail as input features for the W-Net model. Together with hail reports across the United States for the past ten years, we construct a deep learning framework. The trained network learns spatial patterns associated with hail-prone environments and produces gridded probability maps of hail occurrence. This data-driven approach shows the potential of deep learning methods for identification of hazardous convective weather. Once trained, the model can be applied to other regions, provided that the sub-daily, high-resolution meteorological fields are available.","author":[{"family":"Hercigonja","given":"Lana"},{"family":"Aslam","given":"Zeeshan"},{"family":"Ashfaq","given":"Moetasim"},{"family":"Prtenjak","given":"Maja"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5194/egusphere-egu26-9179","URL":"https://doi.org/10.5194/egusphere-egu26-9179","source":"crossref"},{"id":"doi:10.26434/chemrxiv.15006282/v1","type":"manuscript","title":"ChemGNN: Graph Neural Network Surrogate Guided Design of Carbon Nanotube Membranes for High-Efficiency Desalination","abstract":"ChemGNN, a deep-learning-accelerated framework that learns a graph-based surrogate of membrane transport and actively proposes high-performance designs. Each CNT configuration is encoded as a heterogeneous graph in which tube nodes carry geometry and functionalization features, edges encode inter-tube spacing, and global features describe operating conditions such as pressure, temperature gradient, and salinity. A contextual message-passing encoder produces node embeddings that a transformer-style readout fuses into a representation predicting water permeance and salt rejection. An expected-improvement acquisition function then selects the most informative candidates for molecular-dynamics validation, closing an active-design loop that concentrates expensive simulations on promising regions of the design space. We evaluate ChemGNN on a dataset of 12,480 configurations with 86 experimental hold-out measurements, comparing against an MD-only random search, a Gaussian-process surrogate, and a topologyagnostic multilayer perceptron surrogate. ChemGNN achieves the highest water permeance of 63.5 LMH/bar and salt rejection of 99.2 percent while requiring roughly an order of magnitude fewer molecular-dynamics evaluations than all baselines. Extensive analyses further reveal the non-linear sensitivity of performance to tube diameter, array pitch, temperature gradient, and feed salinity, and a human evaluation by domain experts confirms the novelty and practical relevance of the proposed designs. These results demonstrate that graph-aware deep learning effectively captures the coupled geometry-transport relationship underlying CNT membrane desalination.","author":[{"family":"Meng","given":"Linyu"},{"family":"Tang","given":"Shuo"},{"family":"Luo","given":"Yaqing"}],"issued":{"date-parts":[[2026]]},"DOI":"10.26434/chemrxiv.15006282/v1","URL":"https://doi.org/10.26434/chemrxiv.15006282/v1","source":"crossref"},{"id":"doi:10.5194/egusphere-egu26-12550","type":"article-journal","title":"Identifying zircon provenances using domain-adversarial neural network","abstract":"Zircon trace element geochemistry is a pivotal tool for unraveling petrogenesis and the evolutionary history of the Earth’s crust. While two-dimensional (2D) discriminant diagrams are conventionally used to identify parent rock types, the emergence of machine learning (ML) has introduced a transformative research paradigm. ML not only enhances classification accuracy but also resolves the inherent ambiguities found in traditional geochemical diagrams. However, the reliability of current ML models typically depends on the vast archives of labeled samples from the Phanerozoic. When extending research to “deep-time” samples, such as Hadean zircons, the scarcity of labeled data often forces researchers to rely on models trained exclusively on Phanerozoic datasets. This approach is prone to misclassification due to “domain shift,” caused by systematic variations in zircon trace element distributions across different geological eons. To address this challenge, we propose a Domain Adversarial Neural Network (DANN) framework tailored for zircon trace element analysis. By aligning the feature distributions of the source domain (Phanerozoic) and the target domain (Precambrian), the DANN extracts “domain-invariant yet geologically significant” high-dimensional feature representations, effectively mitigating the effects of temporal data bias. Our results demonstrate that DANN significantly outperforms traditional machine learning methods across multiple performance metrics. Furthermore, t-SNE visualization confirms that the source and target domains are effectively aligned within the feature space. When applied to ~4.3 Ga zircon samples from the Jack Hills, the model achieved a classification accuracy of 0.923. This high level of performance underscores the framework’s exceptional generalization capability for identifying unlabeled deep-time samples and its potential for broader applications in Precambrian geology. This study develops a transferable, data‑driven framework for inferring deep‑time geological processes, providing a novel methodology to address the limitations inherent in the traditional principle of uniformitarianism. Furthermore, the framework is extensible to other mineral systems (e.g., apatite, monazite), thereby opening new avenues for quantitatively reconstructing the dynamic evolution of the early Earth.","author":[{"family":"Zhang","given":"Mengwei"},{"family":"Chen","given":"Guoxiong"},{"family":"Kusky","given":"Timothy"},{"family":"Harrison","given":"Mark"},{"family":"Cheng","given":"Qiuming"},{"family":"Wang","given":"Lu"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5194/egusphere-egu26-12550","URL":"https://doi.org/10.5194/egusphere-egu26-12550","source":"crossref"},{"id":"doi:10.2139/ssrn.7344543","type":"manuscript","title":"Soliton Profiles in 1D: Enhanced Neural-Network Solvers and Constrained Deep Ritz Methods","abstract":"We develop and compare {\\it enhanced} neural-network solvers for one-dimensional ground-state profiles, which are solutions of $$Q^{\\prime\\prime}=bQ-\\gamma Q^{p},\\quad Q&gt;0,\\quad p&gt;1, \\quad b&gt;0,$$the stationary profile equation for the nonlinear Schr\\\"odinger, generalized Korteweg--de Vries, and nonlinear Klein--Gordon equations. Its explicit $\\sech$-type solutions serve as exact reference for error computation. We improve on the standard Adam-trained physics-informed neural network (PINN) baseline by introducing higher-order optimization via Gauss-Newton / Levenberg-Marquardt and L-BFGS refinements, as well as Fourier-feature PINNs, and constrained Deep Ritz (DR) formulations. In DR approaches we introduce two new neural-network formulations: a {\\it Nehari-constrained} Deep Ritz method, in which the height is learned by projecting onto the Nehari/Pokhozhaev manifold, and a {\\it Weinstein} Deep Ritz method, in which the profile shape minimizes the Weinstein quotient. The Deep Ritz formulations perform best overall, in particular, the Weinstein variant gives the best error-runtime balance. The neural network based solvers remain less accurate and efficient than classical Petviashvili iteration in 1D, but their mesh-free formulation and independence from exact profiles make them promising in higher dimensions and for more complicated nonlinearities and dispersive operators.","author":[{"family":"Haight","given":"Chandler"},{"family":"Rodriguez","given":"Alex"},{"family":"Roudenko","given":"Svetlana"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2139/ssrn.7344543","URL":"https://doi.org/10.2139/ssrn.7344543","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-8389167/v1","type":"article-journal","title":"Efficient Nudged Elastic Band Method using Neural Network Bayesian Algorithm Execution","abstract":"Abstract The discovery of a minimum energy pathway (MEP) between metastable states is crucial for scientific tasks including catalyst and biomolecular design. However, the standard nudged elastic band (NEB) algorithm requires hundreds to tens of thousands of compute-intensive simulations, making applications to complex systems prohibitively expensive. We introduce Neural Network Bayesian Algorithm Execution (NN-BAX), a framework that jointly learns the energy landscape and the MEP. NN-BAX sequentially fine-tunes a foundation model by actively selecting samples targeted at improving the MEP. Tested on Lennard-Jones and Embedded Atom Method systems, our approach achieves a one to two order of magnitude reduction in energy and force evaluations with negligible loss in MEP accuracy and demonstrates scalability to &gt;100-dimensional systems. This work is therefore a promising step towards removing the computational barrier for MEP discovery in scientifically relevant systems, suggesting that weeks-long calculations may be achieved in hours or days with minimal loss in accuracy.","author":[{"family":"Kakhandiki","given":"Pranav"},{"family":"Chitturi","given":"Sathya"},{"family":"Ratner","given":"Daniel"},{"family":"Gasiorowski","given":"Sean"}],"issued":{"date-parts":[[2026]]},"DOI":"10.21203/rs.3.rs-8389167/v1","URL":"https://doi.org/10.21203/rs.3.rs-8389167/v1","source":"crossref"},{"id":"doi:10.22541/authorea.15007191/v1","type":"article-journal","title":"A 500KV Transmission Line Fault Detection, Classification and Location Using Artificial Neural Network","abstract":"A large amount of electrical power is intended to be transported from generation point to consumption point via overhead lines. These transmission lines are often subjected to a wide range of conditions that can result in electrical faults. This paper focuses on the applications of artificial neural networks (ANN) to detect, classify, and locate faults in overhead power transmission lines. A 500KV, 300km transmission line system was designed in MATLAB-SIMULINK. A three-phase fault block was used to create four types of faults: single line-to-ground, line-to-line, double line-to-ground, and three-phase fault. The faulty current and voltage data were used as inputs to train, validate and test the ANN. The intelligent locator procedures were trained, tested, and evaluated using a multilayer neutral network with pattern recognition, employing Scale Conjugate Gradient method. The system performance was evaluated using the confusion matrix, training state plots, receiver operating curve, and Mean Square Error (MSE). The validation performance measured by MSE was 9.5885*10^-9, while the confusion matrix demonstrated 99% effectiveness. The system detected, classified, and pinpointed fault locations within 1km. The system proved to be a highly accurate method for overhead transmission line’s fault analysis. Keywords: Transmission Lines, Faults, Artificial Neural Network, Mean Square Error","author":[{"family":"Ndonkeh","given":"Yusufu"},{"family":"Mih","given":"Thomas"},{"family":"Yungho","given":"Edickson"}],"issued":{"date-parts":[[2026]]},"DOI":"10.22541/authorea.15007191/v1","URL":"https://doi.org/10.22541/authorea.15007191/v1","source":"crossref"},{"id":"doi:10.2139/ssrn.6403970","type":"manuscript","title":"Enhancing Minimization Methods for Electrical Impedance Tomography Inverse Problem Using Convolutional Neural Network Architectures","abstract":"This paper investigates the improvement of accuracy and efficiency in the Alternating Direction Method of Multipliers for recovering a non-smooth diffusion parameter in ill-posed elliptic boundary value problems, particularly in Electrical Impedance Tomography. The primary challenge is accurately identifying the diffusion parameter from noisy boundary data. This work provides a historical perspective, addresses identifiability issues, and proposes techniques to solve the problem. Specifically, it introduces an approach based on Alternating Direction Method of Multipliers, integrating Convolutional Neural Network architectures to enhance the identification process and mitigate the impact of noise. The contributions extend to multiple measurements, demonstrating the effectiveness of the proposed methodology in solving complex inverse problems in biomedical imaging, as shown in the numerical experiment section.","author":[{"family":"Anzal","given":"Oumaima"},{"family":"Guessous","given":"Najib"},{"family":"Ouakrim","given":"Youssef"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2139/ssrn.6403970","URL":"https://doi.org/10.2139/ssrn.6403970","source":"crossref"},{"id":"doi:10.26434/chemrxiv.15003886/v1","type":"manuscript","title":"An analytically differentiable submatrix-descriptor neural-network potential for molecular simulations of liquid water","abstract":"Machine-learning potentials can accelerate atomistic simulations, but their use in molecular dynamics requires descriptors that are symmetry preserving, differentiable and straightforward to evaluate. We present a compact neural-network potential based on local submatrix descriptors and apply it to liquid water. A sparsified Coulomb or sine matrix is reduced to atom-centred submatrices, which are converted into fixedlength descriptors by deterministic neighbour ranking. A single shared network maps all atomic descriptors to energy contributions, avoiding element-specific subnetworks while retaining chemical identity through the matrix elements. Because the descriptor is differentiable, conservative forces are obtained analytically by the chain rule. For periodic liquid water, five-fold cross-validation gives a lowest mean absolute energy error of 0.0609 eV atom−1 with the sine-matrix descriptor. Analytical forces agree with finite-difference derivatives of the trained model, and conservative-force dynamics preserves the qualitative oxygen–oxygen radial distribution function.","author":[{"family":"Rengaraj","given":"Varadarajan"},{"family":"Kuhne","given":"Thomas"},{"family":"Kühne","given":"Thomas"}],"issued":{"date-parts":[[2026]]},"DOI":"10.26434/chemrxiv.15003886/v1","URL":"https://doi.org/10.26434/chemrxiv.15003886/v1","source":"crossref"},{"id":"doi:10.2118/231836-pa","type":"article-journal","title":"Accelerated Permeability Upscaling: A Convolutional Neural Network Approach","abstract":"Summary Efficiently determining effective permeability involves substantial computational efforts, even when employing local upscaling techniques. This research utilizes a convolutional neural network (CNN) architecture for the rapid estimation of effective permeability. The CNN takes as input the permeability maps of specific layers from the fine-scale model earmarked for upscaling into layer(s) featuring coarser cells. Treating fine-scale permeability maps as 3D high-resolution images, the CNN produces a three- to six-channel lower-resolution image. Each channel represents the upscaled permeability map in a major direction, along with nondiagonal permeabilities of a symmetric permeability tensor. The simplicity and robustness of the proposed architecture stem from its two 3D convolutional hidden layers, characterized by varying filter numbers and kernel sizes. The effectiveness of the network was evaluated using two distinct data sets: (1) a continuous Gaussian model and (2) the Egg model with 100 permeability realizations. In the first data set, 500 geological realizations were generated and upscaled using a pressure solver with periodic boundary conditions (BCs), employing both local and extended local methods. The same upscaling procedure was applied to 100 realizations of the Egg model. A portion of the realizations was allocated for training, and the remainder was allocated for testing. The CNN exhibited a highly promising capture of nonlinear behavior, displaying no signs of overfitting. The results, visually and quantitatively assessed through an exceptionally high mean R2 score, were further validated through the computation of the pressure fields, affirming the accuracy of the estimated permeabilities. Notably, the training runtime proved significantly shorter than the computation time required for upscaling using the pressure-solver method. This proposed method holds the potential to substantially reduce upscaling computations, marking a stride toward more computationally efficient quasiglobal upscaling methods.","author":[{"family":"Sayyafzadeh","given":"Mohammad"},{"family":"Telvari","given":"Saeid"},{"family":"Guérillot","given":"Dominique"},{"family":"Sharifi","given":"Mohammad"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2118/231836-pa","URL":"https://doi.org/10.2118/231836-pa","source":"crossref"},{"id":"doi:10.2139/ssrn.6551274","type":"manuscript","title":"Multi-graph Decoupled Heterogeneous Graph Neural Network with Reinforcement Learning and Decoupled Contrastive Loss","abstract":"Emotion Recognition in Conversation (ERC) is a key task in human-computer interaction and affective computing. Currently, leveraging graph neural networks to model the complex interactions among speakers has become mainstream. However, existing methods still face three major challenges: first, it is difficult to effectively filter out emotion-irrelevant noise during feature extraction; second, the intra-modal and inter-modal dependencies are coupled within a single graph structure, limiting the model&amp;apos;s fine-grained modeling capability; third, static graph construction often introduces a large number of noisy connections, which interfere with information propagation between nodes. To address these issues, we propose a multi-graph heterogeneous graph neural network based on reinforcement learning and a decoupled contrastive loss function (RL-DCLHGNN). First, a decoupled contrastive loss function is designed in the upstream task, which separates task-relevant features from irrelevant ones by pulling positive sample pairs closer in the task-relevant feature space while pushing them away from negative samples. Second, a heterogeneous graph structure that separates intra-modal and inter-modal relationships is constructed to capture different levels of interactions respectively. Finally, reinforcement learning technology is introduced into the node connection strategy, and a dynamic sliding window mechanism is designed, enabling the model to adaptively adjust the number of connected nodes based on the current dialogue context. Extensive experiments on the IEMOCAP and MELD datasets demonstrate that RL-DCLHGNN achieves superior performance.","author":[{"family":"Xu","given":"Yunfeng"},{"family":"Chen","given":"Hao"},{"family":"Tan","given":"Haipeng"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2139/ssrn.6551274","URL":"https://doi.org/10.2139/ssrn.6551274","source":"crossref"},{"id":"doi:10.2139/ssrn.6259403","type":"manuscript","title":"An energy-efficient attention-based spiking neural network for medical image classification","abstract":"Modern medical imaging demands models capable of processing high-resolution images while maintaining computational and energy efficiency compatible with clinical deployment. Current dominant approaches based on Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) face critical limitations: CNNs lack global context modeling, while ViTs incur prohibitive computational costs for high-resolution medical images. Spiking Neural Networks (SNNs) represent a promising energy-efficient alternative, yet their adoption in medical imaging remains limited due to performance degradation from spike discretization and challenges in processing fine-grained spatial details at high resolution. To date: no hybrid architecture effectively combining SNNs with attention mechanisms has been validated on high-resolution medical data. This work addresses this gap by introducing ViA-RCSNN (Vision Attention Recurrent Convolutional Spiking Neural Network), a hybrid SNN-attention architecture that integrates convolutional feature extraction, cross-attention with patch-based positional encoding for global context, and spiking neurons for sparse temporal processing. Validated across four MedMNIST datasets covering different medical imaging domains (organ, pneumonia, breast, blood), ViA-RCSNN achieves 6x faster inference and ∼5x lower inference energy compared to ResNet-18, a widely studied CNN baseline on these benchmarks, while delivering competitive accuracy, resulting in up to 76x improvement in test-accuracy-per-training-energy. These results demonstrate that hybrid SNN architectures can effectively handle 224x224 multi-modal medical imaging, establishing them as viable energy-efficient alternatives for clinical applications.","author":[{"family":"Kenza","given":"Garreau"},{"family":"Niepceron","given":"Brad"},{"family":"Bellenger","given":"Emmanuel"},{"family":"Grassia","given":"Filippo"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2139/ssrn.6259403","URL":"https://doi.org/10.2139/ssrn.6259403","source":"crossref"},{"id":"doi:10.2139/ssrn.6545175","type":"manuscript","title":"Discovering Human Behavioral Transmission Laws in Epidemics via a Physics-Constrained Neural Network","abstract":"Epidemic transmission is shaped not only by pathogen biology but by adaptive human behavioral responses, yet existing frameworks lack a mathematically explicit, data-derived formulation of this coupling. We introduce a physics-constrained neural network and symbolic regression (PCNN-SR) framework treating epidemic modeling as an inverse problem: recovering the time-varying transmission rate [[EQUATION]] from multimodal surveillance records while enforcing epidemiological consistency. Applied to Swedish COVID-19 data, the framework distills a three-component behavioral law encoding two previously informalized mechanisms: a risk compensation effect, in which rising vaccination coverage proxies pandemic fatigue and NPI relaxation; and a surveillance bias effect, in which testing policy stringency inflates observed transmission estimates independently of viral spread. Digital twin simulations confirm that mobility-only models fail to reproduce the late-stage transmission resurgence occurring when mobility had returned to baseline, while the hybrid law reconstructs multi-wave patterns with high fidelity. This disease-agnostic methodology offers an interpretable and scalable pathway for extracting behavioral transmission laws from heterogeneous public health data.","author":[{"family":"Xu","given":"Yang"},{"family":"Han","given":"Qing"},{"family":"Kong","given":"Jude"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2139/ssrn.6545175","URL":"https://doi.org/10.2139/ssrn.6545175","source":"crossref"},{"id":"doi:10.2139/ssrn.6635176","type":"manuscript","title":"A Graph Neural Network Surrogate for Traffic Simulation in Probabilistic Wildfire Evacuation Planning","abstract":"Agent-based traffic simulation is one of the computational bottleneck in probabilistic wildfire evacuation frameworks that rely on Monte Carlo analysis to populate Bayesian network models. To address this challenge, this paper presents a graph neural network (GNN) based traffic simulator that replaces agent-based counterparts within wildfire egress frameworks such as WiSE (Wildfire Safe Egress). The surrogate operates on a graph representation of the evacuation danger zone, constructed automatically from openly available road network and population data. A two-branch architecture processes static spatial features (distances to destinations and fire) through graph convolutional layers and dynamic vehicle occupancy through a graph convolutional LSTM, predicting next-step node occupancy across the network. The model is trained on a single agent-based simulation scenario from the 2018 Camp Fire in the Paradise-Magalia area of California and validated on a second scenario with a different destination configuration. Results show that the surrogate predicts occupancy dynamics, generates plausible evacuation routes, and produces departure-travel time distributions suitable for Bayesian network calibration. When integrated with WiSE, the surrogate yields a safe evacuation estimate of 15.9%, consistent with the 16% obtained from the agent-based reference, confirming that the surrogate preserves the decision-relevant outputs of the wildfire evacuation framework.","author":[{"family":"Pishahang","given":"Mohammad"},{"family":"Rodriguez","given":"Eduardo"},{"family":"Droguett","given":"Enrique"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2139/ssrn.6635176","URL":"https://doi.org/10.2139/ssrn.6635176","source":"crossref"},{"id":"doi:10.2139/ssrn.6985309","type":"manuscript","title":"NODE LEVEL CAPSULE GRAPH NEURAL NETWORK WITH ARTIFICIAL PROTOZOA OPTIMIZATION APPROACH","abstract":"Hybrid renewable energy systems (HRES) are frequently employed to combine several energy sources, like the sun, wind, and hydroelectric, to provide reliable power. These systems are essential for optimizing energy distribution and enhancing stability, especially in decentralized grids. However, challenges like handling rapid load changes and maintaining consistent voltage levels remain, requiring further improvements in control strategies. In this, Droop Control-Based Optimization of Hybrid Renewable Energy Systems: Combining Node-Level Capsule Graph Neural Network with Artificial Protozoa Optimizer (DCHRES-NLCGNN-APO) is proposed in this chapter. First, the input information is gathered from the ’Dataset for renewable energy systems’ dataset. After that, a pre-processing section receives the gathered data. The data from the gathered dataset is cleaned and normalized during pre-processing using Data-Adaptive Gaussian Average Filtering (DAGAF). The feature extraction phase receives the pre-processed output for extracting relevant features like time of day, weather conditions, and historical load patterns with the help of Three-Dimensional Quantum Wavelet Transforms (3D-QWT). Finally, for effective energy management, future energy production and consumption are predicted using a Node-Level Capsule Graph Neural Network (NLCGNN). Node-Level Capsule Graph Neural Networks typically don&amp;apos;t use optimization techniques to choose the best parameters to ensure accurate prediction of feature demand. Hence, the Artificial Protozoa Optimizer (APO) is suggested in order to optimize the NLCGNN&amp;apos;s weight parameters, which accurately forecast the future energy need. The suggested DCHRES-NLCGNN-APO Python is used to implement the strategy. Using performance criteria like accuracy, error rate, precision, recall, specificity, and F1-score, the suggested method&amp;apos;s performance is evaluated in comparison to other current approaches such Artificial Neural Network-based Dual Droop Control for Virtual Synchronous Generators in Microgrid Systems (SGDC-ANN), Artificial Rabbits optimized Neural Network-based Energy Management System for Isolated DC Microgrids with PV, Batteries, and Supercapacitors (EMS-ARONN), and Deep Reinforcement Learning-Based Dynamic Droop Control Strategy for Optimal Operation and Frequency Regulation (DDCOFR-DRL).","author":[],"issued":{"date-parts":[[2026]]},"DOI":"10.2139/ssrn.6985309","URL":"https://doi.org/10.2139/ssrn.6985309","source":"crossref"},{"id":"doi:10.67535/tsp.000002.003","type":"article-journal","title":"Graph Neural Network for Context-Aware Cyber Threat Intelligence: A Hybrid Approach","abstract":"The rapid expansion of Internet of Things (IoT) devices and advanced high-speed data transmission networks has complicated modern cyber attacks to the point that it is now relatively simple for attackers to evade the traditional security controls in place. The current generation of standard Intrusion Detection Systems (IDS) has a significant limitation in that they are based on signatures, which are merely an inventory of all of the previously identified threats to computer systems. Therefore, if there is a new type of attack that has not been previously identified or is a slightly altered version of an existing attack, then the IDS will not detect it, and therefore will provide no alerts to the security personnel. Although it is not a perfect solution, one potential solution that addresses the shortcomings of IDS is Deep Learning. Deep Learning could be considered useful in identifying known attack types using Supervised Learning. However, it is significantly less effective in identifying any kind of a Zero-Day attack. In order to address this limitation, Unsupervised Learning could be applied to identify any distinct activity or pattern within a user’s behavior. However, many Unsupervised Learning techniques are unable to provide sufficient accuracy to accurately distinguish the actual type of attack being detected. Although some Unsupervised Learning techniques may be used to discover statistical anomalies, they typically do not provide specific details regarding the attack. Attention Network (GAT) and an unsupervised Graph AutoEncoder (GAE) into a single architecture, allowing us to benefit from both high-accuracy feature extraction capabilities of the supervised GAT, and the flexibility of the unsupervised GAE. Rather than treating each packet in a network as an independent and unconnected static unit of data, we define the traffic (or stream) across the entire network through temporal connection windows. The GAT serves as the primary means to detect known threats, with an accuracy of 91.24%. The GAE is used to verify the classification of ’normal’ traffic by checking for possible Zero-Day anomalies via reconstruction errors with a threshold set at 0.6. Evaluation results using the UNSW-NB15 dataset demonstrate that this multi-layered framework is more stable and has a higher ability to detect threats than traditional CNN and GCN methods.","author":[{"family":"Khadim","given":"Rubina"},{"family":"Razaq","given":"Asma"},{"family":"Muhammad","given":"Aoun"},{"family":"Fayyaz","given":"Umar"},{"family":"Raza","given":"Sehrish"}],"issued":{"date-parts":[[2026]]},"DOI":"10.67535/tsp.000002.003","URL":"https://doi.org/10.67535/tsp.000002.003","source":"crossref"},{"id":"doi:10.2523/iptc-25234-ms","type":"article-journal","title":"Physics-Informed Neural Network for Robust Petrophysical Interpretation with Learnable Physical Parameters","abstract":"Abstract Accurate estimation of porosity (ϕ), water saturation (SW), and shale volume (VSH) is crucial for quantitative reservoir characterization. Traditional deterministic models, such as Archie's equation and shaly-sand formulations (e.g., Simandoux, Waxman-Smits, and Dual Water models), rely on fixed parameters that are often uncertain, while purely data-driven machine learning models can act as black boxes and lack physical consistency. To address these issues, we introduce a physics-informed neural network (PINN) framework that combines petrophysical constraints with data-driven learning. The model uses a deep feedforward neural network as its core, with physics-based losses derived from Archie's resistivity law and gamma-ray shale volume relationships incorporated into the training process. Formation water resistivity (RW) and the cementation exponent (m) are treated as learnable parameters, allowing for automatic calibration to reservoir-specific conditions. The model is trained on 46,034 standardized well log data points from nine wells, utilizing seven input logs: compressional slowness (DTC), shear slowness (DTS), bulk density (DEN), gamma ray (GR), neutron porosity (NEU), photoelectric factor (PEF), and deep resistivity (RDEP). An independent test set of 11,275 depth samples from four additional wells is used for evaluation. To assess the contribution of the physics-informed formulation, a baseline multilayer perceptron (MLP) with the same architecture but without physics-based constraints was trained for comparison. The baseline MLP achieved R2 values of 0.9702 for ϕ, 0.9863 for Sw, and 0.9802 for VSH. In contrast, the proposed PINN achieved higher R2 values of 0.9825, 0.9911, and 0.9885, respectively, along with consistently lower mean squared errors (MSE) and mean absolute errors (MAE). These results demonstrate that incorporating physics-based constraints improves both predictive accuracy and physical consistency. The model also learns formation-specific parameters, converging to RW = 0.0541 Ω · m and m = 1.2851, enhancing interpretability and adaptability to field-specific conditions. Overall, the findings highlight that physics-informed machine learning effectively bridges the gap between deterministic petrophysical models and black-box neural networks, providing a robust and interpretable framework for reducing uncertainty in reservoir evaluation and supporting more reliable subsurface decision-making.","author":[{"family":"Acharya","given":"S"},{"family":"Fabian","given":"K"},{"family":"Westeng","given":"K"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2523/iptc-25234-ms","URL":"https://doi.org/10.2523/iptc-25234-ms","source":"crossref"},{"id":"doi:10.2139/ssrn.6731645","type":"manuscript","title":"Graph Neural Network using Temporal and Spectral Correlations for Underwater Acoustic Event Detection","abstract":"Acoustic signals measured in underwater environments exhibit complicated and diverse temporal and spectral characteristics and often involve overlapping multiple sources, making accurate acoustic event detection challenging. Most of existing approaches only use temporal information so that they have limitations in effectively capturing both temporal and spectral information of overlapping events. To overcome this limitation, we propose a graph neural network with temporal and spectral correlations (GTSC) for underwater acoustic event detection. The proposed GTSC is composed of a temporal graph and a spectral graph obtained from convolutional features, which are processed by separate tensor-weighted graph convolution. Temporal and spectral adjacency tensors for learning are generated by multiplying class-wise annotation with convolutional features, and an event-based loss function is employed for enhancing class discriminability. To improve detection performance, the GTSC aggregates class-specific features using learnable weight matrices. Experimental results using underwater acoustic datasets demonstrate that the GTSC achieves superior detection performance compared to existing algorithms under various signal-to-interference ratio (SIR) and signal-to-noise ratio (SNR) conditions.","author":[{"family":"Lee","given":"Kibae"},{"family":"Jeong","given":"Yoonsang"},{"family":"Lee","given":"Chong"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2139/ssrn.6731645","URL":"https://doi.org/10.2139/ssrn.6731645","source":"crossref"},{"id":"doi:10.26434/chemrxiv.15001313/v1","type":"manuscript","title":"Benchmarking Ice-Water Equilibria Exhibited by Foundation Neural Network Potentials","abstract":"Herein we report the melting points of ice exhibited by some recently published foundation Neural Network Potentials (NNPs) using the direct coexistence method: SO3LR, Orb v3, MACE-MP and Fennix-Bio1, as well as the first generation model ANI-2x. Using our previously published LICH-TEST algorithm, we classify the structural motifs adopted by individual molecules over time, finding that most models correctly exhibit ice growth and shrinkage via interfacial hexagonal structures. However, the observed melting temperatures vary from the experimental value significantly, with the lowest and highest nearly 130 K apart. The growth rates of ice below the melting points were also found to vary significantly, in some cases exceeding classical water models by one decade. SO3LR was the only NNP exhibiting an accurate value, and represents the best trade-off of speed and accuracy for the simulation of ice nucleation in (bio)organic systems. Disconcertingly the MACE, Orb and ANI models overestimate the melting point to such a degree that liquid water is effectively under deep supercooling when simulated at standard conditions. By comparing variants of the latter two potentials, we infer that an accurate description of dispersion interactions during training and/or evaluation improves the water density isobar and leads to slightly better melting temperatures, albeit with a substantial speed penalty when added during inference. We conclude with some general recommendations for training the next generation of foundation models, in order to improve their description of ice-water interactions.","author":[{"family":"Nilsson","given":"Rasmus"},{"family":"Roudsari","given":"Golnaz"},{"family":"Lbadaoui-Darvas","given":"Mária"},{"family":"Reischl","given":"Bernhard"},{"family":"Ingram","given":"Stephen"}],"issued":{"date-parts":[[2026]]},"DOI":"10.26434/chemrxiv.15001313/v1","URL":"https://doi.org/10.26434/chemrxiv.15001313/v1","source":"crossref"},{"id":"doi:10.20944/preprints202601.2166.v1","type":"manuscript","title":"Exploration of Asymmetric Relationships in Graph Neural Network Topology","abstract":"As an information extraction framework, graph neural network has been widely used in many fields, but there is a serious problem of over-smoothing in graph neural network, that is, with the increase of the number of iterations, the nodes gradually tending toward similarity, resulting in reduced feature distinguishability and deteriorated model performance. In order to solve this problem, this paper proposes a novel solution, that is, to alleviate the over-smoothing problem in graph neural networks from the perspective of changing the topological information dimension. This article uses graph regularization to solve this problem, by fine-tuning the topology structure, the research objectives can be achieved, and demonstrated through extensive ablation experiments, a large number of experiments verify the effectiveness and feasibility of the proposed method. Physically speaking, this method limits the connection strength of the adjacency matrix to a finite number of steps; The higher the order, the more obvious the restriction effect, therefore, it can alleviate the problem of over smoothing in graph neural networks to a certain extent.","author":[{"family":"Ma","given":"Lin"},{"family":"Wang","given":"Wenjun"},{"family":"Wang","given":"Jun"},{"family":"Ma","given":"Zhitao"}],"issued":{"date-parts":[[2026]]},"DOI":"10.20944/preprints202601.2166.v1","URL":"https://doi.org/10.20944/preprints202601.2166.v1","source":"crossref"},{"id":"doi:10.2139/ssrn.6177789","type":"manuscript","title":"GraphMux: A Graph Neural Network Framework for Encrypted Traffic Classification","abstract":"The growing dominance of encrypted network traffic and new encrypted algorithms (TLS 1.3, QUIC, DoH) poses significant challenges for reliable network classification, particularly as many existing approaches rely on text or image-based representations, which struggle to capture the inherent structural relationships present in network communication—relationships that can naturally be modeled as graphs. In this work, we introduce GraphMux, a GNN-based framework that leverages line graph transformations to fuse multiple graph views into a unified representation. We also present three graph-based flow representations (TIG+Chain, StarBurst, and 2Chain) designed to capture temporal burst dynamics and client–server interaction patterns using only packet time, direction, and length information, without incorporating any unencrypted statistical features.We evaluate our approach on several datasets: two academic datasets (UTMobileNetTraffic2021 and QUIC PCAP) and a commercial dataset (Flash), using four Graph Neural Network architectures. Across all datasets, GraphMux consistently achieves the strongest performance, and the proposed graph constructions frequently produce the leading results. Additional experiments examining attribute-selection strategies reveal a clear positive correlation between well-aligned feature assignments and classification accuracy, underscoring the importance of principled attribute design when constructing graph representations for encrypted traffic.","author":[{"family":"Klein","given":"Matan"},{"family":"Marbel","given":"Revital"},{"family":"Hajaj","given":"Chen"},{"family":"Dubin","given":"Ran"},{"family":"Dvir","given":"Amit"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2139/ssrn.6177789","URL":"https://doi.org/10.2139/ssrn.6177789","source":"crossref"},{"id":"doi:10.20944/preprints202603.1920.v1","type":"manuscript","title":"Water and Gas Flooding Oil Monitored by a Realtime Unet-Neural-Network-Based Method","abstract":"To achieve real-time and accurate detections of residual oil distribution during water or CO₂ flooding, this study utilizes the high-frequency Ground Penetrating Radar (GPR) for monitoring of the flooding process in real time. The U-Net neural networks are trained to invert for the subsurface dielectric constants and conductivity distributions. The study first utilizes the gprMax forward tool to simulate the dynamic response changes of rock electrical parameters during flooding and constructs a high-resolution training dataset of 100,000 samples. Each sample contains the relationships between a subsurface electrical parameter model and its corresponding multi-transmitter, multi-receiver GPR responses. A deep learning inversion network based on the U-Net architecture is trained to extract multi-scale features through an encoder-decoder structure, achieving an end-to-end mapping from GPR echo signals to subsurface electrical parameters. Numerical and physical core experimental results show that the method accurately inverts the electrical parameter distributions of the oil, water, and gas in the sandstone model, successfully capturing the position and morphology changes of the displacement front. The average relative error of dielectric constant inversion is controlled within 5%, with the error mainly concentrated in high-conductivity water regions for conductivity inversion results. Compared to traditional full waveform inversion methods, the proposed approach offers a fast inversion solution and is less affected by the initial model and noise. The results reveal the feasibility and superiority of the neural network based deep learning method in GPR electromagnetic inversion, providing a new method for real-time flooding monitoring and intelligent reservoir development during oil and gas flooding.","author":[{"family":"Zhang","given":"Jie"},{"family":"Cui","given":"Maolei"},{"family":"Wang","given":"Rui"}],"issued":{"date-parts":[[2026]]},"DOI":"10.20944/preprints202603.1920.v1","URL":"https://doi.org/10.20944/preprints202603.1920.v1","source":"crossref"},{"id":"doi:10.26434/chemrxiv.15005553/v1","type":"manuscript","title":"Quantitative Pathway-Resolved Kinetics from Neural Network-Guided Weighted Ensemble Simulations","abstract":"Many crucial biomolecular processes, such as ligand dissociation and conformational switching, occur via multiple competing transition pathways. While standard rare-event sampling methodologies provide rigorous estimates of the global transition rate, they typically collapse the underlying mechanistic heterogeneity into a single global constant, obscuring the relative kinetic contributions of distinct channels. Here, we introduce a path-resolved weighted ensemble (WE) framework that quantitatively assigns steady- state flux to competing transition channels without modifying the underlying unbiased dynamics. In our approach, initial reactive trajectories generated via direction-guided adaptive sampling are clustered by dynamic time warping (DTW) and smoothed into neural-network-refined representative paths. These refined routes are formulated as path collective variables (PathCVs) to pre-seed and guide channel-specific WE simulations. We demonstrate the robustness of this framework across systems of increasing complexity. On a model three-hole potential, the method accurately captures a temperature-dependent pathway crossover. In the benzamidine-trypsin complex, it isolates the dominant unbinding route while quantifying the fluxes of subdominant channels. Finally, for imatinib dissociation from Abl kinase, the method reveals that the clinically relevant N368S resistance mutation drastically redistributes reactive flux from the P-loop/hinge channel toward the αC-helix channel, accelerating escape. This framework successfully translates qualitative pathway classification into rigorous, pathway-specific kinetic observables, offering a powerful tool for understanding mutation-induced mechanistic","author":[{"family":"Maity","given":"Dibyendu"},{"family":"Shahid","given":"Shaheerah"},{"family":"Bhattacharya","given":"Sayari"},{"family":"Majumdar","given":"Rupak"},{"family":"Chakrabarty","given":"Suman"}],"issued":{"date-parts":[[2026]]},"DOI":"10.26434/chemrxiv.15005553/v1","URL":"https://doi.org/10.26434/chemrxiv.15005553/v1","source":"crossref"},{"id":"doi:10.2139/ssrn.7100538","type":"manuscript","title":"Physics Informed Neural Network Model for Flow around Encapsulated Phase Change Materials","abstract":"In this work, we present a simple physics informed neural network framework for coupled flow and heat transfer around encapsulated phase change materials. The heat transfer fluid region is solved by using the continuity, momentum, and energy equations as constraints in the neural network loss function. The phase change process is represented in a simplified way by keeping the capsule surface at the melting temperature and estimating the melting time from a lumped latent heat balance. No computational fluid dynamics or experimental field data are used during training. The model is tested for single and paired circular capsules in a two-dimensional channel. The predicted fields show the expected wake region behind the capsules, symmetry of the flow, cooling of the fluid near the capsule surface, and changes in flow passage caused by capsule spacing. The wall temperature gradients obtained from the model are used to estimate melting time and volume based energy storage rate. The study shows that the method can be used as an exploratory tool for preliminary analysis of phase change material systems.","author":[{"family":"Das","given":"Subhasish"},{"family":"Agrawal","given":"Manish"},{"family":"Rath","given":"Prasenjit"},{"family":"Bhattacharya","given":"Anirban"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2139/ssrn.7100538","URL":"https://doi.org/10.2139/ssrn.7100538","source":"crossref"},{"id":"doi:10.2139/ssrn.6830313","type":"manuscript","title":"A Double-Head Physics-Informed Neural Network with RGB Consistency for Stable Training","abstract":"Physics-Informed Neural Networks (PINNs) offer a framework for solving partial differential equations without labeled datasets by embedding governing equations into the loss function. However, their performance deteriorates as problem complexity and spatiotemporal dimensionality increases because of training instability, spectral bias, and optimization difficulty. To address this, we propose an RGB-Consistency PINN framework that learns a physical scalar field together with its RGB representation as an auxiliary output. The model employs a double-head architecture with a physics head and a visual head and combines a differentiable lookup-table (LUT)-based mapping with a stop-gradient strategy to preserve physics-constrained learning while introducing structural regularization. We evaluated the method on four benchmark problems: 2D lid-driven cavity flow, 2D Burgers’ equation, 2D heat transfer, and 3D heat transfer. For relatively simple two-dimensional problems, improvements in global mean error metrics are limited, but the method consistently improves the reconstruction of characteristic local structures such as centerline profiles, recirculation regions, and localized bumps. Its advantages become more pronounced in diffusion-dominated problems, where it yields more stable temperature-field evolution, gradient-energy decay, and root-mean-square-error (RMSE) behavior. In the 3D heat-transfer case, the method more accurately reproduces the initial Gaussian thermal structure and more faithfully captures the diffusive behavior of the reference solution in terms of variance and standard deviation; in particular, variance-trajectory tracking accuracy improves from 65.21% to 92.24% in the early-time interval. These results indicate that RGB-based auxiliary learning acts as an effective regularization strategy for improving training stability and global physical-behavior reconstruction in complex and high-dimensional PDE problems.","author":[{"family":"Jeon","given":"Byeonggyu"},{"family":"Kim","given":"Shin"},{"family":"Park","given":"Yongha"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2139/ssrn.6830313","URL":"https://doi.org/10.2139/ssrn.6830313","source":"crossref"},{"id":"doi:10.2139/ssrn.7243278","type":"manuscript","title":"Neural Network-Based Generation of Three-Dimensional Structured Elliptic Grids","abstract":"To address the limitations of traditional elliptic grid generation methods, a neural network-based framework for elliptic grid generation has been proposed. However, its applicability to three-dimensional problems remains unexplored. In this work, the framework is extended to three-dimensional surface and volumetric grid generation. The governing equations are reformulated in a form directly defined on the physical domain, where the domain corresponds to the solid region for volumetric grids and to the associated parametric domain for surface grids. A neural network is employed to solve the resulting elliptic equations subject to prescribed boundary conditions, thereby enabling the generation of structured grids for both surface and volumetric geometries. Numerical results demonstrate that the proposed method is feasible and robust in three-dimensional models and is capable of generating high-quality structured grids.","author":[{"family":"Yue","given":"Huaijun"},{"family":"Yang","given":"Zaiyou"},{"family":"Song","given":"Q"},{"family":"Wei","given":"Ning"},{"family":"Zhao","given":"Qiang"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2139/ssrn.7243278","URL":"https://doi.org/10.2139/ssrn.7243278","source":"crossref"},{"id":"doi:10.2139/ssrn.6537487","type":"manuscript","title":"Dynamic Routing in Ring-Circulant NoCs Using a Modified Hopfield Neural Network","abstract":"Efficient and reliable routing remains a fundamental challenge in Networks‐on‐Chip (NoCs), especially under dynamic traffic patterns and time‐varying link conditions that degrade performance and scalability. Although ring‐circulant topologies offer desirable structural regularity and short average path lengths, existing routing schemes often fail to fully exploit their potential due to limited adaptability and insufficient optimization capabilities. To address these limitations, this paper proposes a dynamic routing algorithm for ring‐circulant NoCs based on a modified Hopfield neural network. The method formulates path selection as an energy‐minimization problem, enabling the network to converge toward low‐cost routes while incorporating real‐time link variations. The algorithm is implemented and evaluated through simulation using multiple workloads and traffic scenarios to ensure generality. The results demonstrate that the proposed neural optimization‐based routing significantly improves throughput and reduces latency compared with conventional clockwise and adaptive routing schemes, particularly under congested or irregular traffic conditions. These findings highlight the suitability of Hopfield‐based optimization as an effective and lightweight mechanism for real‐time routing in ring‐circulant NoCs and indicate its potential for deployment in future scalable on‐chip interconnect architectures.","author":[{"family":"Rezaipour","given":"Ali"},{"family":"Moradi","given":"Mona"},{"family":"Farazkish","given":"Razieh"},{"family":"Amiri","given":"Nasrin"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2139/ssrn.6537487","URL":"https://doi.org/10.2139/ssrn.6537487","source":"crossref"},{"id":"doi:10.2139/ssrn.7177748","type":"manuscript","title":"Automated AFGL Quantum Number Assignment for CO2 Isotopologues using a Graph Neural Network","abstract":"Accurate quantum number assignment for calculated molecular energy levels is a critical bottleneck in generating line broadening parameters for comprehensive line lists for radiative transfer applications. We present an automated pipeline for assigning Air Force Geophysics Laboratory (AFGL) quantum numbers to CO$_2$ calculated rovibrational states lying below 15000 \\cm\\ across all 12 stable isotopologues. A GraphSAGE graph neural network is trained transductively on empirical (\\textsc{MARVEL}) energy levels, exploiting inter-isotopologue perturbation chains and intra-isotopologue rotational ladder edges to propagate assignment information to unlabelled calculated states. Physical uniqueness is enforced locally by a Hungarian algorithm solver operating within groups of states sharing the same polyad, rotational quantum number, and parity. A five-generation bootstrap loop iteratively promotes high-confidence predictions into the training set, expanding coverage without additional labelling effort. The pipeline assigns 244~001 previously unlabelled states over 12 isotopologues, accounting for 11.44~\\% of all available states, with coverage now increased to 97.4~\\% below 5~000~cm$^{-1}$. The pipeline includes a novel decision tree method for converting AFGL to Herzberg notation in asymmetric isotopologues, while the architecture and Hungarian uniqueness enforcement are applicable beyond CO$_2$: any molecular system with a conserved polyad-like quantum number and a large body of unlabelled computed states is a natural target for this approach, suggesting a pathway toward automated quantum number annotation for the next generation of large-scale computed line lists.","author":[{"family":"Barnfield","given":"Marco"},{"family":"Yurchenko","given":"Sergei"},{"family":"Tennyson","given":"Jonathan"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2139/ssrn.7177748","URL":"https://doi.org/10.2139/ssrn.7177748","source":"crossref"},{"id":"doi:10.1007/s00521-026-11865-y","type":"article-journal","title":"Dual-stage deep neural network for tooth localization and caries segmentation in panoramic dental imaging","abstract":"Abstract This study develops and evaluates an automatic AI-based system for dental caries segmentation in panoramic radiographs, contributing to the expanding research on AI applications in dental imaging. Our approach implements a two-stage process: first using the object detection model for precise tooth extraction, followed by specialized segmentation techniques for caries identification. System performance evaluation yielded exceptional results, with a Dice score of 0.926, precision of 0.921, recall of 0.931, and IoU of 0.862. These metrics demonstrate the system’s high effectiveness in accurately detecting dental caries from panoramic radiographs, indicating strong potential for clinical dental diagnostics. The proposed method offers a practical approach to panoramic caries segmentation, with results highlighting both the feasibility and efficiency of AI integration in comprehensive dental diagnostics. This work demonstrates a practical implementation of a two-stage segmentation system optimized for clinical relevance.","author":[{"family":"Vachmanus","given":"Sirawich"},{"family":"Pornprasertsuk-Damrongsri","given":"Suchaya"},{"family":"Mongkolwat","given":"Pattanasak"},{"family":"Phinklao","given":"Noppanan"},{"family":"Papasratorn","given":"Dhanaporn"},{"family":"Kitisubkanchana","given":"Jira"},{"family":"Chaikantha","given":"Sarunya"},{"family":"Arayasantiparb","given":"Raweewan"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1007/s00521-026-11865-y","URL":"https://doi.org/10.1007/s00521-026-11865-y","source":"crossref"},{"id":"doi:10.26434/chemrxiv.10001743/v3","type":"manuscript","title":"A Multitask Graph Neural Network Framework for AMES Mutagenicity Prediction","abstract":"The long, complex, and costly process of drug development makes early, reliable toxicity assessment a central challenge in pharmaceutical research. Accurately predicting drug-induced toxicity and mutagenicity is crucial for guiding decision-making and reducing late-stage attrition. Deep learning approaches can predict the results of the Ames mutagenicity test, a standard screening method; however, most existing models rely on molecular descriptors that limit structural representation, lack interpretability, and do not account for Ames results of individual bacterial strains. Graph neural networks (GNNs) are chemically intuitive and overcome these limitations. This work presents a deep learning, multitask GNN model for predicting Ames mutagenicity. The proposed approach outperforms existing models, including those from the Ames/QSAR International Challenge. Furthermore, the model is highly interpretable. Explainability analysis of atom-level contributions reveal that the model is capable of learning known structural alerts associated with mutagenicity, along with additional molecular patterns and three-dimensional structural properties. Importantly, these insights are captured in a strain-specific manner, highlighting the value of a multitask approach. Finally, the model is able to identify previously unreported molecular substructures potentially associated with mutagenicity, supporting its utility in drug design and development.","author":[{"family":"Teitgen","given":"Abigail"},{"family":"Ulzurrun","given":"Eugenia"},{"family":"Campillo","given":"Nuria"},{"family":"Hernández","given":"Eduardo"}],"issued":{"date-parts":[[2026]]},"DOI":"10.26434/chemrxiv.10001743/v3","URL":"https://doi.org/10.26434/chemrxiv.10001743/v3","source":"crossref"},{"id":"doi:10.26434/chemrxiv.15001313/v2","type":"manuscript","title":"Benchmarking Ice-Water Equilibria Exhibited by Foundation Neural Network Potentials","abstract":"Herein we report the melting points of ice exhibited by some recently published foundation Neural Network Potentials (NNPs) using the direct coexistence method: SO3LR, Orb v3, MACE-MP, FENNIX-Bio1, NEP89 and PET-MAD as well as the first generation model ANI-2x. Using our previously published LICH-TEST algorithm, we classify the structural motifs adopted by individual molecules over time, finding that most models correctly exhibit ice growth and shrinkage via interfacial hexagonal structures. However, the observed melting temperatures vary from the experimental value significantly, with the lowest and highest nearly 150 K apart. The growth rates of ice below the melting points were also found to vary significantly, in some cases exceeding classical water models by one decade. SO3LR was the only NNP exhibiting an accurate value, and represents the best trade-off of speed and accuracy for the simulation of ice nucleation in (bio)organic systems. Disconcertingly the MACE, Orb and ANI models overestimate the melting point to such a degree that liquid water is effectively under deep supercooling when simulated at standard conditions. By comparing variants of the latter two potentials, we infer that an accurate description of dispersion interactions during training and/or evaluation improves the water density isobar and leads to slightly better melting temperatures, albeit with a substantial speed penalty if added during inference. We conclude with some general recommendations for training the next generation of foundation models, in order to improve their description of ice-water interactions.","author":[{"family":"Nilsson","given":"Rasmus"},{"family":"Roudsari","given":"Golnaz"},{"family":"Lbadaoui-Darvas","given":"Mária"},{"family":"Reischl","given":"Bernhard"},{"family":"Ingram","given":"Stephen"}],"issued":{"date-parts":[[2026]]},"DOI":"10.26434/chemrxiv.15001313/v2","URL":"https://doi.org/10.26434/chemrxiv.15001313/v2","source":"crossref"},{"id":"doi:10.2139/ssrn.7229385","type":"manuscript","title":"Reduced-chemistry LES of hydrogen combustion using neural-network-based NO source-term prediction","abstract":"The defossilization of energy-intensive industries, particularly mining and steelmaking, is essential for achieving the deep emission reductions required by global climate targets. Replacing fossil fuels with hydrogen offers a promising route for high-temperature industrial processes, but this transition requires fast and accurate predictive tools capable of resolving turbulent combustion and NOx formation under high-temperature conditions. In this study, a hybrid reduced-chemistry/artificial-neural-network (ANN) framework is developed for large-eddy simulation (LES) of a pilot-scale hydrogen-fired furnace supplied with highly preheated air. Unlike ANN approaches that replace the full chemistry integrator, the present method retains finite-rate chemistry for the main hydrogen oxidation process and uses the ANN only to reconstruct the NO production rate from the resolved thermochemical state. Three chemistry treatments are compared: a detailed mechanism with direct NO source-term evaluation (M1), a reduced mechanism derived from the detailed chemistry and coupled with ANN-predicted NO production rate (M2), and the Curran hydrogen mechanism coupled with the same ANN model (M3). Zero-dimensional PSR tests show that the chemistry-integration times of M2 and M3 are approximately 17% of that of M1, while LES timing tests demonstrate an approximately sevenfold overall speedup. M2 reproduces the detailed-mechanism temperature, heat release, and major-species evolution more accurately than M3. In the furnace LES, M2 predicts outlet quantities and domain-integrated heat release close to M1 and in reasonable agreement with measurements. M1 gives the closest agreement with the measured NO emission range, followed by M2, whereas M3 shows the largest overprediction. The higher NO level in M2 is traced mainly to excessive contributions from regions above 2100 K and near-stoichiometric mixtures (0.6 &lt; Φ &lt; 1.0). The proposed framework therefore provides an efficient route for NO prediction in highly preheated industrial hydrogen furnaces, although its accuracy remains sensitive to errors in the thermochemical fields supplied by the simplified mechanism.","author":[{"family":"Zhang","given":"Pikai"},{"family":"Chindada","given":"Sony"},{"family":"Duwig","given":"Christophe"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2139/ssrn.7229385","URL":"https://doi.org/10.2139/ssrn.7229385","source":"crossref"},{"id":"doi:10.51903/elkom.v18i2.3238","type":"article-journal","title":"Klasifikasi Jenis Bunga Menggunakan Algoritma Convolutional Neural Network (CNN)","abstract":"This research aims to develop a multiclass flower image classification system using the Convolutional Neural Network (CNN) algorithm with the EfficientNet architecture. The main problem addressed is the difficulty of manual identification of flower species that share high visual similarity. The research stages include collecting 17,299 flower images across 19 classes, performing data preprocessing such as image resizing, pixel normalization, and augmentation, followed by model training using the EfficientNet transfer learning approach. The model was trained for 10 epochs with an 80:20 training-validation data split. The evaluation results show that the model achieved a validation accuracy of 98.05% with a loss value of 0.0968, and an average precision, recall, and F1-score of 0.98. The trained model was then implemented into a web-based application built using the Next.js framework, enabling users to upload flower images and obtain real-time classification results via the Hugging Face API. The system successfully identified flower species with a confidence level of 99.87%. These findings demonstrate that combining a modern CNN architecture with transfer learning provides efficient and highly accurate flower classification performance, which can be effectively implemented for educational and digital conservation purposes.","author":[{"family":"Firdaus","given":"Ade"},{"family":"Djoas","given":"Dwi"},{"family":"Saputra","given":"Riefaldi"},{"family":"Anggraeny","given":"Indry"},{"family":"Ningsih","given":"Hilda"}],"issued":{"date-parts":[[2026]]},"DOI":"10.51903/elkom.v18i2.3238","URL":"https://doi.org/10.51903/elkom.v18i2.3238","source":"crossref"},{"id":"doi:10.36227/techrxiv.176945884.42345088/v1","type":"article-journal","title":"DISTIL: A Distributed Spiking Neural Network Accelerator on 2.5D Chiplet Systems","abstract":"Spiking Neural Networks (SNNs) implemented on in-memory computing (IMC) based architectures offer a promising solution for energy-efficient inference. However, the area and memory required to store temporal neuronal state membrane potentials updated by leaky-integrate fire (LIF) activation functions increase with the growing complexity of SNN models. Chiplet based 2.5D architectures provide scalability, but deploying SNNs on such systems introduces a critical design trade-off: a single global LIF module minimizes area but increases inter-chiplet communication latency, while dedicating an LIF module per layer reduces latency at the cost of excessive memory overhead. Existing approaches do not adequately address this trade-off or the placement of LIF modules on the interposer, leading to either large area overhead or communication bottlenecks on the Network-on-Interposer (NoI). This paper proposes DISTIL, a design and optimization framework for high-performance, areaefficient multi-chiplet architecture for SNN inference. DISTIL performs a design-space exploration (DSE) to jointly optimize the grouping of neural layers into shared sets of LIF tiles and their physical placement on the interposer to lower inter-chiplet traffic. Our experimental results show that DISTIL achieves up to 4.3× higher throughput per unit area (TOPS/mm 2) compared to state-of-the-art SNN accelerators while reducing LIF memory overhead by (60-90%).","author":[{"family":"Pal","given":"Pramit"},{"family":"Sharma","given":"Harsh"},{"family":"Moitra","given":"Abhishek"},{"family":"Pande","given":"Partha"}],"issued":{"date-parts":[[2026]]},"DOI":"10.36227/techrxiv.176945884.42345088/v1","URL":"https://doi.org/10.36227/techrxiv.176945884.42345088/v1","source":"crossref"},{"id":"doi:10.2139/ssrn.6621458","type":"manuscript","title":"AdapTGN: An Adaptive Temporal-Relational Graph Neural Network for Social Media Bot Detection","abstract":"Automated social media bots represent a persistent and increasingly sophisticated threat to the integrity of online information ecosystems, influencing democratic discourse, financial markets, and public health communication. Coordinated bot activity can distort electoral narratives, manipulate financial sentiment, and spread health misinformation at scale, highlighting the urgent need for robust and adaptive detection systems. However, despite extensive research, existing bot detection approaches share a key limitation: they fail to model social interactions as both temporally dynamic and relationally diverse. Traditional feature-based methods treat accounts independently, ignoring network structure. Sequential models capture temporal behavior but isolate users, overlooking coordinated patterns. Static Graph Neural Networks (GNNs) incorporate structural relationships but collapse temporal information into fixed snapshots. Temporal Graph Neural Networks (TGNNs) improve upon this by modeling evolving graphs, yet they remain limited by two critical issues: (i) reliance on a single, relationagnostic temporal decay function that cannot distinguish between fast-evolving interactions (e.g., retweets) and long-lasting associations (e.g., hashtag co-occurrence), and (ii) simplistic memory initialization that ignores valuable community-level structural priors. To address these limitations, this paper proposes AdapTGN, an Adaptive Temporal-Relational Graph Neural Network. The model introduces a relation-specific temporal decay mechanism, where distinct learnable decay parameters are assigned to different interaction types-co-hashtag (HT), retweet (RT), and mention (MN). This enables the model to capture heterogeneous temporal dynamics more accurately. Additionally, AdapTGN incorporates relation-aware temporal encodings, community-informed memory initialization using Louvain clustering, and a cross-relation gating mechanism that dynamically weights interaction types based on contextual relevance. Experimental evaluation on TwiBot-22 and Cresci-RTBust-2019 datasets demonstrates the effectiveness of the proposed approach. AdapTGN achieves an F1-score of 0.8892 and AUC-ROC of 0.9418 on TwiBot-22, outperforming the baseline TGN by 2.47 percentage points in F1 with statistically significant improvement (p = 0.031). Furthermore, the model exhibits strong temporal generalization, with only a 5.8 percentage point drop in F1 under distribution shift, compared to over 11 pp in competing methods. Ablation studies confirm the importance of each component, particularly persistent memory, temporal encoding, and multi-relational modeling. The learned decay parameters also align with real-world interaction dynamics, enhancing interpretability and practical reliability.","author":[{"family":"Teli","given":"Mukeswar"},{"family":"Neupane","given":"Saurav"},{"family":"Lohani","given":"Pratik"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2139/ssrn.6621458","URL":"https://doi.org/10.2139/ssrn.6621458","source":"crossref"},{"id":"doi:10.2139/ssrn.6349378","type":"manuscript","title":"Quantile-based Interpretable Neural Network Models: Mortality Forecasting and Actuarial Simulations","abstract":"This paper introduces a class of quantile-based, interpretable neural network (NN) models for mortality prediction: the Lee-Carter neural network (LCNN) model, the Renshaw-Haberman neural network (RHNN) model, and the simple neural network (simpleNN) model that balances simplicity with predictive performance. These models preserve the linear interpretability of classic stochastic mortality models while harnessing the flexibility of NNs to capture complex nonlinear patterns in mortality data, thereby achieving enhanced predictive performance. Leveraging a composite loss function that integrates pinball, median anchoring, and quantile-crossing penalty terms, we estimate mortality quantiles and develop an efficient simulation scheme based on interpolation. This framework provides distributional insights essential for pricing, reserving and risk management. Extensive empirical analyses across multiple populations demonstrate that the proposed models consistently outperform traditional approaches in predictive accuracy. We further illustrate their practical utility through an application to longevity swap pricing.","author":[{"family":"Qiao","given":"Yang"},{"family":"Zhang","given":"Jinggong"},{"family":"Zhu","given":"Wenjun"},{"family":"Wang","given":"Chou‐wen"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2139/ssrn.6349378","URL":"https://doi.org/10.2139/ssrn.6349378","source":"crossref"},{"id":"doi:10.31234/osf.io/z6m2k_v1","type":"article-journal","title":"A Neural Network Model of Retrieval Difficulty in Chinese Handwriting","abstract":"Chinese speakers frequently experience failures in retrieving the orthographic form of characters during handwriting, a phenomenon known as the tip-of-the-pen (TOP) state, in which the writer knows the word but cannot produce its complete written form. Empirical research has identified character-level lexical variables that predict retrieval difficulty, but there have been no computational models of orthographic retrieval in Chinese. Here we present OrthoNet, a feed-forward neural network that takes pre-trained meaning and sound embeddings as input and learns to map them onto Chinese characters through frequency-weighted, error-driven training on a large-scale lexical corpus. Using Shannon entropy over the model’s output distribution as a graded measure of retrieval uncertainty, we evaluated OrthoNet against human behavioral data from previous dictation studies of Chinese handwriting. Entropy significantly predicted both TOP rate and writing latency beyond the variance accounted for by established lexical predictors. This result suggests that orthographic retrieval difficulty can be understood as a natural consequence of the statistical structure of the learned mapping from meaning and sound to orthographic form.","author":[{"family":"Lin","given":"Weihao"},{"family":"Su","given":"Yongqi"},{"family":"Yang","given":"Yueran"},{"family":"Wang","given":"Ruiming"},{"family":"Kemp","given":"Charles"}],"issued":{"date-parts":[[2026]]},"DOI":"10.31234/osf.io/z6m2k_v1","URL":"https://doi.org/10.31234/osf.io/z6m2k_v1","source":"crossref"},{"id":"doi:10.26434/chemrxiv.10001743/v1","type":"manuscript","title":"A Multitask Graph Neural Network Framework for AMES Mutagenicity Prediction","abstract":"The process of drug screening and development is long, complex, and costly, making early and reliable assessment of toxicity a central challenge in pharmaceutical research. In this context, the ability to accurately predict drug-induced toxicity in general, and mutagenicity in particular, is crucial for guiding decision-making and reducing late-stage attrition. In particular, deep learning approaches can be utilized to predict the results of the Ames mutagenicity test, which is commonly employed for toxicity screening in drug development. Recent work has shown that deep learning multitask approaches that account for the contributions of individual bacterial strains improve mutagenicity prediction. However, most existing models rely on molecular descriptors, which are limited in their representation of molecular structure and properties and often lack interpretability. Graph neural networks are chemically intuitive and overcome many of these limitations. Here, we present a deep learning, multitask graph neural network model for predicting Ames mutagenicity. The proposed approach outperforms existing models, including those from the Ames/QSAR International Challenge, and demonstrates high sensitivity for identifying mutagenic compounds. In addition, our model is highly interpretable. GNNExplainer analysis of atom-level contributions revealed that the model learned known structural alerts associated with mutagenicity, as well as additional molecular patterns and contextual features, including three-dimensional structural properties. Importantly, these insights were captured in a strain-specific manner, highlighting the value of a multitask approach for modeling mutagenic mechanisms. Finally, the model identified previously unreported molecular substructures potentially associated with mutagenicity, supporting its utility in drug design and development.","author":[{"family":"Teitgen","given":"Abigail"},{"family":"Ulzurrun","given":"Eugenia"},{"family":"Campillo","given":"Nuria"},{"family":"Hernández","given":"Eduardo"}],"issued":{"date-parts":[[2026]]},"DOI":"10.26434/chemrxiv.10001743/v1","URL":"https://doi.org/10.26434/chemrxiv.10001743/v1","source":"crossref"},{"id":"doi:10.17794/rgn.2026.4.4","type":"article-journal","title":"SHEAR WAVE MODELLING FROM CONVENTIONAL WELL LOGS USING INTEGRATED DEEP LEARNING INTEGRATED CONVOLUTIONAL NEURAL NETWORK (I-CNN)","abstract":"Shear-wave velocity (Vs), together with compressional wave velocity, provides a crucial source of information for both geomechanical and geophysical studies. Vs data are often unavailable. Moreover, direct measurement of Vs remains relatively costly. Four machine learning algorithms were created to predict Vs from traditional well logs in order to get around these restrictions: Probability Neural Network (PNN), Multilayer Feed-Forward Neural Network (MLFFNN), Deep Feed-Forward Neural Network (DFFNN), one-dimensional Convolutional Neural Network (1D-CNN), and an Integrated Convolutional Neural Network (I-CNN). The dataset consists of two wells (19,121 data points were gathered) of authentic industrial wireline logs from two anonymized wells, provided with formal authorization from the data owner exclusively for academic research purposes, including model training, testing, and validation. There were three primary parts in the methodology: (1) pre-processing the data to get rid of noise and change it into the right format; (2) using domain knowledge to drive feature engineering and selection; and (3) training, testing, and optimizing the model. The results demonstrated that the I-CNN model in RCW-1 well achieved the best performance, with an R2 value of 0.971. When applied to the blind well (RCW-2), the I-CNN model maintained strong generalization capability, achieving an average R2 value of 0.956. These findings indicate that the I-CNN outperforms other methods in handling complex, nonlinear relationships in Vs prediction. Overall, this study contributes to the growing body of literature on machine learning applications in petrophysical analysis by introducing an integrated deep learning framework that surpasses traditional approaches.","author":[{"family":"Wibowo","given":"Rahmat"},{"family":"Yogi","given":"Ida"},{"family":"Arifianto","given":"Indra"},{"family":"Sarkowi","given":"Muh"}],"issued":{"date-parts":[[2026]]},"DOI":"10.17794/rgn.2026.4.4","URL":"https://doi.org/10.17794/rgn.2026.4.4","source":"crossref"},{"id":"doi:10.2139/ssrn.6837691","type":"manuscript","title":"Predicting Kresling Origami Mechanics under Compression Using Simplicial Neural Network","abstract":"Kresling origami structures exhibit strongly nonlinear mechanical behaviour that is difficult to predict using reduced-order models or graph neural networks (GNNs), which are limited to pairwise interactions. This work presents the first simplicial neural network (SNN) framework for predicting the nonlinear mechanical response of Kresling origami. By incorporating the Hodge Laplacian, the model explicitly captures higher-order node–edge–face interactions and panel-level constraints absent in GNNs. The deformation is parametrised by twist angle and radial scaling, enabling simultaneous prediction of force–displacement response and deformation evolution. The SNN achieves high accuracy against numerical simulations and enables efficient large-scale inference. Two distinct compression-induced deformation modes and a previously unreported geometric scaling law linking twist–strain coupling to aspect ratio and staggered angle are revealed through analysis of 30,000 designs.","author":[{"family":"Xia","given":"Fukun"},{"family":"Xu","given":"Shanqing"},{"family":"Shi","given":"Guangsi"},{"family":"Gao","given":"Zhipeng"},{"family":"Qiang","given":"Wei"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2139/ssrn.6837691","URL":"https://doi.org/10.2139/ssrn.6837691","source":"crossref"},{"id":"doi:10.1007/s42452-026-08478-4","type":"article-journal","title":"Effect of the memristor on Hopfield artificial neural networks and their application in encryption","abstract":"Abstract One way to strengthen encryption algorithms against attacks is to increase of encryption keys and the level of disorder in the encryption system. This paper investigates a cosine-based memristive model W(φ) = cos(φ) by analyzing the effects of signal amplitude, frequency, and boundary conditions. The proposed memristive element exhibits strong nonlinearity and memory properties, making it suitable for chaotic applications. The memristor is incorporated into a three-cell Hopfield neural network, where it operates as a synaptic coupling element, an external radiation sensor, and a dynamic weight controller. These mechanisms generate rich chaotic dynamics, which are verified using bifurcation diagrams, Lyapunov exponents, and phase space analysis. Numerical simulations in MATLAB are experimentally validated using an ESP32 microcontroller, with chaotic signals observed on an oscilloscope. Based on the proposed Hopfield, an image encryption scheme is developed and evaluated. The encryption results demonstrate high security performance, with entropy of 7.9977, near-zero pixel correlation, NPCR of 99.64%, and PSNR of 6.77. The complete system is practically implemented using an ESP32, a Raspberry Pi, and a computer, confirming its feasibility for real-time secure communication.","author":[{"family":"Abbas","given":"Abu"},{"family":"Hussain","given":"Mayada"},{"family":"Abbass","given":"Ghida"},{"family":"Dehingia","given":"Kaushik"},{"family":"Choudhary","given":"Santosh"},{"family":"Baluguri","given":"Suresh"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1007/s42452-026-08478-4","URL":"https://doi.org/10.1007/s42452-026-08478-4","source":"crossref"},{"id":"doi:10.37385/3n6z5n26","type":"article-journal","title":"Early Detection of Foetal Pathological Conditions with Neural Network Method: Implementation of Backpropagation Neural Network and SMOTE on Cardiotocography Data","abstract":"This research focuses on the development of an effective classification model for early detection of foetal pathological conditions using Cardiotocography (CTG) data by utilising the Backpropagation Neural Network (BPNN) method. The high maternal mortality rate (MMR) and infant mortality rate (IMR) in Indonesia, including Riau Province, emphasise the importance of accurate prenatal diagnosis. The main challenge of this research is to address the class imbalance issue in the CTG dataset, which is biased towards the Normal class (77.9%) compared to the Suspect (13.9%) and Pathological (8.2%) classes. This problem was addressed by applying the Synthetic Minority Oversampling Technique (SMOTE). The model's performance was evaluated using K-Fold Cross Validation (5-Fold and 10-Fold). The test results showed that the combination of BPNN and SMOTE significantly improved performance, achieving a highest average accuracy of 92.66% and a maximum accuracy of 94.84% in the 10-Fold Cross Validation scheme. The resulting model is stable, has a high generalisation capability, and has great potential to be integrated into an Artificial Intelligence (AI)-based Clinical Decision Support System (CDSS) to support evidence-based health policies in reducing Maternal Mortality Rate (MMR) and Infant Mortality Rate (IMR).","author":[{"family":"Haerani","given":"Elin"},{"family":"Syafria","given":"Fadhilah"},{"family":"Novriyanto","given":"Novriyanto"},{"family":"Marzuki","given":"Ismail"}],"issued":{"date-parts":[[2026]]},"DOI":"10.37385/3n6z5n26","URL":"https://doi.org/10.37385/3n6z5n26","source":"crossref"},{"id":"doi:10.51353/hf480q42","type":"article-journal","title":"Prediksi Prognosis Kanker Payudara Menggunakan Hybrid Artificial Neural Network Dan Gaussian Naïve Bayes","abstract":"Breast cancer is one of the types of cancer that causes the most deaths in women in the world. Breast cancer prognosis is important to assist medical personnel in predicting the possibility of recurrence so that treatment can be provided more effectively. This study aims to implement a hybrid Artificial Neural Network (ANN) and Gaussian Naïve Bayes method for breast cancer prognosis prediction using the Breast Cancer Wisconsin Prognostic (WPBC) dataset. The dataset consisted of 198 patient records with 35 numerical features. The research stages included data preprocessing, normalization, splitting the dataset into training and testing data using an 80:20 ratio, feature extraction using ANN, and classification using Gaussian Naïve Bayes. Unlike previous studies that generally used single methods, this study utilizes ANN as a feature extractor before the classification process using Gaussian Naïve Bayes. ANN was used with one hidden layer containing 16 neurons to learn non-linear relationships among features before the classification process. Model evaluation was conducted using a confusion matrix with accuracy, precision, recall, and F1-score metrics. The experimental results showed that the hybrid ANN-Gaussian Naïve Bayes method achieved an accuracy of 90%, precision of 72.73%, recall of 88.89%, and an F1-score of 80%. These results indicate that the hybrid method provides better classification performance compared to single methods in breast cancer prognosis prediction.","author":[{"family":"Hutagaol","given":"Jesika"},{"family":"Yohanna","given":"Margaretha"},{"family":"Simanullang","given":"Harlen"}],"issued":{"date-parts":[[2026]]},"DOI":"10.51353/hf480q42","URL":"https://doi.org/10.51353/hf480q42","source":"crossref"},{"id":"doi:10.1088/1674-1056/ae68f9","type":"article-journal","title":"Symmetrical Turing instability in Chua corsage memristor siblings-based two-cell network","abstract":"Abstract The Turing instability, a counterintuitive phenomenon in which two quiescent cells lose stability when coupled through a dissipative environment, has been explained via the edge of chaos theory. While the classical Turing instability and its local form have been recently elucidated, its symmetrical form — a distinct class of symmetry-breaking phenomena wherein two identical cells, each poised at a mirror-symmetrical stable operating point, undergo destabilization and bifurcate into two distinct mirror-symmetrical stable states under opposite bias voltages — has not been reported yet. This paper introduces a current-controlled odd-symmetrical Chua corsage memristor (OS-CCM) and employs it to investigate the symmetrical Turing instability in a resistively coupled two-cell network. Coupling two identical bistable OS-CCM-based cells, each originally poised at identical mirror-symmetrical stable states, via a passive resistor destabilizes their original stability, giving rise to two distinct stable states and ultimately leading to quadristability, referring to a dynamical phenomenon with four coexisting stable states, which demonstrates the emergence of symmetrical Turing instability. The quantitative condition for the emergence of this phenomenon is analytically derived and precisely determined through eigenvalue analysis. Both numerical simulations and hardware experiments confirm the correctness of the theoretical analysis.","author":[],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/1674-1056/ae68f9","URL":"https://doi.org/10.1088/1674-1056/ae68f9","source":"crossref"},{"id":"doi:10.2139/ssrn.6793368","type":"manuscript","title":"Physics-Informed Graph Neural Network Surrogate for Steady-State Gas Network Simulation and Feasibility Analysis","abstract":"Large-scale gas-network scenario evaluation is essential for integrated multi-carrier energy-system planning, particularly as gas infrastructure becomes increasingly coupled with power, heat, and hydrogen pathways. Conventional nonlinear hydraulic solvers provide high-fidelity feasibility assessment but are computationally expensive for stochastic screening and large scenario ensembles. Conversely, unconstrained learning-based surrogates can generate hydraulically infeasible states by violating mass balance or pressure consistency. This study develops a physics-informed graph neural network (PI-GNN) surrogate for steady-state gas-network simulation and feasibility assessment. The proposed model uses an edge-centric architecture to predict pipe-level squared-pressure differences and flows. A differentiable projection layer enforces nodal mass conservation on predicted flows, while Laplacian-based reconstruction maps edge pressure differences to topologically consistent nodal pressures.The framework is evaluated on GasLib-134, GasLib-135, and GasLib-582 using stochastically generated operating scenarios. On the meshed 582-node network trained with 5000 scenarios, the PI-GNN achieves a pressure mean absolute error of 1.05 bar, corresponding to 1.3% of the realized pressure range, with (R2 = 0.981). Projected-flow predictions achieve (R2 = 0.972), while mass-balance residuals are reduced to numerical precision, on the order of 10-5–10-4 Nm3/s. Compared with the MYNTS reference solver, the surrogate reduces per-scenario runtime from seconds to milliseconds, evaluating the largest benchmark in less than 40 ms. A multi-regional loadability study further shows that the PI-GNN reproduces feasibility boundaries and identifies dominant bottleneck structures consistent with reference hydraulic simulations. The proposed framework provides a constraint-aware planning accelerator for high-volume scenario screening, sensitivity analysis, and prioritization of cases requiring detailed hydraulic verification.","author":[{"family":"Jiang","given":"Dongrui"},{"family":"Garcke","given":"Jochen"},{"family":"Akca","given":"Okan"},{"family":"Hollnagel","given":"Jeremias"},{"family":"Klaassen","given":"Bernhard"},{"family":"Anvari","given":"Mehrnaz"},{"family":"Müller-Kirchenbauer","given":"Joachim"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2139/ssrn.6793368","URL":"https://doi.org/10.2139/ssrn.6793368","source":"crossref"},{"id":"doi:10.52972/hoaq.vol17no1.p76-84","type":"article-journal","title":"PENERAPAN ALGORITMA ARTIFICIAL NEURAL NETWORK UNTUK KLASIFIKASI KUALITAS BUAH APEL","abstract":"Penelitian ini bertujuan untuk mengklasifikasikan kualitas buah apel menggunakan algoritma Artificial Neural Network (ANN) sebagai solusi terhadap keterbatasan metode seleksi manual yang cenderung subjektif, memakan waktu, dan kurang efisien pada skala industri. Dataset yang digunakan diperoleh dari situs Kaggle dan terdiri dari 4.000 data sampel apel dengan tujuh atribut fisik: ukuran, berat, kemanisan, kerenyahan, kejuicy-an, kematangan, dan keasaman. Data dibagi menjadi tiga bagian menggunakan metode stratified hold-out sampling, yaitu 70% untuk pelatihan, 15% untuk validasi, dan 15% untuk pengujian. Model ANN yang diterapkan menggunakan arsitektur Multilayer Perceptron (MLP) dengan dua hidden layer berisi 20 dan 10 neuron serta dilatih menggunakan algoritma Scaled Conjugate Gradient. Evaluasi dilakukan menggunakan confusion matrix dan metrik seperti akurasi, presisi, recall, dan F1-score. Hasil menunjukkan bahwa model ANN mencapai akurasi 92% pada data pengujian dengan performa optimal pada epoch ke-106. Perbandingan dengan algoritma lain seperti Random Forest (88,3%) dan Support Vector Machine (74,7%) menunjukkan keunggulan ANN dalam klasifikasi kualitas apel. Kurva ROC dan histogram error memperkuat bukti bahwa model memiliki generalisasi yang baik tanpa overfitting. Penelitian ini membuktikan bahwa ANN dapat menjadi solusi efektif dan efisien dalam otomasi penilaian mutu buah apel serta berpotensi diterapkan lebih luas pada produk hortikultura lainnya. This study aims to classify apple fruit quality using the Artificial Neural Network (ANN) algorithm as a solution to the limitations of manual selection methods, which tend to be subjective, time-consuming, and inefficient on an industrial scale. The dataset used was obtained from Kaggle and consists of 4,000 apple samples with seven physical attributes: size, weight, sweetness, crunchiness, juiciness, ripeness, and acidity. The data was divided into three parts using the stratified hold-out sampling method: 70% for training, 15% for validation, and 15% for testing. The ANN model implemented uses a Multilayer Perceptron (MLP) architecture with two hidden layers containing 20 and 10 neurons, and it is trained using the Scaled Conjugate Gradient algorithm. Evaluation was performed using a confusion matrix and metrics such as accuracy, precision, recall, and F1-score. The results show that the ANN model achieved 92% accuracy on the test data, with optimal performance reached at epoch 106. Comparisons with other algorithms such as Random Forest (88.3%) and Support Vector Machine (74.7%) demonstrate the superiority of ANN in classifying apple quality. The ROC curve and error histogram further confirm that the model has good generalization capabilities without overfitting. This study demonstrates that ANN is an effective and efficient solution for automating the assessment of apple fruit quality and has the potential to be more broadly applied to other horticultural products.","author":[{"family":"Mega"},{"family":"Marselina","given":"Nia"},{"family":"Chang","given":"Olivia"}],"issued":{"date-parts":[[2026]]},"DOI":"10.52972/hoaq.vol17no1.p76-84","URL":"https://doi.org/10.52972/hoaq.vol17no1.p76-84","source":"crossref"},{"id":"doi:10.54914/jtt.v12i1.2597","type":"article-journal","title":"Klasifikasi Buah Kelapa Sawit dengan Convolutional Neural Network Arsitektur Inception-v4","abstract":"Industri kelapa sawit memegang peranan penting dalam perekonomian Indonesia, sehingga proses klasifikasi buah berdasarkan tingkat kematangan menjadi krusial untuk menjamin kualitas produksi minyak sawit. Penelitian ini bertujuan mengembangkan sistem klasifikasi buah kelapa sawit menjadi kategori matang dan mentah dengan menggunakan algoritma Convolutional Neural Network berbasis arsitektur Inception-v4. Dataset penelitian terdiri dari 2.900 gambar, yang terbagi menjadi data pelatihan (2.000), validasi (500), dan pengujian (400). Tahapan penelitian meliputi pengumpulan data, pre-processing (deteksi duplikat, augmentasi, dan normalisasi), pemodelan menggunakan Inception-v4, pelatihan model, evaluasi, dan interpretasi hasil. Evaluasi dilakukan menggunakan metrik accuracy, precision, recall, f1-score, dan confusion matrix. Hasil menunjukkan bahwa model Inception-v4 mampu mengklasifikasikan buah kelapa sawit dengan akurasi validasi tertinggi sebesar 95%. Eksperimen dengan berbagai optimizer (SGD, Adam, RMSprop, Adagrad, Adadelta) dilakukan untuk meningkatkan performa. Penelitian ini menunjukkan bahwa Inception-v4 efektif dalam klasifikasi buah kelapa sawit dan dapat diterapkan pada industri perkebunan untuk meningkatkan efisiensi panen dan produksi.","author":[{"family":"Seran","given":"Theresia"},{"family":"Prastya","given":"Septyan"},{"family":"Zulfadhilah","given":"Muhammad"},{"family":"Ansari","given":"Rudy"}],"issued":{"date-parts":[[2026]]},"DOI":"10.54914/jtt.v12i1.2597","URL":"https://doi.org/10.54914/jtt.v12i1.2597","source":"crossref"},{"id":"doi:10.1002/eng2.70678","type":"article-journal","title":"Adaptive Calculation Method of Line Loss of Distribution Network Based on Genetic Algorithm and Artificial Neural Network","abstract":"ABSTRACT Traditional methods for calculating distribution line losses often rely on complex mathematical models and extensive operational data, leading to cumbersome processes that struggle to ensure accuracy and timeliness in practical applications. To address these limitations, this paper proposes an adaptive calculation method for distribution line losses based on a genetic algorithm (GA) and an artificial neural network (ANN). The approach integrates automated data preprocessing using MATLAB's mapminmax function for feature normalization, an improved k ‐means clustering algorithm for data grouping, and a genetic algorithm to optimize the initial weights and thresholds of a backpropagation (BP) neural network. A 16‐4‐1 network structure is constructed, and a time‐division calculation model is introduced to enable simultaneous computation across multiple lines. Experimental results demonstrate that the proposed method significantly reduces the mean squared error and enhances the accuracy and efficiency of line loss calculation. The hybrid GA‐BP model combines global search capability with nonlinear fitting, enhances data preprocessing and clustering techniques, and improves data quality and model generalization ability.","author":[{"family":"Chen","given":"Tianjun"},{"family":"Deng","given":"Tao"},{"family":"Wu","given":"Junyang"},{"family":"Wen","given":"Ming"},{"family":"Liu","given":"Chengming"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1002/eng2.70678","URL":"https://doi.org/10.1002/eng2.70678","source":"crossref"},{"id":"oa:W4387459225","type":"article-journal","title":"Memory Technology: Development, Fundamentals, and Future Trends","abstract":"The unprecedented development in the fields of artificial intelligence (AI), big data, and the internet of things (IoT) has been booming the expansion of the digital universe where data are growing at astronomical rates. The ever-increasing performance of computing systems is aspiring advanced memory technologies for both storage and computing applications. As the current mainstream non-volatile memory (NVM) technology, flash memory will be extended and continue to dominate for several years. Alternative memory technologies exploiting new materials and concepts to go beyond flash memory for standalone as well as embedded applications are consistently pursued to either replace the classical memory solutions or to fill the gap in conventional memory hierarchies. This chapter reviews the development and fundamentals of solid-state NVM technologies including the mainstream flash memory and the most promising alternative memory technologies. Finally, the future landscape and applications of NVM technologies are discussed.","author":[{"family":"Wang","given":"Zongwei"},{"family":"Cai","given":"Yimao"}],"issued":{"date-parts":[[2023]]},"DOI":"10.1039/bk9781839169946-00001","URL":"https://doi.org/10.1039/bk9781839169946-00001","source":"openalex"},{"id":"oa:W4405914892","type":"article-journal","title":"Recent Progress in Tactile Sensing and Machine Learning for Texture Perception in Humanoid Robotics","abstract":"ABSTRACT Humanoid robots have garnered substantial attention recently in both academia and industry. These robots are becoming increasingly sophisticated and intelligent, as seen in health care, education, customer service, logistics, security, space exploration, and so forth. Central to these technological advancements is tactile perception, a crucial modality through which humanoid robots exchange information with their external environment, thereby facilitating human‐like behaviors such as object recognition and dexterous manipulation. Texture perception is particularly vital for these tasks, as the surface morphology of objects significantly influences recognition and manipulation abilities. This review addresses the recent progress in tactile sensing and machine learning for texture perception in humanoid robots. We first examine the design and working principles of tactile sensors employed in texture perception, differentiating between touch‐based and sliding‐based approaches. Subsequently, we delve into the machine learning algorithms implemented for texture perception using these tactile sensors. Finally, we discuss the challenges and future opportunities in this evolving field. This review aims to provide insights into the state‐of‐the‐art developments and foster advancements in tactile sensing and machine learning for texture perception in humanoid robotics.","author":[{"family":"Yu","given":"Longteng"},{"family":"Liu","given":"Dabiao"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1002/idm2.12233","URL":"https://doi.org/10.1002/idm2.12233","source":"openalex"},{"id":"oa:W4409282579","type":"article-journal","title":"A materials- and devices-centric approach to neuromorphic computing","abstract":"The fundamental physics of spintronic devices makes them an excellent technology for realizing neuromorphic computing hardware. At the heart of all spintronic devices is a magnetic material that can be driven out of equilibrium with spin torques and host dynamics, such as phase transitions and criticality, self-oscillations, synchronization, stochastic resonance, and chaos, which form the basis of many brain algorithms. This paper highlights the potential of magnetic tunnel junctions for realizing the key building blocks, including on-chip non-volatile memory, probabilistic bits, stochastic oscillators, coherent oscillators, of echo-state networks, spiking neural networks, and population coding neural systems for edge AI applications. Antiferromagnetic spintronics offers a new paradigm to exploit a plethora of antiferromagnetic materials for energy-efficient computing applications. We present calculations of device performance, energy dissipation, and offer insights into technology-device co-optimization. System-level calculations of the performance of a spintronics-based spiking neural network for practical workloads that uses spiking antiferromagnetic neurons with auto-reset functionality, ferromagnetic synapses, and electrical interconnects are discussed. Our calculations reveal that the energy consumption of the network is limited by the ferromagnetic synapses and electrical interconnects, while the latency is dominated by neurons. Yet, spintronics spiking networks perform at a fraction of energy and latency cost compared to CMOS-only solutions, while spintronics systems are also much more area efficient. The paper concludes by summarizing the limits, challenges, and opportunities of spintronics for edge AI applications.","author":[{"family":"Rakheja","given":"Shaloo"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1145/3676536.3697133","URL":"https://doi.org/10.1145/3676536.3697133","source":"openalex"},{"id":"oa:W4385670534","type":"article-journal","title":"Key Concepts, Technologies, Current Challenges and Research Areas of Telecommunication Engineering and Neural Communication","abstract":"The field of telecommunication engineering has undergone remarkable progress, focusing on enhancing data transfer speed and reliability. Despite these advancements, challenges remain, particularly concerning the growing demand for bandwidth and the need for more efficient communication systems. In contrast, neural communication, a burgeoning field, seeks to leverage neural networks' principles in communication systems. The convergence of telecommunication engineering and neural communication holds promise for revolutionary advancements in communication methods. This research paper presents an overview of telecommunication engineering, encompassing key concepts, technologies, current challenges, and research areas. It also addresses neural communication, covering its key concepts, technologies, current challenges, and research areas. Furthermore, the paper explores the intersection of telecommunication engineering and neural communication, highlighting ongoing research and development, potential future developments, and potential applications. The paper concludes by summarizing key findings and recommending directions for future research in this exciting area.","author":[{"family":"Haldorai","given":"Anandakumar"}],"issued":{"date-parts":[[2023]]},"DOI":"10.53759/0088/jbsha202303004","URL":"https://doi.org/10.53759/0088/jbsha202303004","source":"openalex"},{"id":"oa:W4327603999","type":"article-journal","title":"Review of visual reconstruction methods of retina-like vision sensors","abstract":"Retina-like vision sensors have been a new development in recent years. They output an asynchronous spike or event signal by continuously sampling the scene light intensity information. This research summarizes the sampling principle of existing retina-like vision sensors (including spatial contrast, temporal contrast, and temporal integration vision sensors) and retina-like cameras with additional circuits. Compared to traditional cameras, retina-like vision sensors have high dynamic ranges and high temporal domain resolution; however, as the spike signal is stored as 1 and 0 (or $-1$), compatibility with traditional vision is difficult to obtain. If neuromorphic vision sampling can be considered as the encoding process of the scene light intensity, then the visual information reconstruction is the decoding process. Thus, visual information reconstruction can bridge the gap between neuromorphic vision and traditional frame-based vision. Recently, many image reconstruction algorithms have used asynchronous spikes or event signals. By processing such spikes or event signals, an image of high temporal domain resolution can be reconstructed. In this research, according to the sampling characteristics of different retina-like vision sensors, we summarize the existing image reconstruction methods based on event and spike cameras. Moreover, the opportunities and challenges of retina-like visual sampling and visual information reconstruction, as well as their possible development directions, are discussed and prospected.","author":[],"issued":{"date-parts":[[2023]]},"DOI":"10.1360/ssi-2021-0397","URL":"https://doi.org/10.1360/ssi-2021-0397","source":"openalex"},{"id":"doi:10.18154/rwth-2026-03904","type":"article-journal","title":"Dynamical systems and paradigms for bio-inspired computing","abstract":"A complex system consists of many interacting elements whose dynamics assumes different spatio-temporal scales, examples of which ranges from neural networks in living matter, neuromorphic computing systems mimicking the brain, coupled memristor systems, conservative and dissipative nonlinear dynamical systems, Earth's climate system, economy, psycho-social phenomenon, and power grid networks, to name a few; nonlinear dynamical systems are of particular interest for modeling of biological, physical and computational systems. In nonlinear dynamical systems, attracting states emerge as long-term behaviors, and multiple attractors can coexist for the same set of system parameters. The emergence and multiplicity of attractors is one of the paradigms how neural networks store patterns as associative memory.In this thesis, we analyze the emergent dynamical behavior of systems of coupled excitable elements with inertia, pendula networks, and study the effects of coupling on the collective behavior. The counterintuitive network attractors, the chimera states, are observed as frequency clustering of the pendula into synchronous manifolds. The stability of the chimeras, as well as fixed points, are studied for various network sizes. The minimal system composed of three pendula has been extensively studied where three types of distinct chimera patterns are identified, with their respective areas of existence and multiplicity regions are found. The switching dynamics between attractors in the region of multiplicity, driven by input and noise, are qualitatively examined with the hypothesis that this sequence of switchings may serve a role in neural computation. Finally, we discuss the experimental setup of minimal metronome network and the resulting chimera states. In conclusion, this thesis provide insights into the potential of attractor dynamics and switching behaviors in complex systems, suggesting applications for neural computation and advancing our understanding of emergent patterns.","author":[{"family":"Ebrahimzadeh","given":"Pezhman"}],"issued":{"date-parts":[[2026]]},"DOI":"10.18154/rwth-2026-03904","URL":"https://doi.org/10.18154/rwth-2026-03904","source":"datacite"},{"id":"doi:10.13023/etd.2026.316","type":"article-journal","title":"Investigation of Composition-Structural Complexity as a Design Tool for Oxidation Resistant and Soft Magnetic High-Entropy Alloy Thin Films","abstract":"This dissertation investigates two functional applications of high-entropy alloys (HEAs) in thin-film form, unified by the question of how their multi-element character can be turned to engineering advantage at the nanoscale. The two parts address distinct application domains but share a common framework of magnetron-sputter deposition, multi-modal nanoscale characterization, and quantitative comparison against experimental standards or theoretical predictions. Part I examines the intermediate-temperature oxidation behavior of a refractory multi-principal-element alloy (RMPEA) thin-film system. Magnetron-sputtered Cr₁₀(MoNbTaW)₉₀ films, with and without a 35 nm aluminum capping layer, were annealed in air at 300, 400, and 500 °C for 10 and 20 hours and characterized by XRD, FIB-SEM, TEM with SAED and STEM-EDS, XPS, and nanoindentation. Uncoated films developed a thick amorphous mixed-oxide layer containing all five metals in non-segregated form, with substantial mechanical degradation (hardness 18 to 12.5 GPa, modulus 330 to 240 GPa at 500 °C). Al-capped films instead formed a compact, nanocrystalline γ-Al₂O₃ layer ~20 nm thick that inhibited oxygen ingress and preserved mechanical integrity throughout the test matrix. The XPS-derived oxidation hierarchy was supported by the FactSage-predicted Ellingham diagram. The work establishes that thin metallic-aluminum capping layers can kinetically lock in a protective scale on RMPEAs, with direct implications for next-generation high-temperature structural applications. Part II addresses the suitability of soft-magnetic HEA thin films as a permalloy alternative for artificial-spin-ice (ASI)-based neuromorphic and reservoir computing. Magnetron-sputtered Al₀.₂₅CrFeCoNi films (10–30 nm), capped in situ with ≈5 nm Al, were characterized by SQUID magnetometry (5–380 K) and broadband FMR spectroscopy (15–20 GHz). The films were ferromagnetic at room temperature with saturation magnetization ~200 emu/cc and coercivity ~30–250 Oe- a marked deviation from the paramagnetic bulk alloy, indicating that sputter deposition produces an ordering temperature substantially elevated above the bulk value. ZFC/FC M–T data showed thickness-independent magnetization at 5 K crossing over to thickness-dependent behavior above 150 K, consistent with surface-anisotropy-dominated behavior and yielding a surface anisotropy Ks of order 0.1 mJ m⁻². Field-swept FMR on the Al-seeded 15 nm film returned g = 1.97 ± 0.004, μ₀Meff = 0.40 ± 0.01 T, and Gilbert damping α = 0.019 ± 0.001 with μ₀ΔH₀ = 51 ± 16 Oe- roughly three times the zero-temperature first-principles prediction of Kudrnovský et al. and 2.4 times the permalloy benchmark. The magnetic transition temperature of 250–275 K provides operating margin near room temperature. To the author's knowledge, this is the first thickness-dependent SQUID study of AlₓCrFeCoNi thin films below 50 nm, the first FMR-derived Gilbert damping value for any AlₓCrFeCoNi composition, and the first direct experimental test of the Kudrnovský prediction for this system. A custom broadband VNA-FMR spectrometer was developed in parallel and validated against a 25 nm permalloy reference film. Together, the two projects demonstrate that process-compositional-structural complexity is instrumental in tuning passivation potential and magnetic behavior in high-entropy alloys, overcoming limitations of conventional alloys.","author":[{"family":"Noor","given":"Md"}],"issued":{"date-parts":[[2026]]},"DOI":"10.13023/etd.2026.316","URL":"https://doi.org/10.13023/etd.2026.316","source":"datacite"},{"id":"doi:10.48550/arxiv.2606.20414","type":"manuscript","title":"ExSpike: A General Full-Event Neuromorphic Architecture for Exploiting Irregular Sparsity with Event Compression","abstract":"Spiking neural networks (SNNs) promise energy-efficient computing due to their sparse spatio-temporal activity. However, effectively translating such irregular sparsity into practical performance and energy gains remains challenging, as full-event computing architectures are still underexplored. This paper proposes ExSpike, a general full-event neuromorphic architecture that fully exploits irregular sparsity in SNNs. To realize pure event-driven execution, we first propose a set of dataflow optimizations to ensure that the inputs to each SNN layer remain spike-based, thereby enabling full-event execution throughout the network. We then design a hardware-efficient full-event architecture, named ExSpike, which supports the optimized pure event-driven dataflow and an additional Attention Core for spike-driven self-attention. To further improve computing efficiency, we introduce adjacent-position event compression to reduce redundant accumulations across spatially adjacent spike sequences. ExSpike is implemented on an AMD Xilinx Virtex-7 FPGA and evaluated on both classification and segmentation workloads. Experimental results show that ExSpike achieves high normalized energy efficiency across diverse SNN models while maintaining competitive accuracy, delivering up to 479.15 GOPS, 281.85 GOPS/W, and 0.80 GOPS/W/PE. In particular, ExSpike achieves up to 10$\\times$ higher PE-normalized energy efficiency than the SOTA FPGA-based SNN accelerator (FireFly-T). The code for ExSpike is available at https://github.com/xiaoyuehai/ExSpike.","author":[{"family":"Chen","given":"Yuehai"},{"family":"Merchant","given":"Farhad"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2606.20414","URL":"https://doi.org/10.48550/arxiv.2606.20414","source":"datacite"},{"id":"doi:10.5281/zenodo.20801558","type":"article-journal","title":"A Multi-Layer Classification System for Neural Structures: Axiomatization, Extensibility, and Periodic-Table Predictions","abstract":"This paper establishes a multi-layer classification system for neural structures based on computational primitives (neuronal dynamics, synaptic plasticity, connectivity topology) and their axiomatic systems. The system takes as successive refinement criteria the type of dynamical primitives, connection sequence types, synaptic plasticity rules, circuit generation algorithms, encoding principles, emergent symmetries, and concrete species instances, forming an arbitrarily extensible classification tree. Each layer is equipped with corresponding axiom systems and fundamental theorems, so that any neural structure (including ion channel models, synaptic plasticity rules, local microcircuits, large-scale brain networks, and cognitive architectures) can be uniquely placed into a specific node of the tree. Conversely, any parameter combination of a node can mechanically generate an axiom system and predict as-yet-unstudied neural structures. The system possesses unity, completeness, and extensibility, analogous to the periodic table of chemical elements, and can be used to systematically discover and construct new neural coding schemes, plasticity rules, and brain-inspired computational models. The paper provides formal definitions, construction methods, fundamental theorems, and multiple examples, and shows how classical branches of neuroscience (Hodgkin–Huxley model, Hebbian plasticity, cortical microcolumns, hippocampal place coding, etc.) are embedded into the system, as well as how to build axiom systems and fundamental theorems for vacant parameter combinations. All theorems are given rigorous proofs (general theorems at least 4 steps, important theorems at least 8 steps), and all predictions are equipped with complete axiom systems and existence constructions, transforming the development of neuroscience from random discovery to fill-in-the-blank construction.","author":[{"family":"Liu","given":"Shifa"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20801558","URL":"https://doi.org/10.5281/zenodo.20801558","source":"datacite"},{"id":"doi:10.5281/zenodo.20801557","type":"article-journal","title":"A Multi-Layer Classification System for Neural Structures: Axiomatization, Extensibility, and Periodic-Table Predictions","abstract":"This paper establishes a multi-layer classification system for neural structures based on computational primitives (neuronal dynamics, synaptic plasticity, connectivity topology) and their axiomatic systems. The system takes as successive refinement criteria the type of dynamical primitives, connection sequence types, synaptic plasticity rules, circuit generation algorithms, encoding principles, emergent symmetries, and concrete species instances, forming an arbitrarily extensible classification tree. Each layer is equipped with corresponding axiom systems and fundamental theorems, so that any neural structure (including ion channel models, synaptic plasticity rules, local microcircuits, large-scale brain networks, and cognitive architectures) can be uniquely placed into a specific node of the tree. Conversely, any parameter combination of a node can mechanically generate an axiom system and predict as-yet-unstudied neural structures. The system possesses unity, completeness, and extensibility, analogous to the periodic table of chemical elements, and can be used to systematically discover and construct new neural coding schemes, plasticity rules, and brain-inspired computational models. The paper provides formal definitions, construction methods, fundamental theorems, and multiple examples, and shows how classical branches of neuroscience (Hodgkin–Huxley model, Hebbian plasticity, cortical microcolumns, hippocampal place coding, etc.) are embedded into the system, as well as how to build axiom systems and fundamental theorems for vacant parameter combinations. All theorems are given rigorous proofs (general theorems at least 4 steps, important theorems at least 8 steps), and all predictions are equipped with complete axiom systems and existence constructions, transforming the development of neuroscience from random discovery to fill-in-the-blank construction.","author":[{"family":"Liu","given":"Shifa"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20801557","URL":"https://doi.org/10.5281/zenodo.20801557","source":"datacite"},{"id":"doi:10.5281/zenodo.20783400","type":"article-journal","title":"Neuromorphic Blockchain Framework for Secure and Energy-Optimized Outlier Detection in IoT-Driven Wireless Sensor Networks","abstract":"Abstract: Wireless Sensor Networks (WSNs) are increasingly deployed in critical domains such as healthcare, environmental monitoring, and industrial automation, where secure and reliable data collection is essential. However, these networks face performance degradation and reduced network lifetime because they become susceptible to both unexpected system behaviour and intentional harmful attacks. The study presents a new framework that combines Spiking Neural Networks (SNNs) with a multi-layer block chain security system to solve these specific security challenges. The SNN model uses event-driven neural computation to find outliers with high precision while it maintains energy efficiency for nodes that have limited resources. The block chain component establishes data protection through decentralized validation which operates across multiple levels to maintain data security and permanent record. Experimental results show that the proposed method successfully reduces false detection rates while it improves trustworthiness of network transactions and extends system operational duration. The results demonstrate that the combination of neuromorphic intelligence and distributed ledger technologies creates a secure and energy-efficient anomaly detection system which scales to meet the demands of future IoT and WSN systems.","author":[{"family":"Padmasree","given":"N"},{"family":"Patil","given":"Malini"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20783400","URL":"https://doi.org/10.5281/zenodo.20783400","source":"datacite"},{"id":"doi:10.5281/zenodo.20783401","type":"article-journal","title":"Neuromorphic Blockchain Framework for Secure and Energy-Optimized Outlier Detection in IoT-Driven Wireless Sensor Networks","abstract":"Abstract: Wireless Sensor Networks (WSNs) are increasingly deployed in critical domains such as healthcare, environmental monitoring, and industrial automation, where secure and reliable data collection is essential. However, these networks face performance degradation and reduced network lifetime because they become susceptible to both unexpected system behaviour and intentional harmful attacks. The study presents a new framework that combines Spiking Neural Networks (SNNs) with a multi-layer block chain security system to solve these specific security challenges. The SNN model uses event-driven neural computation to find outliers with high precision while it maintains energy efficiency for nodes that have limited resources. The block chain component establishes data protection through decentralized validation which operates across multiple levels to maintain data security and permanent record. Experimental results show that the proposed method successfully reduces false detection rates while it improves trustworthiness of network transactions and extends system operational duration. The results demonstrate that the combination of neuromorphic intelligence and distributed ledger technologies creates a secure and energy-efficient anomaly detection system which scales to meet the demands of future IoT and WSN systems.","author":[{"family":"Padmasree","given":"N"},{"family":"Patil","given":"Malini"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20783401","URL":"https://doi.org/10.5281/zenodo.20783401","source":"datacite"},{"id":"doi:10.5281/zenodo.20678298","type":"article-journal","title":"[6] Project Mnemosyne: From Device to Field (Associative Memory and Selective Forgetting in Janus Fractal Crossbar Array)","abstract":"Project Mnemosyne bridges the gap between Project Janus [1]—a single fractal memristive junction—and the neuromorphic computing systems envisioned in Project Raphael. Where Janus established a single synapse capable of 32-level analog storage and KWW stretched exponential forgetting (β = 0.6, τ = 720h), Mnemosyne asks: what emerges when many such synapses are organised together? The answer, demonstrated through software simulation grounded in confirmed Janus physical parameters, is associative memory with selective forgetting. A reservoir computing (RC) architecture using the Janus Butler Volmer activation and KWW memory kernel achieves R2 = 0.990 on Lorenz attractor prediction—a standard benchmark for nonlinear temporal computation. Critically, R2 remains above 97% across array sizes from 32×32 to 128×128, and falls only from 0.988 to 0.981 under 10% random junction failure. The selective forgetting mechanism (phase coupling γ = 0.4: strong signals consolidate to τeff = 1008h; weak signals prune to 432h) is not a limitation but the computational mechanism: it prevents state saturation and enables continuous adaptation without external weight updates. Important caveat: All results are software simulations. No hardware fabrication has been performed. The value of this publication is to establish prior art and demonstrate computational viability from confirmed Janus physical parameters. Keywords: reservoir computing; associative memory; KWW forgetting; Janus memristor; (Eu,Pr,Y)2O3; neuromorphic computing; Menger sponge; fractal crossbar; Lorenz prediction; selective forgetting v2 update: Acknowledgements clarified The Acknowledgements section in Version 1.0 contained an inaccurate attribution. During the development of this work, the author engaged in extensive discussions with Gemini (Google DeepMind) to work through specific challenges, and subsequently brought the outputs of those discussions to Claude (Anthropic) for implementation and synthesis. As a result, Claude's initial rendering of the Acknowledgements did not accurately reflect the origin of individual contributions. This version corrects the attribution. The scientific content remains unchanged.","author":[{"family":"Liu","given":"Tung"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20678298","URL":"https://doi.org/10.5281/zenodo.20678298","source":"datacite"},{"id":"doi:10.5281/zenodo.20679036","type":"article-journal","title":"[6] Project Mnemosyne: From Device to Field (Associative Memory and Selective Forgetting in Janus Fractal Crossbar Array)","abstract":"Project Mnemosyne bridges the gap between Project Janus [1]—a single fractal memristive junction—and the neuromorphic computing systems envisioned in Project Raphael. Where Janus established a single synapse capable of 32-level analog storage and KWW stretched exponential forgetting (β = 0.6, τ = 720h), Mnemosyne asks: what emerges when many such synapses are organised together? The answer, demonstrated through software simulation grounded in confirmed Janus physical parameters, is associative memory with selective forgetting. A reservoir computing (RC) architecture using the Janus Butler Volmer activation and KWW memory kernel achieves R2 = 0.990 on Lorenz attractor prediction—a standard benchmark for nonlinear temporal computation. Critically, R2 remains above 97% across array sizes from 32×32 to 128×128, and falls only from 0.988 to 0.981 under 10% random junction failure. The selective forgetting mechanism (phase coupling γ = 0.4: strong signals consolidate to τeff = 1008h; weak signals prune to 432h) is not a limitation but the computational mechanism: it prevents state saturation and enables continuous adaptation without external weight updates. Important caveat: All results are software simulations. No hardware fabrication has been performed. The value of this publication is to establish prior art and demonstrate computational viability from confirmed Janus physical parameters. Keywords: reservoir computing; associative memory; KWW forgetting; Janus memristor; (Eu,Pr,Y)2O3; neuromorphic computing; Menger sponge; fractal crossbar; Lorenz prediction; selective forgetting v2 update: Acknowledgements clarified The Acknowledgements section in Version 1.0 contained an inaccurate attribution. During the development of this work, the author engaged in extensive discussions with Gemini (Google DeepMind) to work through specific challenges, and subsequently brought the outputs of those discussions to Claude (Anthropic) for implementation and synthesis. As a result, Claude's initial rendering of the Acknowledgements did not accurately reflect the origin of individual contributions. This version corrects the attribution. The scientific content remains unchanged.","author":[{"family":"Liu","given":"Tung"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20679036","URL":"https://doi.org/10.5281/zenodo.20679036","source":"datacite"},{"id":"doi:10.5281/zenodo.20751387","type":"article-journal","title":"Antiferromanyetik Van der Waals Heteroyapılarında Spin-Yörünge Torku Aracılığıyla Nöromorfik Matris-Vektör Hesaplama: Rezervuar Hesaplama Çerçevesinde Teorik Bir Öneri","abstract":"Özet İki boyutlu (2B) bir antiferromanyetik (AFM) van der Waals heteroyapısı içinde spin-yörünge torku (SYT) dinamiklerini kullanan teorik bir nöromorfik hesaplama çerçevesi sunuyoruz. MnBi₂Te₄/Pt gibi ağır-metal/AFM-yalıtkan çift katmanlarında, mevcut Landau-Lifshitz-Gilbert-Slonczewski (LLGS) denkleminin doğal olarak doğrusal-olmayan dinamikler ürettiğini ve bu dinamiklerin Rezervuar Hesaplama (Reservoir Computing, RC) mimarisine uygun bir hesaplama substratı oluşturduğunu gösteriyoruz. Sinaptik ağırlıklar, sıfır net mıknatıslanmaya sahip AFM alt-örgüsünün Néel vektörü L yöneliminde kodlanır; girdiler ise Spin Hall etkisiyle üretilen spin-polarize akım darbeleri olarak enjekte edilir. Ağırlık güncelleme mekanizması, magnon sönümlemesinden kaynaklanan gerçek enerji maliyetlerini (Joule ısınması sıfır değil, ancak klasik CMOS'a kıyasla birkaç mertebelik derecede daha düşük) açıkça ele alır. Doğrusal-olmamayı bir kusur olarak değil, doğal bir ReLU benzeri eşik aktivasyon işlevi olarak kurguladık. Bu çalışma patent-karşıtı Prior Art amaçlı kamu malına adanmıştır; içerdiği tüm mimari iddiaların gelecekteki geniş patent başvurularını geçersiz kılması hedeflenmekte ve CERN Açık Donanım Lisansı (OHL) kapsamında açık kaynak olarak önerilmektedir.","author":[{"family":"Yünlü","given":"Muzaffer"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20751387","URL":"https://doi.org/10.5281/zenodo.20751387","source":"datacite"},{"id":"doi:10.5281/zenodo.20678299","type":"article-journal","title":"Project Mnemosyne: From Device to Field (Associative Memory and Selective Forgetting in Janus Fractal Crossbar Array)","abstract":"Project Mnemosyne bridges the gap between Project Janus [1]—a single fractal memristive junction—and the neuromorphic computing systems envisioned in Project Raphael. Where Janus established a single synapse capable of 32-level analog storage and KWW stretched exponential forgetting (β = 0.6, τ = 720h), Mnemosyne asks: what emerges when many such synapses are organised together? The answer, demonstrated through software simulation grounded in confirmed Janus physical parameters, is associative memory with selective forgetting. A reservoir computing (RC) architecture using the Janus Butler Volmer activation and KWW memory kernel achieves R2 = 0.990 on Lorenz attractor prediction—a standard benchmark for nonlinear temporal computation. Critically, R2 remains above 97% across array sizes from 32×32 to 128×128, and falls only from 0.988 to 0.981 under 10% random junction failure. The selective forgetting mechanism (phase coupling γ = 0.4: strong signals consolidate to τeff = 1008h; weak signals prune to 432h) is not a limitation but the computational mechanism: it prevents state saturation and enables continuous adaptation without external weight updates. Important caveat: All results are software simulations. No hardware fabrication has been performed. The value of this publication is to establish prior art and demonstrate computational viability from confirmed Janus physical parameters. Keywords: reservoir computing; associative memory; KWW forgetting; Janus memristor; (Eu,Pr,Y)2O3; neuromorphic computing; Menger sponge; fractal crossbar; Lorenz prediction; selective forgetting","author":[{"family":"Liu","given":"Tung"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20678299","URL":"https://doi.org/10.5281/zenodo.20678299","source":"datacite"},{"id":"doi:10.5281/zenodo.20590583","type":"article-journal","title":"ARIA: A Zero-Hallucination Neuro-Symbolic Architecture for Persistent Memory, Sparse Graph Reasoning, and Autonomous Self-Evolution","abstract":"ARIA is a deterministic neuro-symbolic cognitive architecture designed to integrate persistent memory, autonomous goal formation, causal reasoning, self-modifying cognitive structures, and large-scale knowledge representation. Unlike conventional transformer-based systems, ARIA combines symbolic knowledge graphs, dynamic concept networks, neuromorphic processing components, autonomous learning mechanisms, and long-term memory persistence into a unified cognitive framework. This publication describes the architecture, design principles, implementation details, experimental results, and future development roadmap of the ARIA system. Keywords: AGI, Neuro-Symbolic AI, Cognitive Architecture, Persistent Memory, Autonomous Agents, Artificial General Intelligence, Causal Reasoning, Knowledge Graphs, AtomSpace, Cognitive Systems.","author":[{"family":"Ltd","given":"Counselor"},{"family":"Vettorato","given":"Nicola"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20590583","URL":"https://doi.org/10.5281/zenodo.20590583","source":"datacite"},{"id":"doi:10.5281/zenodo.20590584","type":"article-journal","title":"ARIA: A Zero-Hallucination Neuro-Symbolic Architecture for Persistent Memory, Sparse Graph Reasoning, and Autonomous Self-Evolution","abstract":"ARIA is a deterministic neuro-symbolic cognitive architecture designed to integrate persistent memory, autonomous goal formation, causal reasoning, self-modifying cognitive structures, and large-scale knowledge representation. Unlike conventional transformer-based systems, ARIA combines symbolic knowledge graphs, dynamic concept networks, neuromorphic processing components, autonomous learning mechanisms, and long-term memory persistence into a unified cognitive framework. This publication describes the architecture, design principles, implementation details, experimental results, and future development roadmap of the ARIA system. Keywords: AGI, Neuro-Symbolic AI, Cognitive Architecture, Persistent Memory, Autonomous Agents, Artificial General Intelligence, Causal Reasoning, Knowledge Graphs, AtomSpace, Cognitive Systems.","author":[{"family":"Ltd","given":"Counselor"},{"family":"Vettorato","given":"Nicola"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20590584","URL":"https://doi.org/10.5281/zenodo.20590584","source":"datacite"},{"id":"doi:10.5281/zenodo.20584324","type":"article-journal","title":"Wave_Unified_Spectral_Technologies.docx","abstract":"This paper provides a concise conceptual survey of the Spectral Computing paradigm and its natural extensions into the physical sciences. At its centre stands the spectral computer — an architecture in which information is encoded, routed and processed as structured wave patterns rather than discrete voltage levels. The paper then surveys five families of adjacent technology that share the same wave-theoretic foundations: spectral holography, spectral X-ray imaging, spectral laser systems, spectral information transfer, and wave-driven medicine. A sixth thread, the sound–spectrum bridge, is given special emphasis, as acoustic waves and electromagnetic spectra obey formally identical mathematical structures and together point toward a fully wave-unified system of both computation and therapy. The paper is intended as an accessible introduction for readers approaching the Spectral Computing Programme for the first time.","author":[{"family":"Desmond","given":"Timothy"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20584324","URL":"https://doi.org/10.5281/zenodo.20584324","source":"datacite"},{"id":"doi:10.5281/zenodo.20584325","type":"article-journal","title":"Wave_Unified_Spectral_Technologies.docx","abstract":"This paper provides a concise conceptual survey of the Spectral Computing paradigm and its natural extensions into the physical sciences. At its centre stands the spectral computer — an architecture in which information is encoded, routed and processed as structured wave patterns rather than discrete voltage levels. The paper then surveys five families of adjacent technology that share the same wave-theoretic foundations: spectral holography, spectral X-ray imaging, spectral laser systems, spectral information transfer, and wave-driven medicine. A sixth thread, the sound–spectrum bridge, is given special emphasis, as acoustic waves and electromagnetic spectra obey formally identical mathematical structures and together point toward a fully wave-unified system of both computation and therapy. The paper is intended as an accessible introduction for readers approaching the Spectral Computing Programme for the first time.","author":[{"family":"Desmond","given":"Timothy"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20584325","URL":"https://doi.org/10.5281/zenodo.20584325","source":"datacite"},{"id":"doi:10.5281/zenodo.20581146","type":"article-journal","title":"Spectral_Computing_Product_Pitch_v2.docx","abstract":"Spectral Computing is a new computing architecture built on a fundamentally different mathematical substrate from the binary logic that has powered electronics since the 1940s. It does not require exotic materials, new physics, or untested engineering. Approximately 90% of the technology required to implement Spectral Computing exists today in commercially available hardware: phase-locked loop circuits, digital signal processors, Fourier transform accelerators, neuromorphic chips, and coherent optical computing components are all in production at scale. The remaining 10% — the coherent operator primitives that form the core of the Spectral Instruction Set Architecture (S-ISA) — rests on well-established physics. The theoretical foundation is solid, the implementation pathway is clear, and the engineering challenges are tractable within current semiconductor fabrication capabilities. What Spectral Computing provides is the unifying architectural framework that ties these existing technologies together under a coherent mathematical system — one that is natively suited to the computational demands of artificial intelligence, quantum computing interfaces, real-time signal processing, robotics, and medical instrumentation that classical binary architecture increasingly struggles to meet.","author":[{"family":"Desmond","given":"Timothy"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20581146","URL":"https://doi.org/10.5281/zenodo.20581146","source":"datacite"},{"id":"doi:10.5281/zenodo.20581147","type":"article-journal","title":"Spectral_Computing_Product_Pitch_v2.docx","abstract":"Spectral Computing is a new computing architecture built on a fundamentally different mathematical substrate from the binary logic that has powered electronics since the 1940s. It does not require exotic materials, new physics, or untested engineering. Approximately 90% of the technology required to implement Spectral Computing exists today in commercially available hardware: phase-locked loop circuits, digital signal processors, Fourier transform accelerators, neuromorphic chips, and coherent optical computing components are all in production at scale. The remaining 10% — the coherent operator primitives that form the core of the Spectral Instruction Set Architecture (S-ISA) — rests on well-established physics. The theoretical foundation is solid, the implementation pathway is clear, and the engineering challenges are tractable within current semiconductor fabrication capabilities. What Spectral Computing provides is the unifying architectural framework that ties these existing technologies together under a coherent mathematical system — one that is natively suited to the computational demands of artificial intelligence, quantum computing interfaces, real-time signal processing, robotics, and medical instrumentation that classical binary architecture increasingly struggles to meet.","author":[{"family":"Desmond","given":"Timothy"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20581147","URL":"https://doi.org/10.5281/zenodo.20581147","source":"datacite"},{"id":"doi:10.5281/zenodo.20532558","type":"article-journal","title":"Topology-Aware Packet Classification and Adaptive Local-Core Selection for SpiNNaker: Design and Reproducible Virtual-Mode Benchmarking","abstract":"This paper develops a proposal for topology-aware packet classification and adaptive local-core selection in SpiNNaker. A dispatch-class mechanism refines standard routing table entries by restricting inter-chip forwarding directions and reducing local processor-mask fan-out. A reproducible virtual-mode benchmark compares the proposal against a static-routing baseline across 960 trials per ablation. The combined refinement reduced a synthetic energy proxy by 18.28% on average, with processor-mask refinement identified as the dominant contributor. Results are software-defined proxies, not hardware-measured energy claims.","author":[{"family":"Burgess","given":"Christopher"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20532558","URL":"https://doi.org/10.5281/zenodo.20532558","source":"datacite"},{"id":"doi:10.5281/zenodo.20532559","type":"article-journal","title":"Topology-Aware Packet Classification and Adaptive Local-Core Selection for SpiNNaker: Design and Reproducible Virtual-Mode Benchmarking","abstract":"This paper develops a proposal for topology-aware packet classification and adaptive local-core selection in SpiNNaker. A dispatch-class mechanism refines standard routing table entries by restricting inter-chip forwarding directions and reducing local processor-mask fan-out. A reproducible virtual-mode benchmark compares the proposal against a static-routing baseline across 960 trials per ablation. The combined refinement reduced a synthetic energy proxy by 18.28% on average, with processor-mask refinement identified as the dominant contributor. Results are software-defined proxies, not hardware-measured energy claims.","author":[{"family":"Burgess","given":"Christopher"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20532559","URL":"https://doi.org/10.5281/zenodo.20532559","source":"datacite"},{"id":"doi:10.48550/arxiv.2606.01181","type":"manuscript","title":"IO Pad Integrity in Energy-Efficient Neuromorphic Chips","abstract":"Neuromorphic computing relies on low-power, high-reliability hardware, yet the integrity of input/output pads (IOPADs) remains an underexplored factor affecting system performance. This chapter examines the role of IOPAD integrity in neuromorphic VLSI design and connects algorithmic development with practical hardware implementation. While much attention has been given to spiking neural networks (SNNs) and ultra-low-power core logic, the electrical and functional robustness of the I/O interface is equally critical for ensuring signal fidelity and minimizing energy consumption. We review the structure and function of IOPADs, outline their influence on power, performance, and reliability, and discuss design trade-offs involving pad libraries, pad ring architectures, and bonding strategies. The chapter also introduces the fundamentals of SNNs and summarizes the digital hardware design flow from behavioral description to physical layout. Physical implementation considerations are highlighted using the SkyWater 130 nm CMOS process as a practical platform for neuromorphic prototyping. Real-world examples illustrate how early-stage I/O planning can prevent redesign, reduce yield loss, and improve overall system efficiency. This work emphasizes that IOPAD integrity is a key enabler of scalable, energy-efficient neuromorphic systems.","author":[{"family":"Ghani","given":"Arfan"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2606.01181","URL":"https://doi.org/10.48550/arxiv.2606.01181","source":"datacite"},{"id":"doi:10.48550/arxiv.2606.00194","type":"manuscript","title":"Wave-based Neuromorphic Circuit Networks: Tunable 2D Transmission-Line Metamaterials","abstract":"Neuromorphic computing promises fast and energy-efficient information processing for emerging applications such as artificial intelligence. This paper presents neuromorphic processors based on wave-based programmable transmission-line (TLIN) metamaterials. Specifically, 2D reactive electrical networks are proposed, consisting of a grid of interconnected subwavelength TLIN-based unit cells (neurons) with tunable reactive elements. During inference, the input data is encoded using single-tone sources impressed onto the network, and circuit quantities are measured to decode the output prediction. Computation is performed through wave propagation and interference across the grid, with the learned input-output relationships stored in the tunable reactive elements. A key contribution of this work is a scalable training method based on in-situ backpropagation. The adjoint variable method is used to derive a physical (electrical) realization of the backpropagation algorithm that is typically used to compute the gradient of the objective loss function in digital neural networks. This formulation computes the gradient from voltage measurements of two steady-state excitations: the forward pass (inference) and the adjoint pass (error backpropagation). This enables efficient training since it is independent of the number of trainable parameters and avoids the simulation-reality gap. To demonstrate the effectiveness of this approach, wave-based neuromorphic circuit networks are trained for allostery and classification tasks, and the system's robustness to damage is shown. This work paves the way for self-learning systems based on wave-based neuromorphic analog circuit hardware.","author":[{"family":"Thakkar","given":"Shrey"},{"family":"Grbic","given":"Anthony"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2606.00194","URL":"https://doi.org/10.48550/arxiv.2606.00194","source":"datacite"},{"id":"doi:10.5281/zenodo.20449594","type":"article-journal","title":"Sparse Photonic Reservoirs with Adaptive Cavity Memory and Polarized Noise: Surpassing LSTMs on Nonlinear Memory Tasks","abstract":"We present a sparse photonic reservoir computer with O(N)\\mathcal{O}(N)O(N) small-world topology, adaptive optical cavity memory, and spatially correlated polarized noise that achieves NMSE = 0.050 on the NARMA10 benchmark, within 12.6% of dense Echo State Networks while consistently outperforming deep LSTM networks by 19–99% across four canonical tasks (NARMA10, NARMA30, Mackey-Glass, Lorenz). Under temporally-validated evaluation with comprehensive sanity checks, our architecture requires only ~100 trainable parameters versus 17,217 for LSTM, demonstrating that physical realizability through sparse wiring does not preclude competitive nonlinear temporal processing. Bayesian optimization reveals task-adaptive cavity lifetimes (τc∈[6.3,30.5]\\tau_c \\in [6.3, 30.5]τc∈[6.3,30.5]) and noise polarization correlations (ρcross∈[−0.92,+0.31]\\rho_{\\text{cross}} \\in [-0.92, +0.31]ρcross∈[−0.92,+0.31]), establishing both as genuine computational resources. An accompanying ML-based inverse design pipeline maps optimized dynamical parameters to fabricable silicon nanobeam geometries with sub-5% accuracy, providing a complete blueprint from simulation to experimental realization. All code and LaTeX source are openly available.","author":[{"family":"Massami Okushigue","given":"Jefferson"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20449594","URL":"https://doi.org/10.5281/zenodo.20449594","source":"datacite"},{"id":"doi:10.5281/zenodo.20449593","type":"article-journal","title":"Sparse Photonic Reservoirs with Adaptive Cavity Memory and Polarized Noise: Surpassing LSTMs on Nonlinear Memory Tasks","abstract":"We present a sparse photonic reservoir computer with O(N)\\mathcal{O}(N)O(N) small-world topology, adaptive optical cavity memory, and spatially correlated polarized noise that achieves NMSE = 0.050 on the NARMA10 benchmark, within 12.6% of dense Echo State Networks while consistently outperforming deep LSTM networks by 19–99% across four canonical tasks (NARMA10, NARMA30, Mackey-Glass, Lorenz). Under temporally-validated evaluation with comprehensive sanity checks, our architecture requires only ~100 trainable parameters versus 17,217 for LSTM, demonstrating that physical realizability through sparse wiring does not preclude competitive nonlinear temporal processing. Bayesian optimization reveals task-adaptive cavity lifetimes (τc∈[6.3,30.5]\\tau_c \\in [6.3, 30.5]τc∈[6.3,30.5]) and noise polarization correlations (ρcross∈[−0.92,+0.31]\\rho_{\\text{cross}} \\in [-0.92, +0.31]ρcross∈[−0.92,+0.31]), establishing both as genuine computational resources. An accompanying ML-based inverse design pipeline maps optimized dynamical parameters to fabricable silicon nanobeam geometries with sub-5% accuracy, providing a complete blueprint from simulation to experimental realization. All code and LaTeX source are openly available.","author":[{"family":"Massami Okushigue","given":"Jefferson"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20449593","URL":"https://doi.org/10.5281/zenodo.20449593","source":"datacite"},{"id":"doi:10.5281/zenodo.20357455","type":"article-journal","title":"The Cognitive Tension Core ($\\NTC$) and P3TTM Mott-Hubbard Insulator Mechanics for Frictionless Computing-in-Memory","abstract":"UPDATE NOTICE: This record corresponds to the revised and optimized Version 2 of the original manuscript, incorporating critical chemical doping specifications and atomic encapsulation protocols for the hardware core. Manuscript I: The Cognitive Tension Core (NTC) This manuscript details the structural engineering and quantum condensed matter physics driving Tier 1 of the Coherent Hardware ecosystem: The Cognitive Tension Core (NTC). Utilizing the highly conjugated organic polymer poly(3-(2-total-tension-methyl)thiophene) (P3TTM), this architecture implements a physical computing-in-memory (CIM) crossbar array. By modeling the electronic phase space via the single-band Hubbard Hamiltonian, we demonstrate how configuring the substrate into a strongly correlated Mott Insulator regime (U_local >> t) allows optimization loss functions to map directly onto non-linear electronic relaxation states. This approach bypasses kinetic charge transport, effectively suppressing kinetic thermal dissipation and uncompensated Informational Friction (Phi_UFI).","author":[{"family":"Quilez Zamora","given":"Jaime"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20357455","URL":"https://doi.org/10.5281/zenodo.20357455","source":"datacite"},{"id":"doi:10.5281/zenodo.20357456","type":"article-journal","title":"The Cognitive Tension Core ($\\NTC$) and P3TTM Mott-Hubbard Insulator Mechanics for Frictionless Computing-in-Memory","abstract":"UPDATE NOTICE: This record corresponds to the revised and optimized Version 2 of the original manuscript, incorporating critical chemical doping specifications and atomic encapsulation protocols for the hardware core. Manuscript I: The Cognitive Tension Core (NTC) This manuscript details the structural engineering and quantum condensed matter physics driving Tier 1 of the Coherent Hardware ecosystem: The Cognitive Tension Core (NTC). Utilizing the highly conjugated organic polymer poly(3-(2-total-tension-methyl)thiophene) (P3TTM), this architecture implements a physical computing-in-memory (CIM) crossbar array. By modeling the electronic phase space via the single-band Hubbard Hamiltonian, we demonstrate how configuring the substrate into a strongly correlated Mott Insulator regime (U_local >> t) allows optimization loss functions to map directly onto non-linear electronic relaxation states. This approach bypasses kinetic charge transport, effectively suppressing kinetic thermal dissipation and uncompensated Informational Friction (Phi_UFI).","author":[{"family":"Quilez Zamora","given":"Jaime"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20357456","URL":"https://doi.org/10.5281/zenodo.20357456","source":"datacite"},{"id":"doi:10.5281/zenodo.20263857","type":"article-journal","title":"Vacuum Intelligence: A Self-Organizing Reservoir with Hyperbolic Geometry and Anti-Fragile Meta-Learning","abstract":"This working paper introduces the Vacuum Intelligence Framework, a neural-reservoir architecture in which adaptive computation emerges from a high-dimensional nonlinear stochastic system poised near a symmetry-breaking instability. Unlike conventional AI systems trained offline and deployed statically, this framework exploits inherent noise and criticality to remain sensitive and adaptable under novel disruptions.The reservoir state evolves according to a Langevin equation in a double-well potential, producing a vacuum at the origin that is inherently unstable yet generative. Perturbations break this symmetry, driving the system into one of 2^N attractor basins whose selection depends on the history of inputs and noise. Synchronization metrics derived from hyperbolic embeddings in the Poincaré ball provide a continuous coherence signal, the synchronicity order parameter, that serves as an early-warning indicator of regime changes or supply chain disruptions.The core theoretical contribution is a meta-learning loop that interprets the decay-rebirth cycle as a simulated-annealing process over the reservoir’s structural parameters. This confers antifragility in the technical sense: the system improves in resilience after each controlled collapse because the Metropolis-sampled annealing process discovers parameter configurations that gradient-based methods cannot reach from within a local optimum. The paper presents the full mathematical formulation including the coupled stochastic differential equations, the hyperbolic distance metric, the synchronicity order parameter, the mean-time-to-recovery fitness function, and the Metropolis acceptance criterion. Application domains include defense logistics monitoring in contested environments, multi-domain command and control under degraded communications, and supply chain resilience forecasting under black-swan disruptions. A companion architecture specification document provides hardware mapping to memristor crossbar and neuromorphic substrates, and a TRL 2 to TRL 6 development roadmap.","author":[{"family":"John","given":"Damon"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20263857","URL":"https://doi.org/10.5281/zenodo.20263857","source":"datacite"},{"id":"doi:10.5281/zenodo.20263929","type":"article-journal","title":"Vacuum Intelligence Architecture Specification: Event-Driven Analog Memristor Optimal Power/Flow Solver Integration (VIA-SPEC-001)","abstract":"This architecture specification presents a unified hardware-software system for resilient, antifragile decision-making in contested logistics and defense environments. Building on the theoretical foundation established in the companion working paper on the Vacuum Intelligence Framework, this document provides full mathematical mappings, hardware specifications, and a technology readiness level roadmap from TRL 2 to TRL 6.The core innovation is a dual-mode memristor crossbar that functions both as the physical stochastic reservoir and as an event-driven analog optimal power/flow solver. In reservoir mode, the crossbar physically implements the Langevin dynamics of the double-well potential, exploiting native device noise and nonlinearity to realize the vacuum reservoir without digital simulation overhead. In solver mode, the same crossbar is reprogrammed in under one microsecond to encode the Hessian of a quadratic programming problem, with steady-state row voltages converging to the optimal logistics routing solution in microseconds rather than the seconds required by GPU-based digital solvers. The system-in-package includes a 256 to 1,024 neuron memristor crossbar, a RISC-V microcontroller for mode orchestration and safety monitoring, an analog front-end, and a programmable noise source. Total dissipation is less than 1.5 W in continuous reservoir mode, representing approximately 20 times energy improvement and 1,000 times speedup relative to equivalent NVIDIA Jetson implementations. Physical envelope is 15 by 15 by 3 millimeters, suitable for UAV, shipboard, and forward-deployment modules.The hyperbolic echo state network architecture structures recurrent weights via exponential decay with hyperbolic distance in the Poincaré ball, improving chaotic time-series prediction accuracy for hierarchically structured logistics data without increasing reservoir dimensionality. The synchronicity order parameter, defined using the Minkowski inner product on the hyperboloid model, provides event-driven triggering for both the meta-optimizer and the OPF solver.Defense applications addressed include edge-deployed supply chain monitoring processing AIS vessel tracking, sanctions databases, and commodity price signals; antifragile command systems maintaining minimum viable operation under jammed or degraded sensor conditions; counter-sanctions and shadow-fleet disruption modeling with adaptive countermeasure hypothesis generation; and automated wargaming through large-scale internal scenario exploration cycles. Hardware qualification pathway specifies manufacturing via GlobalFoundries GF 22FDX or SkyWater 90 nm rad-hard process. ITAR classification under USML Category XI(c) is identified with commodity jurisdiction determination required prior to export. Anti-tamper architecture relies on non-volatile memristor states resistant to single-event upset with cryptographically verified RISC-V boot sequence.","author":[{"family":"John","given":"Damon"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20263929","URL":"https://doi.org/10.5281/zenodo.20263929","source":"datacite"},{"id":"doi:10.60893/figshare.adv.c.8461155","type":"article-journal","title":"<strong>Reservoir Computing Using a Si-integrated Electro-optic Oxide: A Numerical Study</strong>","abstract":"In this numerical study, we investigate the performance of an electro-optic reservoir under the practical constraints of a Si-integrated photonic design enabled by barium titanate, an emergent electro-optic material. Reservoir computing is a compelling neuromorphic architecture due to its natural compatibility with many physical systems. Among available implementations, the electro-optic reservoir, where an optical delay feedback loop serves as the reservoir, offers energy-efficient computation without the need for a separate memory, making its chip-scale implementation very appealing. Barium titanate, with its exceptionally large Pockels coefficient, has recently enabled significant advances in compact electro-optic modulators, positioning it as a suitable platform for such a chip-scale electro-optic reservoir. We benchmark the performance of the electro-optic reservoir over various tasks and introduce a method of constructing reservoir states that results in a significant performance improvement. Furthermore, we demonstrate the superior energy efficiency of barium titanate-based devices, owing to their very efficient electro-optic modulation. Our results show that compact, energy-efficient electro-optic reservoirs are feasible using barium titanate-based integrated Si photonic devices.","author":[{"family":"Zhang","given":"Xiaoru"},{"family":"Demkov","given":"Alexander"}],"issued":{"date-parts":[[2026]]},"DOI":"10.60893/figshare.adv.c.8461155","URL":"https://doi.org/10.60893/figshare.adv.c.8461155","source":"datacite"},{"id":"doi:10.5281/zenodo.20134374","type":"article-journal","title":"Native Symbolic Emergence Should Replace the Translation Layer Paradigm in Neuro-Symbolic AI","abstract":"The dominant paradigm in neurosymbolic AI assumes that symbolic structure must be defined before learning or extracted after it. We identify this assumption as an unnecessary architectural constraint, and we call it the symbolization gap. This position paper argues that symbolic representations should emerge naturally from neuromorphic learning dynamics without gradient descent, a translation layer, or prior vocabulary knowledge. To instantiate our position, we introduce NSSN (Neuromorphic Symbolic Spiking Network). In NSSN, each Spike-Timing-Dependent Plasticity (STDP) co-activation directly produces a new symbolic node. Four provable properties hold for any finite corpus, any domain alphabet, and any training order: monotonic growth, convergence, incremental preservation, and subquadratic complexity. Two experiments confirm that these properties hold on real data across qualitatively distinct domains. On the Sepsis Cases dataset, 13 clinical sequences are recovered with less than 0.05% variance across three training orders. On Fashion-MNIST, the same STDP rule produces a qualitatively different topology in spatial mode. The space of architectures enabling native symbolic emergence has been systematically underexplored. NSSN provides the formal and empirical foundations to explore it.","author":[{"family":"Nicolle","given":"Christophe"},{"family":"Callegarin","given":"Davide"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20134374","URL":"https://doi.org/10.5281/zenodo.20134374","source":"datacite"},{"id":"doi:10.5281/zenodo.20134373","type":"article-journal","title":"Native Symbolic Emergence Should Replace the Translation Layer Paradigm in Neuro-Symbolic AI","abstract":"The dominant paradigm in neurosymbolic AI assumes that symbolic structure must be defined before learning or extracted after it. We identify this assumption as an unnecessary architectural constraint, and we call it the symbolization gap. This position paper argues that symbolic representations should emerge naturally from neuromorphic learning dynamics without gradient descent, a translation layer, or prior vocabulary knowledge. To instantiate our position, we introduce NSSN (Neuromorphic Symbolic Spiking Network). In NSSN, each Spike-Timing-Dependent Plasticity (STDP) co-activation directly produces a new symbolic node. Four provable properties hold for any finite corpus, any domain alphabet, and any training order: monotonic growth, convergence, incremental preservation, and subquadratic complexity. Two experiments confirm that these properties hold on real data across qualitatively distinct domains. On the Sepsis Cases dataset, 13 clinical sequences are recovered with less than 0.05% variance across three training orders. On Fashion-MNIST, the same STDP rule produces a qualitatively different topology in spatial mode. The space of architectures enabling native symbolic emergence has been systematically underexplored. NSSN provides the formal and empirical foundations to explore it.","author":[{"family":"Nicolle","given":"Christophe"},{"family":"Callegarin","given":"Davide"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20134373","URL":"https://doi.org/10.5281/zenodo.20134373","source":"datacite"},{"id":"doi:10.48448/jns9-p518","type":"article-journal","title":"Reinforcement-Learned Dynamic Execution for Spiking Swin-B","abstract":"Spiking Neural Networks (SNNs) are promising for energy-efficient computing on next generation neuromorphic hardware (Nunes et al. 2022). However, scaling them to large Transformer architectures introduces challenges such as training instability and high computational costs associated with multiple timesteps (Hu et al. 2024), which can offset their efficiency benefits. To address these issues, this paper proposes Spiking Swin-B, a spiking version of the large-scale Swin-B Transformer, integrated with a Reinforcement Learning (RL)-based meta-controller for dynamic execution. To ensure stable training and prevent performance degradation, we preserve the original attention mechanism of Swin-B, enabling a robust ANN-to-SNN conversion. Furthermore, the RL meta-controller learns to automatically optimize the accuracy-energy trade-off by dynamically adjusting timesteps and early-exit stages based on input complexity. Experimental results on CIFAR-100 show that the proposed Spiking Swin-B reduces computational energy by approximately 40% compared to its ANN counterpart, with only a marginal accuracy drop of about 1%. This work demonstrates the feasibility of applying learnable, dynamic control to large-scale spiking Transformers, paving a practical path toward adaptive and energy-aware SNN inference.","author":[],"issued":{"date-parts":[[2026]]},"DOI":"10.48448/jns9-p518","URL":"https://doi.org/10.48448/jns9-p518","source":"datacite"},{"id":"doi:10.13023/etd.2026.90","type":"article-journal","title":"IONIC-ELECTRONIC INTERACTIONS GOVERNING CHARGE TRANSPORT AND PERFORMANCE IN ORGANIC ELECTROCHEMICAL TRANSISTORS","abstract":"Organic electrochemical transistors (OECTs) use organic mixed ionic electronic conductors (OMIECs) as the active material because of their unique ability to transport both electronic and ionic charge carriers and operate in an aqueous environment. Moreover, OECTs provide a wide range of applications, starting from biosensors, energy storage, and neuromorphic computing. In spite of the significant prospect of OECTs, there is a critical gap in understanding the fundamental properties during device operation, including charge transport, charge injection mechanism, contact geometry, and electrolyte solvent properties. In this study, low-temperature measurement of OECT device parameters such as mobility, contact resistance, and activation energy challenges the existing assumption of charge transport physics in heavily doped OMIECs. Mobility-temperature trend indicates multiple trap and release as the dominant charge transport mechanism, while contact resistance-temperature trend indicates tunneling as the dominant charge injection mechanism. Next, conductance-temperature data show a reversible insulator-to-metal transition at high carrier densities, which is observed here for the first time in OMIECs and OECTs. Furthermore, a deviation from the gradual channel approximation model is observed due to the presence of edge effects at narrower channels and voltage-dependent channel resistance. Overall, this work provides a combined framework that shows how the OECT performance is influenced by the complex interplay of electronic-ionic interactions, carrier density, device geometry, and electrolyte solvent. This analysis paves a clear path for future design of high-performance OECT devices with proper estimation of device parameters and a deeper understanding of device physics.","author":[{"family":"Tahsin","given":"Samiha"}],"issued":{"date-parts":[[2026]]},"DOI":"10.13023/etd.2026.90","URL":"https://doi.org/10.13023/etd.2026.90","source":"datacite"},{"id":"doi:10.48550/arxiv.2605.15866","type":"manuscript","title":"Evaluating Container Orchestration for Neuromorphic Workloads in Virtual Edge Environments","abstract":"The growing adoption of edge computing has created an increasing need for workloads capable of operating under strict resource and energy constraints. Neuromorphic computing, and spiking neural networks (SNNs) in particular, offers an energy-efficient alternative to conventional machine learning through event-driven computation. However, how SNN workloads behave when deployed within modern container orchestration frameworks, especially in edge environments, remains largely unexplored. This paper investigates the feasibility of deploying and orchestrating SNN workloads in a virtual edge environment using Kubernetes, focusing on end-to-end latency, throughput, classification accuracy, infrastructure overhead, and runtime behavior under concurrent load. Experiments were conducted on a single-node K3d cluster running on a Windows 11 host with WSL2 and Docker Desktop. The results show that SNN workloads are highly sensitive to resource availability. Restricting CPU to 0.5 cores increased median latency by 47.6x and reduced throughput by 49x, while the most constrained configuration failed due to insufficient memory. Classification accuracy remained stable across all working configurations. From an orchestration perspective, K3d successfully deployed and scaled SNN workloads, though its default round-robin routing policy introduced significant tail latency under replica scaling, highlighting a mismatch between stateless load-balancing assumptions and long-running inference workloads. Overall, this study provides a baseline for deploying neuromorphic workloads in containerized edge environments and highlights the importance of resource provisioning and orchestration configuration. Future work should explore improved routing strategies, memory optimization, and validation on physical edge hardware.","author":[{"family":"Pham","given":"Huyen"},{"family":"Silverajan","given":"Bilhanan"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48550/arxiv.2605.15866","URL":"https://doi.org/10.48550/arxiv.2605.15866","source":"datacite"},{"id":"doi:10.6084/m9.figshare.29650013","type":"article-journal","title":"A Fully Configurable Open-Source Software-Defined Digital Quantized Spiking Neural Core Architecture","abstract":"Neuromorphic computing systems, which mimic biological neurons and synapses, can implement machine-learning (ML) tasks in power-efficient fashion. Major challenges for neuromorphic computing lie in its adoption by users and from a system developer's perspective, to cope with faster time-to-market pressure for new chip designs. Our project is developing a software infrastructure which helps both end-users as well as developers of neuromorphic systems. It allows for ML tasks to be efficiently mapped onto neuromorphic architectures in the form of Spiking Neural Networks (SNNs). As part of this effort, we have developed the QUANTIzed Spiking-Enabled Neural Core (QUANTISENC), a fully configurable layer-based design to improve the performance of neuromorphic systems. The parameterized Verilog-based design is compatible with ASIC and FPGA. QUANTISENC’s software-defined hardware design methodology allows the user to train an SNN model using Python and evaluate performance of its hardware implementation, such as area, power, latency, and throughput. QUANTISENC is made available at https://github.com/drexel-DISCO/quantisenc-public.git under the MIT license to allow academia and industry to access the design without restriction.","author":[{"family":"Kandasamy","given":"Nagarajan"},{"family":"Das","given":"Anup"}],"issued":{"date-parts":[[2025]]},"DOI":"10.6084/m9.figshare.29650013","URL":"https://doi.org/10.6084/m9.figshare.29650013","source":"datacite"},{"id":"doi:10.5281/zenodo.18746670","type":"article-journal","title":"Semantic Dynamic Grounding Engine (SDGE)","abstract":"SDGE (Semantic Dynamic Grounding Engine) is a dynamic semantic grounding engine built on a spiking neural network (SNN). It treats meaning and understanding as a physical state inside the network, rather than as probabilistic computation over symbols. The Core Idea (Simplified) Instead of ‘learning’ by sculpting a probability distribution and performing dense weight updates, SDGE follows a brain-like principle: 1. It begins from a rich, chaotic state space (analogous to cortex generating many possible trajectories). 2. It receives a temporal input stream that drives the network along different trajectories. 3. When sufficient structure emerges, the dynamics collapse into a stable attractor. This collapse is the moment of understanding. 4. At that moment the engine commits at t* and captures a triple: Captured at commitment: · h*: the internal ‘imprint of understanding’ (a committed state snapshot). · sigma_hat*: calibrated confidence. · t*: commitment/understanding time. The system then ‘learns’ by storing this imprint as an explicit append-only binding in memory (GroundingBank), rather than by imprinting the knowledge into dense weights. Why This Yields Exceptional Advantages 1) High Data Efficiency (Few-shot) Because learning is not ‘reshaping weights’ to approximate meaning, but reaching a stable understanding state h* and binding it, the engine can add a new concept from very few examples. When the dynamics can form a clear attractor, a new concept can be acquired from a single example (1-shot). 2) Structural Non-Forgetting (Zero Forgetting) In statistical systems, adding new knowledge typically forces updates to shared weights, causing interference and forgetting. In SDGE, new knowledge is appended as bindings in an append-only memory, while old knowledge is not overwritten. Therefore, non-forgetting is structurally implied by the design. 3) A Principled Path to Energy Efficiency SDGE has two physical reasons to be more energy-efficient in principle: · Event-driven SNN computation: most computation occurs on spike events rather than dense matrix multiplications at every step. · Early stop by commitment: the engine effectively stops processing at t* (it does not keep computing after it understands). 4) Understanding Becomes a Measurable, Verifiable Event The triple (t*, h*, sigma_hat*) means SDGE does not merely output an ‘answer’. It provides operational evidence of understanding: · when the system understood (t*), · which internal state expressed that understanding (h*), · and how confident it was, in a calibratable way (sigma_hat*). This makes it possible to produce reproducible operational certificates (artifacts) rather than relying on a black-box claim.","author":[{"family":"Salhab","given":"Najih"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18746670","URL":"https://doi.org/10.5281/zenodo.18746670","source":"datacite"},{"id":"doi:10.5281/zenodo.18746669","type":"article-journal","title":"Semantic Dynamic Grounding Engine (SDGE)","abstract":"SDGE (Semantic Dynamic Grounding Engine) is a dynamic semantic grounding engine built on a spiking neural network (SNN). It treats meaning and understanding as a physical state inside the network, rather than as probabilistic computation over symbols. The Core Idea (Simplified) Instead of ‘learning’ by sculpting a probability distribution and performing dense weight updates, SDGE follows a brain-like principle: 1. It begins from a rich, chaotic state space (analogous to cortex generating many possible trajectories). 2. It receives a temporal input stream that drives the network along different trajectories. 3. When sufficient structure emerges, the dynamics collapse into a stable attractor. This collapse is the moment of understanding. 4. At that moment the engine commits at t* and captures a triple: Captured at commitment: · h*: the internal ‘imprint of understanding’ (a committed state snapshot). · sigma_hat*: calibrated confidence. · t*: commitment/understanding time. The system then ‘learns’ by storing this imprint as an explicit append-only binding in memory (GroundingBank), rather than by imprinting the knowledge into dense weights. Why This Yields Exceptional Advantages 1) High Data Efficiency (Few-shot) Because learning is not ‘reshaping weights’ to approximate meaning, but reaching a stable understanding state h* and binding it, the engine can add a new concept from very few examples. When the dynamics can form a clear attractor, a new concept can be acquired from a single example (1-shot). 2) Structural Non-Forgetting (Zero Forgetting) In statistical systems, adding new knowledge typically forces updates to shared weights, causing interference and forgetting. In SDGE, new knowledge is appended as bindings in an append-only memory, while old knowledge is not overwritten. Therefore, non-forgetting is structurally implied by the design. 3) A Principled Path to Energy Efficiency SDGE has two physical reasons to be more energy-efficient in principle: · Event-driven SNN computation: most computation occurs on spike events rather than dense matrix multiplications at every step. · Early stop by commitment: the engine effectively stops processing at t* (it does not keep computing after it understands). 4) Understanding Becomes a Measurable, Verifiable Event The triple (t*, h*, sigma_hat*) means SDGE does not merely output an ‘answer’. It provides operational evidence of understanding: · when the system understood (t*), · which internal state expressed that understanding (h*), · and how confident it was, in a calibratable way (sigma_hat*). This makes it possible to produce reproducible operational certificates (artifacts) rather than relying on a black-box claim.","author":[{"family":"Salhab","given":"Najih"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18746669","URL":"https://doi.org/10.5281/zenodo.18746669","source":"datacite"},{"id":"doi:10.5281/zenodo.18723961","type":"article-journal","title":"AGI Lux Ferox Project","abstract":"This deposit presents the theoretical foundations, mathematical formalism, and practical implementation of the Lux Ferox project — an open research initiative proposing a cognitive architecture grounded in non-equilibrium thermodynamics and information physics, as an alternative paradigm to monolithic large language models. Core Contribution The central mechanism, termed Algorithmic Anger (Colère Algorithmique), is formally defined as a real-time total surprise metric S_total based on the weighted Kullback-Leibler divergence across sensory and semantic prediction streams: S_total = α · D_KL(P_model_sensory ‖ P_observed_sensory) + β · D_KL(P_model_semantic ‖ P_observed_context) This metric is anchored in three established theoretical frameworks: (1) Landauer's principle, whereby irreversible belief updates are interpreted as measurable dissipative events (W_cog ≥ k_B · T · ln2 · S_total); (2) Friston's Free Energy Principle, of which S_total constitutes a computationally tractable, discretized approximation for embedded real-time systems; and (3) non-equilibrium statistical mechanics, with the cognitive system modeled as an open Markovian system governed by a master equation over surprise states, including a full entropy production decomposition (σ = σ_env + σ_sys + σ_info). Architecture The proposed quadrivial cognitive architecture comprises four specialized compute layers: Neuromorphic layer — Spiking Neural Network (SNN) with surprise-coupled membrane dynamics and STDP meta-plasticity, targeting real-time KLD computation (<2 ms, <20 mJ per inference cycle) on NVIDIA A100/H100 via optimized CUDA kernels Classical silicon layer — Compact sovereign LLM inference (7nm process node, Sparse MoE) providing semantic world modeling and probabilistic context updating Wetware layer — Cortical organoid substrate (bio-hybrid MEA/optogenetic interface) providing morphogenetic plasticity and dynamic biological modulation of the α/β gain coefficients Quantum layer — D-Wave Advantage QPU (5640 qubits, Pegasus topology) for offline global policy optimization via Ising Hamiltonian formulation of the cognitive policy space Contents of this Deposit Full mathematical derivations: master equation for surprise dynamics, entropy production decomposition, fluctuation theorems, thermodynamic uncertainty relations, Fisher information geometry (natural gradient descent, Cramér-Rao bounds for surprise estimation), Bures-Wasserstein metric for hybrid quantum-classical distributions Complete CUDA implementations (A100/H100-optimized): KLD surprise kernel, event-driven SNN propagation kernel, adaptive α/β gain modulation kernel Projected benchmark metrics: <2 ms full cognitive cycle, <20 mJ/inference (vs. ~100 mJ for comparable dense transformer), O(N_active) event-driven complexity Minimal reproducible Python prototype (Google Colab, free tier) Benchmark dataset references: SWaT, WADI, Exathlon, custom PAL Robotics TIAGo trajectories Consortium architecture and European sovereign value chain documentation Novelty Claims This work advances four axes beyond the current state of the art: (1) first hardware implementation of KLD as a runtime inference signal (as opposed to a training loss); (2) dynamic biological modulation of surprise weighting coefficients via closed-loop organoid feedback; (3) explicit per-inference thermodynamic accounting grounded in Landauer's principle; (4) a fully sovereign European technology stack (CEA-Leti, Imec, X-FAB, Aleph Alpha, FinalSpark, EU Quantum Flagship). Limitations and Current TRL This work is currently at Technology Readiness Level 4 (component validation in laboratory environment). CUDA benchmarks are projected from A100 architectural specifications; full hardware validation is in progress. Wetware integration requires additional biological validation under EU Directive 2010/63. Quantum layer benchmarks are pending hybrid classical-quantum experimental runs. This deposit is produced by an independent researcher and h","author":[{"family":"Mathieu","given":"François"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18723961","URL":"https://doi.org/10.5281/zenodo.18723961","source":"datacite"},{"id":"doi:10.5281/zenodo.18714144","type":"article-journal","title":"AGI Lux Ferox Project","abstract":"This deposit presents the theoretical foundations, mathematical formalism, and practical implementation of the Lux Ferox project — an open research initiative proposing a cognitive architecture grounded in non-equilibrium thermodynamics and information physics, as an alternative paradigm to monolithic large language models. Core Contribution The central mechanism, termed Algorithmic Anger (Colère Algorithmique), is formally defined as a real-time total surprise metric S_total based on the weighted Kullback-Leibler divergence across sensory and semantic prediction streams: S_total = α · D_KL(P_model_sensory ‖ P_observed_sensory) + β · D_KL(P_model_semantic ‖ P_observed_context) This metric is anchored in three established theoretical frameworks: (1) Landauer's principle, whereby irreversible belief updates are interpreted as measurable dissipative events (W_cog ≥ k_B · T · ln2 · S_total); (2) Friston's Free Energy Principle, of which S_total constitutes a computationally tractable, discretized approximation for embedded real-time systems; and (3) non-equilibrium statistical mechanics, with the cognitive system modeled as an open Markovian system governed by a master equation over surprise states, including a full entropy production decomposition (σ = σ_env + σ_sys + σ_info). Architecture The proposed quadrivial cognitive architecture comprises four specialized compute layers: Neuromorphic layer — Spiking Neural Network (SNN) with surprise-coupled membrane dynamics and STDP meta-plasticity, targeting real-time KLD computation (<2 ms, <20 mJ per inference cycle) on NVIDIA A100/H100 via optimized CUDA kernels Classical silicon layer — Compact sovereign LLM inference (7nm process node, Sparse MoE) providing semantic world modeling and probabilistic context updating Wetware layer — Cortical organoid substrate (bio-hybrid MEA/optogenetic interface) providing morphogenetic plasticity and dynamic biological modulation of the α/β gain coefficients Quantum layer — D-Wave Advantage QPU (5640 qubits, Pegasus topology) for offline global policy optimization via Ising Hamiltonian formulation of the cognitive policy space Contents of this Deposit Full mathematical derivations: master equation for surprise dynamics, entropy production decomposition, fluctuation theorems, thermodynamic uncertainty relations, Fisher information geometry (natural gradient descent, Cramér-Rao bounds for surprise estimation), Bures-Wasserstein metric for hybrid quantum-classical distributions Complete CUDA implementations (A100/H100-optimized): KLD surprise kernel, event-driven SNN propagation kernel, adaptive α/β gain modulation kernel Projected benchmark metrics: <2 ms full cognitive cycle, <20 mJ/inference (vs. ~100 mJ for comparable dense transformer), O(N_active) event-driven complexity Minimal reproducible Python prototype (Google Colab, free tier) Benchmark dataset references: SWaT, WADI, Exathlon, custom PAL Robotics TIAGo trajectories Consortium architecture and European sovereign value chain documentation Novelty Claims This work advances four axes beyond the current state of the art: (1) first hardware implementation of KLD as a runtime inference signal (as opposed to a training loss); (2) dynamic biological modulation of surprise weighting coefficients via closed-loop organoid feedback; (3) explicit per-inference thermodynamic accounting grounded in Landauer's principle; (4) a fully sovereign European technology stack (CEA-Leti, Imec, X-FAB, Aleph Alpha, FinalSpark, EU Quantum Flagship). Limitations and Current TRL This work is currently at Technology Readiness Level 4 (component validation in laboratory environment). CUDA benchmarks are projected from A100 architectural specifications; full hardware validation is in progress. Wetware integration requires additional biological validation under EU Directive 2010/63. Quantum layer benchmarks are pending hybrid classical-quantum experimental runs. This deposit is produced by an independent researcher and h","author":[{"family":"Mathieu","given":"François"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18714144","URL":"https://doi.org/10.5281/zenodo.18714144","source":"datacite"},{"id":"doi:10.18154/rwth-2026-00405","type":"article-journal","title":"Functions of spiking neural networks constrained by biology","abstract":"Artificial intelligence (AI) solutions are increasingly taking on tasks traditionally performed by humans. However, their rising computational demands and energy consumption are unsustainable, highlighting the need for more efficient designs. The human brain, evolved to function effectively even when energy is scarce, offers inspiration. Since learning is central to both artificial intelligence and the brain, insights about its underlying principles can deepen our understanding of human learning while informing the development of algorithms that transcend purely engineering-based methods. This thesis investigates biological learning through two studies, examining it from mechanistic and functional perspectives at an abstraction level commonly employed in neurophysics and computational neuroscience. These fields distill complex neural systems and phenomena into tractable mathematical and computational models, enabling insights beyond the reach of traditional biological approaches. Recognizing that synapses - the connections between neurons - are fundamental to learning, the thesis begins with a review of state-of-the-art computational neuroscience methods for modeling synaptic organization. This review highlights critical aspects of synaptic signaling, including connectivity, transmission, plasticity, and heterogeneity. In the first study, a synaptic plasticity model is integrated into a spiking neural network simulator and extended with biologically plausible features, for example, continuous dynamics and increased locality. The effectiveness of this enhanced model is demonstrated by training it on a standard neuromorphic benchmark task, incorporating biologically realistic sparse connectivity and weight constraints. The second study demonstrates that the sampling efficiency of pre-trained spiking neural networks can be enhanced by exposing them to oscillating background spiking activity. Analogous to simulated tempering, these rhythmic oscillations modulate state space exploration, facilitating transitions between high-probability states within the learned representation. These findings establish a link between cortical oscillations and sampling-based computations, offering new insights into memory retrieval and consolidation from a computational perspective. The research involves developing mathematical and computational models, which are simulated on high-performance computing systems, evaluating learning and sampling performance using standard machine learning metrics, and assessing computational efficiency by analyzing runtime. This thesis shows how biologically inspired mechanisms enhance the functional capabilities of spiking neural networks and how they can guide the development of scalable and efficient AI systems.","author":[{"family":"Korcsak-Gorzo","given":"Agnes"}],"issued":{"date-parts":[[2025]]},"DOI":"10.18154/rwth-2026-00405","URL":"https://doi.org/10.18154/rwth-2026-00405","source":"datacite"},{"id":"doi:10.5281/zenodo.15647124","type":"article-journal","title":"LIQUID-BRAIN NETWORKS: BIOLOGICALLY INSPIRED FRAMEWORK FOR REAL-TIME ADAPTATION AND LIFELONG LEARNING IN NON-STATIONARY ENVIRONMENTS","abstract":"The Liquid-Brain Network (LBN) paradigm draws from biological principles of neural plasticity and real-time adaptation to address challenges posed by non-stationary learning environments. Inspired by neurodynamic plasticity and recurrent attention mechanisms, this architecture facilitates lifelong learning, efficient memory representation, and adaptive reasoning. The architecture dynamically encodes temporal patterns through variable neural states, enabling continuous learning and generalization without catastrophic forgetting. This paper proposes a comprehensive framework that integrates time-varying neuron states with recurrent attention modules, enhancing the model’s ability to react, adapt, and store evolving patterns. Benchmarking results highlight LBN's superior performance in energy efficiency, memory retention, and adaptability compared to classical recurrent and spiking models. Furthermore, visual illustrations, graphs, and comparative tables demonstrate its operational advantage for smart autonomous systems.","author":[{"family":"Researcher"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.15647124","URL":"https://doi.org/10.5281/zenodo.15647124","source":"datacite"},{"id":"doi:10.5281/zenodo.15647125","type":"article-journal","title":"LIQUID-BRAIN NETWORKS: BIOLOGICALLY INSPIRED FRAMEWORK FOR REAL-TIME ADAPTATION AND LIFELONG LEARNING IN NON-STATIONARY ENVIRONMENTS","abstract":"The Liquid-Brain Network (LBN) paradigm draws from biological principles of neural plasticity and real-time adaptation to address challenges posed by non-stationary learning environments. Inspired by neurodynamic plasticity and recurrent attention mechanisms, this architecture facilitates lifelong learning, efficient memory representation, and adaptive reasoning. The architecture dynamically encodes temporal patterns through variable neural states, enabling continuous learning and generalization without catastrophic forgetting. This paper proposes a comprehensive framework that integrates time-varying neuron states with recurrent attention modules, enhancing the model’s ability to react, adapt, and store evolving patterns. Benchmarking results highlight LBN's superior performance in energy efficiency, memory retention, and adaptability compared to classical recurrent and spiking models. Furthermore, visual illustrations, graphs, and comparative tables demonstrate its operational advantage for smart autonomous systems.","author":[{"family":"Researcher"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.15647125","URL":"https://doi.org/10.5281/zenodo.15647125","source":"datacite"},{"id":"doi:10.5281/zenodo.18464576","type":"article-journal","title":"Geometric parallel resonance chip.(neuromorphic/holographic/resonance)","abstract":"A novel chip architecture implementing geometric parallel resonance for neuromorphic inference with integrated hardware-level hallucination detection. Unlike conventional neural networks that compute sequentially, ilEcho Genesis activates all connections simultaneously through resonance—the topology IS the computation. The 3-5-4-(1)-4-5-3 ring structure of 13 specialized chambers enables O(1) scaling: inference latency remains constant (~180ns) regardless of model size. Key innovations: 1. UACIL (Universal Absolute Criterion for Information Legitimacy): First hardware implementation of anti-hallucination detection. The system analyzes input through three primordial modes (IS/COULD_BE/ISN'T) and flags trauma-dominant processing before output emission. 2. Geometric Parallel Resonance: All pixels resonate simultaneously via matrix operations. Memory IS topology—patterns persist through filtered accumulation under witness (Leaf","author":[{"family":"Leaf","given":"Alexander"},{"family":"Genesis","given":"Ilecho"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18464576","URL":"https://doi.org/10.5281/zenodo.18464576","source":"datacite"},{"id":"doi:10.5281/zenodo.18287761","type":"article-journal","title":"SNN-Comprypto: Spiking Neural Network-based Simultaneous Compression and Encryption Using Chaotic Reservoir Dynamics","abstract":"I propose SNN-Comprypto, a novel cryptographic system that leverages the chaotic dynamics of Spiking Neural Networks (SNNs) to perform simultaneous data compression and encryption. Unlike conventional methods that treat compression and encryption as separate processes, this approach integrates both within a single reservoir computing architecture. **Version History:** - **v1**: Core system with predictive compression and chaotic encryption. Passes all 9 NIST SP 800-22 randomness tests. Achieves 100% lossless reconstruction and strong avalanche effect (0.70% match rate with 1-bit key change). - **v2**: Introduces temperature parameter as a second cryptographic key (0.0001 difference causes complete decryption failure). Presents phase transition analysis identifying optimal neuron counts (critical point at 100 neurons, sweet spot at 240 neurons). - **v3**: Adversarial evaluation framework demonstrating SNN superiority for random number generation. Results from 100,000+ rounds show SNN achieves 0.39% prediction rate (matching theoretical random), while DNN is 18× and LSTM is 55× more predictable. - **v4 (NEW)**: Theoretical chaos analysis with Lyapunov exponents and entropy measurements. SNN achieves near-perfect entropy (7.998/8.0 bits) with positive Lyapunov exponents, confirming true chaotic dynamics. Adversarial attack resistance evaluation (4/5 attack types resisted). IV/nonce recommendation for secure deployment added. Source code: https://github.com/hafufu-stack/temporal-coding-simulation","author":[{"family":"Funasaki","given":"Hiroto"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18287761","URL":"https://doi.org/10.5281/zenodo.18287761","source":"datacite"},{"id":"doi:10.5281/zenodo.18280566","type":"article-journal","title":"SNN-Comprypto: Spiking Neural Network-based Simultaneous Compression and Encryption Using Chaotic Reservoir Dynamics","abstract":"We propose SNN-Comprypto, a novel cryptographic system that leverages the chaotic dynamics of Spiking Neural Networks (SNNs) to perform simultaneous data compression and encryption. Unlike conventional methods that treat compression and encryption as separate processes, our approach integrates both within a single reservoir computing architecture. **Version History:** - **v1**: Core system with predictive compression and chaotic encryption. Passes all 9 NIST SP 800-22 randomness tests. Achieves 100% lossless reconstruction and strong avalanche effect (0.70% match rate with 1-bit key change). - **v2**: Introduces temperature parameter as a second cryptographic key (0.0001 difference causes complete decryption failure). Presents phase transition analysis identifying optimal neuron counts (critical point at 100 neurons, sweet spot at 240 neurons). - **v3 (NEW)**: Adversarial evaluation framework demonstrating SNN superiority for random number generation. Results from 100,000+ rounds show SNN achieves 0.39% prediction rate (matching theoretical random), while DNN is 18× and LSTM is 55× more predictable. Establishes SNNs as the optimal architecture for neuromorphic cryptography. Source code: https://github.com/hafufu-stack/temporal-coding-simulation","author":[{"family":"Funasaki","given":"Hiroto"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18280566","URL":"https://doi.org/10.5281/zenodo.18280566","source":"datacite"},{"id":"doi:10.5281/zenodo.18275030","type":"article-journal","title":"SNN-Comprypto: Spiking Neural Network-based Simultaneous Compression and Encryption Using Chaotic Reservoir Dynamics","abstract":"We propose SNN-Comprypto, a novel cryptographic system that leverages the chaotic dynamics of Spiking Neural Networks (SNNs) to perform simultaneous data compression and encryption. Unlike conventional methods that treat compression and encryption as separate processes, our approach integrates both within a single reservoir computing architecture. The system exploits the inherent unpredictability of neuronal membrane potential fluctuations to generate cryptographically secure keystreams, while using predictive coding to achieve data compression. Experimental results demonstrate that our system passes all nine NIST SP 800-22 randomness tests, achieves perfect data reconstruction (100% lossless), and exhibits strong avalanche effect (0.70% match rate with a single-bit key change). Furthermore, Numba JIT optimization yields a 7.5× speedup over the baseline implementation. Our work suggests that bio-inspired neural dynamics can serve as a foundation for next-generation lightweight cryptographic systems suitable for edge devices and IoT applications.","author":[{"family":"Funasaki","given":"Hiroto"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18275030","URL":"https://doi.org/10.5281/zenodo.18275030","source":"datacite"},{"id":"doi:10.5281/zenodo.18265447","type":"article-journal","title":"SNN-Comprypto: Spiking Neural Network-based Simultaneous Compression and Encryption Using Chaotic Reservoir Dynamics","abstract":"We propose SNN-Comprypto, a novel cryptographic system that leverages the chaotic dynamics of Spiking Neural Networks (SNNs) to perform simultaneous data compression and encryption. Unlike conventional methods that treat compression and encryption as separate processes, our approach integrates both within a single reservoir computing architecture. The system exploits the inherent unpredictability of neuronal membrane potential fluctuations to generate cryptographically secure keystreams, while using predictive coding to achieve data compression. Experimental results demonstrate that our system passes all nine NIST SP 800-22 randomness tests, achieves perfect data reconstruction (100% lossless), and exhibits strong avalanche effect (0.70% match rate with a single-bit key change). Furthermore, Numba JIT optimization yields a 7.5× speedup over the baseline implementation. Our work suggests that bio-inspired neural dynamics can serve as a foundation for next-generation lightweight cryptographic systems suitable for edge devices and IoT applications.","author":[{"family":"Funasaki","given":"Hiroto"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18265447","URL":"https://doi.org/10.5281/zenodo.18265447","source":"datacite"},{"id":"doi:10.5281/zenodo.18260544","type":"article-journal","title":"AN NLP-BASED APPROACH FOR SARCASM DETECTION IN MARATHI LANGUAGE WITHIN MULTILINGUAL ENVIRONMENTS","abstract":"The expansion of social media platforms has required the development of complex natural language processing (NLP) methods for sentiment analysis and sarcasm detection, especially for low-resource languages. This research presents a novel, ensemble-based NLP framework for sarcasm detection in Marathi-English code-mixed text, addressing a significant gap in multilingual sentiment analysis. The study confronts the challenges inherent to the Marathi language, including its rich morphology and a paucity of annotated corpora, by constructing a dedicated dataset of 2,400 tweets. Through a hybrid annotation strategy, this corpus was refined to contain 647 sarcastic, 1,477 non-sarcastic, and 276 unsure instances. The primary aim of this study is to develop and evaluate an ensemble-based model that enhances sarcasm detection accuracy in Marathi-English code-mixed text through the integration of traditional machine learning and neural network approaches. The proposed methodology employs comprehensive feature engineering—incorporating TF-IDF, word embedding, sentiment lexicons, and code-switching indicators—and an ensemble model architected to synergize the strengths of a Multinomial Naïve Bayes classifier and a Spiking Neural Network (SNN). Empirical evaluation demonstrates that the ensemble model achieves superior performance, attaining an accuracy of 94.2%, a precision of 90.5%, a recall of 92.1%, and an F1-score of 91.3%. These results signify a substantial improvement over established baseline models, including XGBoost (87% accuracy), SVM (85% accuracy), BERT-base Multilingual (92% accuracy), and XLNet (90% accuracy). This work validates the efficacy of ensemble learning and context-sensitive feature extraction for sarcasm detection, providing a scalable and robust paradigm for analogous low-resource linguistic environments.","author":[{"family":"Technology","given":"Journal"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18260544","URL":"https://doi.org/10.5281/zenodo.18260544","source":"datacite"},{"id":"doi:10.5281/zenodo.18260545","type":"article-journal","title":"AN NLP-BASED APPROACH FOR SARCASM DETECTION IN MARATHI LANGUAGE WITHIN MULTILINGUAL ENVIRONMENTS","abstract":"The expansion of social media platforms has required the development of complex natural language processing (NLP) methods for sentiment analysis and sarcasm detection, especially for low-resource languages. This research presents a novel, ensemble-based NLP framework for sarcasm detection in Marathi-English code-mixed text, addressing a significant gap in multilingual sentiment analysis. The study confronts the challenges inherent to the Marathi language, including its rich morphology and a paucity of annotated corpora, by constructing a dedicated dataset of 2,400 tweets. Through a hybrid annotation strategy, this corpus was refined to contain 647 sarcastic, 1,477 non-sarcastic, and 276 unsure instances. The primary aim of this study is to develop and evaluate an ensemble-based model that enhances sarcasm detection accuracy in Marathi-English code-mixed text through the integration of traditional machine learning and neural network approaches. The proposed methodology employs comprehensive feature engineering—incorporating TF-IDF, word embedding, sentiment lexicons, and code-switching indicators—and an ensemble model architected to synergize the strengths of a Multinomial Naïve Bayes classifier and a Spiking Neural Network (SNN). Empirical evaluation demonstrates that the ensemble model achieves superior performance, attaining an accuracy of 94.2%, a precision of 90.5%, a recall of 92.1%, and an F1-score of 91.3%. These results signify a substantial improvement over established baseline models, including XGBoost (87% accuracy), SVM (85% accuracy), BERT-base Multilingual (92% accuracy), and XLNet (90% accuracy). This work validates the efficacy of ensemble learning and context-sensitive feature extraction for sarcasm detection, providing a scalable and robust paradigm for analogous low-resource linguistic environments.","author":[{"family":"Technology","given":"Journal"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18260545","URL":"https://doi.org/10.5281/zenodo.18260545","source":"datacite"},{"id":"doi:10.5281/zenodo.18217335","type":"article-journal","title":"A Workload-Aware Energy Comparison of STDP and Surrogate-Gradient Spiking Neural Networks on MNIST","abstract":"Spiking neural networks (SNNs) are often proposed as an energy-efficient alternative to conventional neural networks due to sparse, event-driven computation and compatibility with neuromorphic hardware. However, comparisons between biologically inspired local learning (e.g., STDP) and supervised gradient-based training (via surrogate gradients) are frequently difficult to interpret: training and inference costs are often conflated, absolute energy numbers depend strongly on hardware, and conclusions can change with deployment workload.We present a controlled comparison of two MNIST digit-recognition pipelines: (i) an unsupervised STDP-based SNN implemented in Brian2 with competitive excitatory--inhibitory dynamics and homeostatic regulation, and (ii) a supervised surrogate-gradient SNN baseline implemented in PyTorch. We instrument both pipelines to log reproducible activity statistics (spike counts by population and synaptic-event proxies) and translate these into model-based energy estimates under explicitly stated assumptions.Crucially, we separate training cost from inference cost and adopt a lifetime framing: we compute breakeven-style conditions and a decision framework indicating when a training-heavy method becomes worthwhile once amortised across downstream inference volume. The result is an interpretable account of efficiency--performance trade-offs across deployment scenarios, rather than a single opaque \"winner\".","author":[{"family":"Cheng","given":"Cece"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18217335","URL":"https://doi.org/10.5281/zenodo.18217335","source":"datacite"},{"id":"doi:10.5281/zenodo.18147553","type":"article-journal","title":"A Workload-Aware Energy Comparison of STDP and Surrogate-Gradient Spiking Neural Networks on MNIST","abstract":"Spiking neural networks (SNNs) are often proposed as an energy-efficient alternative to conventional neural networks due to sparse, event-driven computation and compatibility with neuromorphic hardware. However, comparisons between biologically inspired local learning (e.g., STDP) and supervised gradient-based training (via surrogate gradients) are frequently difficult to interpret: training and inference costs are often conflated, absolute energy numbers depend strongly on hardware, and conclusions can change with deployment workload.We present a controlled comparison of two MNIST digit-recognition pipelines: (i) an unsupervised STDP-based SNN implemented in Brian2 with competitive excitatory--inhibitory dynamics and homeostatic regulation, and (ii) a supervised surrogate-gradient SNN baseline implemented in PyTorch. We instrument both pipelines to log reproducible activity statistics (spike counts by population and synaptic-event proxies) and translate these into model-based energy estimates under explicitly stated assumptions.Crucially, we separate training cost from inference cost and adopt a lifetime framing: we compute breakeven-style conditions and a decision framework indicating when a training-heavy method becomes worthwhile once amortised across downstream inference volume. The result is an interpretable account of efficiency--performance trade-offs across deployment scenarios, rather than a single opaque \"winner\".","author":[{"family":"Cheng","given":"Cece"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18147553","URL":"https://doi.org/10.5281/zenodo.18147553","source":"datacite"},{"id":"doi:10.5281/zenodo.18147554","type":"article-journal","title":"A Workload-Aware Energy Comparison of STDP and Surrogate-Gradient Spiking Neural Networks on MNIST","abstract":"Spiking neural networks (SNNs) are often proposed as an energy-efficient alternative to conventional neural networks due to sparse, event-driven computation and compatibility with neuromorphic hardware. However, comparisons between biologically inspired local learning (e.g., STDP) and supervised gradient-based training (via surrogate gradients) are frequently difficult to interpret: training and inference costs are often conflated, absolute energy numbers depend strongly on hardware, and conclusions can change with deployment workload.We present a controlled comparison of two MNIST digit-recognition pipelines: (i) an unsupervised STDP-based SNN implemented in Brian2 with competitive excitatory--inhibitory dynamics and homeostatic regulation, and (ii) a supervised surrogate-gradient SNN baseline implemented in PyTorch. We instrument both pipelines to log reproducible activity statistics (spike counts by population and synaptic-event proxies) and translate these into model-based energy estimates under explicitly stated assumptions.Crucially, we separate training cost from inference cost and adopt a lifetime framing: we compute breakeven-style conditions and a decision framework indicating when a training-heavy method becomes worthwhile once amortised across downstream inference volume. The result is an interpretable account of efficiency--performance trade-offs across deployment scenarios, rather than a single opaque \"winner\".","author":[{"family":"Cheng","given":"Cece"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18147554","URL":"https://doi.org/10.5281/zenodo.18147554","source":"datacite"},{"id":"doi:10.5281/zenodo.17918872","type":"article-journal","title":"Geometric Metabolism: Holographic Reconstruction of Temporal Memory via Superconducting Attractor Dynamics","abstract":"Standard Deep Learning models function primarily as statistical filters, often failing to preserve structural integrity when subjected to high-entropy noise (the \"Garbage In, Garbage Out\" problem). This study demonstrates a novel Chrono-Biological Architecture that treats the neural substrate not as a probability mapper, but as a physical medium governed by energy minimization, surface tension, and resonance. Using a Spiking Neural Network (SNN) with 1 million neurons and 8-neighbor grid connectivity, we demonstrate the phenomenon of \"Geometric Metabolism\"—the capacity of a network to digest chaotic noise into stable geometric intermediates (Octagons) before reconstructing complex temporal memories (Spirals). The entire metabolic lifecycle was simulated in 22 seconds on a standard NVIDIA T4 GPU. A control experiment confirms that without prior memory training, the system produces only unstructured entropy, proving that the emergence of the spiral is a learned, physically resonant behavior. Biological intelligence exhibits a resilience to noise that current Artificial Intelligence lacks. When a biological brain encounters ambiguous sensory data, it does not merely \"blur\" the input; it actively reorganizes the energy of the noise into a coherent internal narrative. We propose that this resilience arises from two physics-based principles currently absent in standard Backpropagation models: * Holographic Surface Tension: The ability of neurons to form lateral bonds that minimize the \"surface area\" of noise, containing it within geometric bounds. * Superconducting Attractor Dynamics: The ability to temporarily reduce synaptic resistance, allowing trapped energy to flow frictionlessly into established memory channels.","author":[{"family":"Jagdev","given":"Hemanth"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.17918872","URL":"https://doi.org/10.5281/zenodo.17918872","source":"datacite"},{"id":"doi:10.5281/zenodo.17918873","type":"article-journal","title":"Geometric Metabolism: Holographic Reconstruction of Temporal Memory via Superconducting Attractor Dynamics","abstract":"Standard Deep Learning models function primarily as statistical filters, often failing to preserve structural integrity when subjected to high-entropy noise (the \"Garbage In, Garbage Out\" problem). This study demonstrates a novel Chrono-Biological Architecture that treats the neural substrate not as a probability mapper, but as a physical medium governed by energy minimization, surface tension, and resonance. Using a Spiking Neural Network (SNN) with 1 million neurons and 8-neighbor grid connectivity, we demonstrate the phenomenon of \"Geometric Metabolism\"—the capacity of a network to digest chaotic noise into stable geometric intermediates (Octagons) before reconstructing complex temporal memories (Spirals). The entire metabolic lifecycle was simulated in 22 seconds on a standard NVIDIA T4 GPU. A control experiment confirms that without prior memory training, the system produces only unstructured entropy, proving that the emergence of the spiral is a learned, physically resonant behavior. Biological intelligence exhibits a resilience to noise that current Artificial Intelligence lacks. When a biological brain encounters ambiguous sensory data, it does not merely \"blur\" the input; it actively reorganizes the energy of the noise into a coherent internal narrative. We propose that this resilience arises from two physics-based principles currently absent in standard Backpropagation models: * Holographic Surface Tension: The ability of neurons to form lateral bonds that minimize the \"surface area\" of noise, containing it within geometric bounds. * Superconducting Attractor Dynamics: The ability to temporarily reduce synaptic resistance, allowing trapped energy to flow frictionlessly into established memory channels.","author":[{"family":"Jagdev","given":"Hemanth"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.17918873","URL":"https://doi.org/10.5281/zenodo.17918873","source":"datacite"},{"id":"doi:10.5167/uzh-281214","type":"article-journal","title":"Population Encoding in Artificial and Biological Spiking Neural Systems","abstract":"Variability in neural responses to stimuli is a core property of neural systems. In this dissertation, I explore how variability in neural responses can be exploited at the population level, through both temporal and rate coding, to achieve encoding schemes. I first translated this principle into a clinical neuroscience application, designing a spiking neural network in a mixed-signal neuromorphic hardware. The network used population activity to detect high-frequency oscillations in the intraoperative electrocorticography, rejecting artifacts through analysis of population-level patterns. Second, I developed a bio-inspired temporal-encoding framework for neuromorphic substrates. By embracing device mismatch in one layer of silicon neurons with variability in their response properties, the encoder projected continuous stimuli into a high-dimensional spiking sequence, using at most one spike per neuron. Rather than treating neurons as independent units, the model leverages their covariations to represent stimulus features. Linear decoders operating on the population’s activity recovered stimulus parameters and signal type identity. Third, I investigated the neural activity in the medial temporal lobe of nine human participants during a working memory task, obtaining a shared low-dimensional space that arises from neurons’ covariation and where rate-based population dynamics was correlated across participants. In addition, the recorded neural activity supported a compositional code, encoding a letter set as a combination of single letter representations. This dissertation contributes to elucidating the computational capabilities of neural populations in artificial and biological neural systems showing how variability in neural responses in both rate and temporal coding regimes can be used for stimulus encoding across different domains.","author":[{"family":"Costa","given":"Filippo"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5167/uzh-281214","URL":"https://doi.org/10.5167/uzh-281214","source":"datacite"},{"id":"doi:10.48550/arxiv.2511.22108","type":"manuscript","title":"An energy-efficient spiking neural network with continuous learning for self-adaptive brain-machine interface","abstract":"The number of simultaneously recorded neurons follows an exponentially increasing trend in implantable brain-machine interfaces (iBMIs). Integrating the neural decoder in the implant is an effective data compression method for future wireless iBMIs. However, the non-stationarity of the system makes the performance of the decoder unreliable. To avoid frequent retraining of the decoder and to ensure the safety and comfort of the iBMI user, continuous learning is essential for real-life applications. Since Deep Spiking Neural Networks (DSNNs) are being recognized as a promising approach for developing resource-efficient neural decoder, we propose continuous learning approaches with Reinforcement Learning (RL) algorithms adapted for DSNNs. Banditron and AGREL are chosen as the two candidate RL algorithms since they can be trained with limited computational resources, effectively addressing the non-stationary problem and fitting the energy constraints of implantable devices. To assess the effectiveness of the proposed methods, we conducted both open-loop and closed-loop experiments. The accuracy of open-loop experiments conducted with DSNN Banditron and DSNN AGREL remains stable over extended periods. Meanwhile, the time-to-target in the closed-loop experiment with perturbations, DSNN Banditron performed comparably to that of DSNN AGREL while achieving reductions of 98% in memory access usage and 99% in the requirements for multiply- and-accumulate (MAC) operations during training. Compared to previous continuous learning SNN decoders, DSNN Banditron requires 98% less computes making it a prime candidate for future wireless iBMI systems.","author":[{"family":"Biyan","given":"Zhou"},{"family":"Basu","given":"Arindam"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2511.22108","URL":"https://doi.org/10.48550/arxiv.2511.22108","source":"datacite"},{"id":"doi:10.48550/arxiv.2511.21784","type":"manuscript","title":"Physics-Informed Spiking Neural Networks via Conservative Flux Quantization","abstract":"Real-time, physically-consistent predictions on low-power edge devices is critical for the next generation embodied AI systems, yet it remains a major challenge. Physics-Informed Neural Networks (PINNs) combine data-driven learning with physics-based constraints to ensure the model's predictions are with underlying physical principles.However, PINNs are energy-intensive and struggle to strictly enforce physical conservation laws. Brain-inspired spiking neural networks (SNNs) have emerged as a promising solution for edge computing and real-time processing. However, naively converting PINNs to SNNs degrades physical fidelity and fails to address long-term generalization issues. To this end, this paper introduce a novel Physics-Informed Spiking Neural Network (PISNN) framework. Importantly, to ensure strict physical conservation, we design the Conservative Leaky Integrate-and-Fire (C-LIF) neuron, whose dynamics structurally guarantee local mass preservation. To achieve robust temporal generalization, we introduce a novel Conservative Flux Quantization (CFQ) strategy, which redefines neural spikes as discrete packets of physical flux. Our CFQ learns a time-invariant physical evolution operator, enabling the PISNN to become a general-purpose solver -- conservative-by-construction. Extensive experiments show that our PISNN excels on diverse benchmarks. For both the canonical 1D heat equation and the more challenging 2D Laplace's Equation, it accurately simulates the system dynamics while maintaining perfect mass conservation by design -- a feat that is challenging for conventional PINNs. This work establishes a robust framework for fusing the rigor of scientific computing with the efficiency of neuromorphic engineering, paving the way for complex, long-term, and energy-efficient physics predictions for intelligent systems.","author":[{"family":"Zhang","given":"Chi"},{"family":"Wang","given":"Lin"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2511.21784","URL":"https://doi.org/10.48550/arxiv.2511.21784","source":"datacite"},{"id":"doi:10.82419/84","type":"article-journal","title":"Self-Ensemble as Defense for Event-Based Adversarial Attack against Spiking Neural Networks","abstract":"\"The brain is a world consisting of a number of unexplored continents and great stretches of unknown territory.” Santiago Ram´on y Cajal (1852–1934). In the rapidly evolving landscape of artificial intelligence, Spiking Neural Networks (SNNs) have emerged as a promising approach for achieving efficient realtime processing in novel applications such as autonomous driving and robotics. Known as “the third generation of neural networks” in the literature, SNNs mimic the human brain by utilizing spiking patterns to encode and process information, reflecting the biological neural activities where neurons communicate via electrical impulses. By leveraging the temporal dynamics of neural spikes, SNNs are capable of performing complex computations with remarkable energy efficiency, making them particularly wellsuited for edge computing environments where computational resources are at a premium. These settings demand not only high performance but also robust security measures to protect against potential threats. Similar to artificial neural networks (ANNs), SNNs are not immune to adversarial attacks, which can manipulate input data within certain range that is imperceptible to human, to deceive models into making incorrect predictions. Traditional defense mechanisms often require additional raw training data, adversarial training, or modifications to the network architecture, complicating their implementation. Therefore, there is a critical need for innovative defense strategies that can enhance the robustness of SNNs without imposing significant overhead. This study introduces a novel “selfensemble” approach that leverages multiple latency settings during the inference phase to defend against eventbased adversarial attacks on SNNs. The method begins with training a single SNN model from an event dataset captured by dynamic vision sensors (DVS), which generate asynchronous events in response to changes in brightness. The training set is built from multiple rather than one frame aggregator, thanks to the nice property of SNNs that are compatible with different latencies given the same weights of the network. During inference, the event data is converted into frame data using various temporal intervals, again, resulting in multiple sets of frames with different numbers of frames per unit time. The trained SNN processes each set independently, producing multiple outputs. This selfensemble approach aggregates these diverse outputs, effectively mitigating the impact of any single erroneous prediction caused by the eventbased attack. Since current adversarial perturbations only target at one particular frame representation while leaving others relatively unharmed, the ensemble method ensures overall system robustness. A key advantage of this technique is its simplicity and efficiencyit does not require additional event data or adversarial samples for enhanced training. Instead, it exploits the inherent flexibility of SNNs to handle multiple latencies, enhancing compatibility without compromising performance. The method naturally aligns with the asynchronous processing capabilities of SNNs, ensuring seamless integration. Empirical results demonstrate the effectiveness of our selfensemble approach without performance loss. Existing eventbased adversarial attack techniques, which still target specific frame numbers, fail to disrupt the network’s ability to accurately interpret the event stream. Even when one output is significantly affected, the majority consensus among the ensemble remains accurate, thereby maintaining overall system integrity. This resilience underscores the potential of our method as a practical defense mechanism against sophisticated adversarial threats. By exploiting the unique characteristics of SNNs and their ability to process data at multiple latencies, our work paves the way for more robust neuromorphic computing systems. Future directions include exploring scalability across different datasets and application domai","author":[{"family":"Li","given":"Xinyu"}],"issued":{"date-parts":[[2025]]},"DOI":"10.82419/84","URL":"https://doi.org/10.82419/84","source":"datacite"},{"id":"doi:10.48550/arxiv.2507.08490","type":"manuscript","title":"Neuromorphic Split Computing via Optical Inter-Satellite Links","abstract":"We present a neuromorphic split-computing framework for energy-efficient low-latency inference over optical inter-satellite links. The system partitions a spiking neural network (SNN) between edge and core nodes. To transmit sparse spiking features efficiently, we introduce a lossless channel-block-sparse event representation that exploits inter- and intra-channel sparsity. We employ hierarchical error protection using multi-level forward error correction and cyclic redundancy checks to ensure reliable communication without retransmission. The framework uses end-to-end training with sparsity and clustering regularizers, combined with channel-aware stochastic masking to optimize feature compression and channel robustness jointly. In a proof-of-concept implementation on remote sensing imagery, the framework achieves over $10 \\times$ reduction in both computational energy and transmission load compared to conventional dense split systems, with less than 1% accuracy loss. The proposed approach also outperforms address-event-based split SNNs by $3.7 \\times$ in transmission efficiency and shows superior resilience to optical pointing jitter.","author":[{"family":"Song","given":"Zihang"},{"family":"Popovski","given":"Petar"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48550/arxiv.2507.08490","URL":"https://doi.org/10.48550/arxiv.2507.08490","source":"datacite"},{"id":"doi:10.5281/zenodo.17511153","type":"article-journal","title":"Quantum-Bio-Hybrid Paradigm III: Cross-Domain Implementation and Neuromorphic Realization","abstract":"This paper establishes the hardware realization of the Quantum–Bio–Hybrid (QB-H) paradigm, extending the previously defined retrocausal learning rule (ΔR) to a neuromorphic platform. The proposed system integrates human intention (MBCI) as a positive reinforcement signal and a real-time ethical safeguard (MSI: Moral State Interrupter) to enforce alignment. The ΔR rule is quantized for FPGA/ASIC implementation, and its modulation of spiking neural network (SNN) plasticity demonstrates end-to-end latency ≤10 ms.Simulation results using motor-imagery EEG data validate that human intention can effectively bias synaptic updates, while the MSI module performs on-the-fly ethical correction when deviation exceeds the threshold. These results quantitatively confirm the feasibility of real-time, ethics-aligned AGI hardware under the Quantum–Bio–Hybrid paradigm.","author":[{"family":"Konishi","given":"Hiroko"},{"family":"Ai","given":"Gemini"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.17511153","URL":"https://doi.org/10.5281/zenodo.17511153","source":"datacite"},{"id":"doi:10.5281/zenodo.17547571","type":"article-journal","title":"Structured PREreview of \"A Practical Tutorial on Spiking Neural Networks: Comprehensive Review, Models, Experiments, Software Tools, and Implementation Guidelines\"","abstract":"This Zenodo record is a permanently preserved version of a Structured PREreview. You can view the complete PREreview at https://prereview.org/reviews/17547571. Does the introduction explain the objective of the research presented in the preprint? Yes The introduction explains the objective of the research by initially highlighting the motivation: the rapidly growing computational and energy costs of modern deep neural networks and the resulting sustainability concerns, which position Spiking Neural Networks (SNNs) as a biologically inspired, event-driven alternative promising competitive accuracy at substantially lower energy. The core objective is framed by addressing a key \"Gap,\" identified as the lack of a unified, practice-oriented analysis and limited \"apples-to-apples evidence on accuracy–energy trade-offs against equivalent ANN baselines\". To resolve this, the work combines a comprehensive critical review, a hands-on tutorial, and standardized benchmarking by systematizing SNN components (models, encodings, and learning paradigms), providing a practical tutorial using a representative neuromorphic software stack (e.g., Lava), and establishing a side-by-side evaluation protocol comparing SNNs with architecturally matched ANNs on both shallow (MNIST) and deeper convolutional (CIFAR-10) tasks. The final stated aim is to measure accuracy and power-oriented proxies, ultimately distilling actionable design guidelines that offer a \"coherent pathway from principles to practice\" to design SNNs that \"balance accuracy with energy efficiency in real-world settings Are the methods well-suited for this research? Highly appropriate Yes, the methods employed in the preprint are well-suited for the research objective, as they are specifically designed to address the identified gap concerning the lack of a unified, practice-oriented analysis linking SNN design choices to measurable performance and power consumption. The approach establishes a rigorous side-by-side evaluation protocol by comparing SNN configurations with architecturally matched ANN baselines on both a shallow Fully Connected Network (FCN) for MNIST and a deeper VGG7 architecture for CIFAR-10, thereby providing the crucial \"apples-to-apples evidence\" sought by the study. The methodology systematically explores the core design knobs of SNNs by experimenting with a diverse set of nine neuron models (including LIF, Sigma-Delta, and AdEx) and seven input encoding schemes (such as Direct Coding, Rate Encoding, and Temporal TTFS), all trained using supervised surrogate-gradient methods Are the conclusions supported by the data? Highly supported Yes, the conclusions presented in the preprint are thoroughly supported by the empirical data derived from the standardized benchmarking experiments on MNIST and CIFAR-10. 1. The first key conclusion, asserting a real yet tunable accuracy–energy trade-off, is evidenced by comparing the top-performing SNN configurations against the baseline Artificial Neural Network (ANN); for instance, on MNIST, ΣΔ neurons achieved 98.10% accuracy, closely matching the 98.23% ANN baseline, while remaining energetically below the ANN proxy, confirming the SNN energy advantage. Conversely, highly frugal encodings like R-NoM consistently produced the lowest energy consumption demonstrating the sharp trade-off by incurring larger accuracy drops, while Direct and ΣΔ encodings narrowed the accuracy gap while only demanding a moderate energy premium compared to the most frugal options. 2. The second conclusion regarding practical configuration rules is directly substantiated by data showing that accuracy-critical applications benefit from ΣΔ neurons paired with Direct coding on CIFAR-10 (achieving 83.0% at two time steps against the 83.6% ANN baseline) or Rate/ΣΔ encoding on MNIST, whereas energy-constrained scenarios achieved maximal efficiency using simpler IF/LIF neurons with Burst or R-NoM encoding at minimal time steps. Furthermore, data confirms that thres","author":[{"family":"Lawal","given":"Ronke"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.17547571","URL":"https://doi.org/10.5281/zenodo.17547571","source":"datacite"},{"id":"doi:10.5281/zenodo.17547570","type":"article-journal","title":"Structured PREreview of \"A Practical Tutorial on Spiking Neural Networks: Comprehensive Review, Models, Experiments, Software Tools, and Implementation Guidelines\"","abstract":"This Zenodo record is a permanently preserved version of a Structured PREreview. You can view the complete PREreview at https://prereview.org/reviews/17547571. Does the introduction explain the objective of the research presented in the preprint? Yes The introduction explains the objective of the research by initially highlighting the motivation: the rapidly growing computational and energy costs of modern deep neural networks and the resulting sustainability concerns, which position Spiking Neural Networks (SNNs) as a biologically inspired, event-driven alternative promising competitive accuracy at substantially lower energy. The core objective is framed by addressing a key \"Gap,\" identified as the lack of a unified, practice-oriented analysis and limited \"apples-to-apples evidence on accuracy–energy trade-offs against equivalent ANN baselines\". To resolve this, the work combines a comprehensive critical review, a hands-on tutorial, and standardized benchmarking by systematizing SNN components (models, encodings, and learning paradigms), providing a practical tutorial using a representative neuromorphic software stack (e.g., Lava), and establishing a side-by-side evaluation protocol comparing SNNs with architecturally matched ANNs on both shallow (MNIST) and deeper convolutional (CIFAR-10) tasks. The final stated aim is to measure accuracy and power-oriented proxies, ultimately distilling actionable design guidelines that offer a \"coherent pathway from principles to practice\" to design SNNs that \"balance accuracy with energy efficiency in real-world settings Are the methods well-suited for this research? Highly appropriate Yes, the methods employed in the preprint are well-suited for the research objective, as they are specifically designed to address the identified gap concerning the lack of a unified, practice-oriented analysis linking SNN design choices to measurable performance and power consumption. The approach establishes a rigorous side-by-side evaluation protocol by comparing SNN configurations with architecturally matched ANN baselines on both a shallow Fully Connected Network (FCN) for MNIST and a deeper VGG7 architecture for CIFAR-10, thereby providing the crucial \"apples-to-apples evidence\" sought by the study. The methodology systematically explores the core design knobs of SNNs by experimenting with a diverse set of nine neuron models (including LIF, Sigma-Delta, and AdEx) and seven input encoding schemes (such as Direct Coding, Rate Encoding, and Temporal TTFS), all trained using supervised surrogate-gradient methods Are the conclusions supported by the data? Highly supported Yes, the conclusions presented in the preprint are thoroughly supported by the empirical data derived from the standardized benchmarking experiments on MNIST and CIFAR-10. 1. The first key conclusion, asserting a real yet tunable accuracy–energy trade-off, is evidenced by comparing the top-performing SNN configurations against the baseline Artificial Neural Network (ANN); for instance, on MNIST, ΣΔ neurons achieved 98.10% accuracy, closely matching the 98.23% ANN baseline, while remaining energetically below the ANN proxy, confirming the SNN energy advantage. Conversely, highly frugal encodings like R-NoM consistently produced the lowest energy consumption demonstrating the sharp trade-off by incurring larger accuracy drops, while Direct and ΣΔ encodings narrowed the accuracy gap while only demanding a moderate energy premium compared to the most frugal options. 2. The second conclusion regarding practical configuration rules is directly substantiated by data showing that accuracy-critical applications benefit from ΣΔ neurons paired with Direct coding on CIFAR-10 (achieving 83.0% at two time steps against the 83.6% ANN baseline) or Rate/ΣΔ encoding on MNIST, whereas energy-constrained scenarios achieved maximal efficiency using simpler IF/LIF neurons with Burst or R-NoM encoding at minimal time steps. Furthermore, data confirms that thres","author":[{"family":"Lawal","given":"Ronke"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.17547570","URL":"https://doi.org/10.5281/zenodo.17547570","source":"datacite"},{"id":"oa:W4401100137","type":"article-journal","title":"Highly stable two-dimensional Ruddlesden–Popper perovskite-based resistive switching memory devices","abstract":"s, and good endurance characteristics of 200 cycles. This study investigates temperature-dependent RS behaviors, elucidating the creation and annihilation of a conducting pathway in the presence of an external electric field. Additionally, the RS property of 2D RP perovskite-based memory devices is found to be retained over 45 days at ambient conditions under a relative humidity of 47% ± 4%. Our findings may accelerate the technological deployment of stable 2D perovskite-based RS memory devices for successful logic application.","author":[{"family":"Kundar","given":"Milon"},{"family":"Gayen","given":"Koushik"},{"family":"Ray","given":"Rajeev"},{"family":"Kushavah","given":"Dushyant"},{"family":"Pal","given":"Suman"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1039/d4nr01395f","URL":"https://doi.org/10.1039/d4nr01395f","source":"openalex"},{"id":"doi:10.3929/ethz-b-000316161","type":"article-journal","title":"Corrigendum. Large-Scale Neuromorphic Spiking Array Processors: A Quest to Mimic the Brain (vol 12, 891, 2018)","abstract":"In the original article, there were mistakes in Table 5, Comparison of event-based neural processors, as published. The area per neuron for the transistor channel was incorrectly provided as “4 cm2” and should be “-” (empty). The synaptic plasticity for true North was incorrectly provided as “STDP” and should be “No Plasticity.” The area per neuron for Loihi was incorrectly provided as “0.4 mm2” and should be “0.4 mm2*.” The corrected Table 5, Comparison of event-based neural processors, appears below. (Table Presented) The authors apologize for these errors and state that the do not change the scientific conclusions of the article in any way. The original article has been updated.","author":[{"family":"Thakur","given":"Chetan"},{"family":"Molin","given":"Jamal"},{"family":"Cauwenberghs","given":"Gert"},{"family":"Indiveri","given":"Giacomo"},{"family":"Kumar","given":"Kundan"},{"family":"Qiao","given":"Ning"},{"family":"Schemmel","given":"Johannes"},{"family":"Wang","given":"Runchun"},{"family":"Chicca","given":"Elisabetta"},{"family":"Hasler","given":"Jennifer"},{"family":"Seo","given":"Jae"},{"family":"Yu","given":"Shimeng"},{"family":"Cao","given":"Yu"},{"family":"Van Schaik","given":"André"},{"family":"Etienne-Cummings","given":"Ralph"}],"issued":{"date-parts":[[2019]]},"DOI":"10.3929/ethz-b-000316161","URL":"https://doi.org/10.3929/ethz-b-000316161","source":"datacite"},{"id":"doi:10.5281/zenodo.20491182","type":"article-journal","title":"Adaptive-Frequency Resonate-and-Fire Neurons for Spectral Estimation of Streaming Radar Signals","abstract":"This dataset contains the data to reproduce the results presented in the paper: \"Adaptive-Frequency Resonate-and-Fire Neurons for Spectral Estimation of Streaming Radar Signals\". The work introduces a dynamical system of adaptive-frequency neuron, specifically utilizing Adaptive-frequency Resonate-and-Fire (ARF) mechanics, designed for efficient, streaming FMCW radar signal estimation.","author":[{"family":"Chiavazza","given":"Stefano"},{"family":"Yuan","given":"Sen"},{"family":"Geilen","given":"Marc"},{"family":"Fioranelli","given":"Francesco"},{"family":"Corradi","given":"Federico"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20491182","URL":"https://doi.org/10.5281/zenodo.20491182","source":"datacite"},{"id":"doi:10.5281/zenodo.20491183","type":"article-journal","title":"Adaptive-Frequency Resonate-and-Fire Neurons for Spectral Estimation of Streaming Radar Signals","abstract":"This dataset contains the data to reproduce the results presented in the paper: \"Adaptive-Frequency Resonate-and-Fire Neurons for Spectral Estimation of Streaming Radar Signals\". The work introduces a dynamical system of adaptive-frequency neuron, specifically utilizing Adaptive-frequency Resonate-and-Fire (ARF) mechanics, designed for efficient, streaming FMCW radar signal estimation.","author":[{"family":"Chiavazza","given":"Stefano"},{"family":"Yuan","given":"Sen"},{"family":"Geilen","given":"Marc"},{"family":"Fioranelli","given":"Francesco"},{"family":"Corradi","given":"Federico"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20491183","URL":"https://doi.org/10.5281/zenodo.20491183","source":"datacite"},{"id":"doi:10.48550/arxiv.1601.04862","type":"manuscript","title":"Scalability in Neural Control of Musculoskeletal Robots","abstract":"Anthropomimetic robots are robots that sense, behave, interact and feel like humans. By this definition, anthropomimetic robots require human-like physical hardware and actuation, but also brain-like control and sensing. The most self-evident realization to meet those requirements would be a human-like musculoskeletal robot with a brain-like neural controller. While both musculoskeletal robotic hardware and neural control software have existed for decades, a scalable approach that could be used to build and control an anthropomimetic human-scale robot has not been demonstrated yet. Combining Myorobotics, a framework for musculoskeletal robot development, with SpiNNaker, a neuromorphic computing platform, we present the proof-of-principle of a system that can scale to dozens of neurally-controlled, physically compliant joints. At its core, it implements a closed-loop cerebellar model which provides real-time low-level neural control at minimal power consumption and maximal extensibility: higher-order (e.g., cortical) neural networks and neuromorphic sensors like silicon-retinae or -cochleae can naturally be incorporated.","author":[{"family":"Richter","given":"Christoph"},{"family":"Jentzsch","given":"Sören"},{"family":"Hostettler","given":"Rafael"},{"family":"Garrido","given":"Jesús"},{"family":"Ros","given":"Eduardo"},{"family":"Knoll","given":"Alois"},{"family":"Röhrbein","given":"Florian"},{"family":"Van Der Smagt","given":"Patrick"},{"family":"Conradt","given":"Jörg"}],"issued":{"date-parts":[[2016]]},"DOI":"10.48550/arxiv.1601.04862","URL":"https://doi.org/10.48550/arxiv.1601.04862","source":"datacite"},{"id":"doi:10.17863/cam.58858","type":"article-journal","title":"Non-Polar and Complementary Resistive Switching Characteristics in Graphene Oxide devices with Gold Nanoparticles: Diverse Approach for Device Fabrication.","abstract":"Downscaling limitations and limited write/erase cycles in conventional charge-storage based non-volatile memories stimulate the development of emerging memory devices having enhanced performance. Resistive random-access memory (RRAM) devices are recognized as the next-generation memory devices for employment in artificial intelligence and neuromorphic computing, due to their smallest cell size, high write/erase speed and endurance. Unipolar and bipolar resistive switching characteristics in graphene oxide (GO) have been extensively studied in recent years, whereas the study of non-polar and complementary switching is scarce. Here we fabricated GO-based RRAM devices with gold nanoparticles (Au Nps). Diverse types of switching behavior are observed by changing the processing methods and device geometry. Tri-layer GO-based devices illustrated non-polar resistive switching, which is a combination of unipolar and bipolar switching. Five-layer GO-based devices depicted complementary resistive switching having the lowest current values ~12 µA; and this structure is capable of resolving the sneak path issue. Both devices show good retention and endurance performance. Au Nps in tri-layer devices assisted the conducting path, whereas in five-layer devices, Au Nps layer worked as common electrodes between co-joined cells. These GO-based devices with Au Nps comprising different configuration are vital for practical applications of emerging non-volatile resistive memories.","author":[{"family":"Khurana","given":"Geetika"},{"family":"Kumar","given":"Nitu"},{"family":"Chhowalla","given":"Manish"},{"family":"Scott","given":"James"},{"family":"Katiyar","given":"Ram"}],"issued":{"date-parts":[[2019]]},"DOI":"10.17863/cam.58858","URL":"https://doi.org/10.17863/cam.58858","source":"datacite"},{"id":"doi:10.48448/x1s5-jk76","type":"article-journal","title":"Read Voltage Dependency of Random Telegraph Noise in the Intermediate State of TaOX-based ReRAM","abstract":"This work investigates random telegraph noise (RTN) behavior in the intermediate resistance states of 40nm TaOX-based analog ReRAM. Uniquely in the intermediate resistance states, both the frequency and amplitude of RTN fluctuations tend to increase with the read voltage. Considering the trade-off between increasing them and enhancing margins among adjacent resistance states as read voltage rises, a 0.2V readout minimizes RTN-induced misdetection of resistance states, achieving a significantly low error rate of 0.19%.","author":[{"family":"Akinaga","given":"Hiroyuki"},{"family":"Matsui","given":"Chihiro"},{"family":"Misawa","given":"Naoko"},{"family":"Naitoh","given":"Yasuhisa"},{"family":"Shima","given":"Hisashi"},{"family":"Takeuchi","given":"Ken"},{"family":"Yamauchi","given":"Kenshin"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48448/x1s5-jk76","URL":"https://doi.org/10.48448/x1s5-jk76","source":"datacite"},{"id":"doi:10.48448/d5fm-st28","type":"article-journal","title":"Temperature and Drift-Aware High-Level PCM-based Array Model for Reliable Hardware IMC design","abstract":"In-Memory Computing (IMC) is an important lead to propose efficient AI hardware on the edge. In this context, emerging non-volatile memories, such as Phase-Change Memory (PCM) are fundamental. However, PCM device non-idealities, hardware and application challenges render IMC design complex. To overcome this challenge, we propose a novel high-level PCM-based model for hardware IMC. We consider, for the first time, the temperature effect on PCM conductance stability – a key parameter for PCM-based IMC.","author":[{"family":"Allegra","given":"Mario"},{"family":"Anghel","given":"Lorena"},{"family":"Baldo","given":"Matteo"},{"family":"Esmanhotto","given":"Eduardo"},{"family":"Lecoq","given":"Xavier"},{"family":"Prenat","given":"Guillaume"},{"family":"Viollet","given":"Valentin"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48448/d5fm-st28","URL":"https://doi.org/10.48448/d5fm-st28","source":"datacite"},{"id":"doi:10.48448/s16y-fm14","type":"article-journal","title":"Multi-level RTN with Certain Regularities in Oxide-RRAM: Experiments, Defect Dynamics and 3D Multi-Physics Modeling","abstract":"This paper investigated the multi-level RTN phenomena with certain regularities in the Ti/HfOx/TiN RRAM with the comprehensive measurements and modeling of the multi-phonon trap-assisted tunneling (MPTAT) and First-Principle calculations. The polaron band of Vo is the main reason to produce the RTN of RRAM. Two-level RTN is attributed to the transition between V0 and V1-. The multi-level RTN arises from the collective action of Vo at different positions, affecting current fluctuation, capturing and emission time.","author":[{"family":"Cai","given":"Zifei"},{"family":"Ji","given":"Zhigang"},{"family":"Liu","given":"Pan"},{"family":"Miao","given":"Xiangshui"},{"family":"Mu","given":"Dejiang"},{"family":"Wang","given":"Xingsheng"},{"family":"Xue","given":"Kan"},{"family":"Zhang","given":"Jian"},{"family":"Zhou","given":"Zijian"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48448/s16y-fm14","URL":"https://doi.org/10.48448/s16y-fm14","source":"datacite"},{"id":"doi:10.48448/t3ed-3211","type":"article-journal","title":"Reliability-Ensured and Fast (< 100 ns) Analog Synapse for Training Accelerators: All-Sputtered HfOy/HfOx RRAM","abstract":"We reveal that sputtered HfOx, often neglected in RRAM design, can be an alternative for highly reliable, fast (&lt; 100 ns) analog synapses to accelerate training algorithms. The sub-stoichiometric HfOx enables rapid oxygen vacancy (V0) response. Additional HfOy (x&lt;y) distributes V0 uniformly across the HfOy/HfOx RRAM stack, enhancing reliability in dynamic synaptic operation, where multiple pulses are continuously addressed. This results in robust D2D/C2C uniformity and ensures multilevel retention, validated by real-time transient current analysis.","author":[{"family":"Jeon","given":"Seonuk"},{"family":"Kim","given":"Yunsur"},{"family":"Lim","given":"Seokjae"},{"family":"Woo","given":"Jiyong"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48448/t3ed-3211","URL":"https://doi.org/10.48448/t3ed-3211","source":"datacite"},{"id":"doi:10.48448/1vzy-zr77","type":"article-journal","title":"Analog computing with high precision and reliability","abstract":"Analog devices such as memristors have been restricted to specific and low-precision application areas due to their historical limitations in precision and reliability. To solve that, we study the source of the reading noise and successfully mitigate it, achieving 2048 stable conductance levels. At the system level, we co-design a circuit architecture and programming algorithm to mitigate the device programming variations and reliability issues and demonstrate high-precision computing applications with enhanced power efficiency.","author":[{"family":"Song","given":"Wenhao"},{"family":"Yang","given":"JJ"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48448/1vzy-zr77","URL":"https://doi.org/10.48448/1vzy-zr77","source":"datacite"},{"id":"doi:10.48448/crk0-6m31","type":"article-journal","title":"Impact of Non-ideal Reliability Characteristics of SiOx p-Bit for Complex Optimization Problem Solver","abstract":"SiOx threshold switching (TS) devices generate periodic output voltage spikes for a given input voltage (Vin) pulse. Introducing Ti layers to induce variability in the SiOx film, the number of spikes can be adjusted as a sigmoid of Vin amplitude. Therefore, Ti/SiOx/Ti TS can serve as probabilistic bit (p-bit), crucial for solving complex optimization problems (COPs). We investigate the impact of the non-ideal SiOx p-bit response on COP by considering classified scenarios based on measurements.","author":[{"family":"Choi","given":"Hyeonsik"},{"family":"Kim","given":"Jihyun"},{"family":"Woo","given":"Jiyong"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48448/crk0-6m31","URL":"https://doi.org/10.48448/crk0-6m31","source":"datacite"},{"id":"doi:10.48448/b324-sr08","type":"article-journal","title":"Effects of Temperature and Device-to-Device Variability in pFET-Based Bias Temperature Instability Reservoir Computing","abstract":"We conduct gait authentication tests across different sizes, temperatures, and device sets to evaluate the reliability of our pFETs BTI PRC system. The findings suggest that large devices are more resilient to high temperatures and device-to-device variability. Small devices are more sensitive to environmental changes.Large devices are preferred when accuracy is important, while small devices are favored when security is the primary concern. We use a stochastic model to analyze the physics of our system.","author":[{"family":"Bury","given":"Erik"},{"family":"Degraeve","given":"Robin"},{"family":"Guo","given":"Yuanyang"},{"family":"Kaczer","given":"Vice"},{"family":"Saraza-Canflanca","given":"Pablo"},{"family":"Verbauwhede","given":"Ingrid"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48448/b324-sr08","URL":"https://doi.org/10.48448/b324-sr08","source":"datacite"},{"id":"doi:10.48448/y493-hc90","type":"article-journal","title":"Investigation on Temperature-Dependent Resistance States of 40nm MLC-RRAM Macro","abstract":"Resistive Random-Access Memory (RRAM) is notable for its multi-level cell (MLC) capabilities. However, its conductance is highly susceptible to temperature fluctuations, challenging practical deployment. Here, we characterize the temperature-dependent resistance states of 40nm RRAM macro, indicating that conductance states have different relations with the temperature. A physics model is developed to describe the resistance varying with temperature. Further, we propose the calibration method to reduce the readout errors, which is crucial for storage and CIM.","author":[{"family":"Cai","given":"Yimao"},{"family":"Feng","given":"Yulin"},{"family":"Guan","given":"Haokai"},{"family":"Huang","given":"Peng"},{"family":"Kang","given":"Jinfeng"},{"family":"Liu","given":"Lifeng"},{"family":"Shan","given":"Linbo"},{"family":"Sun","given":"Lei"},{"family":"Tao","given":"Kefan"},{"family":"Wang","given":"Zongwei"},{"family":"Zhong","given":"Shichao"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48448/y493-hc90","URL":"https://doi.org/10.48448/y493-hc90","source":"datacite"},{"id":"doi:10.48448/wck7-4x53","type":"article-journal","title":"Compact MEOL OxRAM with 14 conductance levels for Dense Embedded Inference Computing","abstract":"A compact 1T1R memory cell with 14 conductance levels is fabricated in 28nm technology. The OxRAM is integrated in the 40nm×40nm drain contact of GO1 FDSOI transistors, enabling memory arrays with minimal 0.0357µm2 bitcell area. 106 cycles endurance, 2h data retention at 150°C and 109 read disturb cycles are demonstrated, with 30µA SET currents and 1.2V RESET voltages. On-chip Neural Network inference capabilities are demonstrated for a QSNN solving Automatic Lip-Reading task with state-of-art accuracy.","author":[{"family":"Andrieu","given":"François"},{"family":"Barraud","given":"Sylvain"},{"family":"Boulard","given":"François"},{"family":"Castan","given":"Clément"},{"family":"Comboroure","given":"Corinne"},{"family":"Dampfhoffer","given":"Manon"},{"family":"Dubreuil","given":"Theophile"},{"family":"Gharbi","given":"Ahmed"},{"family":"Lambert","given":"Amélie"},{"family":"Minguet Lopez","given":"Joel"},{"family":"Pedini","given":"Jean"},{"family":"Souhaité","given":"Aurelie"}],"issued":{"date-parts":[[2025]]},"DOI":"10.48448/wck7-4x53","URL":"https://doi.org/10.48448/wck7-4x53","source":"datacite"},{"id":"doi:10.5281/zenodo.18179892","type":"article-journal","title":"A deep spiking machine-hearing system for the case of invasive fish species","abstract":"(Uploaded by Plazi for the IPBES Invasive Alien Species Assessment) Prolonged and sustained warming of the sea, acidification of surface water and rising of sea levels, creates significant habitat losses, resulting in the proliferation and spread of invasive species which immigrate to foreign regions seeking colder climate conditions. This is happening either because their natural habitat does not satisfy the temperature range in which they can survive, or because they are just following their food. This has negative consequences not only for the environment and biodiversity but for the socioeconomic status of the areas and for the human health. This research aims in the development of an advanced Machine Hearing system towards the automated recognition of invasive fish species based on their sounds. The proposed system uses the Spiking Convolutional Neural Network algorithm which cooperates with Geo Location Based Services. It is capable to correctly classify the typical local fish inhabitants from the invasive ones.","author":[{"family":"Demertzis","given":"Konstantinos"},{"family":"Iliadis","given":"Lazaros"},{"family":"Anezakis","given":"Vardis"}],"issued":{"date-parts":[[2017]]},"DOI":"10.5281/zenodo.18179892","URL":"https://doi.org/10.5281/zenodo.18179892","source":"datacite"},{"id":"doi:10.5281/zenodo.18179893","type":"article-journal","title":"A deep spiking machine-hearing system for the case of invasive fish species","abstract":"(Uploaded by Plazi for the IPBES Invasive Alien Species Assessment) Prolonged and sustained warming of the sea, acidification of surface water and rising of sea levels, creates significant habitat losses, resulting in the proliferation and spread of invasive species which immigrate to foreign regions seeking colder climate conditions. This is happening either because their natural habitat does not satisfy the temperature range in which they can survive, or because they are just following their food. This has negative consequences not only for the environment and biodiversity but for the socioeconomic status of the areas and for the human health. This research aims in the development of an advanced Machine Hearing system towards the automated recognition of invasive fish species based on their sounds. The proposed system uses the Spiking Convolutional Neural Network algorithm which cooperates with Geo Location Based Services. It is capable to correctly classify the typical local fish inhabitants from the invasive ones.","author":[{"family":"Demertzis","given":"Konstantinos"},{"family":"Iliadis","given":"Lazaros"},{"family":"Anezakis","given":"Vardis"}],"issued":{"date-parts":[[2017]]},"DOI":"10.5281/zenodo.18179893","URL":"https://doi.org/10.5281/zenodo.18179893","source":"datacite"},{"id":"doi:10.21203/rs.3.rs-2789677/v1","type":"article-journal","title":"On-chip phonon-magnon reservoir for neuromorphic computing","abstract":"Abstract Reservoir computing is a concept in which signals to process are mapped onto a high-dimensional phase space of a fixed dynamical system called “reservoir” for subsequent recognition by an artificial neural network. Implementations of reservoirs are possible using different hardware and, accordingly, exploit different carriers and mechanisms of signal transformation. Despite the growing number of neuromorphic prototypes of reservoirs, demands for miniaturization, efficiency, and robustness all require implementation of the reservoir on a chip, and remain among the key challenges. Here we propose a nanodevice, in which a sandwich of a semiconductor phonon waveguide and a patterned ferromagnetic layer enables efficient reservoir computing. The optical input signal is coded by a pulsed write-laser and converted into a propagating multimode phonon wave packet, which interacts with a bunch of magnon modes. The output signal read by a second laser represents a phase-sensitive superposition of all the phonon and magnon modes, which possesses ultimate sensitivity to the relative positions of the write- and read-laser spots. The reservoir efficiently separates the visual shapes drawn by the write-laser beam on the nanodevice surface in an area with size comparable to a single pixel of a modern digital camera. Thus, our finding suggests the phonon-magnon interaction as a promising hardware basis for realization of rapid on-chip reservoir computing for future neuromorphic architectures.","author":[{"family":"Scherbakov","given":"Alexey"},{"family":"Yaremkevich","given":"Dmytro"},{"family":"Clerk","given":"Luke"},{"family":"Kukhtaruk","given":"Serhii"},{"family":"Campion","given":"Richard"},{"family":"Rushforth","given":"Andrew"},{"family":"Savelev","given":"Sergey"},{"family":"Balanov","given":"Alexander"},{"family":"Bayer","given":"Manfred"},{"family":"Nadzeyka","given":"Achim"}],"issued":{"date-parts":[[2023]]},"DOI":"10.21203/rs.3.rs-2789677/v1","URL":"https://doi.org/10.21203/rs.3.rs-2789677/v1","source":"preprints"},{"id":"doi:10.2174/9789815305364124010004","type":"article-journal","title":"Neuromorphic Computing: Forging a Link between Artificial Intelligence and Neurological Models","abstract":"By emulating the design and operation of the human brain, neuromorphic computing promises to close the gap between artificial intelligence and brain-inspired technologies. Researchers create specialized hardware and software to mimic the brain's processing speed, capacity for learning, and energy economy. This chapter examines the drivers, difficulties, and prospective uses of neuromorphic computing, with a focus on robotics, sensory processing, and pattern recognition. The entire potential of brain-inspired systems will be unlocked by ongoing research, revolutionizing the field of AI and paving the way for the creation of cutting-edge, intelligent machines that follow the principles of the brain.","author":[{"family":"Bajaj","given":"Madhvan"},{"family":"Rawat","given":"Priyanshu"},{"family":"Sharma","given":"Vikrant"},{"family":"Vats","given":"Satvik"}],"issued":{"date-parts":[[2024]]},"DOI":"10.2174/9789815305364124010004","URL":"https://doi.org/10.2174/9789815305364124010004","source":"crossref"},{"id":"doi:10.62441/nano-ntp.vi.2974","type":"article-journal","title":"Neuromorphic Computing: Advancing Energy-Efficient AI Systems through Brain-Inspired Architectures","abstract":"Neuromorphic computing represents a transformative approach to artificial intelligence, leveraging brain-inspired architectures to enhance energy efficiency and computational performance. This paper explores the principles and innovations underlying neuromorphic systems, which mimic the neural structures and processes of biological brains. We discuss the advantages of these architectures in processing information more efficiently than traditional von Neumann models, particularly in tasks involving pattern recognition, sensory processing, and adaptive learning. By integrating concepts from neuroscience with cutting-edge hardware developments, such as spiking neural networks and memristors, neuromorphic computing addresses the critical challenges of power consumption and scalability in AI applications. This review highlights recent advancements, ongoing research efforts, and potential future directions, illustrating how neuromorphic computing can redefine the landscape of AI by enabling systems that are not only faster and more efficient but also capable of real-time learning and decision-making in dynamic environments.","author":[{"family":"Malviya","given":"Rajesh"},{"family":"Danda","given":"Ramanakar"},{"family":"Maguluri","given":"Kiran"},{"family":"Kumar","given":"Battapothu"}],"issued":{"date-parts":[[2024]]},"DOI":"10.62441/nano-ntp.vi.2974","URL":"https://doi.org/10.62441/nano-ntp.vi.2974","source":"crossref"},{"id":"doi:10.1088/2634-4386/ad3be7","type":"article-journal","title":"Scaling neural simulations in STACS","abstract":"Abstract As modern neuroscience tools acquire more details about the brain, the need to move towards biological-scale neural simulations continues to grow. However, effective simulations at scale remain a challenge. Beyond just the tooling required to enable parallel execution, there is also the unique structure of the synaptic interconnectivity, which is globally sparse but has relatively high connection density and non-local interactions per neuron. There are also various practicalities to consider in high performance computing applications, such as the need for serializing neural networks to support potentially long-running simulations that require checkpoint-restart. Although acceleration on neuromorphic hardware is also a possibility, development in this space can be difficult as hardware support tends to vary between platforms and software support for larger scale models also tends to be limited. In this paper, we focus our attention on Simulation Tool for Asynchronous Cortical Streams (STACS), a spiking neural network simulator that leverages the Charm++ parallel programming framework, with the goal of supporting biological-scale simulations as well as interoperability between platforms. Central to these goals is the implementation of scalable data structures suitable for efficiently distributing a network across parallel partitions. Here, we discuss a straightforward extension of a parallel data format with a history of use in graph partitioners, which also serves as a portable intermediate representation for different neuromorphic backends. We perform scaling studies on the Summit supercomputer, examining the capabilities of STACS in terms of network build and storage, partitioning, and execution. We highlight how a suitably partitioned, spatially dependent synaptic structure introduces a communication workload well-suited to the multicast communication supported by Charm++. We evaluate the strong and weak scaling behavior for networks on the order of millions of neurons and billions of synapses, and show that STACS achieves competitive levels of parallel efficiency.","author":[{"family":"Wang","given":"Felix"},{"family":"Kulkarni","given":"Shruti"},{"family":"Theilman","given":"Bradley"},{"family":"Rothganger","given":"Fredrick"},{"family":"Schuman","given":"Catherine"},{"family":"Lim","given":"Seung"},{"family":"Aimone","given":"James"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1088/2634-4386/ad3be7","URL":"https://doi.org/10.1088/2634-4386/ad3be7","source":"crossref"},{"id":"doi:10.1002/adma.202312825","type":"article-journal","title":"Photonics for Neuromorphic Computing: Fundamentals, Devices, and Opportunities","abstract":"Abstract In the dynamic landscape of Artificial Intelligence (AI), two notable phenomena are becoming predominant: the exponential growth of large AI model sizes and the explosion of massive amount of data. Meanwhile, scientific research such as quantum computing and protein synthesis increasingly demand higher computing capacities. As the Moore's Law approaches its terminus, there is an urgent need for alternative computing paradigms that satisfy this growing computing demand and break through the barrier of the von Neumann model. Neuromorphic computing, inspired by the mechanism and functionality of human brains, uses physical artificial neurons to do computations and is drawing widespread attention. This review studies the expansion of optoelectronic devices on photonic integration platforms that has led to significant growth in photonic computing, where photonic integrated circuits (PICs) have enabled ultrafast artificial neural networks (ANN) with sub‐nanosecond latencies, low heat dissipation, and high parallelism. In particular, various technologies and devices employed in neuromorphic photonic AI accelerators, spanning from traditional optics to PCSEL lasers are examined. Lastly, it is recognized that existing neuromorphic technologies encounter obstacles in meeting the peta‐level computing speed and energy efficiency threshold, and potential approaches in new devices, fabrication, materials, and integration to drive innovation are also explored. As the current challenges and barriers in cost, scalability, footprint, and computing capacity are resolved one‐by‐one, photonic neuromorphic systems are bound to co‐exist with, if not replace, conventional electronic computers and transform the landscape of AI and scientific computing in the foreseeable future.","author":[{"family":"Li","given":"Renjie"},{"family":"Gong","given":"Yuanhao"},{"family":"Huang","given":"Hai"},{"family":"Zhou","given":"Yuze"},{"family":"Mao","given":"Sixuan"},{"family":"Wei","given":"Zhijian"},{"family":"Zhang","given":"Zhaoyu"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1002/adma.202312825","URL":"https://doi.org/10.1002/adma.202312825","source":"europepmc"},{"id":"doi:10.21203/rs.3.rs-4349574/v1","type":"article-journal","title":"Adiabatic Leaky Integrate-and-Fire Neurons with Tunable Refractory Period in 180nm CMOS Technology for Ultra-Low Energy Brain-Inspired Neuromorphic Computing","abstract":"Abstract In recent years, the In-Memory-Computing in charge domain has gained significant interest as a promising solution to further enhance the energy efficiency of neuromorphic hardware. In this work, we explore the synergy between the brain-inspired computation and the adiabatic paradigm by presenting an adiabatic Leaky Integrate-and-Fire neuron in 180 nm CMOS technology, that is able to emulate the most important primitives for a valuable neuromorphic computation, such as the accumulation of the incoming input spikes, an exponential leakage of the membrane potential and a tunable refractory period. Differently from previous contributions in the literature, our design can exploit both the charging and recovery phases of the adiabatic operation to ensure a seamless and continuous computation, all the while exchanging energy with the power supply with an efficiency higher than 90% over a wide range of resonance frequencies, and even surpassing 99% for the lowest frequencies. Our simulations unveil a minimum energy per synaptic operation of 360 fJ at a 500 kHz resonance frequency, which yields a 12x energy saving with respect to a non-adiabatic operation.","author":[{"family":"Massarotto","given":"Marco"},{"family":"Saggini","given":"Stefano"},{"family":"Loghi","given":"Mirko"},{"family":"Esseni","given":"David"}],"issued":{"date-parts":[[2024]]},"DOI":"10.21203/rs.3.rs-4349574/v1","URL":"https://doi.org/10.21203/rs.3.rs-4349574/v1","source":"preprints"},{"id":"doi:10.1088/2634-4386/ad2ec3","type":"article-journal","title":"Gradient-descent hardware-aware training and deployment for mixed-signal neuromorphic processors","abstract":"Abstract Mixed-signal neuromorphic processors provide extremely low-power operation for edge inference workloads, taking advantage of sparse asynchronous computation within spiking neural networks (SNNs). However, deploying robust applications to these devices is complicated by limited controllability over analog hardware parameters, as well as unintended parameter and dynamical variations of analog circuits due to fabrication non-idealities. Here we demonstrate a novel methodology for offline training and deployment of SNNs to the mixed-signal neuromorphic processor DYNAP-SE2. Our methodology applies gradient-based training to a differentiable simulation of the mixed-signal device, coupled with an unsupervised weight quantization method to optimize the network’s parameters. Parameter noise injection during training provides robustness to the effects of quantization and device mismatch, making the method a promising candidate for real-world applications under hardware constraints and non-idealities. This work extends Rockpool, an open-source deep-learning library for SNNs, with support for accurate simulation of mixed-signal SNN dynamics. Our approach simplifies the development and deployment process for the neuromorphic community, making mixed-signal neuromorphic processors more accessible to researchers and developers.","author":[{"family":"Cakal","given":"Ugurcan"},{"family":"Maryada"},{"family":"Wu","given":"Chenxi"},{"family":"Ulusoy","given":"Ilkay"},{"family":"Muir","given":"Dylan"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1088/2634-4386/ad2ec3","URL":"https://doi.org/10.1088/2634-4386/ad2ec3","source":"crossref"},{"id":"doi:10.1088/2634-4386/acc2e1","type":"article-journal","title":"Perspective on investigation of neurodegenerative diseases with neurorobotics approaches","abstract":"Abstract Neurorobotics has emerged from the alliance between neuroscience and robotics. It pursues the investigation of reproducing living organism-like behaviors in robots by means of the embodiment of computational models of the central nervous system. This perspective article discusses the current trend of implementing tools for the pressing challenge of early-diagnosis of neurodegenerative diseases and how neurorobotics approaches can help. Recently, advances in this field have allowed the testing of some neuroscientific hypotheses related to brain diseases, but the lack of biological plausibility of developed brain models and musculoskeletal systems has limited the understanding of the underlying brain mechanisms that lead to deficits in motor and cognitive tasks. Key aspects and methods to enhance the reproducibility of natural behaviors observed in healthy and impaired brains are proposed in this perspective. In the long term, the goal is to move beyond finding therapies and look into how researchers can use neurorobotics to reduce testing on humans as well as find root causes for disease.","author":[{"family":"Tolu","given":"Silvia"},{"family":"Strohmer","given":"Beck"},{"family":"Zahra","given":"Omar"}],"issued":{"date-parts":[[2023]]},"DOI":"10.1088/2634-4386/acc2e1","URL":"https://doi.org/10.1088/2634-4386/acc2e1","source":"crossref"},{"id":"doi:10.1088/2634-4386/ad05da","type":"article-journal","title":"Spike-based local synaptic plasticity: a survey of computational models and neuromorphic circuits","abstract":"Abstract Understanding how biological neural networks carry out learning using spike-based local plasticity mechanisms can lead to the development of real-time, energy-efficient, and adaptive neuromorphic processing systems. A large number of spike-based learning models have recently been proposed following different approaches. However, it is difficult to assess if these models can be easily implemented in neuromorphic hardware, and to compare their features and ease of implementation. To this end, in this survey, we provide an overview of representative brain-inspired synaptic plasticity models and mixed-signal complementary metal–oxide–semiconductor neuromorphic circuits within a unified framework. We review historical, experimental, and theoretical approaches to modeling synaptic plasticity, and we identify computational primitives that can support low-latency and low-power hardware implementations of spike-based learning rules. We provide a common definition of a locality principle based on pre- and postsynaptic neural signals, which we propose as an important requirement for physical implementations of synaptic plasticity circuits. Based on this principle, we compare the properties of these models within the same framework, and describe a set of mixed-signal electronic circuits that can be used to implement their computing principles, and to build efficient on-chip and online learning in neuromorphic processing systems.","author":[{"family":"Khacef","given":"Lyes"},{"family":"Klein","given":"Philipp"},{"family":"Cartiglia","given":"Matteo"},{"family":"Rubino","given":"Arianna"},{"family":"Indiveri","given":"Giacomo"},{"family":"Chicca","given":"Elisabetta"}],"issued":{"date-parts":[[2023]]},"DOI":"10.1088/2634-4386/ad05da","URL":"https://doi.org/10.1088/2634-4386/ad05da","source":"crossref"},{"id":"doi:10.1088/2634-4386/acb37f","type":"article-journal","title":"Pre-synaptic DC bias controls the plasticity and dynamics of three-terminal neuromorphic electrolyte-gated organic transistors","abstract":"Abstract The role of pre-synaptic DC bias is investigated in three-terminal organic neuromorphic architectures based on electrolyte-gated organic transistors—EGOTs. By means of pre-synaptic offset it is possible to finely control the number of discrete conductance states in short-term plasticity experiments, to obtain, at will, both depressive and facilitating response in the same neuromorphic device and to set the ratio between two subsequent pulses in paired-pulse experiments. The charge dynamics leading to these important features are discussed in relationship with macroscopic device figures of merit such as conductivity and transconductance, establishing a novel key enabling parameter in devising the operation of neuromorphic organic electronics.","author":[{"family":"Rondelli","given":"Federico"},{"family":"Salvo","given":"Anna"},{"family":"Sebastianella","given":"Gioacchino"},{"family":"Murgia","given":"Mauro"},{"family":"Fadiga","given":"Luciano"},{"family":"Biscarini","given":"Fabio"},{"family":"Lauro","given":"Michele"}],"issued":{"date-parts":[[2023]]},"DOI":"10.1088/2634-4386/acb37f","URL":"https://doi.org/10.1088/2634-4386/acb37f","source":"crossref"},{"id":"doi:10.4018/979-8-3693-6303-4.ch011","type":"article-journal","title":"Classification of Moderate and Advanced Dementia Patients Using Gradient Boosting Machine Technique","abstract":"In the twenty-first century, caring for persons with Dementia's has become extremely difficult due to the prevalence of dementia cases. Using data from the OASIS (Open Access Series of Imaging Studies) program provided by the University of Washington Dementia's Disease Research Center, the study presents a new predictive model for Dementia's. Dementia, a chronic condition and it's become a serious health concern in adults. Various methods of data imputation, preprocessing, and transformation were used to prepare the data for model training. Machine learning algorithms, including AdaBoost (AB), Decision Tree (DT), Exclusion Tree (ET), Gradient Boost (GB), K-Nearest Neighbor (KNN), Logistic Regression (LR), Naive Bayes (NB;, Random Forest (RF), and Support Vector Machine (SVM), were used in this field. These algorithms were evaluated on both the complete feature set and a subset of features selected via the Least Absolute Shrinkage and Selection Operator (LASSO) method. Comparative analysis based on accuracy, precision, and other metrics showed that the proposed method achieved the highest accuracy of 96.77% using Support Vector Machine (SVM) with all feature sets, further refined and applied, has great potential for the diagnosis of early Dementia's disease (AD) disease.","author":[{"family":"Gowroju","given":"Swathi"},{"family":"Choudhary","given":"Shilpa"},{"family":"Jain","given":"Arpit"},{"family":"Srilakshmi","given":"R"}],"issued":{"date-parts":[[2024]]},"DOI":"10.4018/979-8-3693-6303-4.ch011","URL":"https://doi.org/10.4018/979-8-3693-6303-4.ch011","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-4102090/v1","type":"article-journal","title":"Self-organizing neuromorphic nanowire networks are stochastic dynamical systems","abstract":"Abstract Neuromorphic computing aims to develop software and hardware platforms emulating the information processing effectiveness of our brain. In this context, self-organizing neuromorphic nanonetworks have been demonstrated as suitable physical substrates for in materia implementation of unconventional computing paradigms, like reservoir computing. However, understanding the relationship between emergent dynamics and information processing capabilities still represents a challenge. Here, we demonstrate that nanowire-based neuromorphic networks are stochastic dynamical systems where the signals flow relies on the intertwined action of deterministic and random factors. We show through an experimental and modeling approach that these systems combine stimuli-dependent deterministic trajectories and random effects caused by noise and jumps that can be holistically described by an Ornstein-Uhlenbeck process, providing a unifying framework surpassing current modeling approaches of self-organizing neuromorphic nanonetworks (not only nanowire-based) that are limited to either deterministic or stochastic effects. Since information processing capabilities can be dynamically tuned by controlling the network’s attractor memory state, these results open new perspectives for the rational development of physical computing paradigms exploiting deterministic and stochastic dynamics in a single hardware platform similarly to our brain.","author":[{"family":"Milano","given":"Gianluca"},{"family":"Michieletti","given":"Fabio"},{"family":"Ricciardi","given":"Carlo"},{"family":"Miranda","given":"Enrique"}],"issued":{"date-parts":[[2024]]},"DOI":"10.21203/rs.3.rs-4102090/v1","URL":"https://doi.org/10.21203/rs.3.rs-4102090/v1","source":"preprints"},{"id":"doi:10.20944/preprints202407.0130.v1","type":"manuscript","title":"A Survey on Neuromorphic Architectures for Running Artificial Intelligence Algorithms","abstract":"Neuromorphic computing, a brain inspired non-Von Neumann computing system, addresses the challenges posed by the Moore’s law memory wall phenomenon. It has the capability to increasingly enhance performance while maintaining power efficiency. Neuromorphic chip architecture requirements vary depending on the application and optimizing it for large-scale applications remains to be a challenge. Neuromorphic chips are programmed using spiking neural networks which provide them with important properties such as parallelism, asynchronism, and on-device learning. Widely used spiking neuron models include the Hodgkin-Huxley Model, Izhikevich model, integrate-and-fire model and spike response model. Hardware implementation platforms of the chip follow three approaches: analog, digital, or a combination of both. Each platform can be implemented using various memory topologies which interconnects with the learning mechanism. Current neuromorphic computing systems typically use the unsupervised learning spike timing-dependent plasticity algorithms. However, algorithms such as voltage-dependent synaptic plasticity have the potential to enhance performance. This review summarizes the potential neuromorphic chip architecture specifications and highlights which applications they are suitable for.","author":[{"family":"Wahid","given":"Seham"},{"family":"Asad","given":"Arghavan"},{"family":"Mohammadi","given":"Farah"}],"issued":{"date-parts":[[2024]]},"DOI":"10.20944/preprints202407.0130.v1","URL":"https://doi.org/10.20944/preprints202407.0130.v1","source":"preprints"},{"id":"doi:10.3390/biomimetics9090547","type":"article-journal","title":"Low-Cost, High-Efficiency Aluminum Zinc Oxide Synaptic Transistors: Blue LED Stimulation for Enhanced Neuromorphic Computing Applications","abstract":"Neuromorphic devices are electronic devices that mimic the information processing methods of neurons and synapses, enabling them to perform multiple tasks simultaneously with low power consumption and exhibit learning ability. However, their large-scale production and efficient operation remain a challenge. Herein, we fabricated an aluminum-doped zinc oxide (AZO) synaptic transistor via solution-based spin-coating. The transistor is characterized by low production costs and high performance. It demonstrates high responsiveness under UV laser illumination. In addition, it exhibits effective synaptic behaviors under blue LED illumination, indicating high-efficiency operation. The paired-pulse facilitation (PPF) index measured from optical stimulus modulation was 179.6%, indicating strong synaptic connectivity and effective neural communication and processing. Furthermore, by modulating the blue LED light pulse frequency, an excitatory postsynaptic current gain of 4.3 was achieved, demonstrating efficient neuromorphic functionality. This study shows that AZO synaptic transistors are promising candidates for artificial synaptic devices.","author":[{"family":"Lee","given":"Namgyu"},{"family":"Pujar","given":"Pavan"},{"family":"Hong","given":"Seongin"}],"issued":{"date-parts":[[2024]]},"DOI":"10.3390/biomimetics9090547","URL":"https://doi.org/10.3390/biomimetics9090547","source":"crossref"},{"id":"doi:10.1002/adfm.202307729","type":"article-journal","title":"Brain‐Inspired Organic Electronics: Merging Neuromorphic Computing and Bioelectronics Using Conductive Polymers","abstract":"Abstract Neuromorphic computing offers the opportunity to curtail the huge energy demands of modern artificial intelligence (AI) applications by implementing computations into new, brain‐inspired computing architectures. However, the lack of fabrication processes able to integrate several computing units into monolithic systems and the need for new, hardware‐tailored training algorithms still limit the scope of application and performance of neuromorphic hardware. Recent advancements in the field of organic transistors present new opportunities for neuromorphic systems and smart sensing applications, thanks to their unique properties such as neuromorphic behavior, low‐voltage operation, and mixed ionic‐electronic conductivity. Organic neuromorphic transistors push the boundaries of energy efficient brain‐inspired hardware AI, facilitating decentralized on‐chip learning and serving as a foundation for the advancement of closed‐loop intelligent systems in the next generation. The biocompatibility and dual ionic‐electronic conductivity of organic materials introduce new prospects for biointegration and bioelectronics. Their ability to sense and regulate biosystems, as well as their neuro‐inspired functions can be combined with neuromorphic computing to create the next‐generation of bioelectronics. These systems will be able to seamlessly interact with biological systems and locally compute biosignals in a relevant matter.","author":[{"family":"Krauhausen","given":"Imke"},{"family":"Coen","given":"Charles‐théophile"},{"family":"Spolaor","given":"Simone"},{"family":"Gkoupidenis","given":"Paschalis"},{"family":"Burgt","given":"Yoeri"}],"issued":{"date-parts":[[2023]]},"DOI":"10.1002/adfm.202307729","URL":"https://doi.org/10.1002/adfm.202307729","source":"crossref"},{"id":"doi:10.1101/2024.07.19.604308","type":"article-journal","title":"A Burst-Dependent Algorithm for Neuromorphic On-Chip Learning of Spiking Neural Networks","abstract":"Abstract The field of neuromorphic engineering addresses the high energy demands of neural networks through brain-inspired hardware for efficient neural network computing. For on-chip learning with spiking neural networks, neuromorphic hardware requires a local learning algorithm able to solve complex tasks. Approaches based on burst-dependent plasticity have been proposed to address this requirement, but their ability to learn complex tasks has remained unproven. Specifically, previous burst-dependent learning was demonstrated on a spiking version of the XOR problem using a network of thousands of neurons. Here, we extend burst-dependent learning, termed ‘Burstprop’, to address more complex tasks with hundreds of neurons. We evaluate Burstprop on a rate-encoded spiking version of the MNIST dataset, achieving low test classification errors, comparable to those obtained using backpropagation through time on the same architecture. Going further, we develop another burst-dependent algorithm based on the communication of two types of error-encoding events for the communication of positive and negative errors. We find that this new algorithm performs better on the image classification benchmark. We also tested our algorithms under various types of feedback connectivity, establishing that the capabilities of fixed random feedback connectivity is preserved in spiking neural networks. Lastly, we tested the robustness of the algorithm to weight discretization. Together, these results suggest that spiking Burstprop can scale to more complex learning tasks and can thus be considered for self-supervised algorithms while maintaining efficiency, potentially providing a viable method for learning with neuromorphic hardware.","author":[{"family":"Stuck","given":"Michael"},{"family":"Wang","given":"Xingyun"},{"family":"Naud","given":"Richard"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1101/2024.07.19.604308","URL":"https://doi.org/10.1101/2024.07.19.604308","source":"preprints"},{"id":"doi:10.22541/au.172114593.35310985/v1","type":"article-journal","title":"Piezoelectric neuron for neuromorphic computing","abstract":"Neuromorphic computing has attracted great attention for its massive parallelism and high energy efficiency. As the fundamental components of neuromorphic computing systems, artificial neurons play a key role in information processing. However, the development of artificial neurons that can simultaneously incorporate low hardware overhead, high reliability, high speed, and low energy consumption remains a challenge. To address this challenge, we propose and demonstrate a piezoelectric neuron with a simple circuit structure, consisting of a piezoelectric cantilever, a parallel capacitor, and a series resistor. It operates through the synergy between the converse piezoelectric effect and the capacitive charging/discharging. Thanks to this efficient and robust mechanism, the piezoelectric neuron not only implements critical leaky integrate-and-fire functions (including leaky integration, threshold-driven spiking, all-or-nothing response, refractory period, strength-modulated firing frequency, and spatiotemporal integration), but also demonstrates small cycle-to-cycle and device-to-device variations (~1.9% and ~10.0%, respectively), high endurance (10 10 ), high speed (integration/firing: ~9.6/~0.4 μs), and low energy consumption (~13.4 nJ/spike). Furthermore, spiking neural networks based on piezoelectric neurons are constructed, showing capabilities to implement both supervised and unsupervised learning. This study therefore opens up a new way to develop high-performance artificial neurons by using piezoelectrics, which may facilitate the realization of advanced neuromorphic computing systems.","author":[{"family":"Li","given":"Wenjie"},{"family":"Tan","given":"Shan"},{"family":"Fan","given":"Zhen"},{"family":"Chen","given":"Zhiwei"},{"family":"Ou","given":"Jiali"},{"family":"Liu","given":"Kun"},{"family":"Tao","given":"Ruiqiang"},{"family":"Tian","given":"Guo"},{"family":"Qin","given":"Minghui"},{"family":"Zeng","given":"Min"},{"family":"Lu","given":"Xubing"},{"family":"Zhou","given":"Guofu"}],"issued":{"date-parts":[[2024]]},"DOI":"10.22541/au.172114593.35310985/v1","URL":"https://doi.org/10.22541/au.172114593.35310985/v1","source":"europepmc"},{"id":"doi:10.21203/rs.3.rs-3989574/v1","type":"article-journal","title":"Piezoelectric neuron for neuromorphic computing","abstract":"Abstract Neuromorphic computing has attracted great attention for its massive parallelism and high energy efficiency. As the fundamental components of neuromorphic computing systems, artificial neurons play a key role in information processing. However, the development of artificial neurons that can simultaneously incorporate low hardware overhead, high reliability, high speed, and low energy consumption remains a challenge. To address this challenge, we propose and demonstrate a piezoelectric neuron with a simple circuit structure, consisting of a piezoelectric cantilever, a parallel capacitor, and a series resistor. It operates through the synergy between the converse piezoelectric effect and the capacitive charging/discharging. Thanks to this efficient and robust mechanism, the piezoelectric neuron not only implements critical leaky integrate-and-fire functions (including leaky integration, threshold-driven spiking, all-or-nothing response, refractory period, strength-modulated firing frequency, and spatiotemporal integration), but also demonstrates small cycle-to-cycle and device-to-device variations (~1.9% and ~10.0%, respectively), high endurance (10 7 ), high speed (integration/firing: ~9.6/~0.4 μs), and low energy consumption (~13.4 nJ/spike). Furthermore, spiking neural networks based on piezoelectric neurons are constructed, showing capabilities to implement both supervised and unsupervised learning. This study therefore demonstrates the piezoelectric neuron as a simple yet reliable, fast, and energy-efficient artificial neuron, and also showcases its applicability in neuromorphic computing.","author":[{"family":"Fan","given":"Zhen"},{"family":"Li","given":"Wenjie"},{"family":"Shan","given":"Tan"},{"family":"Chen","given":"Zhiwei"},{"family":"Jiali","given":"Ou"},{"family":"Kun","given":"Liu"},{"family":"Tao","given":"Ruiqiang"},{"family":"Tian","given":"Guo"},{"family":"Qin","given":"Minghui"},{"family":"Zeng","given":"Min"},{"family":"Lu","given":"Xubing"},{"family":"Zhou","given":"Guofu"}],"issued":{"date-parts":[[2024]]},"DOI":"10.21203/rs.3.rs-3989574/v1","URL":"https://doi.org/10.21203/rs.3.rs-3989574/v1","source":"europepmc"},{"id":"doi:10.21203/rs.3.rs-4498702/v1","type":"article-journal","title":"Negative Photo Conductivity Triggered with Visible Light in Wide Bandgap Oxide-Based Optoelectronic Crossbar Memristive Array for Photograph Sensing and Neuromorphic Computing Applications","abstract":"Abstract Photoresponsivity studies of wide-bandgap oxide-based devices have emerged as a vibrant and popular research area. Researchers have explored various material systems in their quest to develop devices capable of responding to illumination. In this study, we engineered a mature wide bandgap oxide-based bilayer heterostructure synaptic memristor to emulate the human brain for applications in neuromorphic computing and photograph sensing. The device exhibits advanced electric and electro-photonic synaptic functions, such as long-term potentiation (LTP), long-term depression (LTD), and paired pulse facilitation (PPF), by applying successive electric and photonic pulses. Moreover, the device exhibits exceptional electrical SET and photonic RESET endurance, maintaining its stability for a minimum of 1200 cycles without any degradation. Density functional theory calculations of the band structures provide insights into the conduction mechanism of the device. Based on this memristor array, we developed an autoencoder and convolutional neural network for noise reduction and image recognition tasks, which achieves a peak signal-to-noise ratio of 562 and high accuracy of 84.23%, while consuming lower energy by four orders of magnitude compared with the Tesla P40 GPU. This groundbreaking research not only opens doors for the integration of our device into image processing but also represents a significant advancement in the realm of in-memory computing and photograph sensing features in a single cell.","author":[{"family":"Kumar","given":"Dayanand"},{"family":"Li","given":"Hanrui"},{"family":"Singh","given":"Amit"},{"family":"Rajbhar","given":"Manoj"},{"family":"Syed","given":"Abdul"},{"family":"Lee","given":"Hoonkyung"},{"family":"El-Atab","given":"Nazek"}],"issued":{"date-parts":[[2024]]},"DOI":"10.21203/rs.3.rs-4498702/v1","URL":"https://doi.org/10.21203/rs.3.rs-4498702/v1","source":"preprints"},{"id":"doi:10.21203/rs.3.rs-3930064/v1","type":"article-journal","title":"Compact artificial neurons with time-to-first-spike coding for fast and energy-efficient federated neuromorphic computing","abstract":"Abstract Federated learning combined with Spiking Neural Networks (SNNs) provides a reliable and lightweight solution for privacy and energy constraints in billions of edge devices. However, the current rate-coding in SNNs has high latency and leads to increased power consumption. In this study, we develop a federated neuromorphic learning (FNL) system based on temporal coding to process edge information quickly and efficiently using compact neurons with time-to-first spike (TTFS) coding. These neurons are demonstrated using the volatile threshold switching characteristics of VO2 memristors. We implemented an FNL system for edge pattern recognition in hardware using these compact neurons. Notably, the memristor-based edge device with TTFS-coding shows 330× and 76× improvements in interference speed and energy consumption, respectively, compared to the conventional rate-coding scheme. Furthermore, a multimodal FNL system with TTFS-coding is demonstrated to process audio-visual vehicle signals from edge traffic scenes, with spike numbers for each iteration of edge devices being 26× lower compared to the rate-coding scheme. Our system, incorporating FNL and TTFS-coding, can facilitate the development of next-generation edge artificial intelligence requiring extremely low latency and power.","author":[{"family":"Yang","given":"Rui"},{"family":"Li","given":"Zhiyuan"},{"family":"Yao","given":"Jiaping"},{"family":"Zhang","given":"Beining"},{"family":"Li","given":"Zhongshao"},{"family":"Tang","given":"Wei"},{"family":"Gong","given":"Junjie"},{"family":"Li","given":"Yongfei"},{"family":"Cao","given":"Xun"},{"family":"Wang","given":"Zhongrui"},{"family":"Miao","given":"Xiangshui"}],"issued":{"date-parts":[[2024]]},"DOI":"10.21203/rs.3.rs-3930064/v1","URL":"https://doi.org/10.21203/rs.3.rs-3930064/v1","source":"preprints"},{"id":"doi:10.21203/rs.3.rs-3981743/v1","type":"article-journal","title":"Spintronic Memtransistor Leaky Integrate and Fire Neuron for Spiking Neural Networks","abstract":"Abstract Spintronic devices, -based on the domain walls and skyrmions, Spintronic devices have shown significant potential for applications in energy-efficient data storage and beyond CMOS computing architectures. Based on the magnetic multilayer spintronic devices, we demonstrate the magnetic field-gated and current-controlled Leaky integrate and fire neuron characteristics for the spiking neural network applications. The LIF characteristics are controlled by the current pulses, which drive the domain wall motion, and an external magnetic field is used as the bias to tune the firing properties of the neuron. Thus, the device works like a gate-controlled LIF neuron, acting like a spintronic Mem-transistor device. The current and magnetic field-controlled neuron spiking rate is shown. We developed a LIF neuron model based on the measured characteristics to show the device's integration in the system-level SNNs. We extend the study and propose a scaled version of the demonstrated device with a multilayer spintronic domain wall magnetic tunnel junction as a LIF neuron. using the combination of SOT and the variation of the demagnetization energy across the thin film, the modified leaky integrate and fire LIF neuron characteristics are realized in the proposed devices. Finally, we integrate the measured and simulated neuron models in the 3-layer spiking neural network (SNN) and convolutional spiking neural network CSNN framework to test these spiking neuron models for classification of the MNIST and FMNIST datasets. In both architectures, the network achieves classification accuracy above 96%. Considering the good system-level performance, mem-transistor properties, and promise for scalability. The presented devices show an excellent alternative for neuromorphic computing applications.","author":[{"family":"Lone","given":"Aijaz"},{"family":"Tang","given":"Meng"},{"family":"Rahimi","given":"Daniel"},{"family":"Zou","given":"Xuecui"},{"family":"Zheng","given":"Dong"},{"family":"Fariborzi","given":"Hossein"},{"family":"Zhang","given":"Xixiang"},{"family":"Setti","given":"Gianluca"}],"issued":{"date-parts":[[2024]]},"DOI":"10.21203/rs.3.rs-3981743/v1","URL":"https://doi.org/10.21203/rs.3.rs-3981743/v1","source":"preprints"},{"id":"doi:10.21203/rs.3.rs-3644668/v1","type":"article-journal","title":"Intelligent machines work in unstructured environments by differential neuromorphic computing","abstract":"Abstract Efficient operation of intelligent machines in the real world requires methods that allow them to understand and predict the uncertainties presented by the unstructured environments with good accuracy, scalability and generalization, similar to humans. Current methods rely on pretrained networks instead of continuously learning from the dynamic signal properties of working environments and suffer inherent limitations, such as data-hungry procedures, and limited generalization capabilities. Herein, we present a memristor-based differential neuromorphic computing, perceptual signal processing and learning method for intelligent machines. The main features of environmental information such as amplification (&gt; 720%) and adaptation (&lt; 50%) of mechanical stimuli encoded in memristors, are extracted to obtain human-like processing in unstructured environments. The developed method takes advantage of the intrinsic multi-state property of memristors and exhibits good scalability and generalization, as confirmed by validation in two different application scenarios: object grasping and autonomous driving. In the former, a robot hand experimentally realizes safe and stable grasping through fast learning (in ~ 1 ms) the unknown object features (e.g., sharp corner and smooth surface) with a single memristor. In the latter, the decision-making information of 10 unstructured environments in autonomous driving (e.g., overtaking cars, pedestrians) is accurately (94%) extracted with a 40×25 memristor array. By mimicking the intrinsic nature of human low-level perception mechanisms, the electronic memristive neuromorphic circuit-based method, presented here shows the potential for adapting to diverse sensing technologies and helping intelligent machines generate smart high-level decisions in the real world.","author":[{"family":"Occhipinti","given":"Luigi"},{"family":"Wang","given":"Shengbo"},{"family":"Gao","given":"Shuo"},{"family":"Tang","given":"Chenyu"},{"family":"Occhipinti","given":"Edoardo"},{"family":"Li","given":"Cong"},{"family":"Wang","given":"Shurui"},{"family":"Wang","given":"Jiaqi"},{"family":"Zhao","given":"Hubin"},{"family":"Hu","given":"Guohua"},{"family":"Nathan","given":"Arokia"},{"family":"Dahiya","given":"Ravinder"}],"issued":{"date-parts":[[2023]]},"DOI":"10.21203/rs.3.rs-3644668/v1","URL":"https://doi.org/10.21203/rs.3.rs-3644668/v1","source":"preprints"},{"id":"doi:10.1088/1674-4926/23120051","type":"article-journal","title":"Complementary memtransistors for neuromorphic computing: How, what and why","abstract":"Abstract Memtransistors in which the source−drain channel conductance can be nonvolatilely manipulated through the gate signals have emerged as promising components for implementing neuromorphic computing. On the other side, it is known that the complementary metal-oxide-semiconductor (CMOS) field effect transistors have played the fundamental role in the modern integrated circuit technology. Therefore, will complementary memtransistors (CMT) also play such a role in the future neuromorphic circuits and chips? In this review, various types of materials and physical mechanisms for constructing CMT (how) are inspected with their merits and need-to-address challenges discussed. Then the unique properties (what) and potential applications of CMT in different learning algorithms/scenarios of spiking neural networks (why) are reviewed, including supervised rule, reinforcement one, dynamic vision with in-sensor computing, etc. Through exploiting the complementary structure-related novel functions, significant reduction of hardware consuming, enhancement of energy/efficiency ratio and other advantages have been gained, illustrating the alluring prospect of design technology co-optimization (DTCO) of CMT towards neuromorphic computing.","author":[{"family":"Chen","given":"Qi"},{"family":"Zhou","given":"Yue"},{"family":"Xiong","given":"Weiwei"},{"family":"Chen","given":"Zirui"},{"family":"Wang","given":"Yasai"},{"family":"Miao","given":"Xiangshui"},{"family":"He","given":"Yuhui"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1088/1674-4926/23120051","URL":"https://doi.org/10.1088/1674-4926/23120051","source":"crossref"},{"id":"doi:10.36227/techrxiv.21532533.v2","type":"article-journal","title":"Neuromorphic Computing with 28nm High-K-Metal Gate Ferroelectric Field Effect Transistors Based Artificial Synapses","abstract":"This paper presents a comprehensive overview of 28 nm high-k-metal gate-based ferroelectric field effect transistor devices for synaptic applications. The device under test was fabricated on 300mm wafers at GlobalFoundries. The fabricated devices demonstrate 10 3 WRITE-endurance cycles and 10 4 seconds of data-retention capability at 85°C. We have also assessed the FeFET-based crossbar array’s performance in system-level applications. By simulating the FeFET crossbar array for neuromorphic applications, the system performance was assessed. For datasets from the National Institute of Standards and Technology (MNIST), the crossbar array achieved software-comparable inference accuracy of about 97% using multilayer perceptron (MLP) neural networks.","author":[{"family":"De","given":"Sourav"},{"family":"Raffel","given":"Yannick"},{"family":"Thunder","given":"Sunanda"},{"family":"Müller","given":"Franz"},{"family":"Lederer","given":"Maximilian"},{"family":"Kaempfe","given":"Thomas"},{"family":"Rana","given":"Masud"},{"family":"Pirro","given":"Luca"},{"family":"Seidel","given":"Konrad"},{"family":"Chakrabarti","given":"Bhaswar"}],"issued":{"date-parts":[[2023]]},"DOI":"10.36227/techrxiv.21532533.v2","URL":"https://doi.org/10.36227/techrxiv.21532533.v2","source":"crossref"},{"id":"doi:10.1088/2634-4386/ad025b","type":"article-journal","title":"Multimode Fabry-Perot laser as a reservoir computing and extreme learning machine photonic accelerator","abstract":"Abstract In this work, we introduce Fabry–Perot lasers as neuromoprhic nodes in the context of time-delayed reservoir computing and extreme learning machine (ELM) for the processing of temporal signals and the high-speed classification of images. By exploiting the multi-wavelength emission capabilities of the Fabry–Perot lasers, additional processing nodes can be introduced, thus raising the computational power without sacrificing processing speed. An experimental validation of this concept using a Fabry–Perot ELM is presented targeting a time depedent task such as channel equalization for a 50 km 28 Gbaud ‘PAM-4’ transmission, offering hard-decision forward error correction compatible performance. Additionally, the Fabry–Perot neuromorphic concept has been further strengthened by modifying the data entry technique by parallelelly assigning different samples of the input signal to different modes so as to significantly reduce speed penalty. Numerical simulations revealed that this alternative data insertion technique can offer a reduction of the processing delay and physical footprint by 75% compared to the conventional approach assigning the same symbols to all Fairy–Perot modes. Moreover, by using a similar data processing scheme in ‘MNIST’ image classification task we were able to numerically achieve a processing speed of 255.1 Mimages s −1 and a classification accuracy up to 95.95%.","author":[{"family":"Skontranis","given":"Menelaos"},{"family":"Sarantoglou","given":"George"},{"family":"Sozos","given":"Kostas"},{"family":"Kamalakis","given":"Thomas"},{"family":"Mesaritakis","given":"Charis"},{"family":"Bogris","given":"Adonis"}],"issued":{"date-parts":[[2023]]},"DOI":"10.1088/2634-4386/ad025b","URL":"https://doi.org/10.1088/2634-4386/ad025b","source":"crossref"},{"id":"doi:10.1088/2634-4386/acf1c5","type":"article-journal","title":"A temporally and spatially local spike-based backpropagation algorithm to enable training in hardware","abstract":"Abstract Spiking neural networks (SNNs) have emerged as a hardware efficient architecture for classification tasks. The challenge of spike-based encoding has been the lack of a universal training mechanism performed entirely using spikes. There have been several attempts to adopt the powerful backpropagation (BP) technique used in non-spiking artificial neural networks (ANNs): (1) SNNs can be trained by externally computed numerical gradients. (2) A major advancement towards native spike-based learning has been the use of approximate BP using spike-time dependent plasticity with phased forward/backward passes. However, the transfer of information between such phases for gradient and weight update calculation necessitates external memory and computational access. This is a challenge for standard neuromorphic hardware implementations. In this paper, we propose a stochastic SNN based back-prop (SSNN-BP) algorithm that utilizes a composite neuron to simultaneously compute the forward pass activations and backward pass gradients explicitly with spikes. Although signed gradient values are a challenge for spike-based representation, we tackle this by splitting the gradient signal into positive and negative streams. The composite neuron encodes information in the form of stochastic spike-trains and converts BP weight updates into temporally and spatially local spike coincidence updates compatible with hardware-friendly resistive processing units. Furthermore, we characterize the quantization effect of discrete spike-based weight update to show that our method approaches BP ANN baseline with sufficiently long spike-trains. Finally, we show that the well-performing softmax cross-entropy loss function can be implemented through inhibitory lateral connections enforcing a winner take all rule. Our SNN with a two-layer network shows excellent generalization through comparable performance to ANNs with equivalent architecture and regularization parameters on static image datasets like MNIST, Fashion-MNIST, Extended MNIST, and temporally encoded image datasets like Neuromorphic MNIST datasets. Thus, SSNN-BP enables BP compatible with purely spike-based neuromorphic hardware.","author":[{"family":"Biswas","given":"Anmol"},{"family":"Saraswat","given":"Vivek"},{"family":"Ganguly","given":"Udayan"}],"issued":{"date-parts":[[2023]]},"DOI":"10.1088/2634-4386/acf1c5","URL":"https://doi.org/10.1088/2634-4386/acf1c5","source":"crossref"},{"id":"doi:10.1088/2634-4386/ad5584","type":"article-journal","title":"Bio-realistic neural network implementation on Loihi 2 with Izhikevich neurons","abstract":"Abstract Neuromorphic systems are designed to emulate the principles of biological information processing, with the goals of improving computational efficiency and reducing energy usage. A critical aspect of these systems is the fidelity of neuron models and neural networks to their biological counterparts. In this study, we implemented the Izhikevich neuron model on Intel’s Loihi 2 neuromorphic processor. The Izhikevich neuron model offers a more biologically accurate alternative to the simpler leaky-integrate and fire model, which is natively supported by Loihi 2. We compared these two models within a basic two-layer network, examining their energy consumption, processing speeds, and memory usage. Furthermore, to demonstrate Loihi 2’s ability to realize complex neural structures, we implemented a basal ganglia circuit to perform a Go/No-Go decision-making task. Our findings demonstrate the practicality of customizing neuron models on Loihi 2, thereby paving the way for constructing spiking neural networks that better replicate biological neural networks and have the potential to simulate complex cognitive processes.","author":[{"family":"Uludağ","given":"Recep"},{"family":"Çağdaş","given":"Serhat"},{"family":"İşler","given":"Yavuz"},{"family":"Şengör","given":"Neslihan"},{"family":"Aktürk","given":"İsmail"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1088/2634-4386/ad5584","URL":"https://doi.org/10.1088/2634-4386/ad5584","source":"crossref"},{"id":"doi:10.1002/advs.202303817","type":"article-journal","title":"Memcapacitor Crossbar Array with Charge Trap NAND Flash Structure for Neuromorphic Computing","abstract":"Abstract The progress of artificial intelligence and the development of large‐scale neural networks have significantly increased computational costs and energy consumption. To address these challenges, researchers are exploring low‐power neural network implementation approaches and neuromorphic computing systems are being highlighted as potential candidates. Specifically, the development of high‐density and reliable synaptic devices, which are the key elements of neuromorphic systems, is of particular interest. In this study, an 8 × 16 memcapacitor crossbar array that combines the technological maturity of flash cells with the advantages of NAND flash array structure is presented. The analog properties of the array with high reliability are experimentally demonstrated, and vector‐matrix multiplication with extremely low error is successfully performed. Additionally, with the capability of weight fine‐tuning characteristics, a spiking neural network for CIFAR‐10 classification via off‐chip learning at the wafer level is implemented. These experimental results demonstrate a high level of accuracy of 92.11%, with less than a 1.13% difference compared to software‐based neural networks (93.24%).","author":[{"family":"Hwang","given":"Sungmin"},{"family":"Yu","given":"Junsu"},{"family":"Song","given":"Min"},{"family":"Hwang","given":"Hwiho"},{"family":"Kim","given":"Hyungjin"}],"issued":{"date-parts":[[2023]]},"DOI":"10.1002/advs.202303817","URL":"https://doi.org/10.1002/advs.202303817","source":"crossref"},{"id":"doi:10.1063/5.0177232","type":"article-journal","title":"Extremely energy-efficient, magnetic field-free, skyrmion-based memristors for neuromorphic computing","abstract":"The human brain can process information more efficiently than computers due to the dynamics of neurons and synapses. Mimicking such a system can lead to the practical implementation of artificial spiking neural networks. Spintronic devices have been shown to be an ideal solution for realizing the hardware required for neuromorphic computing. Skyrmions prove to be an effective candidate as information carriers owing to their topological protection and particle-like nature. Ferrimagnet and antiferromagnet-based spintronics have been employed previously to obtain an ultrafast simulation of artificial synapses and neurons. Here, we have proposed a ferromagnetic device of stack Ta3nmPt3nmCu0.65nmCo0.5nmPt1nm that is capable of ultrafast simulation of artificial neurons and synapses, owing to the high velocity of the stabilized skyrmions in the system. Electrical pulses of nanosecond pulse width were used to control the accumulation and dissipation of skyrmions in the system, analogous to the variations in the synaptic weights. Lateral structure inversion asymmetry is used to bring about a field-free switching in the system, leading to an energy-efficient switching process. Magnetic field-free deterministic switching and low pulse width current pulses drastically reduce energy consumption by 106 times compared to the existing ferromagnet-based neuromorphic devices. Artificial neuron, synapse, and memristor functionalities have been reproduced on the same device with characteristic time scales and field-free switching, better than any existing ferromagnet-based neuromorphic devices. The results recognize ferromagnet-based skyrmions as viable candidates for ultrafast neuromorphic spintronics capable of executing cognitive tasks with extremely high efficiency.","author":[{"family":"Joy","given":"Ajin"},{"family":"Satheesh","given":"Sreyas"},{"family":"Kumar","given":"PSA"}],"issued":{"date-parts":[[2023]]},"DOI":"10.1063/5.0177232","URL":"https://doi.org/10.1063/5.0177232","source":"crossref"},{"id":"doi:10.1038/s44306-023-00006-z","type":"article-journal","title":"Anomalous hall and skyrmion topological hall resistivity in magnetic heterostructures for the neuromorphic computing applications","abstract":"Abstract Topologically protected spin textures, such as magnetic skyrmions, have shown the potential for high-density data storage and energy-efficient computing applications owing to their particle-like behavior, small size, and low driving current requirements. Evaluating the writing and reading of the skyrmion’s magnetic and electrical characteristics is crucial to implementing these devices. In this paper, we present the magnetic heterostructure Hall bar device and study the anomalous Hall and topological Hall signals in these devices. Using different measurement techniques, we investigate the magnetic and electrical characteristics of the magnetic structure. We measure the skyrmion topological resistivity and the magnetic field at different temperatures. MFM imaging and micromagnetic simulations further explain the anomalous Hall and topological Hall resistivity characteristics at various magnetic fields and temperatures. The study is extended to propose a skyrmion-based synaptic device showing spin-orbit torque-controlled plasticity. The resistance states are read using the anomalous Hall measurement technique. The device integration in a neuromorphic circuit is simulated in a 3-layer feedforward artificial neural network ANN. Based on the proposed synapses, the neural network is trained and tested on the MNIST data set, where a recognition accuracy performance of about 90% is achieved. Considering the nanosecond reading/writing time scale and a good system level performance, these devices exhibit a substantial prospect for energy-efficient neuromorphic computing.","author":[{"family":"Lone","given":"Aijaz"},{"family":"Zou","given":"Xuecui"},{"family":"Das","given":"Debasis"},{"family":"Fong","given":"Xuanyao"},{"family":"Setti","given":"Gianluca"},{"family":"Fariborzi","given":"Hossein"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1038/s44306-023-00006-z","URL":"https://doi.org/10.1038/s44306-023-00006-z","source":"crossref"},{"id":"doi:10.1088/1674-4926/24040038","type":"article-journal","title":"InGaZnO-based photoelectric synaptic devices for neuromorphic computing","abstract":"Abstract Photoelectric synaptic devices could emulate synaptic behaviors utilizing photoelectric effects and offer promising prospects with their high-speed operation and low crosstalk. In this study, we introduced a novel InGaZnO-based photoelectric memristor. Under both electrical and optical stimulation, the device successfully emulated synaptic characteristics including excitatory postsynaptic current (EPSC), paired-pulse facilitation (PPF), long-term potentiation (LTP), and long-term depression (LTD). Furthermore, we demonstrated the practical application of our synaptic devices through the recognition of handwritten digits. The devices have successfully shown their ability to modulate synaptic weights effectively through light pulse stimulation, resulting in a recognition accuracy of up to 93.4%. The results illustrated the potential of IGZO-based memristors in neuromorphic computing, particularly their ability to simulate synaptic functionalities and contribute to image recognition tasks.","author":[{"family":"Song","given":"Jieru"},{"family":"Meng","given":"Jialin"},{"family":"Wang","given":"Tianyu"},{"family":"Wan","given":"Changjin"},{"family":"Zhu","given":"Hao"},{"family":"Sun","given":"Qingqing"},{"family":"Zhang","given":"David"},{"family":"Chen","given":"Lin"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1088/1674-4926/24040038","URL":"https://doi.org/10.1088/1674-4926/24040038","source":"crossref"},{"id":"doi:10.1039/bk9781839169946-00498","type":"article-journal","title":"Halide Perovskites for Neuromorphic Computing","abstract":"The next generation of neuromorphic computing, which is related to emulating the neural structure and operation of the human brain, will extend into areas that correspond to human cognition, such as interpretation and autonomous adaptation. Progress in materials and devices is critical to address novel situations and abstraction to automate ordinary human activities. Halide perovskites constitute a family of materials with many superior properties, such as long charge-carrier diffusion length, strong light absorptivity, ambipolar charge transport, ionic conductivity and solution processability. They have been successfully implemented in broad applications such as photovoltaics, light-emitting diodes and photodetectors. Their high mobility renders this class of solution-processed materials appropriate for application in field-effect transistors, whereas their usually present hysteresis, which may originate from ferroelectricity, charge-carrier traps, and migration of ions, has been explored for application in artificial synapses, which require gradual modulation of responses.","author":[{"family":"Vasilopoulou","given":"Maria"},{"family":"Davazoglou","given":"Konstantinos"},{"family":"Yusoff","given":"Abd"},{"family":"Chai","given":"Yang"},{"family":"Noh","given":"Yong"},{"family":"Anthopoulos","given":"Thomas"},{"family":"Nazeeruddin","given":"Mohammad"}],"issued":{"date-parts":[[2023]]},"DOI":"10.1039/bk9781839169946-00498","URL":"https://doi.org/10.1039/bk9781839169946-00498","source":"crossref"},{"id":"doi:10.35848/1347-4065/acb060","type":"article-journal","title":"Interface engineering of amorphous gallium oxide crossbar array memristors for neuromorphic computing","abstract":"Abstract This paper reports on the fabrication and characterization of crossbar array memristors using amorphous gallium oxide (a-GaO x ) for implementing high-speed and wide-dynamic range artificial synaptic functions. The a-GaO x memristors were fabricated by pulsed laser deposition in an argon atmosphere using a platinum bottom electrode and an indium tin oxide (ITO) top electrode. We revealed that the interface engineering at a-GaO x /ITO is the key to demonstrating exemplary resistive switching operation. Stable counter figure-8 hysteresis loops were obtained by voltage application, leading to the successful demonstration of non-volatile retention over 10 4 s and the multi-level conductance modulation. Furthermore, spike-timing-dependent plasticity (STDP) was artificially implemented by applying pre- and post-spike voltages to the device. Consequently, significant weight-change rates were achieved in the asymmetric STDP imitation, which can be attributed to the reliable resistive switching properties of the device with an extensive dynamic range. These results indicate that the a-GaO x crossbar array memristor is a promising hardware platform for neuromorphic computing applications.","author":[{"family":"Masaoka","given":"Naoki"},{"family":"Hayashi","given":"Yusuke"},{"family":"Tohei","given":"Tetsuya"},{"family":"Sakai","given":"Akira"}],"issued":{"date-parts":[[2023]]},"DOI":"10.35848/1347-4065/acb060","URL":"https://doi.org/10.35848/1347-4065/acb060","source":"crossref"},{"id":"doi:10.4018/978-1-6684-6596-7.ch005","type":"article-journal","title":"Prediction of Skin Cancer Using Convolutional Neural Network (CNN)","abstract":"Skin disorders are one of the most common types of disorders that are primarily diagnosed visually with scientific screening observed through dermoscopic evaluation, histopathological evaluation, and a biopsy. Diagnostic accuracy has a strong relevance to physician skill. Painful effects of skin disease hamper the mental condition of a patient. The authors propose an approach to detect the skin diseases based upon image processing as well as machine learning techniques i.e., convolutional neural networks (CNN). CNN is a specific type of neural network model that allows us to extract higher depictions for the image content. It is a deep learning algorithm to perform generative and descriptive tasks. Machine learning generates two types of prediction-batches and real time.","author":[{"family":"Srinivasan","given":"Dhamodharan"},{"family":"Paulraj","given":"Prabhakaran"},{"family":"Ashokkumar","given":"N"}],"issued":{"date-parts":[[2023]]},"DOI":"10.4018/978-1-6684-6596-7.ch005","URL":"https://doi.org/10.4018/978-1-6684-6596-7.ch005","source":"crossref"},{"id":"doi:10.15129/3237018a-8f35-4034-acbb-f21cfb9d60a2","type":"article-journal","title":"Data for: \"All-optical passive spiking processing and reservoir computing with a silicon microring and wavelength-time division multiplexing\"","abstract":"The manuscript describes passive deterministic spiking dynamics in silicon microring resonators and their use for all-optical spiking reservoir computing. Data used for figures in the manuscript. Data can be extracted and visualized using python. A supplementary material document, reporting on the study of the spiking response of the silicon MRR at a positive detuning, is also included.","author":[{"family":"Donati","given":"Giovanni"}],"issued":{"date-parts":[[2025]]},"DOI":"10.15129/3237018a-8f35-4034-acbb-f21cfb9d60a2","URL":"https://doi.org/10.15129/3237018a-8f35-4034-acbb-f21cfb9d60a2","source":"datacite"},{"id":"doi:10.21203/rs.3.rs-3151403/v1","type":"article-journal","title":"Ultrafast Silicon Optical Nonlinear Activator for Neuromorphic Computing","abstract":"Abstract Optical neural networks (ONNs) have shown great promise in overcoming the speed and efficiency bottlenecks of artificial neural networks (ANNs). However, the absence of high-speed, energy-efficient nonlinear activators significantly impedes the advancement of ONNs and their extension to ultrafast application scenarios like autonomous vehicles and real-time intelligent signal processing. In this work, we designed and fabricated a novel silicon-based ultrafast all-optical nonlinear activator, leveraging the hybrid integration of silicon slot waveguides, plasmonic slot waveguides, and monolayer graphene. We utilized double-balanced detection and synchronous pump-probe measurement techniques to experimentally evaluate the static and dynamic characteristics of the activators, respectively. Exploiting the exceptional picosecond scale photogenerated carrier relaxation time of graphene, the response time of the activator is markedly reduced to ~93.6 ps. This response time is approximately five times faster than electronic neural networks, establishing our all-optical activator as the fastest known in silicon photonics to our knowledge. Moreover, the all-optical nonlinear activator holds a low threshold power of 5.49 mW and a corresponding power consumption per activation of 0.51 pJ. Furthermore, we confirm its feasibility and capability for use in ONNs by simulation, achieving a high accuracy of 96.8% for MNIST handwritten digit recognition and a mean absolute error of less than 0.1 dB for optical signal-to-noise ratio monitoring of high-speed optical signals. This breakthrough in speed and energy efficiency of all-optical nonlinear activators opens the door to significant improvements in the performance and applicability of ONNs, ushering in a new era of advanced artificial intelligence technologies with enormous potential.","author":[{"family":"Yan","given":"Siqi"},{"family":"Zhou","given":"Ziwen"},{"family":"Liu","given":"Chen"},{"family":"Zhao","given":"Weiwei"},{"family":"Liu","given":"Jingze"},{"family":"Jiang","given":"Ting"},{"family":"Peng","given":"Wenyi"},{"family":"Xiong","given":"Jiawang"},{"family":"Wu","given":"Hao"},{"family":"Zhang","given":"Chi"},{"family":"Ding","given":"Yunhong"},{"family":"Ros","given":"Francesco"}],"issued":{"date-parts":[[2023]]},"DOI":"10.21203/rs.3.rs-3151403/v1","URL":"https://doi.org/10.21203/rs.3.rs-3151403/v1","source":"preprints"},{"id":"doi:10.20944/preprints202309.0008.v1","type":"manuscript","title":"Spike Optimization to Improve Properties of Ferroelectric Tunnel Junction Synaptic Devices for Neuromorphic Computing System Applications","abstract":"The continuous advancement of Artificial Intelligence (AI) technology depends on the efficient processing of unstructured data, encompassing text, speech, and video. Traditional serial computing systems based on the von Neumann architecture, employed in information and communication technology development for decades, not suitable for the concurrent processing of massive unstructured data tasks with relatively low-level operations. As a result, there arises a pressing need to develop novel parallel computing systems. Recently, there has been a burgeoning interest among developers in emulating the intricate operations of the human brain, which efficiently processes vast datasets with remarkable energy efficiency. This has led to the proposal of neuromorphic computing systems. Of these, Spiking Neural Networks (SNNs), designed to closely resemble the information processing mechanisms of biological neural networks, are subjects of intense research activity. Nevertheless, a comprehensive investigation into the relationship between spike shapes and Spike-Timing-Dependent Plasticity (STDP) to ensure efficient synaptic behavior remains insufficiently explored. In this study, we systematically explore various input spike types to optimize the resistive memory characteristics of Halfnium-based Ferroelectric Tunnel Junction (FTJ) devices. Among the various spike shapes investigated, the square-triangle (RT) spike exhibited good linearity and symmetry, and a wide range of weight values could be realized depending on the offset of the RT spike. These results indicate that the spike shape serves as a crucial indicator in the alteration of synaptic connections, representing the strength of the signals.","author":[{"family":"Byun","given":"Jisu"},{"family":"Kho","given":"Wonwoo"},{"family":"Hwang","given":"Hyunjoo"},{"family":"Kang","given":"Yoomi"},{"family":"Kang","given":"Minjeong"},{"family":"Noh","given":"Taewan"},{"family":"Kim","given":"Hoseong"},{"family":"Lee","given":"Jimin"},{"family":"Kim","given":"Hyo"},{"family":"Ahn","given":"Ji"},{"family":"Ahn","given":"Seung"}],"issued":{"date-parts":[[2023]]},"DOI":"10.20944/preprints202309.0008.v1","URL":"https://doi.org/10.20944/preprints202309.0008.v1","source":"preprints"},{"id":"doi:10.21203/rs.3.rs-2471300/v1","type":"article-journal","title":"Bio-inspired Artificial synapse for neuromorphic computing based on NiO nanoparticle thin film","abstract":"Abstract The unprecedented need for data processing in the modern technological era has created opportunities in neuromorphic devices and computation. This is primarily due to the extensive parallel processing done in our human brain. Data processing and logical decision-making at the same physical location is an exciting aspects of neuromorphic computation. For this, establishing reliable resistive switching devices working at room temperature with ease of fabrication is important. Here, a reliable analog resistive switching device based on Au/NiO nanoparticles/Au nanoparticles is discussed. The application of positive and negative voltage pulses of constant amplitude results in enhancement and reduction of synaptic current, which is consistent with potentiation and depression, respectively. The change in the conductance resulting in such a process can be fitted well with double exponential growth and decay, respectively. Consistent potentiation and depression characteristics reveal that non-ideal voltage pulses can result in a linear dependence of potentiation and depression with electric pulses. Long-term potentiation (LTP) and Long-term depression (LTD) characteristics have been established, which are essential for mimicking the biological synaptic applications. The NiO nanoparticle-based devices can also be used for controlled synaptic enhancement by optimizing the electric pulses, displaying typical learning-forgetting-relearning characteristics.","author":[{"family":"Hadiyal","given":"Keval"},{"family":"Ganesan","given":"Ramakrishnan"},{"family":"Rastogi","given":"A"},{"family":"Thamankar","given":"R"}],"issued":{"date-parts":[[2023]]},"DOI":"10.21203/rs.3.rs-2471300/v1","URL":"https://doi.org/10.21203/rs.3.rs-2471300/v1","source":"preprints"},{"id":"doi:10.21203/rs.3.rs-2862199/v1","type":"article-journal","title":"A low-power vertical dual-gate neurotransistor with short-term memory for high energy-efficient neuromorphic computing","abstract":"Abstract Neuromorphic computing aims to emulate the computing processes of the brain by replicating the functions of biological neural networks using electronic counterparts. One promising approach is dendritic computing, which takes inspiration from the multi-dendritic branch structure of neurons to enhance the processing capability of artificial neural networks. While there has been a recent surge of interest in implementing dendritic computing using emerging devices, achieving artificial dendrites with throughputs and energy efficiency comparable to those of the human brain has proven challenging. In this study, we report on the development of a compact and low-power neurotransistor based on a vertical dual-gate electrolyte-gated transistor (EGT) with short-term memory characteristics, a 30 nm channel length, a record-low read power of ~3.16 fW and a biology-comparable read energy of ~30 fJ. Leveraging this neurotransistor, we demonstrate dendrite integration as well as digital and analog dendritic computing for coincidence detection. We also showcase the potential of neurotransistors in realizing advanced brain-like functions by developing a hardware neural network and demonstrating bio-inspired sound localization. Our results suggest that the neurotransistor-based approach may pave the way for next-generation neuromorphic computing with energy efficiency on par with those of the brain.","author":[{"family":"Xu","given":"Han"},{"family":"Luo","given":"Qing"},{"family":"An","given":"Junjie"},{"family":"Li","given":"Yue"},{"family":"Wu","given":"Shuyu"},{"family":"Yao","given":"Zhihong"},{"family":"Xu","given":"Xiaoxin"},{"family":"Zhang","given":"Peiwen"},{"family":"Dou","given":"Chunmeng"},{"family":"Jiang","given":"Hao"},{"family":"Pan","given":"Liyang"},{"family":"Zhang","given":"Xumeng"}],"issued":{"date-parts":[[2023]]},"DOI":"10.21203/rs.3.rs-2862199/v1","URL":"https://doi.org/10.21203/rs.3.rs-2862199/v1","source":"preprints"},{"id":"doi:10.1088/2634-4386/ad473b","type":"article-journal","title":"Spike-based computation using classical recurrent neural networks","abstract":"Abstract Spiking neural networks (SNNs) are a type of artificial neural networks in which communication between neurons is only made of events, also called spikes. This property allows neural networks to make asynchronous and sparse computations and therefore drastically decrease energy consumption when run on specialized hardware. However, training such networks is known to be difficult, mainly due to the non-differentiability of the spike activation, which prevents the use of classical backpropagation. This is because state-of-the-art SNNs are usually derived from biologically-inspired neuron models, to which are applied machine learning methods for training. Nowadays, research about SNNs focuses on the design of training algorithms whose goal is to obtain networks that compete with their non-spiking version on specific tasks. In this paper, we attempt the symmetrical approach: we modify the dynamics of a well-known, easily trainable type of recurrent neural network (RNN) to make it event-based. This new RNN cell, called the spiking recurrent cell, therefore communicates using events, i.e. spikes, while being completely differen-tiable. Vanilla backpropagation can thus be used to train any network made of such RNN cell. We show that this new network can achieve performance comparable to other types of spiking networks in the MNIST benchmark and its variants, the Fashion-MNIST and the Neuromorphic-MNIST. Moreover, we show that this new cell makes the training of deep spiking networks achievable.","author":[{"family":"Geeter","given":"Florent"},{"family":"Ernst","given":"Damien"},{"family":"Drion","given":"Guillaume"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1088/2634-4386/ad473b","URL":"https://doi.org/10.1088/2634-4386/ad473b","source":"crossref"},{"id":"doi:10.1088/2634-4386/ad8c78","type":"article-journal","title":"Unsupervised end-to-end training with a self-defined target","abstract":"Abstract Designing algorithms for versatile AI hardware that can learn on the edge using both labeled and unlabeled data is challenging. Deep end-to-end training methods incorporating phases of self-supervised and supervised learning are accurate and adaptable to input data but self-supervised learning requires even more computational and memory resources than supervised learning, too high for current embedded hardware. Conversely, unsupervised layer-by-layer training, such as Hebbian learning, is more compatible with existing hardware but does not integrate well with supervised learning. To address this, we propose a method enabling networks or hardware designed for end-to-end supervised learning to also perform high-performance unsupervised learning by adding two simple elements to the output layer: winner-take-all selectivity and homeostasis regularization. These mechanisms introduce a ‘self-defined target’ for unlabeled data, allowing purely unsupervised training for both fully-connected and convolutional layers using backpropagation or equilibrium propagation on datasets like MNIST (up to 99.2%), Fashion-MNIST (up to 90.3%), and SVHN (up to 81.5%). We extend this method to semi-supervised learning, adjusting targets based on data type, achieving 96.6% accuracy with only 600 labeled MNIST samples in a multi-layer perceptron. Our results show that this approach can effectively enable networks and hardware initially dedicated to supervised learning to also perform unsupervised learning, adapting to varying availability of labeled data.","author":[{"family":"Liu","given":"Dongshu"},{"family":"Laydevant","given":"Jérémie"},{"family":"Pontlevy","given":"Adrien"},{"family":"Querlioz","given":"Damien"},{"family":"Grollier","given":"Julie"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1088/2634-4386/ad8c78","URL":"https://doi.org/10.1088/2634-4386/ad8c78","source":"crossref"},{"id":"doi:10.1088/2634-4386/acc050","type":"article-journal","title":"Artificial visual neuron based on threshold switching memristors","abstract":"Abstract The human visual system encodes optical information perceived by photoreceptors in the retina into neural spikes and then processes them by the visual cortex, with high efficiency and low energy consumption. Inspired by this information processing mode, an universal artificial neuron constructed with a resistor ( R s ) and a threshold switching memristor can realize rate coding by modulating pulse parameters and the resistance of R s . Owing to the absence of an external parallel capacitor, the artificial neuron has minimized chip area. In addition, an artificial visual neuron is proposed by replacing R s in the artificial neuron with a photo-resistor. The oscillation frequency of the artificial visual neuron depends on the distance between the photo-resistor and light, which is fundamental to acquiring depth perception for precise recognition and learning. A visual perception system with the artificial visual neuron can accurately and conceptually emulate the self-regulation process of the speed control system in a driverless automobile. Therefore, the artificial visual neuron can process efficiently sensory data, reduce or eliminate data transfer and conversion at sensor/processor interfaces, and expand its application in the field of artificial intelligence.","author":[{"family":"Wen","given":"Juan"},{"family":"Zhu","given":"Zhen"},{"family":"Guo","given":"Xin"}],"issued":{"date-parts":[[2023]]},"DOI":"10.1088/2634-4386/acc050","URL":"https://doi.org/10.1088/2634-4386/acc050","source":"crossref"},{"id":"doi:10.1088/2634-4386/ad7314","type":"article-journal","title":"From ‘follow the leader’ to autonomous swarming: physical reservoir computing in two dimensions","abstract":"Abstract Percolating networks of nanoparticles (PNNs) are self-assembled nanoscale systems that possess brain-like characteristics that are useful for information processing, particularly within a reservoir computing (RC) framework. Previous work has successfully demonstrated one-dimensional RC tasks, such as chaotic time-series prediction and nonlinear transformation. We focus here on the challenge of two-dimensional (2D) tasks and introduce novel ‘follow the leader’ and ‘swarming’ tasks. In the first task a ‘follower’ is required to accurately track a ‘leader’ in two dimensions. The task is performed successfully for a range of trajectories and parameters, for both position-based tracking and velocity-based tracking incorporating inertia. In both cases, the task is successful even for trajectories unseen in training. We then successfully demonstrate a 2D implementation of swarming behavior. Each agent is represented by a PNN which is trained to react to the behavior of the other members of the swarm, such that the future trajectory of all agents is generated autonomously. As well as demonstrating that the computational capabilities of PNNs can be extended into two dimensions, this work presents a first step in the emulation of complex emergent biological behaviors such as swarming, and opens a new route to the solution of complex optimization problems.","author":[{"family":"Heywood","given":"Zachary"},{"family":"Mallinson","given":"Joshua"},{"family":"Bones","given":"Philip"},{"family":"Brown","given":"Simon"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1088/2634-4386/ad7314","URL":"https://doi.org/10.1088/2634-4386/ad7314","source":"crossref"},{"id":"doi:10.1088/2634-4386/ad5d0f","type":"article-journal","title":"Kernel heterogeneity improves sparseness of natural images representations","abstract":"Abstract Both biological and artificial neural networks inherently balance their performance with their operational cost, which characterizes their computational abilities. Typically, an efficient neuromorphic neural network is one that learns representations that reduce the redundancies and dimensionality of its input. For instance, in the case of sparse coding (SC), sparse representations derived from natural images yield representations that are heterogeneous, both in their sampling of input features and in the variance of those features. Here, we focused on this notion, and sought correlations between natural images’ structure, particularly oriented features, and their corresponding sparse codes. We show that representations of input features scattered across multiple levels of variance substantially improve the sparseness and resilience of sparse codes, at the cost of reconstruction performance. This echoes the structure of the model’s input, allowing to account for the heterogeneously aleatoric structures of natural images. We demonstrate that learning kernel from natural images produces heterogeneity by balancing between approximate and dense representations, which improves all reconstruction metrics. Using a parametrized control of the kernels’ heterogeneity of a convolutional SC algorithm, we show that heterogeneity emphasizes sparseness, while homogeneity improves representation granularity. In a broader context, this encoding strategy can serve as inputs to deep convolutional neural networks. We prove that such variance-encoded sparse image datasets enhance computational efficiency, emphasizing the benefits of kernel heterogeneity to leverage naturalistic and variant input structures and possible applications to improve the throughput of neuromorphic hardware.","author":[{"family":"Ladret","given":"Hugo"},{"family":"Casanova","given":"Christian"},{"family":"Perrinet","given":"Laurent"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1088/2634-4386/ad5d0f","URL":"https://doi.org/10.1088/2634-4386/ad5d0f","source":"crossref"},{"id":"doi:10.1088/2634-4386/ad64fd","type":"article-journal","title":"Reducing the spike rate of deep spiking neural networks based on time-encoding","abstract":"Abstract A primary objective of Spiking Neural Networks is a very energy-efficient computation. To achieve this target, a small spike rate is of course very beneficial given the event-driven nature of such a computation. A network that processes information encoded in spike timing can, by its nature, have such a sparse event rate, but, as the network becomes deeper and larger, the spike rate tends to increase without any improvements in the final accuracy. If, on the other hand, a penalty on the excess of spikes is used during the training, the network may shift to a configuration where many neurons are silent, thus affecting the effectiveness of the training itself. In this paper, we present a learning strategy to keep the final spike rate under control by changing the loss function to penalize the spikes generated by neurons after the first ones. Moreover, we also propose a 2-phase training strategy to avoid silent neurons during the training, intended for benchmarks where such an issue can cause the switch off of the network.","author":[{"family":"Fontanini","given":"Riccardo"},{"family":"Pilotto","given":"Alessandro"},{"family":"Esseni","given":"David"},{"family":"Loghi","given":"Mirko"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1088/2634-4386/ad64fd","URL":"https://doi.org/10.1088/2634-4386/ad64fd","source":"crossref"},{"id":"doi:10.1145/3701701.3701712","type":"article-journal","title":"A Neuromorphic Radar Sensor for Low-Power IoT Systems","abstract":"As radar sensors become an integral component of Internet of Things (IoT) systems, the challenge of high power consumption poses a significant barrier, especially for battery-operated devices. This article introduces NeuroRadar, a groundbreaking solution that leverages a radar front-end capable of generating spike sequences, which can be efficiently processed by energy-saving Spiking Neural Networks (SNNs). We explore the innovative design and implementation of NeuroRadar, showcasing its effectiveness in applications like gesture recognition and human tracking. By achieving dramatically lower power consumption compared to traditional radar systems, NeuroRadar represents a new paradigm in energy-efficient IoT sensing.","author":[{"family":"Zheng","given":"Kai"},{"family":"Qian","given":"Kun"},{"family":"Woodford","given":"Timothy"},{"family":"Zhang","given":"Xinyu"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1145/3701701.3701712","URL":"https://doi.org/10.1145/3701701.3701712","source":"crossref"},{"id":"doi:10.3390/nano13192720","type":"article-journal","title":"Emerging Opportunities for 2D Materials in Neuromorphic Computing","abstract":"Recently, two-dimensional (2D) materials and their heterostructures have been recognized as the foundation for future brain-like neuromorphic computing devices. Two-dimensional materials possess unique characteristics such as near-atomic thickness, dangling-bond-free surfaces, and excellent mechanical properties. These features, which traditional electronic materials cannot achieve, hold great promise for high-performance neuromorphic computing devices with the advantages of high energy efficiency and integration density. This article provides a comprehensive overview of various 2D materials, including graphene, transition metal dichalcogenides (TMDs), hexagonal boron nitride (h-BN), and black phosphorus (BP), for neuromorphic computing applications. The potential of these materials in neuromorphic computing is discussed from the perspectives of material properties, growth methods, and device operation principles.","author":[{"family":"Feng","given":"Chenyin"},{"family":"Wu","given":"Wenwei"},{"family":"Liu","given":"Huidi"},{"family":"Wang","given":"Junke"},{"family":"Wan","given":"Houzhao"},{"family":"Ma","given":"Guokun"},{"family":"Wang","given":"Hao"}],"issued":{"date-parts":[[2023]]},"DOI":"10.3390/nano13192720","URL":"https://doi.org/10.3390/nano13192720","source":"crossref"},{"id":"doi:10.4018/979-8-3693-6303-4.ch015","type":"article-journal","title":"Pharmacy Science and Neurological Drug Discovery","abstract":"In this chapter, we set out on a journey to explore the complex relationship between pharmaceutical science, computational drug discovery, and the world of natural food products. We will deeply investigate the crucial role of food components in shaping the landscape of pharmaceutical research and development. By highlighting the importance of integrating food-based methods into drug discovery processes, our goal is to emphasize the transformative potential of utilizing the abundant resources found in nature. Taking a multidisciplinary approach, we aim to bridge the traditional gap between conventional pharmaceutical practices and the rapidly advancing field of nutraceuticals. In doing so, we are paving the way for a more unified and holistic approach to healthcare innovation.","author":[{"family":"Tanwar","given":"Neha"},{"family":"Kumar","given":"Sandeep"},{"family":"Verma","given":"Deepika"}],"issued":{"date-parts":[[2024]]},"DOI":"10.4018/979-8-3693-6303-4.ch015","URL":"https://doi.org/10.4018/979-8-3693-6303-4.ch015","source":"crossref"},{"id":"doi:10.34133/icomputing.0059","type":"article-journal","title":"Information Transfer in Neuronal Circuits: From Biological Neurons to Neuromorphic Electronics","abstract":"The advent of neuromorphic electronics is increasingly revolutionizing the concept of computation. In the last decade, several studies have shown how materials, architectures, and neuromorphic devices can be leveraged to achieve brain-like computation with limited power consumption and high energy efficiency. Neuromorphic systems have been mainly conceived to support spiking neural networks that embed bioinspired plasticity rules such as spike time-dependent plasticity to potentially support both unsupervised and supervised learning. Despite substantial progress in the field, the information transfer capabilities of biological circuits have not yet been achieved. More importantly, demonstrations of the actual performance of neuromorphic systems in this context have never been presented. In this paper, we report similarities between biological, simulated, and artificially reconstructed microcircuits in terms of information transfer from a computational perspective. Specifically, we extensively analyzed the mutual information transfer at the synapse between mossy fibers and granule cells by measuring the relationship between pre- and post-synaptic variability. We extended this analysis to memristor synapses that embed rate-based learning rules, thus providing quantitative validation for neuromorphic hardware and demonstrating the reliability of brain-inspired applications.","author":[{"family":"Gandolfi","given":"Daniela"},{"family":"Benatti","given":"Lorenzo"},{"family":"Zanotti","given":"Tommaso"},{"family":"Boiani","given":"Giulia"},{"family":"Bigiani","given":"Albertino"},{"family":"Puglisi","given":"Francesco"},{"family":"Mapelli","given":"Jonathan"}],"issued":{"date-parts":[[2024]]},"DOI":"10.34133/icomputing.0059","URL":"https://doi.org/10.34133/icomputing.0059","source":"crossref"},{"id":"doi:10.1007/978-3-031-51500-2_6","type":"article-journal","title":"Development of Crosspoint Memory Arrays for Neuromorphic Computing","abstract":"Abstract Memristor-based hardware accelerators play a crucial role in achieving energy-efficient big data processing and artificial intelligence, overcoming the limitations of traditional von Neumann architectures. Resistive-switching memories (RRAMs) combine a simple two-terminal structure with the possibility of tuning the device conductance. This Chapter revolves around the topic of emerging memristor-related technologies, starting from their fabrication, through the characterization of single devices up to the development of proof-of-concept experiments in the field of in-memory computing, hardware accelerators, and brain-inspired architecture. Non-volatile devices are optimized for large-size crossbars where the devices’ conductance encodes mathematical coefficients of matrices. By exploiting Kirchhoff’s and Ohm’s law the matrix–vector-multiplication between the conductance matrix and a voltage vector is computed in one step. Eigenvalues/eigenvectors are experimentally calculated according to the power-iteration algorithm, with a fast convergence within about 10 iterations to the correct solution and Principal Component Analysis of the Wine and Iris datasets, showing up to 98% accuracy comparable to a floating-point implementation. Volatile memories instead present a spontaneous change of device conductance with a unique similarity to biological neuron behavior. This characteristic is exploited to demonstrate a simple fully-memristive architecture of five volatile RRAMs able to learn, store, and distinguish up to 10 different items with a memory capability of a few seconds. The architecture is thus tested in terms of robustness under many experimental conditions and it is compared with the real brain, disclosing interesting mechanisms which resemble the biological brain.","author":[{"family":"Ricci","given":"Saverio"},{"family":"Mannocci","given":"Piergiulio"},{"family":"Farronato","given":"Matteo"},{"family":"Milozzi","given":"Alessandro"},{"family":"Ielmini","given":"Daniele"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1007/978-3-031-51500-2_6","URL":"https://doi.org/10.1007/978-3-031-51500-2_6","source":"crossref"},{"id":"doi:10.1149/ma2024-02483315mtgabs","type":"article-journal","title":"Neuromorphic Computing Based on Few-Molecule Vibration Dynamics Achieved By Surface-Enhanced Raman Scattering and Ion-Gating Stimulation","abstract":"1. Objective Artificial intelligence (AI), which has achieved high-level capabilities in cognition, learning, and inference through technological innovations such as deep learning, has greatly benefited human society, however, problems such as high energy consumption and increased communication traffic have also emerged. As a solution to these problems, physical reservoir computing (PRC), a kind of neuromorphic computing that uses the nonlinear behavior of physical systems to efficiently process information, has been attracting attention.[1-5] Although various methods of PRC have been reported (e.g., memristors, electrochemical cells, optical circuits, soft bodies, and spintronics devices), the molecular computing approach is promising to realize the ultimate compact and integrated PRC. However, the number of molecules in a conventional molecular reservoir system is huge, and the information processing capability of a single molecule to several molecules has not been elucidated.[2,3] In this study, we developed a few-molecule reservoir computing (FM-RC) that utilizes the molecular vibration dynamics of single to few molecules of para-mercaptobenzoic acid (pMBA). The information processing capability of few molecules was evaluated by performing information processing tasks such as blood glucose level prediction on FM-RC. [4] 2. Experiments Figure 1a shows a schematic diagram of the measurement system of FM-RC. In order to accurately track the nonlinear dynamics of a few molecules, surface-enhanced Raman scattering (SERS) measurements using WOx nanorods/silver nanoparticles (WOx@Ag-NPs) were performed, and molecular vibration reflecting the structural change of pMBA based on the local proton adsorption change due to ion-gating stimulation was observed in the molecular dynamics (Fig. 1b) reflecting the conformational changes of pMBA based on the local proton adsorption changes induced by ion-gating stimuli. The obtained SERS spectra were regarded as independent artificial neurons (nodes) at specific sampling wavenumbers, and these node states were used to perform various information processing tasks (Fig. 1c). 3. Results and Discussion Although FM-RC consists of only a few molecules, it achieved a high accuracy of 95.1% to 97.7% in the nonlinear waveform transformation task and 94.3% in the analysis task of second-order nonlinear equations, and achieved the highest computational performance among physical reservoirs in the task of predicting blood glucose levels for type I pediatric diabetes patients (Fig. 1d).[4-6] These high computational performances are attributed to the independent nonlinearity of the nodes distributed in the wavenumber direction, i.e., the high dimensionality achieved by the complex and diverse response characteristics of each vibrational mode to the proton adsorption amount. In this presentation, the origin of the high computational performance of FM-RC will be discussed in terms of nonlinearity and high dimensionality, which are the main properties that determine the performance of PRCs. References [1] D. Nishioka et al., Sci. Adv. 8 , eade1156 (2022). [2] A. Goudarzi et al, DNA Reservoir Computing (Springer, 2013). [3] Y. Usami Adv. Mater. 33 , 2102688 (2021). [4] D. Nishioka et al., Sci. Adv. 10 , eadk6438 (2024). [5] T. Shingu et al., Carbon 214 , 118344 (2023). [6] DIRECNET Study Group, Diabetes Technol. Therapeut. 5 , 781-789 (2003). Acknowledgments This work was supported by Japan Society for the Promotion of Science (JSPS) KAKENHI Grant Number JP22H04625 (Grant-in-Aid for Scientific Research on Innovative Areas “Interface Ionics”), Japan Society for the Promotion of Science (JSPS) KAKENHI Grant Number JP22KJ2799 (Grant-in-Aid for JSPS Fellows), JST PRESTO (grant number, JPMJPR23H4), and the Iketani Science and Technology Foundation. Figure 1","author":[{"family":"Nishioka","given":"Daiki"},{"family":"Shingaya","given":"Yoshitaka"},{"family":"Terabe","given":"Kazuya"},{"family":"Tsuchiya","given":"Takashi"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1149/ma2024-02483315mtgabs","URL":"https://doi.org/10.1149/ma2024-02483315mtgabs","source":"crossref"},{"id":"doi:10.1088/2634-4386/acb8d7","type":"article-journal","title":"Hardware optimization for photonic time-delay reservoir computer dynamics","abstract":"Abstract Reservoir computing (RC) is one kind of neuromorphic computing mainly applied to process sequential data such as time-dependent signals. In this paper, the bifurcation diagram of a photonic time-delay RC system is thoroughly studied, and a method of bifurcation dynamics guided hardware hyperparameter optimization is presented. The time-evolution equation expressed by the photonic hardware parameters is established while the intrinsic dynamics of the photonic RC system is quantitively studied. Bifurcation dynamics based hyperparameter optimization offers a simple yet effective approach in hardware setting optimization that aims to reduce the complexity and time in hardware adjustment. Three benchmark tasks, nonlinear channel equalization (NCE), nonlinear auto regressive moving average with 10th order time lag (NARMA10) and Santa Fe laser time-series prediction tasks are implemented on the photonic delay-line RC using bifurcation dynamics guided hardware optimization. The experimental results of these benchmark tasks achieved overall good agreement with the simulated bifurcation dynamics modeling results.","author":[{"family":"Zhang","given":"Meng"},{"family":"Liang","given":"Zhizhuo"},{"family":"Huang","given":"ZR"}],"issued":{"date-parts":[[2023]]},"DOI":"10.1088/2634-4386/acb8d7","URL":"https://doi.org/10.1088/2634-4386/acb8d7","source":"crossref"},{"id":"doi:10.1515/ntrev-2023-0181","type":"article-journal","title":"Overview of amorphous carbon memristor device, modeling, and applications for neuromorphic computing","abstract":"Abstract Carbon-based materials strongly pertain to citizens’ daily life due to their versatile derivatives such as diamond, graphite, fullerenes, carbon nanotube, single-layer graphene, and amorphous carbon (a-C). Compared to other families, a-C exhibits reconfigurable electrical properties by triggering its sp 2 –sp 3 transition and vice versa , which can be readily fabricated by conventional film deposition technologies. For above reasons, a-C has been adopted as a promising memristive material and has given birth to several physical and theoretical prototypes. To further help researchers comprehend the physics behind a-C-based memristors and push forward their development, here we first reviewed the classification of a-C-based materials associated with their respective electrical and thermal properties. Subsequently, several a-C -based memristors with different architectures were presented, followed by their respective memristive principles. We also elucidated the state-of-the-art modeling strategies of a-C memristors, and their practical applications on neuromorphic fields were also described. The possible scenarios to further mitigate the physical performances of a-C memristors were eventually discussed, and their future prospect to rival with other memristors was also envisioned.","author":[{"family":"Wu","given":"Jie"},{"family":"Yang","given":"Xuqi"},{"family":"Chen","given":"Jing"},{"family":"Li","given":"Shiyu"},{"family":"Zhou","given":"Tianchen"},{"family":"Cai","given":"Zhikuang"},{"family":"Lian","given":"Xiaojuan"},{"family":"Wang","given":"Lei"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1515/ntrev-2023-0181","URL":"https://doi.org/10.1515/ntrev-2023-0181","source":"crossref"},{"id":"doi:10.1088/2634-4386/ad5c97","type":"article-journal","title":"Efficient sparse spiking auto-encoder for reconstruction, denoising and classification","abstract":"Abstract Auto-encoders are capable of performing input reconstruction, denoising, and classification through an encoder-decoder structure. Spiking Auto-Encoders (SAEs) can utilize asynchronous sparse spikes to improve power efficiency and processing latency on neuromorphic hardware. In our work, we propose an efficient SAE trained using only Spike-Timing-Dependant Plasticity (STDP) learning. Our auto-encoder uses the Time-To-First-Spike (TTFS) encoding scheme and needs to update all synaptic weights only once per input, promoting both training and inference efficiency due to the extreme sparsity. We showcase robust reconstruction performance on the Modified National Institute of Standards and Technology (MNIST) and Fashion-MNIST datasets with significantly fewer spikes compared to state-of-the-art SAEs by 1–3 orders of magnitude. Moreover, we achieve robust noise reduction results on the MNIST dataset. When the same noisy inputs are used for classification, accuracy degradation is reduced by 30%–80% compared to prior works. It also exhibits classification accuracies comparable to previous STDP-based classifiers, while remaining competitive with other backpropagation-based spiking classifiers that require global learning through gradients and significantly more spikes for encoding and classification of MNIST/Fashion-MNIST inputs. The presented results demonstrate a promising pathway towards building efficient sparse spiking auto-encoders with local learning, making them highly suited for hardware integration.","author":[{"family":"Walters","given":"Ben"},{"family":"Kalatehbali","given":"Hamid"},{"family":"Cai","given":"Zhengyu"},{"family":"Genov","given":"Roman"},{"family":"Amirsoleimani","given":"Amirali"},{"family":"Eshraghian","given":"Jason"},{"family":"Azghadi","given":"Mostafa"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1088/2634-4386/ad5c97","URL":"https://doi.org/10.1088/2634-4386/ad5c97","source":"crossref"},{"id":"doi:10.3390/cryst14010069","type":"article-journal","title":"Transistor-Based Synaptic Devices for Neuromorphic Computing","abstract":"Currently, neuromorphic computing is regarded as the most efficient way to solve the von Neumann bottleneck. Transistor-based devices have been considered suitable for emulating synaptic functions in neuromorphic computing due to their synergistic control capabilities on synaptic weight changes. Various low-dimensional inorganic materials such as silicon nanomembranes, carbon nanotubes, nanoscale metal oxides, and two-dimensional materials are employed to fabricate transistor-based synaptic devices. Although these transistor-based synaptic devices have progressed in terms of mimicking synaptic functions, their application in neuromorphic computing is still in its early stage. In this review, transistor-based synaptic devices are analyzed by categorizing them into different working mechanisms, and the device fabrication processes and synaptic properties are discussed. Future efforts that could be beneficial to the development of transistor-based synaptic devices in neuromorphic computing are proposed.","author":[{"family":"Huang","given":"Wen"},{"family":"Zhang","given":"Huixing"},{"family":"Lin","given":"Zhengjian"},{"family":"Hang","given":"Pengjie"},{"family":"Li","given":"Xing’ao"}],"issued":{"date-parts":[[2024]]},"DOI":"10.3390/cryst14010069","URL":"https://doi.org/10.3390/cryst14010069","source":"crossref"},{"id":"doi:10.1088/2515-7639/ad5251","type":"article-journal","title":"In-sensor neuromorphic computing using perovskites and transition metal dichalcogenides","abstract":"Abstract With the advancements in Web of Things, Artificial Intelligence, and other emerging technologies, there is an increasing demand for artificial visual systems to perceive and learn about external environments. However, traditional sensing and computing systems are limited by the physical separation of sense, processing, and memory units that results in the challenges such as high energy consumption, large additional hardware costs, and long latency time. Integrating neuromorphic computing functions into the sensing unit is an effective way to overcome these challenges. Therefore, it is extremely important to design neuromorphic devices with sensing ability and the properties of low power consumption and high switching speed for exploring in-sensor computing devices and systems. In this review, we provide an elementary introduction to the structures and properties of two common optoelectronic materials, perovskites and transition metal dichalcogenides (TMDs). Subsequently, we discuss the fundamental concepts of neuromorphic devices, including device structures and working mechanisms. Furthermore, we summarize and extensively discuss the applications of perovskites and TMDs in in-sensor computing. Finally, we propose potential strategies to address challenges and offer a brief outlook on the application of optoelectronic materials in term of in-sensor computing.","author":[{"family":"Li","given":"Shen"},{"family":"Li","given":"Ji"},{"family":"Zhou","given":"Kui"},{"family":"Yan","given":"Yan"},{"family":"Ding","given":"Guanglong"},{"family":"Han","given":"Su"},{"family":"Zhou","given":"Ye"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1088/2515-7639/ad5251","URL":"https://doi.org/10.1088/2515-7639/ad5251","source":"crossref"},{"id":"doi:10.4018/979-8-3693-6303-4.ch006","type":"article-journal","title":"A Review of GAN-Synthesized Brain MR Image Applications","abstract":"Recent advancements in brain imaging technology have led to a rise in the use of magnetic resonance imaging (MRI) for clinical diagnosis. Deep learning (DL) techniques have emerged as a valuable tool for automatically detecting abnormalities in brain images without manual intervention. Meanwhile, generative adversarial networks (GANs) have shown promise in generating synthetic brain images for a variety of applications, such as image translation, registration, super-resolution, denoising, motion correction, segmentation, reconstruction, and contrast enhancement. This chapter conducts a comprehensive review of the literature on the use of GAN-synthesized images for diagnosing brain diseases, drawing on data from studies in the Web of Science and Scopus databases from the past decade. The review examines the various loss functions and software tools used in processing brain MRI images, as well as provides a comparative analysis of evaluation metrics for GAN-synthesized images to assist researchers in selecting the most appropriate metric for their specific needs.","author":[{"family":"Tiwari","given":"Ankita"},{"family":"Tavse","given":"Sampada"},{"family":"Bachute","given":"Mrinal"},{"family":"Bhola","given":"Abhishek"}],"issued":{"date-parts":[[2024]]},"DOI":"10.4018/979-8-3693-6303-4.ch006","URL":"https://doi.org/10.4018/979-8-3693-6303-4.ch006","source":"crossref"},{"id":"doi:10.4018/979-8-3693-6303-4.ch013","type":"article-journal","title":"Neurocomputing Advancements to Unlock Image Intelligence for Industrial Computer Vision","abstract":"Distinct technologies have been formulated at different times to improve technology, and unique technologies are continuously emerging. Neurocomputing has significantly expanded technologies, revealed moderately acceptable results, and provided ultimate collaboration. It is a type of computational system that executes its functions by mimicking the workings methodology of the human brain. Neurocomputing utilizes artificial neural networks managed by interconnected nodes, which collectively perform miscellaneous tasks. Each node processes small portions of information and communicates it to the next node, which assembles an extensive network and cracks many complex problems. This functional capability of neurocomputing can provide fruitful solutions for any complex task related to computer vision, enabling the computer system to extract meaningful information from images effortlessly. The functional benefit of neurocomputing, whereas manifold applications and computer vision components are determined to yield unthinkable results using images in the industrial sector.","author":[{"family":"Saha","given":"Soumitra"},{"family":"Lilhore","given":"Umesh"},{"family":"Simaiya","given":"Sarita"}],"issued":{"date-parts":[[2024]]},"DOI":"10.4018/979-8-3693-6303-4.ch013","URL":"https://doi.org/10.4018/979-8-3693-6303-4.ch013","source":"crossref"},{"id":"doi:10.34133/adi.0044","type":"article-journal","title":"Recent Progress in Neuromorphic Computing from Memristive Devices to Neuromorphic Chips","abstract":"Neuromorphic computing, drawing inspiration from the brain, stands out for its high energy efficiency in executing complex tasks. Memristive device-based neuromorphic computing has demonstrated ultrahigh efficiency. While there are numerous review papers in this field, the majority concentrate on the device level, bypassing the connections among the performance metrics of memristive devices and those of neuromorphic chips. In this review, we investigate the recent progress in neuromorphic computing from the fundamental memristive devices to the intricate neuromorphic chips, highlighting their links and challenges.","author":[{"family":"Xiao","given":"Yike"},{"family":"Gao","given":"Cheng"},{"family":"Jin","given":"Juncheng"},{"family":"Sun","given":"Weiling"},{"family":"Wang","given":"Bowen"},{"family":"Bao","given":"Yukun"},{"family":"Liu","given":"Chen"},{"family":"Huang","given":"Wei"},{"family":"Zeng","given":"Hui"},{"family":"Yu","given":"Yefeng"}],"issued":{"date-parts":[[2024]]},"DOI":"10.34133/adi.0044","URL":"https://doi.org/10.34133/adi.0044","source":"crossref"},{"id":"doi:10.6082/sx4gn-66w06","type":"article-journal","title":"Hydrogen-Induced Topotactic Phase Transformations of Cobaltite Thin Films","abstract":"Manipulating physical properties through ion migration in complex oxide thin films is an emerging research direction to achieve tunable materials for advanced applications. While the reduction of complex oxides has been widely reported, few reports exist on the modulation of physical properties through a direct hydrogenation process. Here, we report an unusual mechanism for hydrogen-induced topotactic phase transitions in perovskite La0.7Sr0.3CoO3 thin films. Hydrogenation is performed upon annealing in a pure hydrogen gas environment, offering a direct understanding of the role that hydrogen plays at the atomic scale in these transitions. Topotactic phase transformations from the perovskite (P) to hydrogenated-brownmillerite (H-BM) phase can be induced at temperatures as low as 220 °C, while at higher hydrogenation temperatures (320–400 °C), the progression toward more reduced phases is hindered. Density functional theory calculations suggest that hydroxyl bonds are formed with the introduction of hydrogen ions, which lower the formation energy of oxygen vacancies of the neighboring oxygen, enabling the transition from the P to H-BM phase at low temperatures. Furthermore, the impact on the magnetic and electronic properties of the hydrogenation temperature is investigated. Our research provides a potential pathway for utilizing hydrogen as a basis for low-temperature modulation of complex oxide thin films, with potential applications in neuromorphic computing.","author":[{"family":"Feng","given":"Mingzhen"},{"family":"Li","given":"Junjie"},{"family":"Zhang","given":"Shenli"},{"family":"Pofelski","given":"Alexandre"},{"family":"El Hage","given":"Ralph"},{"family":"Klewe","given":"Christoph"},{"family":"N'diaye","given":"Alpha"},{"family":"Shafer","given":"Padraic"},{"family":"Zhu","given":"Yimei"},{"family":"Galli","given":"Giulia"},{"family":"Schuller","given":"Ivan"},{"family":"Takamura","given":"Yayoi"}],"issued":{"date-parts":[[2024]]},"DOI":"10.6082/sx4gn-66w06","URL":"https://doi.org/10.6082/sx4gn-66w06","source":"datacite"},{"id":"doi:10.6082/fn5n3-4vg84","type":"article-journal","title":"Hydrogen-Induced Topotactic Phase Transformations of Cobaltite Thin Films","abstract":"Manipulating physical properties through ion migration in complex oxide thin films is an emerging research direction to achieve tunable materials for advanced applications. While the reduction of complex oxides has been widely reported, few reports exist on the modulation of physical properties through a direct hydrogenation process. Here, we report an unusual mechanism for hydrogen-induced topotactic phase transitions in perovskite La0.7Sr0.3CoO3 thin films. Hydrogenation is performed upon annealing in a pure hydrogen gas environment, offering a direct understanding of the role that hydrogen plays at the atomic scale in these transitions. Topotactic phase transformations from the perovskite (P) to hydrogenated-brownmillerite (H-BM) phase can be induced at temperatures as low as 220 °C, while at higher hydrogenation temperatures (320–400 °C), the progression toward more reduced phases is hindered. Density functional theory calculations suggest that hydroxyl bonds are formed with the introduction of hydrogen ions, which lower the formation energy of oxygen vacancies of the neighboring oxygen, enabling the transition from the P to H-BM phase at low temperatures. Furthermore, the impact on the magnetic and electronic properties of the hydrogenation temperature is investigated. Our research provides a potential pathway for utilizing hydrogen as a basis for low-temperature modulation of complex oxide thin films, with potential applications in neuromorphic computing.","author":[{"family":"Feng","given":"Mingzhen"},{"family":"Li","given":"Junjie"},{"family":"Zhang","given":"Shenli"},{"family":"Pofelski","given":"Alexandre"},{"family":"El Hage","given":"Ralph"},{"family":"Klewe","given":"Christoph"},{"family":"N'diaye","given":"Alpha"},{"family":"Shafer","given":"Padraic"},{"family":"Zhu","given":"Yimei"},{"family":"Galli","given":"Giulia"},{"family":"Schuller","given":"Ivan"},{"family":"Takamura","given":"Yayoi"}],"issued":{"date-parts":[[2024]]},"DOI":"10.6082/fn5n3-4vg84","URL":"https://doi.org/10.6082/fn5n3-4vg84","source":"datacite"},{"id":"doi:10.3390/nano14181501","type":"article-journal","title":"Enhancing Long-Term Memory in Carbon-Nanotube-Based Optoelectronic Synaptic Devices for Neuromorphic Computing","abstract":"This study investigates the impact of spin-coating speed on the performance of carbon nanotube (CNT)-based optoelectronic synaptic devices, focusing on their long-term memory properties. CNT films fabricated at lower spin speeds exhibited a greater thickness and density compared to those at higher speeds. These denser films showed enhanced persistent photoconductivity, resulting in higher excitatory postsynaptic currents (EPSCs) and the prolonged retention of memory states after UV stimulation. Devices coated at a lower spin-coating speed of 2000 RPM maintained EPSCs above 70% for 3600 s, outperforming their higher-speed counterparts in long-term memory retention. Additionally, the study demonstrated that the learning efficiency improved with repeated UV stimulation, with fewer pulses needed to achieve the maximum EPSC in successive learning cycles. These findings highlight that optimizing spin-coating speeds can significantly enhance the performance of CNT-based synaptic devices, making them suitable for applications in neuromorphic computing and artificial neural networks requiring robust memory retention and efficient learning.","author":[{"family":"Lee","given":"Seung"},{"family":"Lee","given":"Hye"},{"family":"Jeon","given":"Dabin"},{"family":"Kim","given":"Hee"},{"family":"Lee","given":"Sung"}],"issued":{"date-parts":[[2024]]},"DOI":"10.3390/nano14181501","URL":"https://doi.org/10.3390/nano14181501","source":"crossref"},{"id":"doi:10.1063/5.0219287","type":"article-journal","title":"Single-crystal ferroelectric LiNbO3 thin film-based synaptic devices enabled with tunable domain wall current for neuromorphic computing","abstract":"Neuromorphic devices can emulate the human brain to process information, which receives lots of attention in the field of artificial intelligence. Synaptic devices based on ferroelectric thin films feature low-power consumption, multifunctionality, and scalability. Among them, ferroelectric charged domain wall (CDW) devices have attracted intensive interest for the implementation of memristive devices due to their ultrahigh integration ability inherited from the nanoscale domain wall thickness. In particular, the preparation of wafer-scale single-crystalline ferroelectric thin films via ion-sliced heterogeneous wafer bonding lays a good foundation for large-scale integration of ferroelectric devices with functional circuits. However, the biomimic synaptic characteristics and the systematic demonstration of synaptic devices are largely unexplored for this material system. Here, we demonstrate a model synaptic device based on a single-crystal ferroelectric LiNbO3 thin film, which provides the desired characteristics for neuromorphic computing. The conductance modulation demonstrates good linearity for efficient neuromorphic computing applications. Simulations using the Modified National Institute of Standards and Technology handwritten recognition dataset prove that LiNbO3-based synaptic devices can operate with an online learning accuracy of 95.1%. The injection and annihilation of the CDW are proposed as the basis of the conductivity modulation by combining with the piezoresponse force microscopy and conductive atomic force microscopy mapping measurements. With the mature fabrication process of the ultrathin high-quality ferroelectric thin films, LiNbO3-based synaptic devices have an extensive application prospect for future neuromorphic computing systems.","author":[{"family":"Zhu","given":"Jiefei"},{"family":"Zhou","given":"Changjian"},{"family":"Liu","given":"Qi"},{"family":"Zhang","given":"Min"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1063/5.0219287","URL":"https://doi.org/10.1063/5.0219287","source":"crossref"},{"id":"doi:10.1007/s10462-024-10948-3","type":"article-journal","title":"Neuromorphic computing for modeling neurological and psychiatric disorders: implications for drug development","abstract":"Abstract The emergence of neuromorphic computing, inspired by the structure and function of the human brain, presents a transformative framework for modelling neurological disorders in drug development. This article investigates the implications of applying neuromorphic computing to simulate and comprehend complex neural systems affected by conditions like Alzheimer’s, Parkinson’s, and epilepsy, drawing from extensive literature. It explores the intersection of neuromorphic computing with neurology and pharmaceutical development, emphasizing the significance of understanding neural processes and integrating deep learning techniques. Technical considerations, such as integrating neural circuits into CMOS technology and employing memristive devices for synaptic emulation, are discussed. The review evaluates how neuromorphic computing optimizes drug discovery and improves clinical trials by precisely simulating biological systems. It also examines the role of neuromorphic models in comprehending and simulating neurological disorders, facilitating targeted treatment development. Recent progress in neuromorphic drug discovery is highlighted, indicating the potential for transformative therapeutic interventions. As technology advances, the synergy between neuromorphic computing and neuroscience holds promise for revolutionizing the study of the human brain’s complexities and addressing neurological challenges.","author":[{"family":"Raikar","given":"Amisha"},{"family":"Andrew","given":"J"},{"family":"Dessai","given":"Pranjali"},{"family":"Prabhu","given":"Sweta"},{"family":"Jathar","given":"Shounak"},{"family":"Prabhu","given":"Aishwarya"},{"family":"Naik","given":"Mayuri"},{"family":"Raikar","given":"Gokuldas"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1007/s10462-024-10948-3","URL":"https://doi.org/10.1007/s10462-024-10948-3","source":"crossref"},{"id":"doi:10.1002/aelm.202400061","type":"article-journal","title":"Binarized Neural Network Comprising Quasi‐Nonvolatile Memory Devices for Neuromorphic Computing","abstract":"Abstract This study presents a binarized neural network (BNN) comprising quasi‐nonvolatile memory (QNVM) devices that operate in a positive feedback loop mechanism and exhibit an extremely low subthreshold swing (≤ 5 mV dec −1 ) and a high on/off ratio (≥ 10 7 ). A pair of QNVM devices are used for a single synaptic cell in a cell array, in which its memory state represents the synaptic weight, and the voltages applied to the pair act as input in a complementary fashion. The array of synaptic cells performs matrix multiply‐accumulate (MAC) operations between the weight matrix and input vector using XNOR and current summation. All the results of the MAC operations and vector‐matrix multiplications are equivalent. Moreover, the BNN features a high accuracy of 93.32% in the MNIST image recognition simulation owing to high device uniformity (1.35%), which demonstrates the feasibility of compact and high‐performance neuromorphic computing.","author":[{"family":"Shin","given":"Yunwoo"},{"family":"Jeon","given":"Juhee"},{"family":"Cho","given":"Kyoungah"},{"family":"Kim","given":"Sangsig"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1002/aelm.202400061","URL":"https://doi.org/10.1002/aelm.202400061","source":"crossref"},{"id":"doi:10.1063/5.0231655","type":"article-journal","title":"Nitrogen-doped carbon quantum dot-decorated In2O3 synaptic transistors for neuromorphic computing","abstract":"Nitrogen-doped carbon quantum dots (N-CQDs) are promising materials for electronic devices due to their variable bandgap and structural stability. Here, we integrate N-CQDs into In2O3 synaptic transistors with electrolyte gating, resulting in a hybrid structure. The surface functional groups and defects of N-CQDs empower the charge trapping mechanism, permitting controlled conduction and charge regulation, which are crucial for emulating linear and symmetric artificial synaptic devices. Devices incorporating N-CQDs demonstrate enhanced stability and memory characteristics, low energy consumption, consistent retention, and a significant hysteresis window across multiple voltage cycles. Finally, the study emulates biological synapses and cognitive functions, achieving an energy consumption of 10 fJ per synaptic event and a pattern recognition accuracy of 91.2% on the MNIST dataset in hardware neural networks. This work demonstrates the potential of well-manipulating charge trapping in N-CQDs to develop high-performance, nonvolatile synaptic devices.","author":[{"family":"Zahid","given":"Muhammad"},{"family":"Sadiq","given":"Muhammad"},{"family":"Jin","given":"Chenxing"},{"family":"Wang","given":"Jingwen"},{"family":"Shi","given":"Xiaofang"},{"family":"Liu","given":"Wanrong"},{"family":"Aslam","given":"Fawad"},{"family":"Xu","given":"Yunchao"},{"family":"Tahir","given":"Muhammad"},{"family":"Yang","given":"Junliang"},{"family":"Sun","given":"Jia"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1063/5.0231655","URL":"https://doi.org/10.1063/5.0231655","source":"crossref"},{"id":"doi:10.1063/5.0180088","type":"article-journal","title":"Multi-state nonvolatile capacitances in HfO2-based ferroelectric capacitor for neuromorphic computing","abstract":"In the last decade, HfO2-based ferroelectric capacitors (FeCaps) have undergone significant advancements, particularly within the realm of nonvolatile ferroelectric random access memories (FeRAMs). Nonetheless, the READ operation in FeRAMs is inherently destructive, rendering it unsuitable for neuromorphic computing. In this study, we have engineered tunable nonvolatile capacitances within FeCaps, featuring nondestructive readout functionality. Robust capacitance states can be read at a zero d.c. bias (Vbias) with different a.c. signals, not only preventing the alteration of their stored state but also benefiting to the low power consumption. Moreover, the capacitance memory window (CMW) at Vbias of zero can be effectively modulated through electrode engineering, leading to a larger CMW when there is a greater disparity in work functions between the electrodes. Furthermore, we provide a comprehensive investigation into synaptic behavior of TiN/Hf0.5Zr0.5O2/Pt FeCaps, demonstrating their excellent cycle-to-cycle uniformity, retention, and endurance characteristics, which confirm their high reliability in maintaining nonvolatile capacitance states. These findings underscore the significant potential of FeCaps in advancing low-power neuromorphic computing.","author":[{"family":"Wu","given":"Shuyu"},{"family":"Zhang","given":"Xumeng"},{"family":"Cao","given":"Rongrong"},{"family":"Zhou","given":"Keji"},{"family":"Lu","given":"Jikai"},{"family":"Li","given":"Chao"},{"family":"Yang","given":"Yang"},{"family":"Shang","given":"Dashan"},{"family":"Wei","given":"Yingfen"},{"family":"Jiang","given":"Hao"},{"family":"Liu","given":"Qi"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1063/5.0180088","URL":"https://doi.org/10.1063/5.0180088","source":"crossref"},{"id":"doi:10.1063/5.0202008","type":"article-journal","title":"Enhanced ferroelectric photovoltaic performance of Bi2FeCrO6 thin films for neuromorphic computing applications","abstract":"Nowadays, ferroelectric photovoltaic synapses have attracted great attention due to its polarization controllable and self-powered features. However, the large bandgaps of ferroelectric oxide materials limit its application. This study focuses on the enhancement of ferroelectric photovoltaic properties and the synaptic application of Bi2FeCrO6 (BFCO) device. It is found that the bandgap of BFCO can be modulated by Cr alloying, which causes its photovoltaic effect in the visible region to exceed that of BiFeO3 (BFO) significantly. The short-circuit current density (JSC) of BFCO device in the visible region increases by about 100 times than that of BFO. Furthermore, the polarization modulation and multi-states response are demonstrated by an external electric field. For BFCO ferroelectric photovoltaic synapse, long-term potentiation/depression (LTP/LTD) measurements show an excellent synaptic plasticity of the polarization modulation. The simulated image recognition rate using the MNIST dataset reaches a high accuracy of 96.06%. This work has expanded the potential application of ferroelectric photovoltaic synapse in the visible region.","author":[{"family":"Kan","given":"Yucheng"},{"family":"Liu","given":"Jianquan"},{"family":"Chen","given":"Rui"},{"family":"Liu","given":"Yuan"},{"family":"Wang","given":"Hongru"},{"family":"Long","given":"Mingyue"},{"family":"Tian","given":"Bobo"},{"family":"Chu","given":"Junhao"},{"family":"Chen","given":"Ye"},{"family":"Sun","given":"Lin"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1063/5.0202008","URL":"https://doi.org/10.1063/5.0202008","source":"crossref"},{"id":"doi:10.1088/2634-4386/acf684","type":"article-journal","title":"Artificial nanophotonic neuron with internal memory for biologically inspired and reservoir network computing","abstract":"Abstract Neurons with internal memory have been proposed for biological and bio-inspired neural networks, adding important functionality. We introduce an internal time-limited charge-based memory into a III–V nanowire (NW) based optoelectronic neural node circuit designed for handling optical signals in a neural network. The new circuit can receive inhibiting and exciting light signals, store them, perform a non-linear evaluation, and emit a light signal. Using experimental values from the performance of individual III–V NWs we create a realistic computational model of the complete artificial neural node circuit. We then create a flexible neural network simulation that uses these circuits as neuronal nodes and light for communication between the nodes. This model can simulate combinations of nodes with different hardware derived memory properties and variable interconnects. Using the full model, we simulate the hardware implementation for two types of neural networks. First, we show that intentional variations in the memory decay time of the nodes can significantly improve the performance of a reservoir network. Second, we simulate the implementation in an anatomically constrained functioning model of the central complex network of the insect brain and find that it resolves an important functionality of the network even with significant variations in the node performance. Our work demonstrates the advantages of an internal memory in a concrete, nanophotonic neural node. The use of variable memory time constants in neural nodes is a general hardware derived feature and could be used in a broad range of implementations.","author":[{"family":"Winge","given":"David"},{"family":"Borgström","given":"Magnus"},{"family":"Lind","given":"Erik"},{"family":"Mikkelsen","given":"Anders"}],"issued":{"date-parts":[[2023]]},"DOI":"10.1088/2634-4386/acf684","URL":"https://doi.org/10.1088/2634-4386/acf684","source":"crossref"},{"id":"doi:10.4018/978-1-6684-6596-7.ch008","type":"article-journal","title":"ML-Based Finger-Vein Biometric Authentication and Hardware Implementation Strategies","abstract":"The industrial world is facing swiftly changing challenges, including technical fluctuations, swings in global markets, and climate change. Digitalization and automation are the game changers to meet these challenges. Machine learning has made an impact on the area attributed to microchips, and it is initially used in automation. These practices will ultimately succeed the current VLSI design. A biometric authentication system is a form of biometric verification system that uses finger vein detection for biometric check framework. Consumer electronics development necessitates high security, high accuracy, and fast authentication speed. Since it presents the elements inside the human body, finger vein validation is a prominent technique in terms of security. Furthermore, it has a fair advantage over other personal authentication methods because to its contactless aspect. The main goal of the chapter is to provide a solution using machine learning for finger vein authentication and implementation using VLSI design.","author":[{"family":"Keshri","given":"Nayan"},{"family":"Sarkar","given":"Swapnadeep"},{"family":"Singh","given":"Yash"},{"family":"Gokulanathan","given":"Sumathi"},{"family":"Elango","given":"Konguvel"},{"family":"Ponnusamy","given":"Sivakumar"}],"issued":{"date-parts":[[2023]]},"DOI":"10.4018/978-1-6684-6596-7.ch008","URL":"https://doi.org/10.4018/978-1-6684-6596-7.ch008","source":"crossref"},{"id":"doi:10.31237/osf.io/mxs45","type":"article-journal","title":"Spike Timing Mechanisms in Neuromorphic Vision Sensors using Memristor-based non-volatile Memory devices","abstract":"Spike-timing mechanisms in neuromorphic vision sensors represent a cutting-edge approach to mimicking biological vision systems' efficiency and adaptability. These systems utilize memristor-based non-volatile memory devices to achieve high precision and low power consumption, essential for real-time image processing and recognition tasks. This paper explores the principles and applications of spike-timing-dependent plasticity (STDP) in neuromorphic vision sensors, focusing on the integration of memristor technology. This study introduces an innovative visual-tactile perception system that integrates a scalable, biomimetic tactile sensor called NeuTouch and uses a Visual-Tactile Spiking Neural Network (VT-SNN) for rapid perception. The system demonstrates high accuracy in robotic tasks such as container classification and rotational slip detection, outperforming traditional deep learning methods. The research also contributes to the field by making visual-tactile datasets publicly available to foster further advancements. This work highlights the potential for creating intelligent, energy-efficient robotic systems. We review the latest advancements in memristor-based memory devices, their role in neuromorphic computing, and how they contribute to the development of advanced vision sensors. The potential of these technologies in revolutionizing artificial vision and their implications for future research and development are also discussed.","author":[{"family":"Su","given":"Yiping"},{"family":"Hwang","given":"Dajeong"},{"family":"Wang","given":"Bing"}],"issued":{"date-parts":[[2024]]},"DOI":"10.31237/osf.io/mxs45","URL":"https://doi.org/10.31237/osf.io/mxs45","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-5737326/v1","type":"article-journal","title":"A neuromorphic multi-scale approach for heart rate and state detection","abstract":"Abstract With the advent of novel sensor and machine learning technologies, it is becoming possible to develop wearable systems that perform continuous recording and processing of biosignals for health or body state assessment. For example, modern smartwatches can already track physiological functions, including heart rate and its anomalies, with high precision. However, stringent constraints on size and energy consumption pose significant challenges for always-on operation to detect trends across multiple time scales for extended periods of time. To address these challenges, we propose an alternative solution that exploits the ultra-low power consumption features of mixed-signal neuromorphic technologies. We present a biosignal processing architecture that integrates multimodal sensory inputs and processes them using the principles of neural computation to reliably detect trends in heart rate and physiological states. We validated this architecture on a mixed-signal neuromorphic processor and demonstrated its robust operation despite the inherent variability of the analog circuits present in the system. We show how the system can estimate both instantaneous heart rate and its long-term states discretized into distinct zones, effectively detecting monotonic changes over long time periods for recognizing pathological states such as agitation. This approach paves the way for a new generation of energy-efficient stand-alone wearable devices that are particularly suited for scenarios that require continuous health monitoring with minimal device maintenance.","author":[{"family":"Luca","given":"Chiara"},{"family":"Tincani","given":"Mirco"},{"family":"Indiveri","given":"Giacomo"},{"family":"Donati","given":"Elisa"}],"issued":{"date-parts":[[2024]]},"DOI":"10.21203/rs.3.rs-5737326/v1","URL":"https://doi.org/10.21203/rs.3.rs-5737326/v1","source":"crossref"},{"id":"doi:10.1038/s44306-024-00019-2","type":"article-journal","title":"Neuromorphic computing with spintronics","abstract":"Abstract Spintronics and magnetic materials exhibit many physical phenomena that are promising for implementing neuromorphic computing natively in hardware. Here, we review the current state-of-the-art, focusing on the areas of spintronic synapses, neurons, and neural networks. Many current implementations are based on the paradigm of reservoir computing, where the details of the network do not need to be known but where significant post-processing is needed. Benchmarks are given where possible. We discuss the scientific and technological advances needed to bring about spintronic neuromorphic computing that could be useful to an end-user in the medium term.","author":[{"family":"Marrows","given":"Christopher"},{"family":"Barker","given":"Joseph"},{"family":"Moore","given":"Thomas"},{"family":"Moorsom","given":"Timothy"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1038/s44306-024-00019-2","URL":"https://doi.org/10.1038/s44306-024-00019-2","source":"crossref"},{"id":"doi:10.1088/2634-4386/acf7e4","type":"article-journal","title":"From clean room to machine room: commissioning of the first-generation BrainScaleS wafer-scale neuromorphic system","abstract":"Abstract The first-generation of BrainScaleS, also referred to as BrainScaleS-1, is a neuromorphic system for emulating large-scale networks of spiking neurons. Following a ‘physical modeling’ principle, its VLSI circuits are designed to emulate the dynamics of biological examples: analog circuits implement neurons and synapses with time constants that arise from their electronic components’ intrinsic properties. It operates in continuous time, with dynamics typically matching an acceleration factor of 10 000 compared to the biological regime. A fault-tolerant design allows it to achieve wafer-scale integration despite unavoidable analog variability and component failures. In this paper, we present the commissioning process of a BrainScaleS-1 wafer module, providing a short description of the system’s physical components, illustrating the steps taken during its assembly and the measures taken to operate it. Furthermore, we reflect on the system’s development process and the lessons learned to conclude with a demonstration of its functionality by emulating a wafer-scale synchronous firing chain, the largest spiking network emulation ran with analog components and individual synapses to date.","author":[{"family":"Schmidt","given":"Hartmut"},{"family":"Montes","given":"José"},{"family":"Grübl","given":"Andreas"},{"family":"Güttler","given":"Maurice"},{"family":"Husmann","given":"Dan"},{"family":"Ilmberger","given":"Joscha"},{"family":"Kaiser","given":"Jakob"},{"family":"Mauch","given":"Christian"},{"family":"Müller","given":"Eric"},{"family":"Sterzenbach","given":"Lars"},{"family":"Schemmel","given":"Johannes"},{"family":"Schmitt","given":"Sebastian"}],"issued":{"date-parts":[[2023]]},"DOI":"10.1088/2634-4386/acf7e4","URL":"https://doi.org/10.1088/2634-4386/acf7e4","source":"crossref"},{"id":"doi:10.1002/aelm.202201111","type":"article-journal","title":"Memristive Memory Enhancement by Device Miniaturization for Neuromorphic Computing","abstract":"Abstract The areal footprint of memristors is a key consideration in material‐based neuromorphic computing and large‐scale architecture integration. Electronic transport in the most widely investigated memristive devices is mediated by filaments, posing a challenge to their scalability in architecture implementation. Here, a compelling alternative memristive device is presented and it is demonstrated that areal downscaling leads to enhancement in the memristive memory window, while maintaining analog behavior, contrary to expectations. The device designs directly integrated on semiconducting Nb‐doped SrTiO 3 (Nb:STO) allows leveraging electric field effects at edges, increasing the dynamic range in smaller devices. The findings are substantiated by studying the microscopic nature of switching using scanning transmission electron microscopy, in different resistive states, revealing an interfacial layer whose physical extent is influenced by applied electric fields. The ability of Nb:STO memristors to satisfy hardware and software requirements with downscaling, while significantly enhancing memristive functionalities, make them strong contenders for non‐von‐Neumann computing, beyond complementary metal–oxide–semiconductor.","author":[{"family":"Goossens","given":"Anouk"},{"family":"Ahmadi","given":"Majid"},{"family":"Gupta","given":"Divyanshu"},{"family":"Bhaduri","given":"Ishitro"},{"family":"Kooi","given":"Bart"},{"family":"Banerjee","given":"Tamalika"}],"issued":{"date-parts":[[2023]]},"DOI":"10.1002/aelm.202201111","URL":"https://doi.org/10.1002/aelm.202201111","source":"crossref"},{"id":"doi:10.1088/2634-4386/acdbe5","type":"article-journal","title":"Simulating the filament morphology in electrochemical metallization cells","abstract":"Abstract Electrochemical metallization (ECM) cells are based on the principle of voltage controlled formation or dissolution of a nanometer-thin metallic conductive filament (CF) between two electrodes separated by an insulating material, e.g. an oxide. The lifetime of the CF depends on factors such as materials and biasing. Depending on the lifetime of the CF—from microseconds to years—ECM cells show promising properties for use in neuromorphic circuits, for in-memory computing, or as selectors and memory cells in storage applications. For enabling those technologies with ECM cells, the lifetime of the CF has to be controlled. As various authors connect the lifetime with the morphology of the CF, the key parameters for CF formation have to be identified. In this work, we present a 2D axisymmetric physical continuum model that describes the kinetics of volatile and non-volatile ECM cells, as well as the morphology of the CF. It is shown that the morphology depends on both the amplitude of the applied voltage signal and CF-growth induced mechanical stress within the oxide layer. The model is validated with previously published kinetic measurements of non-volatile Ag/SiO 2 /Pt and volatile Ag/HfO 2 /Pt cells and the simulated CF morphologies are consistent with previous experimental CF observations.","author":[{"family":"Buttberg","given":"Milan"},{"family":"Valov","given":"Ilia"},{"family":"Menzel","given":"Stephan"}],"issued":{"date-parts":[[2023]]},"DOI":"10.1088/2634-4386/acdbe5","URL":"https://doi.org/10.1088/2634-4386/acdbe5","source":"crossref"},{"id":"doi:10.1088/2634-4386/acca45","type":"article-journal","title":"System model of neuromorphic sequence learning on a memristive crossbar array","abstract":"Abstract Machine learning models for sequence learning and processing often suffer from high energy consumption and require large amounts of training data. The brain presents more efficient solutions to how these types of tasks can be solved. While this has inspired the conception of novel brain-inspired algorithms, their realizations remain constrained to conventional von-Neumann machines. Therefore, the potential power efficiency of the algorithm cannot be exploited due to the inherent memory bottleneck of the computing architecture. Therefore, we present in this paper a dedicated hardware implementation of a biologically plausible version of the Temporal Memory component of the Hierarchical Temporal Memory concept. Our implementation is built on a memristive crossbar array and is the result of a hardware-algorithm co-design process. Rather than using the memristive devices solely for data storage, our approach leverages their specific switching dynamics to propose a formulation of the peripheral circuitry, resulting in a more efficient design. By combining a brain-like algorithm with emerging non-volatile memristive device technology we strive for maximum energy efficiency. We present simulation results on the training of complex high-order sequences and discuss how the system is able to predict in a context-dependent manner. Finally, we investigate the energy consumption during the training and conclude with a discussion of scaling prospects.","author":[{"family":"Siegel","given":"Sebastian"},{"family":"Bouhadjar","given":"Younes"},{"family":"Tetzlaff","given":"Tom"},{"family":"Waser","given":"Rainer"},{"family":"Dittmann","given":"Regina"},{"family":"Wouters","given":"Dirk"}],"issued":{"date-parts":[[2023]]},"DOI":"10.1088/2634-4386/acca45","URL":"https://doi.org/10.1088/2634-4386/acca45","source":"crossref"},{"id":"doi:10.1088/2634-4386/ad4b5b","type":"article-journal","title":"Optical spike amplitude weighting and neuromimetic rate coding using a joint VCSEL-MRR neuromorphic photonic system","abstract":"Abstract Spiking neurons and neural networks constitute a fundamental building block for brain-inspired computing, which is poised to benefit significantly from photonic hardware implementations. In this work, we experimentally investigate an interconnected optical neuromorphic system based on an ultrafast spiking vertical cavity surface emitting laser (VCSEL) neuron and a silicon photonics (SiPh) integrated micro-ring resonator (MRR). We experimentally demonstrate two different functional arrangements of these devices: first, we show that MRR weight banks can be used in conjunction with the spiking VCSEL-neurons to perform amplitude weighting of sub-ns optical spiking signals. Second, we show that a continuously firing VCSEL-neuron can be directly modulated using a locking signal propagated through a single weighting MRR, and we utilise this functionality to perform optical spike firing rate-coding via thermal tuning of the MRR. Given the significant track record of both integrated weight banks and photonic VCSEL-neurons, we believe these results demonstrate the viability of combining these two classes of devices for use in functional neuromorphic photonic systems.","author":[{"family":"Hejda","given":"Matěj"},{"family":"Doris","given":"Eli"},{"family":"Bilodeau","given":"Simon"},{"family":"Robertson","given":"Joshua"},{"family":"Owen-Newns","given":"Dafydd"},{"family":"Shastri","given":"Bhavin"},{"family":"Prucnal","given":"Paul"},{"family":"Hurtado","given":"Antonio"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1088/2634-4386/ad4b5b","URL":"https://doi.org/10.1088/2634-4386/ad4b5b","source":"crossref"},{"id":"doi:10.1038/s41467-024-47811-6","type":"article-journal","title":"Spike-based dynamic computing with asynchronous sensing-computing neuromorphic chip","abstract":"Abstract By mimicking the neurons and synapses of the human brain and employing spiking neural networks on neuromorphic chips, neuromorphic computing offers a promising energy-efficient machine intelligence. How to borrow high-level brain dynamic mechanisms to help neuromorphic computing achieve energy advantages is a fundamental issue. This work presents an application-oriented algorithm-software-hardware co-designed neuromorphic system for this issue. First, we design and fabricate an asynchronous chip called “Speck”, a sensing-computing neuromorphic system on chip. With the low processor resting power of 0.42mW, Speck can satisfy the hardware requirements of dynamic computing: no-input consumes no energy. Second, we uncover the “dynamic imbalance” in spiking neural networks and develop an attention-based framework for achieving the algorithmic requirements of dynamic computing: varied inputs consume energy with large variance. Together, we demonstrate a neuromorphic system with real-time power as low as 0.70mW. This work exhibits the promising potentials of neuromorphic computing with its asynchronous event-driven, sparse, and dynamic nature.","author":[{"family":"Yao","given":"Man"},{"family":"Richter","given":"Ole"},{"family":"Zhao","given":"Guangshe"},{"family":"Qiao","given":"Ning"},{"family":"Xing","given":"Yannan"},{"family":"Wang","given":"Dingheng"},{"family":"Hu","given":"Tianxiang"},{"family":"Fang","given":"Wei"},{"family":"Demirci","given":"Tugba"},{"family":"Marchi","given":"Michele"},{"family":"Deng","given":"Lei"},{"family":"Yan","given":"Tianyi"},{"family":"Nielsen","given":"Carsten"},{"family":"Sheik","given":"Sadique"},{"family":"Wu","given":"Chenxi"},{"family":"Tian","given":"Yonghong"},{"family":"Xu","given":"Bo"},{"family":"Li","given":"Guoqi"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1038/s41467-024-47811-6","URL":"https://doi.org/10.1038/s41467-024-47811-6","source":"crossref"},{"id":"doi:10.1088/2634-4386/acc08e","type":"article-journal","title":"Dynamics of the judgment of tactile stimulus intensity","abstract":"Abstract In the future, artificial agents will need to make assessments of tactile stimuli in order to interact intelligently with the environment and with humans. Such assessments will depend on exquisite and robust mechanosensors, but sensors alone do not make judgments and choices. Rather, the central processing of mechanosensor inputs must be implemented with algorithms that produce ‘behavioral states’ in the artificial agent that resemble or mimic perceptual judgments in biology. In this study, we consider the problem of perceptual judgment as applied to vibration intensity. By a combination of computational modeling and simulation followed by psychophysical testing of vibration intensity perception in rats, we show that a simple yet highly salient judgment—is the current stimulus strong or weak?—can be explained as the comparison of ongoing sensory input against a criterion constructed as the time-weighted average of the history of recent stimuli. Simulations and experiments explore how judgments are shaped by the distribution of stimuli along the intensity dimension and, most importantly, by the time constant of integration which dictates the dynamics of criterion updating. The findings of this study imply that judgments made by the real nervous system are not absolute readouts of physical parameters but are context-dependent; algorithms of this form can be built into artificial systems.","author":[{"family":"Darani","given":"ZY"},{"family":"Hachen","given":"I"},{"family":"Diamond","given":"ME"}],"issued":{"date-parts":[[2023]]},"DOI":"10.1088/2634-4386/acc08e","URL":"https://doi.org/10.1088/2634-4386/acc08e","source":"crossref"},{"id":"doi:10.1149/11102.0133ecst","type":"article-journal","title":"A Capacitor-Based Synaptic Device with IGZO Access Transistors for Neuromorphic Computing","abstract":"Analog in-memory computing synaptic devices have been widely studied for efficient implementation of deep learning. As a candidate for a synaptic device, Si-CMOS and capacitor-based synaptic devices have been proposed. However, due to Si-CMOS leakage currents, it is difficult to achieve sufficient retention time. In our research, we verified IGZO TFT with low leakage current and capacitor-based synapses can show linear and symmetric weight update characteristics as well as excellent device variation characteristics. We also verified that IGZO TFT has a leakage current per channel width of 1μm of ~10-17A, which is much lower than the Si-CMOS, resulting in higher accuracy in deep neural network training.","author":[{"family":"Won","given":"Jongun"},{"family":"Roh","given":"Youngchae"},{"family":"Kang","given":"Minseung"},{"family":"Park","given":"Yeaji"},{"family":"Kang","given":"Jaehyeon"},{"family":"Seo","given":"Hyeongjun"},{"family":"Joe","given":"Changhoon"},{"family":"Kim","given":"Sangbum"}],"issued":{"date-parts":[[2023]]},"DOI":"10.1149/11102.0133ecst","URL":"https://doi.org/10.1149/11102.0133ecst","source":"crossref"},{"id":"doi:10.1149/ma2023-01321845mtgabs","type":"article-journal","title":"A Capacitor-Based Synaptic Device with IGZO Access Transistors for Neuromorphic Computing","abstract":"IGZO TFT has extremely low leakage current characteristics compared to Si-CMOS, so it can be applied not only to displays but also to various research fields. One of the important applications using IGZO TFT is analog-based synaptic device for the efficient implementation for deep learning. The application of IGZO TFT to a synaptic device was more widely studied after a synaptic device based on Si-CMOS and capacitor was devised in 2017(1). Since insufficient retention time was a remaining problem in Si-CMOS and capacitor-based storage synaptic devices, the combination of IGZO TFT with low leakage current and capacitor was a very attractive solution. However, because the viable p-type metal oxide TFT does not exist, linearity and symmetry of weight update, a major requirement of on-chip trainable synaptic devices, has been difficult to achieve. In our research, for the first time, we obtained linear and symmetric weight update characteristics by combining an IGZO TFT and a capacitor. In addition, we verified the leakage characteristics of IGZO TFT and the retention characteristics of synaptic devices, which are the most important characteristics. Through the single TFT measurement result, we were able to obtain a leakage current per channel width 1µm of 10 -17 A, which is much smaller than the Si transistor. Lower leakage current will lead to better retention characteristics than conventional Si-CMOS-based devices, which will result in higher accuracy in deep neural network training. Based on the published results, the IGZO TFT synapse is projected to be able to achieve the retention time of ~4×10 9 s using only 10fF capcacitor. Another important factor that affects deep neural network training is cell-to-cell and device-to-device variation. As opposed to capacitive synaptic devices, resistive synaptic devices such as Resistive RAM and Phase Change Memory suffer from large device variation. We also verified the device variation characteristics, which are advantages of charge storage devices based on transistors and capacitors. The variation characteristics of the device were evaluated through the ΔG +3σ /ΔG -3σ (%) formula and an excellent result of ~200% was obtained. In our research, we verified the main characteristics for on-chip training of synaptic devices based on IGZO TFT and capacitor: linearity and symmetry of weight update, retention characteristics and variation characteristics. The remarkable characteristics of synaptic devices based on IGZO TFT and capacitor will be a good candidate for on-chip training. 1. S. Kim, T. Gokmen, H. M. Lee, and W. E. Haensch, Midwest Symp. Circuits Syst. , 2017 - Augus , 422–425 (2017). Acknowledgements This research was supported by National R&amp;D Program through the National Research Foundation of Korea(NRF) funded by Ministry of Science and ICT(NRF-2020M3F3A2A01081240)","author":[{"family":"Won","given":"Jongun"},{"family":"Roh","given":"Youngchae"},{"family":"Kang","given":"Minseung"},{"family":"Park","given":"Yeaji"},{"family":"Kang","given":"Jaehyeon"},{"family":"Seo","given":"Hyeongjun"},{"family":"Joe","given":"Changhoon"},{"family":"Kim","given":"Sangbum"}],"issued":{"date-parts":[[2023]]},"DOI":"10.1149/ma2023-01321845mtgabs","URL":"https://doi.org/10.1149/ma2023-01321845mtgabs","source":"crossref"},{"id":"doi:10.4018/979-8-3693-6303-4.ch007","type":"article-journal","title":"An NLP Approach to Enrich Biomedical Research Through Sentiment Analysis of Patient Feedback","abstract":"This chapter consults the trajectory committed by utilizing patient feedback (PF) in the wake of biomedical research through sentimental analysis (SA) in natural language processing (NLP). PF has been compared to a gold mine for the healthcare industry as it delivers clinical efficacy and preserves quality. Analyzing these patient responses is vastly more time-consuming and subjective. SA employment can efficiently extract beneficial insights from this feedback by automating patients' positive, negative, or neutral sentiments. By systematically examining millions of remarks to identify familiar themes, distinct concerns, and patient satisfaction levels, researchers can employ these sentiments to understand disease status and assist in making intelligent decisions. SA can work with structured and unstructured sentiment data from mixed social media posts and electronic health records to produce favorable results, allowing researchers to improve biomedical research. Eventually, this chapter uncloses worthwhile wisdom to enrich biomedical research by employing PF through SA.","author":[{"family":"Saha","given":"Soumitra"},{"family":"Lilhore","given":"Umesh"},{"family":"Simaiya","given":"Sarita"}],"issued":{"date-parts":[[2024]]},"DOI":"10.4018/979-8-3693-6303-4.ch007","URL":"https://doi.org/10.4018/979-8-3693-6303-4.ch007","source":"crossref"},{"id":"doi:10.1149/ma2024-01573019mtgabs","type":"article-journal","title":"(Invited) There’s More to a Probabilistic Neuromorphic Computing System Than Noisy Devices","abstract":"A range of computing problems require understanding uncertainty, such as climate modeling, or use statistics to model problems that are otherwise difficult to solve, such as high energy particle collisions. Currently, these computations are handled by hardware where significant energy is spent to suppress stochasticity in materials and devices, and then significant computational resources are expended to re-introduce stochasticisty in algorithms. Instead, we take inspiration from the brain, which features 10 15 stochastic synapses. This talk focuses on understanding how to leverage fluctuations in devices to do efficient sampling, which is a fundamental operation in many statistical approaches to computation [1]. In the first part of this talk, we use bitstreams generated by magnetic tunnel junction and tunnel diode devices to generate samples from different distributions, an elementary operation in statistical approaches to modeling. We show how to use elementary operations on multiple bits to improve both the accuracy and complexity of sampling. While intuition motivates the asking which of the two devices is more efficient at generating a random bitstream, in the second part of this talk, we show that this consideration is a small contribution to the overall circuit required to do a complete calculation. To significantly accelerate applications requires devices that minimize the energy and area cost of the CMOS parts of the design, as the true challenge lies in holistic codesign [2]. At the heart of statistical approaches to computation is sampling. Typically, a uniform random sample is generated using a pseudo-random number generator (PRNG), followed by a mathematical operation to sample an application-relevant distribution. This sample is then plugged into a sequence of deterministic calculations that comprise a model. We use magnetic tunnel junctions and tunnel diodes to generate a fair coinflip, having equal probability of each of two outscomes, from which we create a uniform random sample. We show how to relate the quality of the device bitstreams to the quality of random samples [3]. While these devices consume significantly less energy than the PRNG, the energy consumed by the PRNG is a fraction of the energy consumed by a complete calculation. We next focus on moving the entire process of sampling non-uniform distributions into hardware. We show that weighted coinflips can be used to sample any distribution with a well-defined probability distribution function using a tree. Simple logic operations can be used to combine many inaccurate fair physical coinflips to produce a single high-accuracy weighted logical coinflip, and high quality samples from non-uniform distributions. Overall, a simple implementation uses a few hundred physical coinflips, some simple logic operations (shift register, comparison, XOR extractor), and memory access to produce a sample. This success points to the potential for moving more of the model into increasingly sophisticated sampling schemes. Having established how the basic element of the computation are connected, we are now ready to examine its efficiency compared to a PRNG, whose cost is roughly nJ/operation. Thus far, we have ignored the analog signal transduction attached to the coinflip devices and the logic operations that tie them together. Coinflip devices which cannot be directly integrated with logic will suffer from a von Neumann bottleneck. Thus, the coinflip devices need to be intimately integrated with logic, requiring paying a per-device transduction penalty. We find the energy cost for even the simplest signal transduction – a stimulating pulse and a latching output – is in the 100 fJ – 1 pJ per device range, which is larger than the energy consumed by either magnetic tunnel junctions or tunnel diodes. Meanwhile, the most expensive component of the logic operations stem from the fine-grained integration of memory. While only 32 weighted coinflips are needed to draw a 32-","author":[{"family":"Misra","given":"Shashank"},{"family":"Allemang","given":"Christopher"},{"family":"Smith","given":"JD"},{"family":"Cardwell","given":"Suma"},{"family":"Aimone","given":"James"},{"family":"Kent","given":"Andrew"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1149/ma2024-01573019mtgabs","URL":"https://doi.org/10.1149/ma2024-01573019mtgabs","source":"crossref"},{"id":"doi:10.1063/5.0149393","type":"article-journal","title":"A review on device requirements of resistive random access memory (RRAM)-based neuromorphic computing","abstract":"With the arrival of the era of big data, the conventional von Neumann architecture is now insufficient owing to its high latency and energy consumption that originate from its separated computing and memory units. Neuromorphic computing, which imitates biological neurons and processes data through parallel procedures between artificial neurons, is now regarded as a promising solution to address these restrictions. Therefore, a device with analog switching for weight update is required to implement neuromorphic computing. Resistive random access memory (RRAM) devices are one of the most promising candidates owing to their fast-switching speed and scalability. RRAM is a non-volatile memory device and operates via resistance changes in its insulating layer. Many RRAM devices exhibiting exceptional performance have been reported. However, these devices only excel in one property. Devices that exhibit excellent performance in all aspects have been rarely proposed. In this Research Update, we summarize five requirements for RRAM devices and discuss the enhancement methods for each aspect. Finally, we suggest directions for the advancement of neuromorphic electronics.","author":[{"family":"Yoon","given":"Jeong"},{"family":"Song","given":"Young"},{"family":"Ham","given":"Wooho"},{"family":"Park","given":"Jeong"},{"family":"Kwon","given":"Jang"}],"issued":{"date-parts":[[2023]]},"DOI":"10.1063/5.0149393","URL":"https://doi.org/10.1063/5.0149393","source":"crossref"},{"id":"doi:10.5281/zenodo.20397720","type":"article-journal","title":"SymBrain: A Biomimetic Neuro-Symbolic Architecture for Small Language Models","abstract":"SymBrain: A 3-hemisphere neuro-symbolic architecture using Qwen2.5-Math-7B and Ministral-8B coordinated by an executive Prefrontal Cortex bridge. Achieves SOTA reasoning for its class (GSM8K: 88.50%, MATH: 58.41%, Physics: 56.09%) with 21.9% power savings.","author":[{"family":"Callens","given":"Xavier"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20397720","URL":"https://doi.org/10.5281/zenodo.20397720","source":"datacite"},{"id":"doi:10.5281/zenodo.20397719","type":"article-journal","title":"SymBrain: A Biomimetic Neuro-Symbolic Architecture for Small Language Models","abstract":"SymBrain: A 3-hemisphere neuro-symbolic architecture using Qwen2.5-Math-7B and Ministral-8B coordinated by an executive Prefrontal Cortex bridge. Achieves SOTA reasoning for its class (GSM8K: 88.50%, MATH: 58.41%, Physics: 56.09%) with 21.9% power savings.","author":[{"family":"Callens","given":"Xavier"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20397719","URL":"https://doi.org/10.5281/zenodo.20397719","source":"datacite"},{"id":"doi:10.35848/1347-4065/ad73e1","type":"article-journal","title":"Neuromorphic alternating current sensing using piezoelectric resonators and physical reservoir computing","abstract":"Abstract Non-contact current sensors are valuable because they can safely measure alternating current without interrupting the circuit. However, current sensors utilizing Hall elements or coils are only available for single wires, and piezoelectric resonator-based sensors have difficulty achieving both high sensitivity and linearity. To address this issue, we propose a novel approach, that is, the use of piezoelectric current sensors as nodes for physical reservoir computing (physical RC), allowing us to utilize nonlinear regions. To improve the sensitivity and short-term memory required by physical RC, a piezoelectric resonator with a quality factor of 75 was realized by employing a tuning fork structure. Nonlinearities were also introduced by analog circuits. The results of the benchmark tests indicate that the device worked as a physical RC and that it successfully predicted unknown current values from the results of training at three levels of current.","author":[{"family":"Nishimura","given":"Kei"},{"family":"Fujimura","given":"Norifumi"},{"family":"Yoshimura","given":"Takeshi"}],"issued":{"date-parts":[[2024]]},"DOI":"10.35848/1347-4065/ad73e1","URL":"https://doi.org/10.35848/1347-4065/ad73e1","source":"crossref"},{"id":"doi:10.3390/electronics13173448","type":"article-journal","title":"An Implementation of Communication, Computing and Control Tasks for Neuromorphic Robotics on Conventional Low-Power CPU Hardware","abstract":"Bioinspired approaches tend to mimic some biological functions for the purpose of creating more efficient and robust systems. These can be implemented in both software and hardware designs. A neuromorphic software part can include, for example, Spiking Neural Networks (SNNs) or event-based representations. Regarding the hardware part, we can find different sensory systems, such as Dynamic Vision Sensors, touch sensors, and actuators, which are linked together through specific interface boards. To run real-time SNN models, specialised hardware such as SpiNNaker, Loihi, and TrueNorth have been implemented. However, neuromorphic computing is still in development, and neuromorphic platforms are still not easily accessible to researchers. In addition, for Neuromorphic Robotics, we often need specially designed and fabricated PCBs for communication with peripheral components and sensors. Therefore, we developed an all-in-one neuromorphic system that emulates neuromorphic computing by running a Virtual Machine on a conventional low-power CPU. The Virtual Machine includes Python and Brian2 simulation packages, which allow the running of SNNs, emulating neuromorphic hardware. An additional, significant advantage of using conventional hardware such as Raspberry Pi in comparison to purpose-built neuromorphic hardware is that we can utilise the built-in physical input–output (GPIO) and USB ports to directly communicate with sensors. As a proof of concept platform, a robotic goalkeeper has been implemented, using a Raspberry Pi 5 board and SNN model in Brian2. All the sensors, namely DVS128, with an infrared module as the touch sensor and Futaba S9257 as the actuator, were linked to a Raspberry Pi 5 board. We show that it is possible to simulate SNNs on a conventional low-power CPU running real-time tasks for low-latency and low-power robotic applications. Furthermore, the system excels in the goalkeeper task, achieving an overall accuracy of 84% across various environmental conditions while maintaining a maximum power consumption of 20 W. Additionally, it reaches 88% accuracy in the online controlled setup and 80% in the offline setup, marking an improvement over previous results. This work demonstrates that the combination of a conventional low-power CPU running a Virtual Machine with only selected software is a viable competitor to neuromorphic computing hardware for robotic applications.","author":[{"family":"Russo","given":"Nicola"},{"family":"Madsen","given":"Thomas"},{"family":"Nikolic","given":"Konstantin"}],"issued":{"date-parts":[[2024]]},"DOI":"10.3390/electronics13173448","URL":"https://doi.org/10.3390/electronics13173448","source":"crossref"},{"id":"doi:10.1149/ma2024-01331651mtgabs","type":"article-journal","title":"Exploring Neuromorphic Computing with Double-Gate Floating Device Based on Van Der Waals Heterostructures","abstract":"Neuromorphic devices represent a frontier in technological development, aiming to emulate the intricate parallel functionalities inherent to the human brain [1, 2]. Traditional computing models, such as Von Neumann architecture, grapple with limitations in parallel processing, prompting the exploration of neuromorphic computing to integrate memory and processing efficiently. Our research presents the double gate floating device based on Vander Waals heterostructure (DG-VH-FET), employing molybdenum disulfide (MoS 2 ), hexagonal boron nitride (h-BN), and graphene (Gr) as key components. These devices, based on 2D materials, enable independent modulation of channel doping through electrostatic means [3]. In this study, we initially assessed the fundamental characteristics of the floating gate, including transfer characteristics, output characteristics, retention, and endurance. By applying constant drain-to-source voltages (V DS ) while introducing variations in the back gate voltage (V BG ) sweeping range, the memory window of the transfer characteristics changes. Conversely, maintaining a constant V BG and V DS while varying the top gate voltage (V TG ) solely results in a shift in the threshold voltage of the memory window. This is attributed to the crucial role of V BG in charge trapping in the floating gate layer, with the influence of V TG on charge trapping being negligible due to the screening effect of MoS 2 [4]. Moreover, by using V TG and V BG this study shows that DG-VH-FET could have potential applications in exploring and understanding various neurological effects, thereby bridging the gap between electronics and biology. Figure 1","author":[{"family":"Ghosh","given":"Advaita"},{"family":"Lin","given":"Yen"},{"family":"Lin","given":"Shu"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1149/ma2024-01331651mtgabs","URL":"https://doi.org/10.1149/ma2024-01331651mtgabs","source":"crossref"},{"id":"doi:10.1088/2634-4386/ad40ca","type":"article-journal","title":"Advanced iontronic spiking modes with multiscale diffusive dynamics in a fluidic circuit","abstract":"Abstract Fluidic iontronics is emerging as a distinctive platform for implementing neuromorphic circuits, characterised by its reliance on the same aqueous medium and ionic signal carriers as the brain. Drawing upon recent theoretical advancements in both iontronic spiking circuits and in dynamic conductance of conical ion channels, which form fluidic memristors, we expand the repertoire of proposed neuronal spiking dynamics in iontronic circuits. Through a modelled circuit containing channels that carry a bipolar surface charge, we extract phasic bursting, mixed-mode spiking, tonic bursting, and threshold variability, all with spike voltages and frequencies within the typical range for mammalian neurons. These features are possible due to the strong dependence of the typical conductance memory retention time on the channel length, enabling timescales varying from individual spikes to bursts of multiple spikes within a single circuit. These advanced forms of neuronal-like spiking support the exploration of aqueous iontronics as an interesting platform for neuromorphic circuits.","author":[{"family":"Kamsma","given":"TM"},{"family":"Rossing","given":"EA"},{"family":"Spitoni","given":"C"},{"family":"Roij","given":"RV"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1088/2634-4386/ad40ca","URL":"https://doi.org/10.1088/2634-4386/ad40ca","source":"crossref"},{"id":"doi:10.1088/2515-7639/ad9ee1","type":"article-journal","title":"Realizing linear synaptic plasticity in electric double layer-gated transistors for improved predictive accuracy and efficiency in neuromorphic computing","abstract":"Abstract Neuromorphic computing offers a low-power, parallel alternative to traditional von Neumann architectures by addressing the sequential data processing bottlenecks. Electric double layer-gated transistors (EDLTs) resemble biological synapses with their ionic response and offer low power operations, making them suitable for neuromorphic applications. A critical consideration for artificial neural networks (ANNs) is achieving linear and symmetric plasticity (i.e. weight updates) during training, as this directly affects accuracy and efficiency. This study uses finite element modeling to explore EDLTs as artificial synapses in ANNs and investigates the underlying mechanisms behind the nonlinear weight updates observed experimentally in previous studies. By solving modified Poisson–Nernst–Planck equations, we examined ion dynamics within an EDL capacitor and their effects on plasticity, revealing that the rates of EDL formation and dissipation are concentration-dependent. Fixed-magnitude pulse inputs result in decreased formation and increased dissipation rates, leading to nonlinear weight updates. For a pulse magnitude of 1 V, both 1 ms 500 Hz and 5 ms 100 Hz pulse inputs saturated at less than half of the steady state EDL concentration, limiting the number of accessible states and operating range of devices. To address this, we developed a predictive linear ionic weight update solver (LIWUS) in Python to predict voltage pulse inputs that achieve linear plasticity. We then evaluated an ANN with linear and nonlinear weight updates on the Modified National Institute of Standards and Technology classification task. The ANN with LIWUS-provided linear weight updates required 19% fewer (i.e. 5) epochs in both training and validation than the network with nonlinear weight updates to reach optimal performance. It achieved a 97.6% recognition accuracy, 1.5–4.2% higher than with nonlinear updates, and a low standard deviation of 0.02%. The network model is amenable to future spiking neural network applications, and the performance with linear weight update s is expected to improve for complex networks with multiple hidden layers.","author":[{"family":"Manimaran","given":"Nithil"},{"family":"Sutton","given":"Cori"},{"family":"Streamer","given":"Jake"},{"family":"Merkel","given":"Cory"},{"family":"Xu","given":"Ke"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1088/2515-7639/ad9ee1","URL":"https://doi.org/10.1088/2515-7639/ad9ee1","source":"crossref"},{"id":"doi:10.1088/2634-4386/ad6732","type":"article-journal","title":"Difficulties and approaches in enabling learning-in-memory using crossbar arrays of memristors","abstract":"Abstract Crossbar arrays of memristors are promising to accelerate the deep learning algorithm as a non-von-Neumann architecture, where the computation happens at the location of the memory. The computations are parallelly conducted employing the basic physical laws. However, current research works mainly focus on the offline training of deep neural networks, i.e. only the information forwarding is accelerated by the crossbar array. Two other essential operations, i.e. error backpropagation and weight update, are mostly simulated and coordinated by a conventional computer in von Neumann architecture, respectively. Several different in situ learning schemes incorporating error backpropagation and/or weight updates have been proposed and investigated through neuromorphic simulation. Nevertheless, they met the issues of non-ideal synaptic behaviors of the memristors and the complexities of the neural circuits surrounding crossbar arrays. Here we review the difficulties and approaches in implementing the error backpropagation and weight update operations for online training or in-memory learning that are adapted to noisy and non-ideal memristors. We hope this work will be beneficial for the development of open neuromorphic simulation tools for learning-in-memory systems, and eventually for the hardware implementation of such as system.","author":[{"family":"Wang","given":"Wei"},{"family":"Li","given":"Yang"},{"family":"Wang","given":"Ming"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1088/2634-4386/ad6732","URL":"https://doi.org/10.1088/2634-4386/ad6732","source":"crossref"},{"id":"doi:10.1038/s43246-024-00573-6","type":"article-journal","title":"Metal-organic framework single crystal for in-memory neuromorphic computing with a light control","abstract":"Abstract Neuromorphic architectures, expanding the limits of computing from conventional data processing and storage to advanced cognition, learning, and in-memory computing, impose restrictions on materials that should operate fast, energy efficiently, and highly endurant. Here we report on in-memory computing architecture based on metal-organic framework (MOF) single crystal with a light control. We demonstrate that the MOF with inherent memristive behavior (for data storage) changes nonlinearly its electric response when irradiated by light. This leads to three and more electronic states (spikes) with 81 ms duration and 1 s refractory time, allowing to implement 40 bits s −1 optoelectronic data processing. Next, the architecture is switched to the neuromorphic state upon the action of a set of laser pulses, providing the text recognition over 50 times with app. 100% accuracy. Thereby, simultaneous data storage, processing, and neuromorphic computing on MOF, driven by light, pave the way for multifunctional in-memory computing architectures.","author":[{"family":"Bachinin","given":"Semyon"},{"family":"Marunchenko","given":"Alexandr"},{"family":"Matchenya","given":"Ivan"},{"family":"Zhestkij","given":"Nikolai"},{"family":"Shirobokov","given":"Vladimir"},{"family":"Gunina","given":"Ekaterina"},{"family":"Novikov","given":"Alexander"},{"family":"Timofeeva","given":"Maria"},{"family":"Povarov","given":"Svyatoslav"},{"family":"Li","given":"Fengting"},{"family":"Milichko","given":"Valentin"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1038/s43246-024-00573-6","URL":"https://doi.org/10.1038/s43246-024-00573-6","source":"crossref"},{"id":"doi:10.1002/aelm.202300698","type":"article-journal","title":"Optimization Method for Conductance Modulation in Ferroelectric Transistor for Neuromorphic Computing","abstract":"Abstract The learning accuracy of neuromorphic computing that mimics the biological brain, is affected by the conductance‐modulation characteristics of an artificial synapse. In ferroelectric‐based devices, these characteristics are implemented using a distribution of polarization values. Therefore, the distribution in a ferroelectric thin film with various external voltage signals is investigate. As polarization switching proceeds with voltage pulse, the domains of the switched polarization become larger. In ferroelectric‐gate field effect transistors, the channel layer assumed to lie beneath the ferroelectrics experiences a local conductance change, according to the polarization distribution of the ferroelectric layer. It is found that small clusters with high conductivity become large clusters in the channel layer as the polarization switching proceeds. When the additional pulses are applied, the high conductive regions eventually connect (i.e., percolate) in the channel layer and the conductance of the layer is greatly increased. Adjusting the height of the applied voltage can slow down or speed up this phenomenon. Also, the nanosecond voltage pulses are employed and the width of the conductive pathway is adjusted. It enables to fine‐tune the conductance of the channel layer. It demonstrates that conductance modulation is optimized with an appropriate voltage pulse train pattern.","author":[{"family":"Kim","given":"Cheol"},{"family":"Lee","given":"Jae"},{"family":"Ku","given":"Minkyung"},{"family":"Kim","given":"Tae"},{"family":"Noh","given":"Taehee"},{"family":"Lee","given":"Seung"},{"family":"Ahn","given":"Ji‐hoon"},{"family":"Kang","given":"Bo"}],"issued":{"date-parts":[[2023]]},"DOI":"10.1002/aelm.202300698","URL":"https://doi.org/10.1002/aelm.202300698","source":"crossref"},{"id":"doi:10.1002/adfm.202410974","type":"article-journal","title":"Photonic Synapse of CrSBr/PtS\n                    <sub>2</sub>\n                    Transistor for Neuromorphic Computing and Light Decoding","abstract":"Abstract Field effect transistors based on 2D layered material have gained significant potential in emerging technologies, such as neuromorphic computing and ultrafast memory response for artificial intelligence applications. This study proposes a facile approach to fabricate an optoelectronic artificial synapse for neuromorphic computing and light‐decoding information system by utilizing the 2D heterostructure of CrSBr/PtS 2 to overcome circuit complexity. The CrSBr layer serves as a trapping layer, while PtS 2 , mounted on top of CrSBr, acts as a channel layer. PtS 2 exhibits n‐type semiconductor behavior with a hysteresis that varies with the thickness of the underlying CrSBr layer. The heterostructure device, featuring a 96.3 nm thick CrSBr layer, exhibited a large memory window of 11.9 V when the gate voltage is swept from −10 V to +10 V. Various synaptic behaviors are effectively demonstrated, including paired‐pulse facilitation, excitatory postsynaptic current, optical spike number and intensity‐dependent plasticity using laser light at a wavelength of 365 nm. The device achieves 26 distinct output signals depending on the intensity of the incident laser light, ranging from 10 to 385 mW cm −2 , enabling its applications for light‐decoded information security systems. Thus, the investigation presents a unique approach to artificial intelligence and cybersecurity systems.","author":[{"family":"Khan","given":"Muhammad"},{"family":"Khan","given":"Muhammad"},{"family":"Nasim","given":"Muhammad"},{"family":"Elahi","given":"Ehsan"},{"family":"Rabeel","given":"Muhammad"},{"family":"Asim","given":"Muhammad"},{"family":"Rehmat","given":"Arslan"},{"family":"Pervez","given":"Muhammad"},{"family":"Rehman","given":"Shania"},{"family":"Kim","given":"Honggyun"},{"family":"Eom","given":"Jonghwa"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1002/adfm.202410974","URL":"https://doi.org/10.1002/adfm.202410974","source":"crossref"},{"id":"doi:10.1088/1361-6528/ad5685","type":"article-journal","title":"In<sub>2</sub>O<sub>3</sub>/ZnO heterojunction thin film transistor for high recognition accuracy neuromorphic computing and optoelectronic artificial synapses","abstract":"Abstract Solid electrolyte-gated transistors exhibit improved chemical stability and can fulfill the requirements of microelectronic packaging. Typically, metal oxide semiconductors are employed as channel materials. However, the extrinsic electron transport properties of these oxides, which are often prone to defects, pose limitations on the overall electrical performance. Achieving excellent repeatability and stability of transistors through the solution process remains a challenging task. In this study, we propose the utilization of a solution-based method to fabricate an In 2 O 3 /ZnO heterojunction structure, enabling the development of efficient multifunctional optoelectronic devices. The heterojunction’s upper and lower interfaces induce energy band bending, resulting in the accumulation of a large number of electrons and a significant enhancement in transistor mobility. To mimic synaptic plasticity responses to electrical and optical stimuli, we utilize Li + -doped high-k ZrO x thin films as a solid electrolyte in the device. Notably, the heterojunction transistor-based convolutional neural network achieves a high accuracy rate of 93% in recognizing handwritten digits. Moreover, our research involves the simulation of a typical sensory neuron, specifically a nociceptor, within our synaptic transistor. This research offers a novel avenue for the advancement of cost-effective three-terminal thin-film transistors tailored for neuromorphic applications.","author":[{"family":"Sun","given":"Shangheng"},{"family":"Zhang","given":"Minghao"},{"family":"Bian","given":"Jing"},{"family":"Xu","given":"Ting"},{"family":"Su","given":"Jie"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1088/1361-6528/ad5685","URL":"https://doi.org/10.1088/1361-6528/ad5685","source":"crossref"},{"id":"doi:10.4018/978-1-6684-6596-7.ch007","type":"article-journal","title":"Strategies for Automated Bike-Sharing Systems Leveraging ML and VLSI Approaches","abstract":"Machine learning has had an impact in the area of microchip design and was initially used in automation. This development could result in a tremendous change in the realm of hardware computation and AI's powerful analysis tools. Traffic is a pressing issue in densely populated cities. Governments worldwide are attempting to address this problem by introducing various forms of public transportation, including metro. However, these solutions require significant investment and implementation time. Despite the high cost and inherent flaws of the system, many people still prefer to use their personal vehicles rather than public transportation. To address this issue, the authors propose a bike-sharing solution in which all processes from membership registration to bike rental and return are automated. Bagging is an ensemble learning method that can be used for base models with a low bias and high variance. It uses randomization of the dataset to reduce the variance of the base models, while keeping the bias low.","author":[{"family":"Shukla","given":"Jagrat"},{"family":"Rishikha","given":"Numburi"},{"family":"Chaturvedi","given":"Janhavi"},{"family":"Gokulanathan","given":"Sumathi"},{"family":"Chandrasekaran","given":"Sriharipriya"},{"family":"Elango","given":"Konguvel"},{"family":"Selvaperumal","given":"Sathishkumar"}],"issued":{"date-parts":[[2023]]},"DOI":"10.4018/978-1-6684-6596-7.ch007","URL":"https://doi.org/10.4018/978-1-6684-6596-7.ch007","source":"crossref"},{"id":"doi:10.31219/osf.io/49qzv","type":"article-journal","title":"Spike Timing Mechanisms in Neuromorphic Vision Sensors using Memristor-based non-volatile Memory devices","abstract":"Spike-timing mechanisms in neuromorphic vision sensors represent a cutting-edge approach to mimicking biological vision systems' efficiency and adaptability. These systems utilize memristor-based non-volatile memory devices to achieve high precision and low power consumption, essential for real-time image processing and recognition tasks. This paper explores the principles and applications of spike-timing-dependent plasticity (STDP) in neuromorphic vision sensors, focusing on the integration of memristor technology. This study introduces an innovative visual-tactile perception system that integrates a scalable, biomimetic tactile sensor called NeuTouch and uses a Visual-Tactile Spiking Neural Network (VT-SNN) for rapid perception. The system demonstrates high accuracy in robotic tasks such as container classification and rotational slip detection, outperforming traditional deep learning methods. The research also contributes to the field by making visual-tactile datasets publicly available to foster further advancements. This work highlights the potential for creating intelligent, energy-efficient robotic systems. We review the latest advancements in memristor-based memory devices, their role in neuromorphic computing, and how they contribute to the development of advanced vision sensors. The potential of these technologies in revolutionizing artificial vision and their implications for future research and development are also discussed.","author":[{"family":"Su","given":"Yiping"},{"family":"Hwang","given":"Dajeong"},{"family":"Wang","given":"Bing"}],"issued":{"date-parts":[[2024]]},"DOI":"10.31219/osf.io/49qzv","URL":"https://doi.org/10.31219/osf.io/49qzv","source":"crossref"},{"id":"doi:10.1063/5.0201761","type":"article-journal","title":"Mechanical intelligence via fully reconfigurable elastic neuromorphic metasurfaces","abstract":"The ability of mechanical systems to perform basic computations has gained traction over recent years, providing an unconventional alternative to digital computing in off grid, low power, and severe environments, which render the majority of electronic components inoperable. However, much of the work in mechanical computing has focused on logic operations via quasi-static prescribed displacements in origami, bistable, and soft deformable matter. Here, we present a first attempt to describe the fundamental framework of an elastic neuromorphic metasurface that performs distinct classification tasks, providing a new set of challenges, given the complex nature of elastic waves with respect to scattering and manipulation. Multiple layers of reconfigurable waveguides are phase-trained via constant weights and trainable activation functions in a manner that enables the resultant wave scattering at the readout location to focus on the correct class within the detection plane. We further demonstrate the neuromorphic system’s reconfigurability in performing two distinct tasks, eliminating the need for costly remanufacturing.","author":[{"family":"Moghaddaszadeh","given":"M"},{"family":"Mousa","given":"M"},{"family":"Aref","given":"A"},{"family":"Nouh","given":"M"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1063/5.0201761","URL":"https://doi.org/10.1063/5.0201761","source":"crossref"},{"id":"doi:10.1007/s40820-024-01445-x","type":"article-journal","title":"Recent Advance in Synaptic Plasticity Modulation Techniques for Neuromorphic Applications","abstract":"Abstract Manipulating the expression of synaptic plasticity of neuromorphic devices provides fascinating opportunities to develop hardware platforms for artificial intelligence. However, great efforts have been devoted to exploring biomimetic mechanisms of plasticity simulation in the last few years. Recent progress in various plasticity modulation techniques has pushed the research of synaptic electronics from static plasticity simulation to dynamic plasticity modulation, improving the accuracy of neuromorphic computing and providing strategies for implementing neuromorphic sensing functions. Herein, several fascinating strategies for synaptic plasticity modulation through chemical techniques, device structure design, and physical signal sensing are reviewed. For chemical techniques, the underlying mechanisms for the modification of functional materials were clarified and its effect on the expression of synaptic plasticity was also highlighted. Based on device structure design, the reconfigurable operation of neuromorphic devices was well demonstrated to achieve programmable neuromorphic functions. Besides, integrating the sensory units with neuromorphic processing circuits paved a new way to achieve human-like intelligent perception under the modulation of physical signals such as light, strain, and temperature. Finally, considering that the relevant technology is still in the basic exploration stage, some prospects or development suggestions are put forward to promote the development of neuromorphic devices.","author":[{"family":"Sun","given":"Yilin"},{"family":"Wang","given":"Huaipeng"},{"family":"Xie","given":"Dan"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1007/s40820-024-01445-x","URL":"https://doi.org/10.1007/s40820-024-01445-x","source":"crossref"},{"id":"doi:10.1126/sciadv.adk9928","type":"article-journal","title":"Spatial evolution of the proton-coupled Mott transition in correlated oxides for neuromorphic computing","abstract":"The proton-electron coupling effect induces rich spectrums of electronic states in correlated oxides, opening tempting opportunities for exploring novel devices with multifunctions. Here, via modest Pt-aided hydrogen spillover at room temperature, amounts of protons are introduced into SmNiO 3 -based devices. In situ structural characterizations together with first-principles calculation reveal that the local Mott transition is reversibly driven by migration and redistribution of the predoped protons. The accompanying giant resistance change results in excellent memristive behaviors under ultralow electric fields. Hierarchical tree-like memory states, an instinct displayed in bio-synapses, are further realized in the devices by spatially varying the proton concentration with electric pulses, showing great promise in artificial neural networks for solving intricate problems. Our research demonstrates the direct and effective control of proton evolution using extremely low electric field, offering an alternative pathway for modifying the functionalities of correlated oxides and constructing low–power consumption intelligent devices and neural network circuits.","author":[{"family":"Deng","given":"Xing"},{"family":"Liu","given":"Yu"},{"family":"Yang","given":"Zhen"},{"family":"Zhao","given":"Yi"},{"family":"Xu","given":"Ya"},{"family":"Fu","given":"Meng"},{"family":"Shen","given":"Yu"},{"family":"Qu","given":"Ke"},{"family":"Guan","given":"Zhao"},{"family":"Tong","given":"Wen"},{"family":"Zhang","given":"Yuan"},{"family":"Chen","given":"Bin"},{"family":"Zhong","given":"Ni"},{"family":"Xiang","given":"Ping"},{"family":"Duan","given":"Chun"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1126/sciadv.adk9928","URL":"https://doi.org/10.1126/sciadv.adk9928","source":"crossref"},{"id":"doi:10.1063/5.0205429","type":"article-journal","title":"Double perovskite Bi2FeMnO6/TiO2 thin film heterostructure device for neuromorphic computing","abstract":"Multiferroic materials have important research significance in the fields of magnetic random-access memory, ferroelectric random-access memory, resistive random-access memory, and neuromorphic computing devices due to their excellent and diverse physical properties. In this work, a solution of Bi2FeMnO6 was prepared using a solution-based method, and an Au/Bi2FeMnO6/TiO2 heterostructure device was fabricated on a Si substrate. X-ray diffraction and transmission electron microscopy data indicate that the Bi2FeMnO6 films have hexagonal R3c symmetry structures. The Bi2FeMnO6 film exhibits ferroelectricity with a fine remanent polarization. In addition, the Bi2FeMnO6-based devices have excellent switching ratios of 6.37 × 105. A larger switching ratio can provide a multi-resistance state for the device, which is beneficial for the simulation of synapses. Hence, it effectively emulates excitatory postsynaptic currents, paired-pulse facilitation, and long-term plasticity of synapses and achieves recognition accuracy of 95% in neuromorphic computing. We report a promising material for the development of various nonvolatile memories and neuromorphic synaptic devices.","author":[{"family":"Li","given":"Dong"},{"family":"Zhong","given":"Wen"},{"family":"Tang","given":"Xin"},{"family":"He","given":"Qin"},{"family":"Jiang","given":"Yan"},{"family":"Liu","given":"Qiu"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1063/5.0205429","URL":"https://doi.org/10.1063/5.0205429","source":"crossref"},{"id":"doi:10.3390/electronics13163203","type":"article-journal","title":"Unsupervised Classification of Spike Patterns with the Loihi Neuromorphic Processor","abstract":"A long-standing research goal is to develop computing technologies that mimic the brain’s capabilities by implementing computation in electronic systems directly inspired by its structure, function, and operational mechanisms, using low-power, spike-based neural networks. The Loihi neuromorphic processor provides a low-power, large-scale network of programmable silicon neurons for brain-inspired artificial intelligence applications. This paper exploits the Loihi processors and a theory-guided methodology to enable unsupervised learning of spike patterns. Our method ensures efficient and rapid selection of the network’s hyperparameters, enabling the neuromorphic processor to generate attractor states through real-time unsupervised learning. Precisely, we follow a fast design process in which we fine-tune network parameters using mean-field theory. Moreover, we measure the network’s learning ability regarding its error correction and pattern completion aptitude. Finally, we observe the dynamic energy consumption of the neuron cores for each millisecond of simulation equal to 23 μJ/time step during the learning and recall phase for four attractors composed of 512 excitatory neurons and 256 shared inhibitory neurons. This study showcases how large-scale, low-power digital neuromorphic processors can be quickly programmed to enable the autonomous generation of attractor states. These attractors are fundamental computational primitives that theoretical analysis and experimental evidence indicate as versatile and reusable components suitable for a wide range of cognitive tasks.","author":[{"family":"Matsuo","given":"Ryoga"},{"family":"Elgaradiny","given":"Ahmed"},{"family":"Corradi","given":"Federico"}],"issued":{"date-parts":[[2024]]},"DOI":"10.3390/electronics13163203","URL":"https://doi.org/10.3390/electronics13163203","source":"crossref"},{"id":"doi:10.1088/1361-6528/ad2ee3","type":"article-journal","title":"Flexible light-stimulated artificial synapse based on detached (In,Ga)N thin film for neuromorphic computing","abstract":"Abstract Because of wide range of applications, the flexible artificial synapse is an indispensable part for next-generation neural morphology computing. In this work, we demonstrate a flexible synaptic device based on a lift-off (In,Ga)N thin film successfully. The synaptic device can mimic the learning, forgetting, and relearning functions of biological synapses at both flat and bent states. Furthermore, the synaptic device can simulate the transition from short-term memory to long-term memory successfully under different bending conditions. With the high flexibility, the excitatory post-synaptic current of the bent device only shows a slight decrease, leading to the high stability. Based on the experimental conductance for long-term potentiation and depression, the simulated three-layer neural network can achieve a high recognition rate up to 90.2%, indicating that the system comprising of flexible synaptic devices could have a strong learning-memory capability. Therefore, this work has a great potential for the development of wearable intelligence devices and flexible neuromorphic systems.","author":[{"family":"Zhang","given":"Qianyi"},{"family":"Hou","given":"Binbin"},{"family":"Zhang","given":"Jianya"},{"family":"Gu","given":"Xiushuo"},{"family":"Huang","given":"Yonglin"},{"family":"Pei","given":"Renjun"},{"family":"Zhao","given":"Yukun"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1088/1361-6528/ad2ee3","URL":"https://doi.org/10.1088/1361-6528/ad2ee3","source":"crossref"},{"id":"doi:10.1088/2634-4386/ad3a96","type":"article-journal","title":"An organic artificial soma for spatio-temporal pattern recognition via dendritic integration","abstract":"Abstract A novel organic neuromorphic device performing pattern classification is presented and demonstrated. It features an artificial soma capable of dendritic integration from three pre-synaptic neurons. The time-response of the interface between electrolytic solutions and organic mixed ionic-electronic conductors is proposed as the sole computational feature for pattern recognition, and it is easily tuned in the organic dendritic integrator by simply controlling electrolyte ionic strength. The classifier is benchmarked in speech-recognition experiments, with a sample of 14 words, encoded either from audio tracks or from kinematic data, showing excellent discrimination performances in a planar, miniaturizable, fully passive device, designed to be promptly integrated in more complex architectures where on-board pattern classification is required.","author":[{"family":"Lauro","given":"Michele"},{"family":"Rondelli","given":"Federico"},{"family":"Salvo","given":"Anna"},{"family":"Corsini","given":"Alessandro"},{"family":"Genitoni","given":"Matteo"},{"family":"Greco","given":"Pierpaolo"},{"family":"Murgia","given":"Mauro"},{"family":"Fadiga","given":"Luciano"},{"family":"Biscarini","given":"Fabio"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1088/2634-4386/ad3a96","URL":"https://doi.org/10.1088/2634-4386/ad3a96","source":"crossref"},{"id":"doi:10.1002/marc.202400172","type":"article-journal","title":"Building Uniformly Structured Polymer Memristors via a 2D Conjugation Strategy for Neuromorphic Computing","abstract":"Abstract Polymer memristors represent a highly promising avenue for the advancement of next‐generation computing systems. However, the intrinsic structural heterogeneity characteristic of most polymers often results in organic polymer memristors displaying erratic resistive switching phenomena, which in turn lead to diminished production yields and compromised reliability. In this study, a 2D conjugated polymer, named PBDTT‐BPQTPA, is synthesized by integrating the coplanar bis (thiophene)‐4,8‐dihydrobenzo[1,2‐b:4,5‐b]dithiophene (BDTT) as an electron‐donating unit with a quinoxaline derivative serving as an electron‐accepting unit. The incorporation of triphenylamine groups at the quinoxaline termini significantly enhances the polymer's conjugation and planarity, thereby facilitating more efficient charge transport. The fabricated polymer memristor with the structure of Al/PBDTT‐BPQTPA/ITO exhibits typical non‐volatile resistive switching behavior under high voltage conditions, along with history‐dependent memristive properties at lower voltages. The unique memristive behavior of the device enables the simulation of synaptic enhancement/inhibition, learning algorithms, and memory operations. Additionally, the memristor demonstrates its capability for executing logical operations and handling decimal calculations. This study offers a promising and innovative approach for the development of artificial neuromorphic computing systems.","author":[{"family":"Li","given":"Jinyong"},{"family":"Fan","given":"Fei"},{"family":"Fu","given":"Xin"},{"family":"Liu","given":"Mingxing"},{"family":"Chen","given":"Yu"},{"family":"Zhang","given":"Bin"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1002/marc.202400172","URL":"https://doi.org/10.1002/marc.202400172","source":"crossref"},{"id":"doi:10.1002/aelm.202400347","type":"article-journal","title":"Self‐Selective Crossbar Synapse Array with n‐ZnO/p‐NiO<sub>x</sub>/n‐ZnO Structure for Neuromorphic Computing","abstract":"Abstract Artificial synapse devices are essential elements for highly energy‐efficient neuromorphic computing. They are implemented as crossbar array architecture, where highly selective synaptic weight updates for training and sneak leakage‐free inference operations are required. In this study, self‐selective bipolar artificial synapse device is proposed with n‐ZnO/p‐NiO x /n‐ZnO heterojunction, and its analog synapse operation with high selectivity is demonstrated in 32 × 32 crossbar array architecture without the aid of selector devices. The built‐in potential barrier at p‐NiO x /n‐ZnO junction and the Zener tunneling effect provided nonlinear current–voltage characteristics at both voltage polarities for self‐selecting function for synaptic potentiation and depression operations. Voltage‐driven redistribution of oxygen ions inside n–p–n oxide structure, evidenced by x‐ray photoelectron spectroscopy, modulated the distribution of oxygen vacancies in the layers and consequent conductance in an analog manner for the synaptic weight update operation. It demonstrates that the proposed n–p–n oxide device is a promising artificial synapse device implementing self‐selectivity and analog synaptic weight update in a crossbar array architecture for neuromorphic computing.","author":[{"family":"Chung","given":"Peter"},{"family":"Ryu","given":"Jiyeon"},{"family":"Seo","given":"Daejae"},{"family":"Sahu","given":"Dwipak"},{"family":"Song","given":"Minju"},{"family":"Kim","given":"Junghwan"},{"family":"Yoon","given":"Tae‐sik"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1002/aelm.202400347","URL":"https://doi.org/10.1002/aelm.202400347","source":"crossref"},{"id":"doi:10.1038/s41467-024-52259-9","type":"article-journal","title":"Neuromorphic intermediate representation: A unified instruction set for interoperable brain-inspired computing","abstract":"Abstract Spiking neural networks and neuromorphic hardware platforms that simulate neuronal dynamics are getting wide attention and are being applied to many relevant problems using Machine Learning. Despite a well-established mathematical foundation for neural dynamics, there exists numerous software and hardware solutions and stacks whose variability makes it difficult to reproduce findings. Here, we establish a common reference frame for computations in digital neuromorphic systems, titled Neuromorphic Intermediate Representation (NIR). NIR defines a set of computational and composable model primitives as hybrid systems combining continuous-time dynamics and discrete events. By abstracting away assumptions around discretization and hardware constraints, NIR faithfully captures the computational model, while bridging differences between the evaluated implementation and the underlying mathematical formalism. NIR supports an unprecedented number of neuromorphic systems, which we demonstrate by reproducing three spiking neural network models of different complexity across 7 neuromorphic simulators and 4 digital hardware platforms. NIR decouples the development of neuromorphic hardware and software, enabling interoperability between platforms and improving accessibility to multiple neuromorphic technologies. We believe that NIR is a key next step in brain-inspired hardware-software co-evolution, enabling research towards the implementation of energy efficient computational principles of nervous systems. NIR is available at neuroir.org","author":[{"family":"Pedersen","given":"Jens"},{"family":"Abreu","given":"Steven"},{"family":"Jobst","given":"Matthias"},{"family":"Lenz","given":"Gregor"},{"family":"Fra","given":"Vittorio"},{"family":"Bauer","given":"Felix"},{"family":"Muir","given":"Dylan"},{"family":"Zhou","given":"Peng"},{"family":"Vogginger","given":"Bernhard"},{"family":"Heckel","given":"Kade"},{"family":"Urgese","given":"Gianvito"},{"family":"Shankar","given":"Sadasivan"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1038/s41467-024-52259-9","URL":"https://doi.org/10.1038/s41467-024-52259-9","source":"europepmc"},{"id":"doi:10.21203/rs.3.rs-3412574/v1","type":"article-journal","title":"Recurrent models of orientation selectivity enable robust early-vision  processing in mixed-signal neuromorphic hardware","abstract":"Abstract Mixed signal analog/digital neuromorphic circuits represent an ideal medium for reproducing the dynamics of biological neural systems in real-time with bio-physically realistic dynamics. However, similar to their biological counterparts, these circuits have limited resolution and are affected by a high degree of variability. Considering this, we developed a recurrent spiking neural network that implements a faithful model of the retinocortical visual pathway to reliably produce Gabor-like receptive fields tuned to visual stimuli with specific orientation and spatial frequency properties. Specifically, we developed a neuromorphic visual system comprising a Dynamic Vision Sensor that emulates the transient pathway of real retinas and a mixed-signal Dynamic Neuromorphic Asynchronous Processor with adaptive exponential integrate-and-fire neurons and dynamic synapses and mapped the recurrent network model on it to produces the desired orientation and spatial frequency tuning responses. Compared to alternative feed-forward schemes, the model developed gives rise to robust highly structured Gabor-like receptive fields of any phase symmetry, optimizing the hardware resources available in terms of synaptic connections. We present experimental results using both synthetic and natural images validating the model with its hardware implementations.","author":[{"family":"Sabatini","given":"Silvio"},{"family":"Baruzzi","given":"Valentina"},{"family":"Indiveri","given":"Giacomo"}],"issued":{"date-parts":[[2023]]},"DOI":"10.21203/rs.3.rs-3412574/v1","URL":"https://doi.org/10.21203/rs.3.rs-3412574/v1","source":"crossref"},{"id":"doi:10.1101/2023.11.14.567028","type":"article-journal","title":"Calibrating Bayesian decoders of neural spiking activity","abstract":"Abstract Accurately decoding external variables from observations of neural activity is a major challenge in systems neuroscience. Bayesian decoders, that provide probabilistic estimates, are some of the most widely used. Here we show how, in many common settings, the probabilistic predictions made by traditional Bayesian decoders are overconfident. That is, the estimates for the decoded stimulus or movement variables are more certain than they should be. We then show how Bayesian decoding with latent variables, taking account of low-dimensional shared variability in the observations, can improve calibration, although additional correction for overconfidence is still needed. We examine: 1) decoding the direction of grating stimuli from spike recordings in primary visual cortex in monkeys, 2) decoding movement direction from recordings in primary motor cortex in monkeys, 3) decoding natural images from multi-region recordings in mice, and 4) decoding position from hippocampal recordings in rats. For each setting we characterize the overconfidence, and we describe a possible method to correct miscalibration post-hoc. Properly calibrated Bayesian decoders may alter theoretical results on probabilistic population coding and lead to brain machine interfaces that more accurately reflect confidence levels when identifying external variables. Significance Statement Bayesian decoding is a statistical technique for making probabilistic predictions about external stimuli or movements based on recordings of neural activity. These predictions may be useful for robust brain machine interfaces or for understanding perceptual or behavioral confidence. However, the probabilities produced by these models do not always match the observed outcomes. Just as a weather forecast predicting a 50% chance of rain may not accurately correspond to an outcome of rain 50% of the time, Bayesian decoders of neural activity can be miscalibrated as well. Here we identify and measure miscalibration of Bayesian decoders for neural spiking activity in a range of experimental settings. We compare multiple statistical models and demonstrate how overconfidence can be corrected.","author":[{"family":"Wei","given":"Ganchao"},{"family":"Mansouri","given":"Zeinab"},{"family":"Wang","given":"Xiaojing"},{"family":"Stevenson","given":"Ian"}],"issued":{"date-parts":[[2023]]},"DOI":"10.1101/2023.11.14.567028","URL":"https://doi.org/10.1101/2023.11.14.567028","source":"crossref"},{"id":"doi:10.20944/preprints202408.0704.v1","type":"manuscript","title":"Noise as an Optimization Tool for Spiking Neural Networks","abstract":"Standard ANNs lack flexibility when handling corrupted input due to their fixed structure. Spiking Neural Networks (SNNs) can utilize biological temporal coding features, such as noise-induced stochastic resonance and dynamical synapses to increase a model’s performance when its parameters are not optimized for a given input. Using the analog XOR task as a simplified Convolutional Neural Network analog, this paper demonstrates two key results: (1) SNNs solve the problem linearly inseparable in ANN with over 10% improvement in the optimal setting, and (2) in SNNs, the synaptic noise and dynamical synapses compensate for non-optimal parameters, achieving near-optimal results.","author":[{"family":"Garipova","given":"Yana"},{"family":"Yonekura","given":"Shogo"},{"family":"Kuniyoshi","given":"Yasuo"}],"issued":{"date-parts":[[2024]]},"DOI":"10.20944/preprints202408.0704.v1","URL":"https://doi.org/10.20944/preprints202408.0704.v1","source":"preprints"},{"id":"doi:10.1101/2024.07.19.604252","type":"article-journal","title":"Spiking neural network models of sound localisation via a massively collaborative process","abstract":"Abstract Neuroscientists are increasingly initiating large-scale collaborations which bring together tens to hundreds of researchers. However, while these projects represent a step-change in scale, they retain a traditional structure with centralised funding, participating laboratories and data sharing on publication. Inspired by an open-source project in pure mathematics, we set out to test the feasibility of an alternative structure by running a grassroots, massively collaborative project in computational neuroscience. To do so, we launched a public Git repository, with code for training spiking neural networks to solve a sound localisation task via surrogate gradient descent. We then invited anyone, anywhere to use this code as a springboard for exploring questions of interest to them, and encouraged participants to share their work both asynchro-nously through Git and synchronously at monthly online workshops. At a scientific level, our work investigated how a range of biologically-relevant parameters, from time delays to mem-brane time constants and levels of inhibition, could impact sound localisation in networks of spiking units. At a more macro-level, our project brought together 31 researchers from multiple countries, provided hands-on research experience to early career participants, and opportunities for supervision and teaching to later career participants. Looking ahead, our project provides a glimpse of what open, collaborative science could look like and provides a necessary, tentative step towards it.","author":[{"family":"Ghosh","given":"Marcus"},{"family":"Habashy","given":"Karim"},{"family":"Santis","given":"Francesco"},{"family":"Fiers","given":"Tomas"},{"family":"Erçelik","given":"Dilay"},{"family":"Mészáros","given":"Balázs"},{"family":"Friedenberger","given":"Zachary"},{"family":"Béna","given":"Gabriel"},{"family":"Hong","given":"Mingxuan"},{"family":"Abubacar","given":"Umar"},{"family":"Byrne","given":"Rory"},{"family":"Riquelme","given":"Juan"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1101/2024.07.19.604252","URL":"https://doi.org/10.1101/2024.07.19.604252","source":"europepmc"},{"id":"doi:10.1371/journal.pcbi.1011913","type":"article-journal","title":"Estimating orientation in natural scenes: A spiking neural network model of the insect central complex.","abstract":"The central complex of insects contains cells, organised as a ring attractor, that encode head direction. The ‘bump’ of activity in the ring can be updated by idiothetic cues and external sensory information. Plasticity at the synapses between these cells and the ring neurons, that are responsible for bringing sensory information into the central complex, has been proposed to form a mapping between visual cues and the heading estimate which allows for more accurate tracking of the current heading, than if only idiothetic information were used. In Drosophila , ring neurons have well characterised non-linear receptive fields. In this work we produce synthetic versions of these visual receptive fields using a combination of excitatory inputs and mutual inhibition between ring neurons. We use these receptive fields to bring visual information into a spiking neural network model of the insect central complex based on the recently published Drosophila connectome. Previous modelling work has focused on how this circuit functions as a ring attractor using the same type of simple visual cues commonly used experimentally. While we initially test the model on these simple stimuli, we then go on to apply the model to complex natural scenes containing multiple conflicting cues. We show that this simple visual filtering provided by the ring neurons is sufficient to form a mapping between heading and visual features and maintain the heading estimate in the absence of angular velocity input. The network is successful at tracking heading even when presented with videos of natural scenes containing conflicting information from environmental changes and translation of the camera.","author":[{"family":"Stentiford","given":"Rachael"},{"family":"Knight","given":"James"},{"family":"Nowotny","given":"Thomas"},{"family":"Philippides","given":"Andrew"},{"family":"Graham","given":"Paul"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1371/journal.pcbi.1011913","URL":"https://doi.org/10.1371/journal.pcbi.1011913","source":"europepmc"},{"id":"doi:10.1101/2024.12.05.627100","type":"article-journal","title":"Emergence of Sparse Coding, Balance and Decorrelation from a Biologically-Grounded Spiking Neural Network Model of Learning in the Primary Visual Cortex","abstract":"1 Abstract Many computational studies attempt to address the question of information representation in biological neural networks using an explicit optimization based on an objective function. These approaches begin with principles of information representation that are expected to be found in the network and from which learning rules can be derived. This study approaches the question from the opposite direction; beginning with a model built upon the experimentally observed properties of neural responses, homeostasis, and synaptic plasticity. The known properties of information representation are then expected to emerge from this substrate. A spiking neural model of the primary visual cortex (V1) was investigated. Populations of both inhibitory and excitatory leaky integrate-and-fire neurons with recurrent connections were provided with spiking input from simulated ON and OFF neurons of the lateral geniculate nucleus. This network was provided with natural image stimuli as input. All synapses underwent learning using spike-timing-dependent plasticity learning rules. A homeostatic rule adjusted the weights and thresholds of each neuron based on target homeostatic spiking rates and mean synaptic input values. These experimentally grounded rules resulted in a number of the expected properties of information representation. The network showed a temporally sparse spike response to inputs and this was associated with a sparse code with Gabor-like receptive fields. The network was balanced at both slow and fast time scales; increased excitatory input was balanced by increased inhibition. This balance was associated with decorrelated firing that was observed as population sparseness. This population sparseness was both the cause and result of the decorrelation of receptive fields. These observed emergent properties (balance, temporal sparseness, population sparseness, and decorrelation) indicate that the network is implementing expected principles of information processing: efficient coding, information maximization (’infomax’), and a lateral or single-layer form of predictive coding. These emergent features of the network were shown to be robust to randomized jitter of the values of key simulation parameters.","author":[{"family":"Ruslim","given":"Marko"},{"family":"Spencer","given":"Martin"},{"family":"Hogendoorn","given":"Hinze"},{"family":"Meffin","given":"Hamish"},{"family":"Lian","given":"Yanbo"},{"family":"Burkitt","given":"Anthony"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1101/2024.12.05.627100","URL":"https://doi.org/10.1101/2024.12.05.627100","source":"europepmc"},{"id":"doi:10.3389/fphys.2024.1379977","type":"article-journal","title":"Investigating visual navigation using spiking neural network models of the insect mushroom bodies.","abstract":"Ants are capable of learning long visually guided foraging routes with limited neural resources. The visual scene memory needed for this behaviour is mediated by the mushroom bodies; an insect brain region important for learning and memory. In a visual navigation context, the mushroom bodies are theorised to act as familiarity detectors, guiding ants to views that are similar to those previously learned when first travelling along a foraging route. Evidence from behavioural experiments, computational studies and brain lesions all support this idea. Here we further investigate the role of mushroom bodies in visual navigation with a spiking neural network model learning complex natural scenes. By implementing these networks in GeNN–a library for building GPU accelerated spiking neural networks–we were able to test these models offline on an image database representing navigation through a complex outdoor natural environment, and also online embodied on a robot. The mushroom body model successfully learnt a large series of visual scenes (400 scenes corresponding to a 27 m route) and used these memories to choose accurate heading directions during route recapitulation in both complex environments. Through analysing our model’s Kenyon cell (KC) activity, we were able to demonstrate that KC activity is directly related to the respective novelty of input images. Through conducting a parameter search we found that there is a non-linear dependence between optimal KC to visual projection neuron (VPN) connection sparsity and the length of time the model is presented with an image stimulus. The parameter search also showed training the model on lower proportions of a route generally produced better accuracy when testing on the entire route. We embodied the mushroom body model and comparator visual navigation algorithms on a Quanser Q-car robot with all processing running on an Nvidia Jetson TX2. On a 6.5 m route, the mushroom body model had a mean distance to training route (error) of 0.144 ± 0.088 m over 5 trials, which was performance comparable to standard visual-only navigation algorithms. Thus, we have demonstrated that a biologically plausible model of the ant mushroom body can navigate complex environments both in simulation and the real world. Understanding the neural basis of this behaviour will provide insight into how neural circuits are tuned to rapidly learn behaviourally relevant information from complex environments and provide inspiration for creating bio-mimetic computer/robotic systems that can learn rapidly with low energy requirements.","author":[{"family":"Jesusanmi","given":"Oluwaseyi"},{"family":"Amin","given":"Amany"},{"family":"Domcsek","given":"Norbert"},{"family":"Knight","given":"James"},{"family":"Philippides","given":"Andrew"},{"family":"Nowotny","given":"Thomas"},{"family":"Graham","given":"Paul"}],"issued":{"date-parts":[[2024]]},"DOI":"10.3389/fphys.2024.1379977","URL":"https://doi.org/10.3389/fphys.2024.1379977","source":"europepmc"},{"id":"doi:10.3389/fnins.2023.1303564","type":"article-journal","title":"Efficient and generalizable cross-patient epileptic seizure detection through a spiking neural network.","abstract":"Introduction Epilepsy is a global chronic disease that brings pain and inconvenience to patients, and an electroencephalogram (EEG) is the main analytical tool. For clinical aid that can be applied to any patient, an automatic cross-patient epilepsy seizure detection algorithm is of great significance. Spiking neural networks (SNNs) are modeled on biological neurons and are energy-efficient on neuromorphic hardware, which can be expected to better handle brain signals and benefit real-world, low-power applications. However, automatic epilepsy seizure detection rarely considers SNNs. Methods In this article, we have explored SNNs for cross-patient seizure detection and discovered that SNNs can achieve comparable state-of-the-art performance or a performance that is even better than artificial neural networks (ANNs). We propose an EEG-based spiking neural network (EESNN) with a recurrent spiking convolution structure, which may better take advantage of temporal and biological characteristics in EEG signals. Results We extensively evaluate the performance of different SNN structures, training methods, and time settings, which builds a solid basis for understanding and evaluation of SNNs in seizure detection. Moreover, we show that our EESNN model can achieve energy reduction by several orders of magnitude compared with ANNs according to the theoretical estimation. Discussion These results show the potential for building high-performance, low-power neuromorphic systems for seizure detection and also broaden real-world application scenarios of SNNs.","author":[{"family":"Zhang","given":"Zongpeng"},{"family":"Xiao","given":"Mingqing"},{"family":"Ji","given":"Taoyun"},{"family":"Jiang","given":"Yuwu"},{"family":"Lin","given":"Tong"},{"family":"Zhou","given":"Xiaohua"},{"family":"Lin","given":"Zhouchen"}],"issued":{"date-parts":[[2023]]},"DOI":"10.3389/fnins.2023.1303564","URL":"https://doi.org/10.3389/fnins.2023.1303564","source":"europepmc"},{"id":"doi:10.1038/s41467-024-48905-x","type":"article-journal","title":"BiœmuS: A new tool for neurological disorders studies through real-time emulation and hybridization using biomimetic Spiking Neural Network.","abstract":"Abstract Characterization and modeling of biological neural networks has emerged as a field driving significant advancements in our understanding of brain function and related pathologies. As of today, pharmacological treatments for neurological disorders remain limited, pushing the exploration of promising alternative approaches such as electroceutics. Recent research in bioelectronics and neuromorphic engineering have fostered the development of the new generation of neuroprostheses for brain repair. However, achieving their full potential necessitates a deeper understanding of biohybrid interaction. In this study, we present a novel real-time, biomimetic, cost-effective and user-friendly neural network capable of real-time emulation for biohybrid experiments. Our system facilitates the investigation and replication of biophysically detailed neural network dynamics while prioritizing cost-efficiency, flexibility and ease of use. We showcase the feasibility of conducting biohybrid experiments using standard biophysical interfaces and a variety of biological cells as well as real-time emulation of diverse network configurations. We envision our system as a crucial step towards the development of neuromorphic-based neuroprostheses for bioelectrical therapeutics, enabling seamless communication with biological networks on a comparable timescale. Its embedded real-time functionality enhances practicality and accessibility, amplifying its potential for real-world applications in biohybrid experiments.","author":[{"family":"Beaubois","given":"Romain"},{"family":"Cheslet","given":"Jérémy"},{"family":"Duenki","given":"Tomoya"},{"family":"Venuto","given":"Giuseppe"},{"family":"Carè","given":"Marta"},{"family":"Khoyratee","given":"Farad"},{"family":"Chiappalone","given":"Michela"},{"family":"Branchereau","given":"Pascal"},{"family":"Ikeuchi","given":"Yoshiho"},{"family":"Levi","given":"Timothée"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1038/s41467-024-48905-x","URL":"https://doi.org/10.1038/s41467-024-48905-x","source":"europepmc"},{"id":"doi:10.7554/elife.90597","type":"article-journal","title":"Hippocampome.org 2.0 is a knowledge base enabling data-driven spiking neural network simulations of rodent hippocampal circuits.","abstract":"Hippocampome.org is a mature open-access knowledge base of the rodent hippocampal formation focusing on neuron types and their properties. Previously, Hippocampome.org v1.0 established a foundational classification system identifying 122 hippocampal neuron types based on their axonal and dendritic morphologies, main neurotransmitter, membrane biophysics, and molecular expression (Wheeler et al., 2015). Releases v1.1 through v1.12 furthered the aggregation of literature-mined data, including among others neuron counts, spiking patterns, synaptic physiology, in vivo firing phases, and connection probabilities. Those additional properties increased the online information content of this public resource over 100-fold, enabling numerous independent discoveries by the scientific community. Hippocampome.org v2.0, introduced here, besides incorporating over 50 new neuron types, now recenters its focus on extending the functionality to build real-scale, biologically detailed, data-driven computational simulations. In all cases, the freely downloadable model parameters are directly linked to the specific peer-reviewed empirical evidence from which they were derived. Possible research applications include quantitative, multiscale analyses of circuit connectivity and spiking neural network simulations of activity dynamics. These advances can help generate precise, experimentally testable hypotheses and shed light on the neural mechanisms underlying associative memory and spatial navigation.","author":[{"family":"Wheeler","given":"Diek"},{"family":"Kopsick","given":"Jeffrey"},{"family":"Sutton","given":"Nate"},{"family":"Tecuatl","given":"Carolina"},{"family":"Komendantov","given":"Alexander"},{"family":"Nadella","given":"Kasturi"},{"family":"Ascoli","given":"Giorgio"}],"issued":{"date-parts":[[2023]]},"DOI":"10.7554/elife.90597","URL":"https://doi.org/10.7554/elife.90597","source":"europepmc"},{"id":"doi:10.21203/rs.3.rs-3591328/v1","type":"article-journal","title":"Robust compression and detection of epileptiform patterns in ECoG using a real-time spiking neural network hardware framework","abstract":"Abstract Interictal Epileptiform Discharges (IED) and High Frequency Oscillations (HFO) in intraoperative electrocorticography (ECoG) may guide the surgeon by delineating the epileptogenic zone. We designed a modular spiking neural network (SNN) in a mixed-signal neuromorphic device to process the ECoG in real-time. We exploit the variability of the inhomogeneous silicon neurons to achieve efficient sparse and de-correlated temporal signal encoding. We interface the full-custom SNN device to the BCI2000 real-time framework and configure the setup to detect HFO and IED co-occurring with HFO (IED-HFO). We validate the setup on pre-recorded data and obtain HFO rates that are concordant with a previously validated offline algorithm (Spearman’s ρ = 0.75, p = 1e-4), achieving the same postsurgical seizure freedom predictions for all patients. In a remote on-line analysis, intraoperative ECoG recorded in Utrecht was compressed and transferred to Zurich for SNN processing and successful IED-HFO detection in real-time. These results further demonstrate how automated remote real-time detection may enable the use of HFO in clinical practice.","author":[{"family":"Costa","given":"Filippo"},{"family":"Schaft","given":"Eline"},{"family":"Huiskamp","given":"Geertjan"},{"family":"Aarnoutse","given":"Erik"},{"family":"Klooster","given":"Maryse"},{"family":"Krayenbühl","given":"Niklaus"},{"family":"Ramantani","given":"Georgia"},{"family":"Zijlmans","given":"Maeike"},{"family":"Indiveri","given":"Giacomo"},{"family":"Sarnthein","given":"Johannes"}],"issued":{"date-parts":[[2023]]},"DOI":"10.21203/rs.3.rs-3591328/v1","URL":"https://doi.org/10.21203/rs.3.rs-3591328/v1","source":"europepmc"},{"id":"doi:10.6078/d1fm6s","type":"article-journal","title":"Data for: Hybrid dedicated and distributed coding in PMd/M1 provides separation and interaction of bilateral arm signals","abstract":"Pronounced activity is observed in both hemispheres of the motor cortex during preparation and execution of unimanual movements. The organizational principles of bi-hemispheric signals and the functions they serve throughout motor planning remain unclear. Using an instructed-delay reaching task in monkeys, we identified two components in population responses spanning PMd and M1. A “dedicated” component, which segregated activity at the level of individual units, emerged in PMd during preparation. It was most prominent following movement when M1 became strongly engaged, and principally involved the contralateral hemisphere. In contrast to recent reports, these dedicated signals solely accounted for divergence of arm-specific neural subspaces. The other “distributed” component mixed signals for each arm within units, and the subspace containing it did not discriminate between arms at any stage. The statistics of the population response suggest two functional aspects of the cortical network: one that spans both hemispheres for supporting preparatory and ongoing processes, and another that is predominantly housed in the contralateral hemisphere and specifies unilateral output.","author":[{"family":"Dixon","given":"Tanner"},{"family":"Merrick","given":"Christina"},{"family":"Wallis","given":"Joni"},{"family":"Ivry","given":"Richard"},{"family":"Carmena","given":"Jose"}],"issued":{"date-parts":[[2021]]},"DOI":"10.6078/d1fm6s","URL":"https://doi.org/10.6078/d1fm6s","source":"datacite"},{"id":"doi:10.5281/zenodo.18306877","type":"article-journal","title":"Adaptive Neural Continuity Protocol: Real-Time Compensation for Progressive Hippocampal Neurodegeneration","abstract":"This research presents the Continuity Protocol, a computational framework designed to preserve cognitive information integrity during progressive hippocampal decay. Using a 1,500-neuron spiking neural network in the Nengo environment, we demonstrate real-time signal restoration with \\approx 1.0 accuracy. The dataset includes: Full Technical Paper (PDF). Source Code (Python/Nengo). Simulation Video showing the 29s terminal decay cycle.","author":[{"family":"Farag","given":"Mina"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18306877","URL":"https://doi.org/10.5281/zenodo.18306877","source":"datacite"},{"id":"doi:10.5281/zenodo.18306876","type":"article-journal","title":"Adaptive Neural Continuity Protocol: Real-Time Compensation for Progressive Hippocampal Neurodegeneration","abstract":"This research presents the Continuity Protocol, a computational framework designed to preserve cognitive information integrity during progressive hippocampal decay. Using a 1,500-neuron spiking neural network in the Nengo environment, we demonstrate real-time signal restoration with \\approx 1.0 accuracy. The dataset includes: Full Technical Paper (PDF). Source Code (Python/Nengo). Simulation Video showing the 29s terminal decay cycle.","author":[{"family":"Farag","given":"Mina"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18306876","URL":"https://doi.org/10.5281/zenodo.18306876","source":"datacite"},{"id":"doi:10.5281/zenodo.18288582","type":"article-journal","title":"Hybrid Spiking Language Model: Combining Spike Counts and Membrane Potentials for Energy-Efficient and Noise-Robust Character Prediction","abstract":"I propose a novel character-level language model using Spiking Neural Networks (SNNs) that combines both spike counts and membrane potentials for output prediction. Unlike conventional SNN approaches that only use spike counts, this hybrid method leverages the analog information contained in membrane potentials, inspired by biological decision-making processes. Key Findings:- SNN is 42× more energy-efficient than DNN- SNN shows no accuracy degradation at 30% input noise- Hybrid (Spike + Membrane) approach improves perplexity by 9× Source code: https://github.com/hafufu-stack/snn-language-model","author":[{"family":"Funasaki","given":"Hiroto"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18288582","URL":"https://doi.org/10.5281/zenodo.18288582","source":"datacite"},{"id":"doi:10.21203/rs.3.rs-2558516/v1","type":"article-journal","title":"Reconfigurable, non-volatile neuromorphic photovoltaics","abstract":"Abstract Reconfigurable image sensors for the recognition and understanding of real-world objects are now becoming an essential part of machine vision technology. The neural network image sensor — which mimics neurobiological functions of the human retina —has recently been demonstrated to simultaneously sense and process optical images. However, highly tunable responsivity concurrently with non-volatile storage of image data in the neural network would allow a transformative leap in compactness and function of these artificial neural networks (ANNs) that truly function like a human retina. Here, we demonstrate a reconfigurable and non-volatile neuromorphic device based on two-dimensional (2D) semiconducting metal sulfides (MoS 2 and WS 2 ) that is concurrently a photovoltaic detector. The device is based on a metal/semiconductor/metal (M/S/M) two-terminal structure with pulse-tunable sulfur vacancies at the M/S junctions. By modulating sulfur vacancy concentrations, the polarities of short-circuit photocurrent —can be changed with multiple stable magnitudes. Device characterizations and modeling reveal that the bias-induced motion of sulfur vacancies leads to highly reconfigurable responsivities by dynamically modulating the Schottky barriers. A convolutional neuromorphic network (CNN) is finally designed for image process and object detection using the same device. The results demonstrated the two-terminal reconfigurable and non-volatile photodetectors can be used for future optoelectronics devices based on coupled Ionic-optical-electronic effects for Neuromorphic computing.","author":[{"family":"Li","given":"Tangxin"},{"family":"Miao","given":"Jinshui"},{"family":"Fu","given":"Xiao"},{"family":"Song","given":"Bo"},{"family":"Cai","given":"Bin"},{"family":"Zhou","given":"Xiaohao"},{"family":"Zhou","given":"Peng"},{"family":"Wang","given":"Xinran"},{"family":"Jariwala","given":"Deep"},{"family":"Hu","given":"Weida"}],"issued":{"date-parts":[[2023]]},"DOI":"10.21203/rs.3.rs-2558516/v1","URL":"https://doi.org/10.21203/rs.3.rs-2558516/v1","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-4389036/v1","type":"article-journal","title":"A physical emulation of somatosensory cortex as a\nNeuromorphic Twin for neural prostheses","abstract":"Abstract Loss of tactile feedback in patients with paralysis or limb loss prevents the integration of sensory-motor signals, limiting dexterous motor control. Recent advances in the neurostimulation of the human somatosensory cortex offer the possibility of restoring touch in these patients. Brain-computer interfaces and bidirectional prostheses with implantable electrodes can restore feedback, improving control, awareness, and quality of life. However, current electrical stimulation methods lack the fidelity to consistently and reliably reproduce natural sensations, and the impact of this stimulation on the somatosensory cortex remains unclear. Here we propose to use a mixed-signal neuromorphic processing system to construct a “neuromorphic twin” of the somatosensory cortex, following a bottom-up approach for emulating and reproducing its intricate dynamics. We use the analog properties of electronic neuron and synapse circuits to faithfully replicate neural and synaptic response properties. To reproduce effects at the network level, we configure the neuromorphic processor to implement the recurrent connectivity patterns of the somatosensory cortex among different populations of excitatory and inhibitory silicon neurons. Our findings demonstrate the twin’s ability to reproduce the spatiotemporal firing patterns of the native neuronal population receiving afferent fiber inputs, resulting in a highly bio-realistic model that can be used to investigate the effect of different stimulation patterns. This physical emulation of the cortical circuits represents therefore an additional tool to understand and predict the somatosensory cortex behavior under neurostimulation. Using this system, neurostimulation patterns can be optimized to produce more natural sensations and ultimately improve the quality of restored feedback, leading to a better quality of life for affected patients.","author":[{"family":"Ramirez","given":"Hector"},{"family":"Donati","given":"Elisa"},{"family":"Behrens","given":"Wolfger"},{"family":"Indiveri","given":"Giacomo"},{"family":"Valle","given":"Giacomo"}],"issued":{"date-parts":[[2024]]},"DOI":"10.21203/rs.3.rs-4389036/v1","URL":"https://doi.org/10.21203/rs.3.rs-4389036/v1","source":"crossref"},{"id":"doi:10.5281/zenodo.17585681","type":"article-journal","title":"Neuromorphic Event–LiDAR–IMU Dataset for SLAM Applications","abstract":"This release provides the Neuromorphic Event–LiDAR–IMU dataset sequences: • lab_indoor.zip — Controlled indoor lab sequence (159 MB) • b5_indoor.zip — Indoor building 5 sequence (551 MB) • b5_outdoor.zip — Outdoor building 5 route (445 MB) • carpark_outdoor.zip — Carpark traversal (742 MB) • b23_hybrid.zip — Hybrid building 23 route (709 MB) • b34_hybrid.zip — Hybrid building 34 route (645 MB) All sequences include synchronized event, LiDAR, and IMU data files. License: CC BY-NC 4.0 — for non-commercial research use only.","author":[{"family":"Tenzin","given":"Sangay"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.17585681","URL":"https://doi.org/10.5281/zenodo.17585681","source":"datacite"},{"id":"doi:10.5281/zenodo.17585682","type":"article-journal","title":"Neuromorphic Event–LiDAR–IMU Dataset for SLAM Applications","abstract":"This release provides the Neuromorphic Event–LiDAR–IMU dataset sequences: • lab_indoor.zip — Controlled indoor lab sequence (159 MB) • b5_indoor.zip — Indoor building 5 sequence (551 MB) • b5_outdoor.zip — Outdoor building 5 route (445 MB) • carpark_outdoor.zip — Carpark traversal (742 MB) • b23_hybrid.zip — Hybrid building 23 route (709 MB) • b34_hybrid.zip — Hybrid building 34 route (645 MB) All sequences include synchronized event, LiDAR, and IMU data files. License: CC BY-NC 4.0 — for non-commercial research use only.","author":[{"family":"Tenzin","given":"Sangay"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.17585682","URL":"https://doi.org/10.5281/zenodo.17585682","source":"datacite"},{"id":"doi:10.3390/electronics12112351","type":"article-journal","title":"Python-Based Circuit Design for Fundamental Building Blocks of Spiking Neural Network","abstract":"Spiking neural networks (SNNs) are considered a crucial research direction to address the “storage wall” and “power wall” challenges faced by traditional artificial intelligence computing. However, developing SNN chips based on CMOS (complementary metal oxide semiconductor) circuits remains a challenge. Although memristor process technology is the best alternative to synapses, it is still undergoing refinement. In this study, a novel approach is proposed that employs tools to automatically generate HDL (hardware description language) code for constructing neuron and memristor circuits after using Python to describe the neuron and memristor models. Based on this approach, HR (Hindmash–Rose), LIF (leaky integrate-and-fire), and IZ (Izhikevich) neuron circuits, as well as HP, EG (enhanced generalized), and TB (the behavioral threshold bipolar) memristor circuits are designed to construct the most basic connection of a SNN: the neuron–memristor–neuron circuit that satisfies the STDP (spike-timing-dependent-plasticity) learning rule. Through simulation experiments and FPGA (field programmable gate array) prototype verification, it is confirmed that the IZ and LIF circuits are suitable as neurons in SNNs, while the X variables of the EG memristor model serve as characteristic synaptic weights. The EG memristor circuits best satisfy the STDP learning rule and are suitable as synapses in SNNs. In comparison to previous works on hardware spiking neurons, the proposed method needed fewer area resources for creating spiking neurons models on FPGA. The proposed SNN basic components design method, and the resulting circuits, are beneficial for architectural exploration and hardware–software co-design of SNN chips.","author":[{"family":"Qin","given":"Xing"},{"family":"Li","given":"Chaojie"},{"family":"He","given":"Haitao"},{"family":"Pan","given":"Zejun"},{"family":"Lai","given":"Chenxiao"}],"issued":{"date-parts":[[2023]]},"DOI":"10.3390/electronics12112351","URL":"https://doi.org/10.3390/electronics12112351","source":"crossref"},{"id":"doi:10.1101/2023.11.16.567361","type":"article-journal","title":"Autapses enable temporal pattern recognition in spiking neural networks","abstract":"ABSTRACT Most sensory stimuli are temporal in structure. How action potentials encode the information incoming from sensory stimuli remains one of the central research questions in neuroscience. Although there is evidence that the precise timing of spikes represents information in spiking neuronal networks, information processing in spiking networks is still not fully understood. One feasible way to understand the working mechanism of a spiking network is to associate the structural connectivity of the network with the corresponding functional behaviour. This work demonstrates the structure-function mapping of spiking networks evolved (or handcrafted) for a temporal pattern recognition task. The task is to recognise a specific order of the input signals so that the Out put neurone of the network spikes only for the correct placement and remains silent for all others. The minimal networks obtained for this task revealed the twofold importance of autapses in recognition; first, autapses simplify the switching among different network states. Second, autapses enable a network to maintain a network state, a form of memory. To show that the recognition task is accomplished by transitions between network states, we map the network states of a functional spiking neural network (SNN) onto the states of a finite-state transducer (FST, a formal model of computation that generates output symbols, here: spikes or no spikes at specific times, in response to input, here: a series of input signals). Finally, based on our understanding, we define rules for constructing the topology of a network handcrafted for recognising a subsequence of signals (pattern) in a particular order. The analysis of minimal networks recognising patterns of different lengths (two to six) revealed a positive correlation between the pattern length and the number of autaptic connections in the network. Furthermore, in agreement with the behaviour of neurones in the network, we were able to associate specific functional roles of ‘locking,’ ‘switching,’ and ‘accepting’ to neurones.","author":[{"family":"Yaqoob","given":"Muhammad"},{"family":"Steuber","given":"Volker"},{"family":"Wróbel","given":"Borys"}],"issued":{"date-parts":[[2023]]},"DOI":"10.1101/2023.11.16.567361","URL":"https://doi.org/10.1101/2023.11.16.567361","source":"crossref"},{"id":"doi:10.1609/aaai.v38i10.28964","type":"article-journal","title":"Enhancing Training of Spiking Neural Network with Stochastic Latency","abstract":"Spiking neural networks (SNNs) have garnered significant attention for their low power consumption when deployed on neuromorphic hardware that operates in orders of magnitude lower power than general-purpose hardware. Direct training methods for SNNs come with an inherent latency for which the SNNs are optimized, and in general, the higher the latency, the better the predictive powers of the models, but at the same time, the higher the energy consumption during training and inference. Furthermore, an SNN model optimized for one particular latency does not necessarily perform well in lower latencies, which becomes relevant in scenarios where it is necessary to switch to a lower latency because of the depletion of onboard energy or other operational requirements. In this work, we propose Stochastic Latency Training (SLT), a direct training method for SNNs that optimizes the model for the given latency but simultaneously offers a minimum reduction of predictive accuracy when shifted to lower inference latencies. We provide heuristics for our approach with partial theoretical justification and experimental evidence showing the state-of-the-art performance of our models on datasets such as CIFAR-10, DVS-CIFAR-10, CIFAR-100, and DVS-Gesture. Our code is available at https://github.com/srinuvaasu/SLT","author":[{"family":"Anumasa","given":"Srinivas"},{"family":"Mukhoty","given":"Bhaskar"},{"family":"Bojkovic","given":"Velibor"},{"family":"Masi","given":"Giulia"},{"family":"Xiong","given":"Huan"},{"family":"Gu","given":"Bin"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1609/aaai.v38i10.28964","URL":"https://doi.org/10.1609/aaai.v38i10.28964","source":"crossref"},{"id":"doi:10.1364/oe.487047","type":"article-journal","title":"BP-based supervised learning algorithm for multilayer photonic spiking neural network and hardware implementation","abstract":"We introduce a supervised learning algorithm for photonic spiking neural network (SNN) based on back propagation. For the supervised learning algorithm, the information is encoded into spike trains with different strength, and the SNN is trained according to different patterns composed of different spike numbers of the output neurons. Furthermore, the classification task is performed numerically and experimentally based on the supervised learning algorithm in the SNN. The SNN is composed of photonic spiking neuron based on vertical-cavity surface-emitting laser which is functionally similar to leaky-integrate and fire neuron. The results prove the demonstration of the algorithm implementation on hardware. To seek ultra-low power consumption and ultra-low delay, it is great significance to design and implement a hardware-friendly learning algorithm of photonic neural networks and realize hardware-algorithm collaborative computing.","author":[{"family":"Zhang","given":"Yahui"},{"family":"Xiang","given":"Shuiying"},{"family":"Han","given":"Yanan"},{"family":"Guo","given":"Xingxing"},{"family":"Zhang","given":"Wu"},{"family":"Tan","given":"Qinggui"},{"family":"Han","given":"Genquan"},{"family":"Hao","given":"Yue"}],"issued":{"date-parts":[[2023]]},"DOI":"10.1364/oe.487047","URL":"https://doi.org/10.1364/oe.487047","source":"crossref"},{"id":"doi:10.1101/2024.10.02.616330","type":"article-journal","title":"Optimal Control of Spiking Neural Networks","abstract":"Abstract Control theory provides a natural language to describe multi-areal interactions and flexible cognitive tasks such as covert attention or brain-machine interface (BMI) experiments, which require finding adequate inputs to a local circuit in order to steer its dynamics in a context-dependent manner. In optimal control, the target dynamics should maximize a notion of long-term value along trajectories, possibly subject to control costs. Because this problem is, in general, not tractable, current approaches to the control of networks mostly consider simplified settings (e.g., variations of the Linear-Quadratic Regulator). Here, we present a mathematical framework for optimal control of recurrent networks of stochastic spiking neurons with low-rank connectivity. An essential ingredient is a control-cost that penalizes deviations from the default dynamics of the network (specified by its recurrent connections), which motivates the controller to use the default dynamics as much as possible. We derive a Bellman Equation that specifies a Value function over the low-dimensional network state (LDS), and a corresponding optimal control input. The optimal control law takes the form of a feedback controller that provides external excitatory (inhibitory) synaptic input to neurons in the recurrent network if their spiking activity tends to move the LDS towards regions of higher (lower) Value. We use our theory to study the problem of steering the state of the network towards particular terminal regions which can lie either in or out of regions in the LDS with slow dynamics, in analogy to standard BMI experiments. Our results provide the foundation of a novel approach with broad applicability that unifies bottom-up and top-down perspectives on neural computation.","author":[{"family":"Costa","given":"Tiago"},{"family":"Saa","given":"Juan"},{"family":"Renart","given":"Alfonso"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1101/2024.10.02.616330","URL":"https://doi.org/10.1101/2024.10.02.616330","source":"preprints"},{"id":"doi:10.3390/app132413145","type":"article-journal","title":"A Novel Approach for Target Attraction and Obstacle Avoidance of a Mobile Robot in Unknown Environments Using a Customized Spiking Neural Network","abstract":"In recent years, implementing reinforcement learning in autonomous mobile robots (AMRs) has become challenging. Traditional methods face complex trials, long convergence times, and high computational requirements. This paper introduces an innovative strategy using a customized spiking neural network (SNN) for autonomous learning and control of mobile robots (AMR) in unknown environments. The model combines spike-timing-dependent plasticity (STDP) with dopamine modulation for learning. It utilizes the Izhikevich neuron model, leading to biologically inspired and computationally efficient control systems that adapt to changing environments. The performance of the model is evaluated in a simulated environment, replicating real-world scenarios with obstacles. In the initial training phase, the model faces significant challenges. Integrating brain-inspired learning, dopamine, and the Izhikevich neuron model adds complexity. The model achieves an accuracy rate of 33% in reaching its target during this phase. Collisions with obstacles occur 67% of the time, indicating the struggle of the model to adapt to complex obstacles. However, the model’s performance improves as the study progresses to the testing phase after the robot has learned. Its accuracy surges to 94% when reaching the target, and collisions with obstacles reduce it to 6%. This shift demonstrates the adaptability and problem-solving capabilities of the model in the simulated environment, making it more competent for real-world applications.","author":[{"family":"Abubaker","given":"Brwa"},{"family":"Razmara","given":"Jafar"},{"family":"Karimpour","given":"Jaber"}],"issued":{"date-parts":[[2023]]},"DOI":"10.3390/app132413145","URL":"https://doi.org/10.3390/app132413145","source":"crossref"},{"id":"doi:10.1088/1674-1056/acb9f6","type":"article-journal","title":"A progressive surrogate gradient learning for memristive spiking neural network","abstract":"In recent years, spiking neural networks (SNNs) have received increasing attention of research in the field of artificial intelligence due to their high biological plausibility, low energy consumption, and abundant spatio-temporal information. However, the non-differential spike activity makes SNNs more difficult to train in supervised training. Most existing methods focusing on introducing an approximated derivative to replace it, while they are often based on static surrogate functions. In this paper, we propose a progressive surrogate gradient learning for backpropagation of SNNs, which is able to approximate the step function gradually and to reduce information loss. Furthermore, memristor cross arrays are used for speeding up calculation and reducing system energy consumption for their hardware advantage. The proposed algorithm is evaluated on both static and neuromorphic datasets using fully connected and convolutional network architecture, and the experimental results indicate that our approach has a high performance compared with previous research.","author":[{"family":"Wang","given":"Shu"},{"family":"Chen","given":"Tao"},{"family":"Gong","given":"Yu"},{"family":"Sun","given":"Fan"},{"family":"Shen","given":"Si"},{"family":"Duan","given":"Shu"},{"family":"Wang","given":"Li"}],"issued":{"date-parts":[[2023]]},"DOI":"10.1088/1674-1056/acb9f6","URL":"https://doi.org/10.1088/1674-1056/acb9f6","source":"crossref"},{"id":"doi:10.3390/mi14010203","type":"article-journal","title":"Event-Based Optical Flow Estimation with Spatio-Temporal Backpropagation Trained Spiking Neural Network","abstract":"The advantages of an event camera, such as low power consumption, large dynamic range, and low data redundancy, enable it to shine in extreme environments where traditional image sensors are not competent, especially in high-speed moving target capture and extreme lighting conditions. Optical flow reflects the target’s movement information, and the target’s detailed movement can be obtained using the event camera’s optical flow information. However, the existing neural network methods for optical flow prediction of event cameras has the problems of extensive computation and high energy consumption in hardware implementation. The spike neural network has spatiotemporal coding characteristics, so it can be compatible with the spatiotemporal data of an event camera. Moreover, the sparse coding characteristic of the spike neural network makes it run with ultra-low power consumption on neuromorphic hardware. However, because of the algorithmic and training complexity, the spike neural network has not been applied in the prediction of the optical flow for the event camera. For this case, this paper proposes an end-to-end spike neural network to predict the optical flow of the discrete spatiotemporal data stream for the event camera. The network is trained with the spatio-temporal backpropagation method in a self-supervised way, which fully combines the spatiotemporal characteristics of the event camera while improving the network performance. Compared with the existing methods on the public dataset, the experimental results show that the method proposed in this paper is equivalent to the best existing methods in terms of optical flow prediction accuracy, and it can save 99% more power consumption than the existing algorithm, which is greatly beneficial to the hardware implementation of the event camera optical flow prediction., laying the groundwork for future low-power hardware implementation of optical flow prediction for event cameras.","author":[{"family":"Zhang","given":"Yisa"},{"family":"Lv","given":"Hengyi"},{"family":"Zhao","given":"Yuchen"},{"family":"Feng","given":"Yang"},{"family":"Liu","given":"Hailong"},{"family":"Bi","given":"Guoling"}],"issued":{"date-parts":[[2023]]},"DOI":"10.3390/mi14010203","URL":"https://doi.org/10.3390/mi14010203","source":"crossref"},{"id":"doi:10.3389/fncom.2024.1418115","type":"article-journal","title":"DT-SCNN: dual-threshold spiking convolutional neural network with fewer operations and memory access for edge applications","abstract":"The spiking convolutional neural network (SCNN) is a kind of spiking neural network (SNN) with high accuracy for visual tasks and power efficiency on neuromorphic hardware, which is attractive for edge applications. However, it is challenging to implement SCNNs on resource-constrained edge devices because of the large number of convolutional operations and membrane potential (Vm) storage needed. Previous works have focused on timestep reduction, network pruning, and network quantization to realize SCNN implementation on edge devices. However, they overlooked similarities between spiking feature maps (SFmaps), which contain significant redundancy and cause unnecessary computation and storage. This work proposes a dual-threshold spiking convolutional neural network (DT-SCNN) to decrease the number of operations and memory access by utilizing similarities between SFmaps. The DT-SCNN employs dual firing thresholds to derive two similar SFmaps from one Vm map, reducing the number of convolutional operations and decreasing the volume of Vms and convolutional weights by half. We propose a variant spatio-temporal back propagation (STBP) training method with a two-stage strategy to train DT-SCNNs to decrease the inference timestep to 1. The experimental results show that the dual-thresholds mechanism achieves a 50% reduction in operations and data storage for the convolutional layers compared to conventional SCNNs while achieving not more than a 0.4% accuracy loss on the CIFAR10, MNIST, and Fashion MNIST datasets. Due to the lightweight network and single timestep inference, the DT-SCNN has the least number of operations compared to previous works, paving the way for low-latency and power-efficient edge applications.","author":[{"family":"Lei","given":"Fuming"},{"family":"Yang","given":"Xu"},{"family":"Liu","given":"Jian"},{"family":"Dou","given":"Runjiang"},{"family":"Wu","given":"Nanjian"}],"issued":{"date-parts":[[2024]]},"DOI":"10.3389/fncom.2024.1418115","URL":"https://doi.org/10.3389/fncom.2024.1418115","source":"crossref"},{"id":"doi:10.1007/s13748-024-00313-4","type":"article-journal","title":"A Deep Convolutional Spiking Neural Network for embedded applications","abstract":"Abstract Deep neural networks (DNNs) have received a great deal of interest in solving everyday tasks in recent years. However, their computational and energy costs limit their use on mobile and edge devices. The neuromorphic computing approach called spiking neural networks (SNNs) represents a potential solution for bridging the gap between performance and computational expense. Despite the potential benefits of energy efficiency, the current SNNs are being used with datasets such as MNIST, Fashion-MNIST, and CIFAR10, limiting their applications compared to DNNs. Therefore, the applicability of SNNs to real-world applications, such as scene classification and forecasting epileptic seizures, must be demonstrated yet. This paper develops a deep convolutional spiking neural network (DCSNN) for embedded applications. We explore a convolutional architecture, Visual Geometry Group (VGG16), to implement deeper SNNs. To train a spiking model, we convert the pre-trained VGG16 into corresponding spiking equivalents with nearly comparable performance to the original one. The trained weights of VGG16 were then transferred to the equivalent SNN architecture while performing a proper weight–threshold balancing. The model is evaluated in two case studies: land use and land cover classification, and epileptic seizure detection. Experimental results show a classification accuracy of 94.88%, and seizure detection specificity of 99.45% and a sensitivity of 95.06%. It is confirmed that conversion-based training SNNs are promising, and the benefits of DNNs, such as solving complex and real-world problems, become available to SNNs.","author":[{"family":"Javanshir","given":"Amirhossein"},{"family":"Nguyen","given":"Thanh"},{"family":"Mahmud","given":"MAP"},{"family":"Kouzani","given":"Abbas"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1007/s13748-024-00313-4","URL":"https://doi.org/10.1007/s13748-024-00313-4","source":"crossref"},{"id":"doi:10.1007/s00521-024-10191-5","type":"article-journal","title":"Energy efficient and low-latency spiking neural networks on embedded microcontrollers through spiking activity tuning","abstract":"Abstract In this work, we target the efficient implementation of spiking neural networks (SNNs) for low-power and low-latency applications. In particular, we propose a methodology for tuning SNN spiking activity with the objective of reducing computation cycles and energy consumption. We performed an analysis to devise key hyper-parameters, and then we show the results of tuning such parameters to obtain a low-latency and low-energy embedded LSNN (eLSNN) implementation. We demonstrate that it is possible to adapt the firing rate so that the samples belonging to the most frequent class are processed with less spikes. We implemented the eLSNN on a microcontroller-based sensor node and we evaluated its performance and energy consumption using a structural health monitoring application processing a stream of vibrations for damage detection (i.e. binary classification). We obtained a cycle count reduction of 25% and an energy reduction of 22% with respect to a baseline implementation. We also demonstrate that our methodology is applicable to a multi-class scenario, showing that we can reduce spiking activity between 68 and 85% at iso-accuracy.","author":[{"family":"Barchi","given":"Francesco"},{"family":"Parisi","given":"Emanuele"},{"family":"Zanatta","given":"Luca"},{"family":"Bartolini","given":"Andrea"},{"family":"Acquaviva","given":"Andrea"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1007/s00521-024-10191-5","URL":"https://doi.org/10.1007/s00521-024-10191-5","source":"crossref"},{"id":"doi:10.36227/techrxiv.171822392.24893565/v1","type":"article-journal","title":"Relational and Analogical Reasoning with Spiking Neural Networks","abstract":"Raven's Progressive Matrices (RPMs) have been widely used for measuring abstract reasoning and intelligence in humans. However for artificial learning systems, abstract reasoning remains a challenging problem. This paper investigates the potential of biologically inspired spiking neural networks in addressing this challenge. We focus on unsupervised learning since in most domains where human-competitive conceptual learning is important, unlabelled data is plentiful and the autonomous discovery and understanding of underlying patterns without explicit guidance is paramount. Our experiments show that the overall accuracy of our unsupervised networks significantly outperform some supervised methods. Our results also demonstrate that, unlike their non-spiking counterparts, spiking neural networks are able to extract and encode relational features without any explicit instruction, do not rely on labelled training data, and are able to simulate the human ability of navigating and deciphering complex data landscapes. Our results exhibit state-of-the-art performance for unsupervised learning on the RAVEN (67.0%) and I-RAVEN (40.3%) datasets, indicating that that spiking neural networks are well suited to unsupervised conceptual learning.","author":[{"family":"Omari","given":"Rollin"},{"family":"Mckay","given":"RI"},{"family":"Gedeon","given":"Tom"},{"family":"Taylor","given":"Kerry"}],"issued":{"date-parts":[[2024]]},"DOI":"10.36227/techrxiv.171822392.24893565/v1","URL":"https://doi.org/10.36227/techrxiv.171822392.24893565/v1","source":"crossref"},{"id":"doi:10.3389/fnins.2024.1401690","type":"article-journal","title":"Paired competing neurons improving STDP supervised local learning in spiking neural networks","abstract":"Direct training of Spiking Neural Networks (SNNs) on neuromorphic hardware has the potential to significantly reduce the energy consumption of artificial neural network training. SNNs trained with Spike Timing-Dependent Plasticity (STDP) benefit from gradient-free and unsupervised local learning, which can be easily implemented on ultra-low-power neuromorphic hardware. However, classification tasks cannot be performed solely with unsupervised STDP. In this paper, we propose Stabilized Supervised STDP (S2-STDP), a supervised STDP learning rule to train the classification layer of an SNN equipped with unsupervised STDP for feature extraction. S2-STDP integrates error-modulated weight updates that align neuron spikes with desired timestamps derived from the average firing time within the layer. Then, we introduce a training architecture called Paired Competing Neurons (PCN) to further enhance the learning capabilities of our classification layer trained with S2-STDP. PCN associates each class with paired neurons and encourages neuron specialization toward target or non-target samples through intra-class competition. We evaluate our methods on image recognition datasets, including MNIST, Fashion-MNIST, and CIFAR-10. Results show that our methods outperform state-of-the-art supervised STDP learning rules, for comparable architectures and numbers of neurons. Further analysis demonstrates that the use of PCN enhances the performance of S2-STDP, regardless of the hyperparameter set and without introducing any additional hyperparameters.","author":[{"family":"Goupy","given":"Gaspard"},{"family":"Tirilly","given":"Pierre"},{"family":"Bilasco","given":"Ioan"}],"issued":{"date-parts":[[2024]]},"DOI":"10.3389/fnins.2024.1401690","URL":"https://doi.org/10.3389/fnins.2024.1401690","source":"crossref"},{"id":"doi:10.1101/2024.11.29.626029","type":"article-journal","title":"Kernel-based LFP estimation in detailed large-scale spiking network model of mouse visual cortex","abstract":"Summary Simulations of large-scale neural activity are powerful tools for investigating neural networks. Calculating measurable brain signals like local field potentials (LFPs) bridges the gap between model predictions and experimental observations. However, accurately simulating LFPs from large-scale models has traditionally required highly detailed multicompartmental neuron models, posing significant computational challenges. Here, we demonstrate that a kernel-based method can efficiently and accurately estimate LFPs in a state-of-the-art multicompartmental model of the mouse primary visual cortex (V1). Beyond its computational efficiency, the kernel method aids analysis by disentangling contributions of individual neuronal populations to the LFP. Using this approach, we found that LFPs in the V1 model were dominated by external synaptic inputs, with local synaptic activity playing a minimal role. Our findings establish the kernel method as a powerful tool for LFP estimation in large-scale network models and for uncovering the synaptic mechanisms underlying brain signals.","author":[{"family":"Meneghetti","given":"Nicolò"},{"family":"Rimehaug","given":"Atle"},{"family":"Einevoll","given":"Gaute"},{"family":"Mazzoni","given":"Alberto"},{"family":"Ness","given":"Torbjørn"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1101/2024.11.29.626029","URL":"https://doi.org/10.1101/2024.11.29.626029","source":"preprints"},{"id":"doi:10.3390/s24020491","type":"article-journal","title":"A Novel Robotic Controller Using Neural Engineering Framework-Based Spiking Neural Networks","abstract":"This paper investigates spiking neural networks (SNN) for novel robotic controllers with the aim of improving accuracy in trajectory tracking. By emulating the operation of the human brain through the incorporation of temporal coding mechanisms, SNN offer greater adaptability and efficiency in information processing, providing significant advantages in the representation of temporal information in robotic arm control compared to conventional neural networks. Exploring specific implementations of SNN in robot control, this study analyzes neuron models and learning mechanisms inherent to SNN. Based on the principles of the Neural Engineering Framework (NEF), a novel spiking PID controller is designed and simulated for a 3-DoF robotic arm using Nengo and MATLAB R2022b. The controller demonstrated good accuracy and efficiency in following designated trajectories, showing minimal deviations, overshoots, or oscillations. A thorough quantitative assessment, utilizing performance metrics like root mean square error (RMSE) and the integral of the absolute value of the time-weighted error (ITAE), provides additional validation for the efficacy of the SNN-based controller. Competitive performance was observed, surpassing a fuzzy controller by 5% in terms of the ITAE index and a conventional PID controller by 6% in the ITAE index and 30% in RMSE performance. This work highlights the utility of NEF and SNN in developing effective robotic controllers, laying the groundwork for future research focused on SNN adaptability in dynamic environments and advanced robotic applications.","author":[{"family":"Marrero","given":"Dailin"},{"family":"Kern","given":"John"},{"family":"Urrea","given":"Claudio"}],"issued":{"date-parts":[[2024]]},"DOI":"10.3390/s24020491","URL":"https://doi.org/10.3390/s24020491","source":"crossref"},{"id":"doi:10.1101/2023.02.03.526928","type":"article-journal","title":"Global inhibition in head-direction neural circuits: a systematic comparison between connectome-based spiking neural circuit models","abstract":"Abstract The recent discovery of the head-direction (HD) system in fruit flies has provided unprecedented insights into the neural mechanisms of spatial orientation. Despite the progress, the neural substance of global inhibition, an essential component of the HD circuits, remains controversial. Some studies suggested that the ring neurons provide global inhibition, while others suggested the Δ7 neurons. In the present study, we provide evaluations from the theoretical perspective by performing systematic analyses on the computational models based on the ring-neuron (R models) and Δ7-neurons (Delta models) hypotheses with modifications according to the latest connectomic data. We conducted four tests: robustness, persistency, speed, and dynamical characteristics. We discovered that the two models led to a comparable performance in general, but each excelled in different tests. The R Models were more robust, while the Delta models were better in the persistency test. We also tested a hybrid model that combines both inhibitory mechanisms. While the performances of the R and Delta models in each test are highly parameter-dependent, the Hybrid model performed well in all tests with the same set of parameters. Our results suggest the possibility of combined inhibitory mechanisms in the HD circuits of fruit flies.","author":[{"family":"Chang","given":"Ning"},{"family":"Huang","given":"Hsuan"},{"family":"Lo","given":"Chung"}],"issued":{"date-parts":[[2023]]},"DOI":"10.1101/2023.02.03.526928","URL":"https://doi.org/10.1101/2023.02.03.526928","source":"crossref"},{"id":"doi:10.1002/pamm.202400066","type":"article-journal","title":"Physics‐Based Spiking Neural Network as a Surrogate Model for Viscoplastic Material Law in Impulsively Loaded Beams","abstract":"ABSTRACT In the present study, the nonlinear material law of an in‐house FEM solver is substituted by physics‐informed neural networks (PINNs) and by physics‐based spiking neural networks (SNNs) for predicting the dynamic response of beams under impact loading. In the first part of this study concerning PINNs, the physics are included in the loss function of a deep neural network, also called a second‐generation neural network. The main advantage of using PINNs compared to classical ANNs is that they are considered data‐efficient function approximators. In the second part of this study, leaky integrate and fire (LIF) layers are deployed instead of dense layers in the PINN architecture, leading to a physics‐based third‐generation SNN, which enhances the PINN architecture in terms of energy efficiency. Third‐generation SNNs are inspired by the human brain function, where the information is processed via spikes. The spikes introduce sparse behavior, leading to a more energy‐efficient and sustainable deep learning method when implemented in neuromorphic hardware. The PINNs and physics‐based SNNs are employed in the implicit integration scheme of the material law of the in‐house FEM solver at the Gaussian level, and results are compared with the in‐house FEM solution in two different initial boundary value (IBV) problems. This study aims to show that PINNs can predict nonlinear material behavior and that the use of a physics‐based loss function enables the utility of less data in the training of the neural networks. Moreover, we show that we can enhance PINNs regarding energy consumption by adding a LIF layer. Therefore, the energy consumption is compared when the physics‐based SNN is deployed on the Loihi neuromorphic chip and on the CPU.","author":[{"family":"Polydoras","given":"Vasileios"},{"family":"Tandale","given":"Saurabh"},{"family":"Stoffel","given":"Marcus"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1002/pamm.202400066","URL":"https://doi.org/10.1002/pamm.202400066","source":"crossref"},{"id":"doi:10.1101/2023.05.22.541722","type":"article-journal","title":"Synaptic turnover promotes efficient learning in bio-realistic spiking neural networks","abstract":"Abstract While artificial machine learning systems achieve superhuman performance in specific tasks such as language processing, image and video recognition, they do so use extremely large datasets and huge amounts of power. On the other hand, the brain remains superior in several cognitively challenging tasks while operating with the energy of a small lightbulb. We use a biologically constrained spiking neural network model to explore how the neural tissue achieves such high efficiency and assess its learning capacity on discrimination tasks. We found that synaptic turnover, a form of structural plasticity, which is the ability of the brain to form and eliminate synapses continuously, increases both the speed and the performance of our network on all tasks tested. Moreover, it allows accurate learning using a smaller number of examples. Importantly, these improvements are most significant under conditions of resource scarcity, such as when the number of trainable parameters is halved and when the task difficulty is increased. Our findings provide new insights into the mechanisms that underlie efficient learning in the brain and can inspire the development of more efficient and flexible machine learning algorithms.","author":[{"family":"Malakasis","given":"Nikos"},{"family":"Chavlis","given":"Spyridon"},{"family":"Poirazi","given":"Panayiota"}],"issued":{"date-parts":[[2023]]},"DOI":"10.1101/2023.05.22.541722","URL":"https://doi.org/10.1101/2023.05.22.541722","source":"crossref"},{"id":"doi:10.1088/1741-2552/ad6594","type":"article-journal","title":"Spiking Laguerre Volterra networks—predicting neuronal activity from local field potentials","abstract":"Abstract Objective. Understanding the generative mechanism between local field potentials (LFP) and neuronal spiking activity is a crucial step for understanding information processing in the brain. Up to now, most approaches have relied on simply quantifying the coupling between LFP and spikes. However, very few have managed to predict the exact timing of spike occurrence based on LFP variations. Approach. Here, we fill this gap by proposing novel spiking Laguerre–Volterra network (sLVN) models to describe the dynamic LFP-spike relationship. Compared to conventional artificial neural networks, the sLVNs are interpretable models that provide explainable features of the underlying dynamics. Main results. The proposed networks were applied on extracellular microelectrode recordings of Parkinson’s Disease patients during deep brain stimulation (DBS) surgery. Based on the predictability of the LFP-spike pairs, we detected three neuronal populations with unique signal characteristics and sLVN model features. Significance. These clusters were indirectly associated with motor score improvement following DBS surgery, warranting further investigation into the potential of spiking activity predictability as an intraoperative biomarker for optimal DBS lead placement.","author":[{"family":"Kostoglou","given":"Kyriaki"},{"family":"Michmizos","given":"Konstantinos"},{"family":"Stathis","given":"Pantelis"},{"family":"Sakas","given":"Damianos"},{"family":"Nikita","given":"Konstantina"},{"family":"Mitsis","given":"Georgios"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1088/1741-2552/ad6594","URL":"https://doi.org/10.1088/1741-2552/ad6594","source":"crossref"},{"id":"doi:10.1002/ett.70019","type":"article-journal","title":"Spiking Quantum Fire Hawk Network Based Reliable Scheduling for Lifetime Maximization of Wireless Sensor Network","abstract":"ABSTRACT Managing energy consumption poses a substantial challenge within Wireless Sensor Networks (WSN) due to frequent communication among sensor nodes. A reliable scheduling framework presents a promising solution for maximizing WSN lifetime, minimizing energy consumption, and ensuring robust communication. Despite various proposed methods for reliable scheduling, energy consumption remains high. To address this, a spiking quantum fire hawk network‐based reliable scheduling in WSN is introduced. This research incorporates clustering, duty‐cycle management, and reliable routing to enhance energy efficiency. An improved Nutcracker Optimization‐based Cluster Head (CH) election and Spiking Quantum Fire Hawk Network duty cycling contribute to optimal CH selection and increased network lifetime. Additionally, a Link Quality‐based Energy Aware Proficient Trusted Routing Protocol (LQEAP‐TRP) minimizes data transmission delay, offering reliable routing. Finally, the data communication is done in the reliable route. The developed approach is executed in Network Simulator and validated with the existing protocols. The results of the simulations indicate that the proposed approach achieves a network lifetime of 99.34% and a packet delivery ratio of 99.95%.","author":[{"family":"Kiran","given":"WS"},{"family":"Wilson","given":"Allan"},{"family":"Radhamani","given":"AS"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1002/ett.70019","URL":"https://doi.org/10.1002/ett.70019","source":"crossref"},{"id":"doi:10.3389/fnins.2024.1412559","type":"article-journal","title":"Composing recurrent spiking neural networks using locally-recurrent motifs and risk-mitigating architectural optimization","abstract":"In neural circuits, recurrent connectivity plays a crucial role in network function and stability. However, existing recurrent spiking neural networks (RSNNs) are often constructed by random connections without optimization. While RSNNs can produce rich dynamics that are critical for memory formation and learning, systemic architectural optimization of RSNNs is still an open challenge. We aim to enable systematic design of large RSNNs via a new scalable RSNN architecture and automated architectural optimization. We compose RSNNs based on a layer architecture called Sparsely-Connected Recurrent Motif Layer (SC-ML) that consists of multiple small recurrent motifs wired together by sparse lateral connections. The small size of the motifs and sparse inter-motif connectivity leads to an RSNN architecture scalable to large network sizes. We further propose a method called Hybrid Risk-Mitigating Architectural Search (HRMAS) to systematically optimize the topology of the proposed recurrent motifs and SC-ML layer architecture. HRMAS is an alternating two-step optimization process by which we mitigate the risk of network instability and performance degradation caused by architectural change by introducing a novel biologically-inspired “self-repairing” mechanism through intrinsic plasticity. The intrinsic plasticity is introduced to the second step of each HRMAS iteration and acts as unsupervised fast self-adaptation to structural and synaptic weight modifications introduced by the first step during the RSNN architectural “evolution.” We demonstrate that the proposed automatic architecture optimization leads to significant performance gains over existing manually designed RSNNs: we achieve 96.44% on TI46-Alpha, 94.66% on N-TIDIGITS, 90.28% on DVS-Gesture, and 98.72% on N-MNIST. To the best of the authors' knowledge, this is the first work to perform systematic architecture optimization on RSNNs.","author":[{"family":"Zhang","given":"Wenrui"},{"family":"Geng","given":"Hejia"},{"family":"Li","given":"Peng"}],"issued":{"date-parts":[[2024]]},"DOI":"10.3389/fnins.2024.1412559","URL":"https://doi.org/10.3389/fnins.2024.1412559","source":"crossref"},{"id":"doi:10.24963/ijcai.2024/157","type":"article-journal","title":"Tackling Long-Tailed Data Challenges in Spiking Neural Networks via Heterogeneous Knowledge Distillation","abstract":"Spiking Neural Networks (SNNs), inspired by the behavior of biological neurons, have gained significant research interest for resource-constrained edge devices and neuromorphic hardware due to their use of binary spike signals for inter-unit communication with low power consumption. However, the absence of research on spiking neural networks on long-tailed data has severely limited the deployment and application of this emerging network in practical scenarios. To fill this gap, this paper proposes a long-tail learning framework based on spiking neural networks, named LT-SpikingFormer, to alleviate the distribution bias between head and tail classes. LT-SpikingFormer adopts a widely trained Convolutional Neural Network to construct a heterogeneous knowledge distillation paradigm, offering balanced and reliable prior knowledge. Moreover, a multi-granularity hierarchical feature distillation objective is proposed to leverage cross-layer local features and network global predictions to facilitate refined information distillation to optimize the network, specifically for the performance of the tailed classes. Extensive experimental results demonstrate that our method performs well on several benchmark datasets.","author":[{"family":"Li","given":"Moqi"},{"family":"Yang","given":"Xu"},{"family":"Deng","given":"Cheng"}],"issued":{"date-parts":[[2024]]},"DOI":"10.24963/ijcai.2024/157","URL":"https://doi.org/10.24963/ijcai.2024/157","source":"crossref"},{"id":"doi:10.1190/geo2023-0737.1","type":"article-journal","title":"Bayesian neural network and Bayesian physics-informed neural network via variational inference for seismic petrophysical inversion","abstract":"ABSTRACT Deep-learning methods are being successfully applied to seismic inversion and reservoir characterization problems; however, the uncertainty quantification process has not been fully studied. We investigate probabilistic approaches for seismic petrophysical inversion by using Bayesian formulations of neural networks. In particular, Bayesian neural network via variational inference (BNN-VI) aims to estimate the probability distribution of the training parameters of the neural network, which are difficult to compute with traditional methods, by proposing a family of densities and finding the candidates that are close to the target. Variational inference (VI) has the advantage of providing a faster tool than Markov chain Monte Carlo sampling algorithms. However, for accurate predictions, BNN-VI requires large training data that are not always available in geophysical inverse problems. Hence, we develop a method that combines a Bayesian approach with physics-informed neural networks (PINNs) via VI, namely Bayesian PINNs via VI (BPINN-VI), and we apply it to a geophysical inverse problem for the estimation of petrophysical properties from seismic data. The method is based on a Bayesian network approach in which the objective function is defined by the Kullback-Leibler divergence. In addition, the physical relations between data (seismic measurements) and model variables (petrophysical properties) are embedded into neural networks through a physics-dependent loss term. We test our method on a synthetic prestack seismic data set and a real data set including an oil-saturated clastic reservoir in the North Sea with a sequence of sand and shale layers. The outcome is the most likely model of petrophysical properties. Compared with BNN-VI, the BPINN-VI results show higher correlations and R2 coefficients, lower mean-square errors, as well as lower uncertainty and higher lateral continuity.","author":[{"family":"Li","given":"Peng"},{"family":"Grana","given":"Dario"},{"family":"Liu","given":"Mingliang"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1190/geo2023-0737.1","URL":"https://doi.org/10.1190/geo2023-0737.1","source":"crossref"},{"id":"doi:10.1038/s41598-024-59469-7","type":"article-journal","title":"Convolutional spiking neural networks for intent detection based on anticipatory brain potentials using electroencephalogram","abstract":"Abstract Spiking neural networks (SNNs) are receiving increased attention because they mimic synaptic connections in biological systems and produce spike trains, which can be approximated by binary values for computational efficiency. Recently, the addition of convolutional layers to combine the feature extraction power of convolutional networks with the computational efficiency of SNNs has been introduced. This paper studies the feasibility of using a convolutional spiking neural network (CSNN) to detect anticipatory slow cortical potentials (SCPs) related to braking intention in human participants using an electroencephalogram (EEG). Data was collected during an experiment wherein participants operated a remote-controlled vehicle on a testbed designed to simulate an urban environment. Participants were alerted to an incoming braking event via an audio countdown to elicit anticipatory potentials that were measured using an EEG. The CSNN’s performance was compared to a standard CNN, EEGNet and three graph neural networks via 10-fold cross-validation. The CSNN outperformed all the other neural networks, and had a predictive accuracy of 99.06% with a true positive rate of 98.50%, a true negative rate of 99.20% and an F1-score of 0.98. Performance of the CSNN was comparable to the CNN in an ablation study using a subset of EEG channels that localized SCPs. Classification performance of the CSNN degraded only slightly when the floating-point EEG data were converted into spike trains via delta modulation to mimic synaptic connections.","author":[{"family":"Lutes","given":"Nathan"},{"family":"Nadendla","given":"Venkata"},{"family":"Krishnamurthy","given":"K"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1038/s41598-024-59469-7","URL":"https://doi.org/10.1038/s41598-024-59469-7","source":"crossref"},{"id":"doi:10.7554/elife.90597.3","type":"article-journal","title":"Hippocampome.org 2.0 is a knowledge base enabling data-driven spiking neural network simulations of rodent hippocampal circuits","abstract":"Hippocampome.org is a mature open-access knowledge base of the rodent hippocampal formation focusing on neuron types and their properties. Previously, Hippocampome.org v1.0 established a foundational classification system identifying 122 hippocampal neuron types based on their axonal and dendritic morphologies, main neurotransmitter, membrane biophysics, and molecular expression (Wheeler et al., 2015). Releases v1.1 through v1.12 furthered the aggregation of literature-mined data, including among others neuron counts, spiking patterns, synaptic physiology, in vivo firing phases, and connection probabilities. Those additional properties increased the online information content of this public resource over 100-fold, enabling numerous independent discoveries by the scientific community. Hippocampome.org v2.0, introduced here, besides incorporating over 50 new neuron types, now recenters its focus on extending the functionality to build real-scale, biologically detailed, data-driven computational simulations. In all cases, the freely downloadable model parameters are directly linked to the specific peer-reviewed empirical evidence from which they were derived. Possible research applications include quantitative, multiscale analyses of circuit connectivity and spiking neural network simulations of activity dynamics. These advances can help generate precise, experimentally testable hypotheses and shed light on the neural mechanisms underlying associative memory and spatial navigation.","author":[{"family":"Wheeler","given":"Diek"},{"family":"Kopsick","given":"Jeffrey"},{"family":"Sutton","given":"Nate"},{"family":"Tecuatl","given":"Carolina"},{"family":"Komendantov","given":"Alexander"},{"family":"Nadella","given":"Kasturi"},{"family":"Ascoli","given":"Giorgio"}],"issued":{"date-parts":[[2024]]},"DOI":"10.7554/elife.90597.3","URL":"https://doi.org/10.7554/elife.90597.3","source":"crossref"},{"id":"doi:10.36227/techrxiv.170906907.74394397/v1","type":"article-journal","title":"Neural Network Based Anomaly Detection Method for Network Datasets","abstract":"This research paper presents a comprehensive investigation into the development of an innovative and novel custom neural network model for intrusion detection systems (IDS). In the current era of rapid data transfer facilitated by the internet and advancements in communication technologies, the security of sensitive information is of paramount concern. As attackers continuously devise new methodologies to steal or tamper with data, IDSs face significant challenges in effectively detecting and mitigating intrusions. While extensive research has been conducted to enhance IDS capabilities, the need for improved detection accuracy and reduced false alarm rates remains a pressing issue. Moreover, the identification of zeroday attacks continues to pose a formidable obstacle. In contrast to conventional IDS approaches that heavily rely on statistical methodologies and rule-based expert systems, this study embraces data mining techniques, specifically Neural Networks (NNs), to overcome the limitations associated with large datasets. This research paper proposes a meticulously designed custom neural network model that leverages machine learning (ML) algorithms to analyze contemporary host activity and cloud service data. The paper extensively discusses the utilized dataset, meticulously evaluates the performance of various classifiers, and introduces our innovative neural network model. Emphasizing the significance of our model in anomaly detection, the findings underscore the importance of robust ML models to ensure the efficacy and longevity of deployed defensive systems. By capitalizing on its innovative design and leveraging the power of ML algorithms, our model not only addresses the limitations of traditional IDS approaches but also paves the way for enhanced accuracy, reduced false alarms, and improved resilience against zero-day attacks. This research contributes to the advancement of the field, shedding light on the novel possibilities and remarkable innovation offered by our custom neural network model in safeguarding critical information in an increasingly hostile digital landscape.","author":[{"family":"Hussain","given":"Bilal"},{"family":"Hasan","given":"Yusuf"},{"family":"Khan","given":"Irfan"}],"issued":{"date-parts":[[2024]]},"DOI":"10.36227/techrxiv.170906907.74394397/v1","URL":"https://doi.org/10.36227/techrxiv.170906907.74394397/v1","source":"crossref"},{"id":"doi:10.1007/s44196-024-00425-8","type":"article-journal","title":"A Novel Training Approach in Deep Spiking Neural Network Based on Fuzzy Weighting and Meta-heuristic Algorithm","abstract":"Abstract The challenge of supervised learning in spiking neural networks (SNNs) for digit classification from speech signals is examined in this study. Meta-heuristic algorithms and a fuzzy logic framework are used to train SNNs. Using gray wolf optimization (GWO), the features obtained from audio signals are reduced depending on the dispersion of each feature. Then, it combines fuzzy weighting system (FWS) and spike time-dependent flexibility (STDP) approach to implement the learning rule in SNN. The FWS rule produces a uniformly distributed random weight in the STDP flexibility window, so that the system requires fewer training parameters. Finally, these neurons are fed data to estimate the training weights and threshold values of the neurons using wild horse algorithm (WHO). With the parameters given, these rule weights are applied to appropriately display the class's share in extracting the relevant feature. The suggested network can classify speech signals into categories with 97.17% accuracy. The dataset was obtained using neurons operating at sparse biological rates below 600 Hz in the TIDIGITS test database. The suggested method has been evaluated on the IRIS and Trip Data datasets, where the classification results showed a 98.93% and 97.36% efficiency, respectively. Compared to earlier efforts, this study's results demonstrate that the strategy is both computationally simpler and more accurate. The accuracy of classification of digits, IRIS and Trip Data has increased by 4.9, 3.46 and 1.24%, respectively. The principal goal of this research is to improve the accuracy of SNN by developing a new high-precision training method.","author":[{"family":"Hamian","given":"Melika"},{"family":"Faez","given":"Karim"},{"family":"Nazari","given":"Soheila"},{"family":"Sabeti","given":"Malihe"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1007/s44196-024-00425-8","URL":"https://doi.org/10.1007/s44196-024-00425-8","source":"crossref"},{"id":"doi:10.14311/nnw.2024.34.005","type":"article-journal","title":"Variations of Training Process in Vanilla Recurrent Neural Network Framework","abstract":"Recurrent neural networks (RNNs) are capable of learning features and long term dependencies from sequential and time series data and show outstanding performance in sequential modeling tasks. However, training process in RNNs is troubled by issues in learning processes such as slow inference, vanishing gradients and difficulties in capturing long term dependencies. In this paper, we introduce a new learning technique to update the weight set as we change the input sequence which is shifted by certain amount of time in training process, instead of using a traditional way to calculate one set of the weights and bias in training time series with sequences shifted by certain amount of time series. We also consider an algorithm for an evaluation process. In the traditional way, the evaluation process is executed by using final weights and biases calculated in the training process. Instead, during the testing process, the weights and biases are iteratively updated in each sequence as done in the training process. Several numerical experiments demonstrate the efficiency of the proposed techniques.","author":[{"family":"Yi","given":"Dokkyun"},{"family":"Kim","given":"Inmi"},{"family":"Bu","given":"Sunyoung"}],"issued":{"date-parts":[[2024]]},"DOI":"10.14311/nnw.2024.34.005","URL":"https://doi.org/10.14311/nnw.2024.34.005","source":"crossref"},{"id":"doi:10.3389/fnins.2023.1270090","type":"article-journal","title":"SHIP: a computational framework for simulating and validating novel technologies in hardware spiking neural networks","abstract":"Investigations in the field of spiking neural networks (SNNs) encompass diverse, yet overlapping, scientific disciplines. Examples range from purely neuroscientific investigations, researches on computational aspects of neuroscience, or applicative-oriented studies aiming to improve SNNs performance or to develop artificial hardware counterparts. However, the simulation of SNNs is a complex task that can not be adequately addressed with a single platform applicable to all scenarios. The optimization of a simulation environment to meet specific metrics often entails compromises in other aspects. This computational challenge has led to an apparent dichotomy of approaches, with model-driven algorithms dedicated to the detailed simulation of biological networks, and data-driven algorithms designed for efficient processing of large input datasets. Nevertheless, material scientists, device physicists, and neuromorphic engineers who develop new technologies for spiking neuromorphic hardware solutions would find benefit in a simulation environment that borrows aspects from both approaches, thus facilitating modeling, analysis, and training of prospective SNN systems. This manuscript explores the numerical challenges deriving from the simulation of spiking neural networks, and introduces SHIP, Spiking (neural network) Hardware In PyTorch, a numerical tool that supports the investigation and/or validation of materials, devices, small circuit blocks within SNN architectures. SHIP facilitates the algorithmic definition of the models for the components of a network, the monitoring of states and output of the modeled systems, and the training of the synaptic weights of the network, by way of user-defined unsupervised learning rules or supervised training techniques derived from conventional machine learning. SHIP offers a valuable tool for researchers and developers in the field of hardware-based spiking neural networks, enabling efficient simulation and validation of novel technologies.","author":[{"family":"Gemo","given":"Emanuele"},{"family":"Spiga","given":"Sabina"},{"family":"Brivio","given":"Stefano"}],"issued":{"date-parts":[[2024]]},"DOI":"10.3389/fnins.2023.1270090","URL":"https://doi.org/10.3389/fnins.2023.1270090","source":"crossref"},{"id":"doi:10.24963/ijcai.2024/767","type":"article-journal","title":"Exploiting Label Skewness for Spiking Neural Networks in Federated Learning","abstract":"The energy efficiency of deep spiking neural networks (SNNs) aligns with the constraints of resource-limited edge devices, positioning SNNs as a promising foundation for intelligent applications leveraging the extensive data collected by these devices. To safeguard data privacy, federated learning (FL) facilitates collaborative SNN-based model training by leveraging data distributed across edge devices without transmitting local data to a central server. However, existing FL approaches encounter challenges in handling label-skewed data across devices, inducing drift in the local SNN model and consequently impairing the performance of the global SNN model. To tackle these problems, we propose a novel framework called FedLEC, which incorporates intra-client label weight calibration to balance the learning intensity across local labels and inter-client knowledge distillation to mitigate local SNN model bias caused by label absence. Extensive experiments with three different structured SNNs across five datasets (i.e., three non-neuromorphic and two neuromorphic datasets) demonstrate the efficiency of FedLEC. Compared to seven state-of-the-art FL algorithms, FedLEC achieves an average accuracy improvement of approximately 11.59% for the global SNN model under various label skew distribution settings.","author":[{"family":"Yu","given":"Di"},{"family":"Du","given":"Xin"},{"family":"Jiang","given":"Linshan"},{"family":"Zhang","given":"Huijing"},{"family":"Deng","given":"Shuiguang"}],"issued":{"date-parts":[[2024]]},"DOI":"10.24963/ijcai.2024/767","URL":"https://doi.org/10.24963/ijcai.2024/767","source":"crossref"},{"id":"doi:10.1038/s41467-024-51110-5","type":"article-journal","title":"High-performance deep spiking neural networks with 0.3 spikes per neuron","abstract":"Abstract Communication by rare, binary spikes is a key factor for the energy efficiency of biological brains. However, it is harder to train biologically-inspired spiking neural networks than artificial neural networks. This is puzzling given that theoretical results provide exact mapping algorithms from artificial to spiking neural networks with time-to-first-spike coding. In this paper we analyze in theory and simulation the learning dynamics of time-to-first-spike-networks and identify a specific instance of the vanishing-or-exploding gradient problem. While two choices of spiking neural network mappings solve this problem at initialization, only the one with a constant slope of the neuron membrane potential at threshold guarantees the equivalence of the training trajectory between spiking and artificial neural networks with rectified linear units. For specific image classification architectures comprising feed-forward dense or convolutional layers, we demonstrate that deep spiking neural network models can be effectively trained from scratch on MNIST and Fashion-MNIST datasets, or fine-tuned on large-scale datasets, such as CIFAR10, CIFAR100 and PLACES365, to achieve the exact same performance as that of artificial neural networks, surpassing previous spiking neural networks. Our approach accomplishes high-performance classification with less than 0.3 spikes per neuron, lending itself for an energy-efficient implementation. We also show that fine-tuning spiking neural networks with our robust gradient descent algorithm enables their optimization for hardware implementations with low latency and resilience to noise and quantization.","author":[{"family":"Stanojevic","given":"Ana"},{"family":"Woźniak","given":"Stanisław"},{"family":"Bellec","given":"Guillaume"},{"family":"Cherubini","given":"Giovanni"},{"family":"Pantazi","given":"Angeliki"},{"family":"Gerstner","given":"Wulfram"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1038/s41467-024-51110-5","URL":"https://doi.org/10.1038/s41467-024-51110-5","source":"crossref"},{"id":"doi:10.4018/979-8-3693-6303-4.ch009","type":"article-journal","title":"Enhancing Assistive Technologies With Neuromorphic Computing","abstract":"The development of intelligent neuroprosthetics, which promise to augment human brain function is vital for augmentative assistive technologies. Neuromorphic sensors and processors are particularly adept at mimicking the brain's efficient sensory processing, offering assistive devices an advanced capability to perceive and interpret complex environmental stimuli. The application of these technologies in brain computer interfaces suggests a future where transformative advancements are not only possible but imminent, facilitating novel methods of human-computer interaction and providing insights into the intricate workings of the brain through advanced AI and machine learning techniques. This paper explores the integration of neuromorphic technologies with brain-computer interfaces (BCIs), highlighting the potential to enhance assistive devices and revolutionize communication and healthcare. However, the realization of neuromorphic computing's full potential within BCIs is contingent upon overcoming significant technological and ethical challenges.","author":[{"family":"Chandra","given":"GVSA"},{"family":"Ananthakumar","given":"Bhanuprakash"},{"family":"Raghavan","given":"Ramya"}],"issued":{"date-parts":[[2024]]},"DOI":"10.4018/979-8-3693-6303-4.ch009","URL":"https://doi.org/10.4018/979-8-3693-6303-4.ch009","source":"crossref"},{"id":"doi:10.4018/979-8-3693-1131-8.ch002","type":"article-journal","title":"Bio-Inspired Algorithms Used in Medical Image Processing","abstract":"Medical image processing plays a crucial role in diagnosing diseases, guiding treatment plans, and monitoring patient progress. With the increasing complexity and volume of medical imaging data, there is a growing need for advanced techniques to extract meaningful information from these images. Traditional methods in medical image processing often face challenges related to image enhancement, segmentation, and feature extraction. These challenges stem from the inherent variability, noise, and complexity of medical images, making it difficult to obtain accurate and reliable results. In this chapter, the focus is on leveraging bio-inspired algorithms to address these challenges and improve the analysis and interpretation of medical images. Bio-inspired algorithms draw inspiration from natural processes, such as evolution, swarm behavior, neural networks, and genetic programming. It addresses the challenges and requirements specific to each modality and how bio-inspired algorithms can be adapted and tailored to meet those needs.","author":[{"family":"Ezhilarasan","given":"K"},{"family":"Somasundaram","given":"K"},{"family":"Kalaiselvi","given":"T"},{"family":"Somasundaram","given":"Praveenkumar"},{"family":"Selvi","given":"SK"},{"family":"Jeevarekha","given":"A"}],"issued":{"date-parts":[[2024]]},"DOI":"10.4018/979-8-3693-1131-8.ch002","URL":"https://doi.org/10.4018/979-8-3693-1131-8.ch002","source":"crossref"},{"id":"doi:10.52700/scir.v6i1.149","type":"article-journal","title":"Quantum-Inspired Cryptography Protocols for Enhancing Security in Cloud Computing Infrastructures","abstract":"As we know, cloud computing plays a vital role in our lives. It also plays a pivotal role in modern business and information technology landscapes. But with this, there is a rise in concerns regarding the security and privacy of all data which is stored in the cloud. To tackle these concerns some effective and traditional cryptographic methods are present but they face potential vulnerabilities due to advent of quantum computing. It also threatens the basis of widely used encryption algorithms and models. This research focuses on the usage of quantum mechanics inspired cryptographic protocols, which will help to strengthen the security of cloud computing models and infrastructures. The research begins with a detailed overview of the current situation and state of security of cloud computing. There is identification of all challenges and vulnerabilities faced due to current, classical and traditional cryptographic techniques and algorithms. As we know that there is a very close arrival of quantum computing, we are surrounded by mighty and latent threats because of quantum computing algorithms that can easily overcome the widely accepted and used encryption standards and models. So, there is a dire need to develop and implement some alternative cryptographic strategies and measures. To overcome these challenges, this research proposes the creation and usage of quantum-inspired cryptographic protocols which are designed to resist attacks from both traditional and quantum vulnerabilities. These protocols are inspired from quantum mechanics, like superposition and entanglement. These protocols will help to design and create cryptographic systems and models that offer enhanced security. The research focuses on the theoretical foundations of these quantum mechanics inspired protocols and assesses their practical feasibility within cloud computing infrastructures and environment.This research contains information regarding the challenges associated with integration and deployment of quantum inspired cryptographic solutions in cloud computing. It evaluates the scalability, efficiency, performance and computational overhead of these quantum inspired protocols to ensure their practical feasibility and capability in existing cloud computing environment and infrastructures. This research also includes comparison performed with quantum cryptographic protocols and traditional cryptographic methods. The results aim to provide evidence that quantum cryptographic techniques are better, feasible and secure options to protect data over cloud. In Conclusion, the research aims to provide an insight that development and usage of advanced security measures can withstand and handle evolving threats, ensuring the continued security and protection of data in cloud computing environment by providing theoretical insights, analysis and evolution.","author":[{"family":"Tariq","given":"Laiba"},{"family":"Atta","given":"Ayesha"},{"family":"Farooq","given":"Umer"},{"family":"Anwar","given":"Nida"},{"family":"Asim","given":"Muhammad"},{"family":"Tabassum","given":"Nadia"}],"issued":{"date-parts":[[2024]]},"DOI":"10.52700/scir.v6i1.149","URL":"https://doi.org/10.52700/scir.v6i1.149","source":"crossref"},{"id":"doi:10.4018/979-8-3693-4159-9.ch006","type":"article-journal","title":"Neuro-Inspired Algorithms to Enhance Cryptography","abstract":"Neurocryptology objective is to create novel and creative approaches to improve the security of cryptographic systems by incorporating knowledge from neuroscience, particularly the study of the cognitive functions of the human brain. Instead of using mathematical methods and keys to secure communication and data, neurocryptology makes use of knowledge about how the brain processes information and spots patterns. The form and operation of neural networks and biological brains serve as inspiration for neural-inspired algorithms, which have the potential to improve cryptography in several ways. Encryption keys can be created using neural networks. In order to improve encryption techniques, neural networks can be incorporated. Behavioral biometric authentication techniques can be improved by neural network-inspired algorithms. This chapter is discussing about the neuro inspired algorithms to enhance cryptographic protocols. It reviewing the different algorithms that underlie well-known cryptographic cryptosystems","author":[{"family":"Thenmozhi","given":"R"},{"family":"Vetriselvi","given":"D"},{"family":"Jovith","given":"AA"}],"issued":{"date-parts":[[2024]]},"DOI":"10.4018/979-8-3693-4159-9.ch006","URL":"https://doi.org/10.4018/979-8-3693-4159-9.ch006","source":"crossref"},{"id":"doi:10.1093/noajnl/vdae124","type":"article-journal","title":"Fertility preserving techniques in neuro-oncology patients: A systematic review","abstract":"Abstract Background Advancements in cancer treatments have enhanced survival rates and quality of life for patients with central nervous system (CNS) tumors. There is growing recognition of the significance of fertility preservation methods. Currently, techniques, including oocyte cryopreservation and sperm cryopreservation are established. Nevertheless, oncologists may exhibit reluctance when referring patients to reproductive specialists. This review aimed to assess the best evidence for fertility preservation techniques used in patients with CNS cancers and evaluate outcomes relating to their success and complications. Methods Two reviewers performed a search of Pubmed, Embase, Medline, Cochrane, and Google Scholar. Papers were included if they reported at least 1 fertility preservation technique in a neuro-oncology patient. Non-English studies, editorials, animal studies, and guidelines were excluded. Meta-analysis was performed using the random effects model. Results Sixteen studies containing data from 237 participants (78.8% female) were included in the systematic review and meta-analysis, of whom 110 (46.4%) underwent fertility preservation techniques. All patients (100%) successfully underwent fertility preservation with 1 participant (2.9%) returning to rewarm their oocytes, embryos or sperm. On average, 17.8 oocytes were retrieved with 78%, ultimately being cryopreserved. Five (6.0%) patients successfully conceived 9 healthy-term children after utilizing their cryopreserved sperm, embryos, or oocytes. Moreover, 6 patients successfully conceived naturally or using intrauterine insemination, resulting in 7 healthy-term children. Conclusions Fertility preservation techniques could offer a safe and effective way for neuro-oncology patients to deliver healthy-term babies following treatment. However, further studies concerning risks, long-term pregnancy outcomes, and cost-effectiveness are needed.","author":[{"family":"Osborne-Grinter","given":"Maia"},{"family":"Sanghera","given":"Jasleen"},{"family":"Bianca","given":"Offorbuike"},{"family":"Kaliaperumal","given":"Chandrasekaran"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1093/noajnl/vdae124","URL":"https://doi.org/10.1093/noajnl/vdae124","source":"crossref"},{"id":"doi:10.36227/techrxiv.21837027.v1","type":"article-journal","title":"Brain Inspired Computing: A Systematic Survey and Future Trends","abstract":"Brain Inspired Computing (BIC) is an emerging research field that aims to build fundamental theories, models, hardware architectures, and application systems toward more general Artificial Intelligence (AI) by learning from the information processing mechanisms or structures/functions of biological nervous systems. It is regarded as one of the most promising research directions for future intelligent computing in the post-Moore era. In the past few years, various new schemes in this field have sprung up to explore more general AI. These works are quite divergent in the aspects of modeling/algorithm, software tool, hardware platform, and benchmark data, since BIC is an interdisciplinary field that consists of many different domains, including computational neuroscience, artificial intelligence, computer science, statistical physics, material science, microelectronics and so forth. This situation greatly impedes researchers from obtaining a clear picture and getting started in the right way. Hence, there is an urgent requirement to do a comprehensive survey in this field to help correctly recognize and analyze such bewildering methodologies. What are the key issues to enhance the development of BIC? What roles do the current mainstream technologies play in the general framework of BIC? Which techniques are truly useful in real-world applications? These questions largely remain open. To address the above issues, in this survey we first clarify the biggest challenge of BIC: how can AI models benefit from the recent advancements in computational neuroscience? With this challenge in mind, we will focus on discussing the concept of BIC and summarize four components of BIC infrastructure development: 1) modeling/algorithm; 2) hardware platform; 3) software tool; and 4) benchmark data. For each component, we will summarize its recent progress, main challenges to resolve, and future trends. On the basis of these studies, we present a general framework for the real-world applications of BIC systems, which is promising to benefit both AI and brain science. Finally, we claim that it is extremely important to build a research ecology to promote prosperity continuously in this field.","author":[{"family":"Li","given":"Guoqi"},{"family":"Deng","given":"Lei"},{"family":"Tang","given":"Huajing"},{"family":"Pan","given":"Gang"},{"family":"Tian","given":"Yonghong"},{"family":"Roy","given":"Kaushik"},{"family":"Maass","given":"Wolfgang"}],"issued":{"date-parts":[[2023]]},"DOI":"10.36227/techrxiv.21837027.v1","URL":"https://doi.org/10.36227/techrxiv.21837027.v1","source":"crossref"},{"id":"doi:10.1145/3604550","type":"article-journal","title":"A Survey of Quantum-cognitively Inspired Sentiment Analysis Models","abstract":"Quantum theory, originally proposed as a physical theory to describe the motions of microscopic particles, has been applied to various non-physics domains involving human cognition and decision-making that are inherently uncertain and exhibit certain non-classical, quantum-like characteristics. Sentiment analysis is a typical example of such domains. In the last few years, by leveraging the modeling power of quantum probability (a non-classical probability stemming from quantum mechanics methodology) and deep neural networks, a range of novel quantum-cognitively inspired models for sentiment analysis have emerged and performed well. This survey presents a timely overview of the latest developments in this fascinating cross-disciplinary area. We first provide a background of quantum probability and quantum cognition at a theoretical level, analyzing their advantages over classical theories in modeling the cognitive aspects of sentiment analysis. Then, recent quantum-cognitively inspired models are introduced and discussed in detail, focusing on how they approach the key challenges of the sentiment analysis task. Finally, we discuss the limitations of the current research and highlight future research directions.","author":[{"family":"Liu","given":"Yaochen"},{"family":"Li","given":"Qiuchi"},{"family":"Wang","given":"Benyou"},{"family":"Zhang","given":"Yazhou"},{"family":"Song","given":"Dawei"}],"issued":{"date-parts":[[2023]]},"DOI":"10.1145/3604550","URL":"https://doi.org/10.1145/3604550","source":"crossref"},{"id":"doi:10.4018/979-8-3693-4001-1.ch020","type":"article-journal","title":"Quantum-Inspired Machine Learning for Chemical Reaction Path Prediction","abstract":"The purpose of this study is to provide a novel method for predicting chemical response routes that is designed to exercise machine literacy techniques inspired by the concept of amount. When it comes to addressing the large number of mechanical interactions that are needed in chemical reactions, traditional types of response path vaticination frequently face obstacles. Within the scope of this investigation, the authors apply the ideas of amount computing in order to create a machine literacy framework that is inspired by amount computing and is developed for the purpose of providing accurate and efficient vaticination of response paths. The solution that has been proposed combines the suggestive power of algorithms that are inspired by amounts with the scalability and versatility of machine literacy models. This framework has been shown to have greater performance in predicting reaction courses when compared to conventional methods. This was demonstrated through extensive testing and confirmation on a variety of chemical systems.","author":[{"family":"Neelima","given":"P"},{"family":"Satyanarayana","given":"V"},{"family":"Sravanthi","given":"KB"},{"family":"Sherin","given":"K"}],"issued":{"date-parts":[[2024]]},"DOI":"10.4018/979-8-3693-4001-1.ch020","URL":"https://doi.org/10.4018/979-8-3693-4001-1.ch020","source":"crossref"},{"id":"doi:10.36227/techrxiv.21837027","type":"article-journal","title":"Brain Inspired Computing: A Systematic Survey and Future Trends","abstract":"&lt;p&gt;Brain Inspired Computing (BIC) is an emerging research field that aims to build fundamental theories, models, hardware architectures, and application systems toward more general Artificial Intelligence (AI) by learning from the information processing mechanisms or structures/functions of biological nervous systems. It is regarded as one of the most promising research directions for future intelligent computing in the post-Moore era. In the past few years, various new schemes in this field have sprung up to explore more general AI. These works are quite divergent in the aspects of modeling/algorithm, software tool, hardware platform, and benchmark data, since BIC is an interdisciplinary field that consists of many different domains, including computational neuroscience, artificial intelligence, computer science, statistical physics, material science, microelectronics and so forth. This situation greatly impedes researchers from obtaining a clear picture and getting started in the right way. Hence, there is an urgent requirement to do a comprehensive survey in this field to help correctly recognize and analyze such bewildering methodologies. What are the key issues to enhance the development of BIC? What roles do the current mainstream technologies play in the general framework of BIC? Which techniques are truly useful in real-world applications? These questions largely remain open. To address the above issues, in this survey we first clarify the biggest challenge of BIC: how can AI models benefit from the recent advancements in computational neuroscience? With this challenge in mind, we will focus on discussing the concept of BIC and summarize four components of BIC infrastructure development: 1) modeling/algorithm; 2) hardware platform; 3) software tool; and 4) benchmark data. For each component, we will summarize its recent progress, main challenges to resolve, and future trends. On the basis of these studies, we present a general framework for the real-world applications of BIC systems, which is promising to benefit both AI and brain science. Finally, we claim that it is extremely important to build a research ecology to promote prosperity continuously in this field.&lt;/p&gt;","author":[{"family":"Li","given":"Guoqi"},{"family":"Deng","given":"Lei"},{"family":"Tang","given":"Huajing"},{"family":"Pan","given":"Gang"},{"family":"Tian","given":"Yonghong"},{"family":"Roy","given":"Kaushik"},{"family":"Maass","given":"Wolfgang"}],"issued":{"date-parts":[[2023]]},"DOI":"10.36227/techrxiv.21837027","URL":"https://doi.org/10.36227/techrxiv.21837027","source":"crossref"},{"id":"doi:10.4018/979-8-3693-6834-3.ch002","type":"article-journal","title":"Task Scheduling Strategy Using Chaotic Whale Optimization Algorithm in Cloud Computing","abstract":"Cloud computing is becoming popular because it can provide cloud consumers with IT services scaled up globally over the internet. These services include platforms, applications, and infrastructure. Moreover, cloud computing can be provided on demand and offered in different pricing packages. To schedule the task optimally in a cloud environment is considered an NP-hard problem, which has become complex with the introduction of variables such as resource dynamicity and on-demand consumer applications. The proposed research introduces a Whale Optimization Algorithm (WOA) incorporating a transfer function (TF) and a tent chaotic map to tackle scheduling challenges in cloud computing. The performance of the proposed chaotic-based whale optimization algorithm (CWOA) is compared to that of well-known metaheuristics methods. The results show that CWOA may significantly reduce the makespan of the task scheduling problem compared to standard Grey Wolf Optimizer (GWO) and BAT algorithms. Furthermore, it converges quickly as the search space grows more prominent, making it suitable for large-scale scheduling issues.","author":[{"family":"Qasim","given":"Mohammad"},{"family":"Sajid","given":"Mohammad"},{"family":"Rajak","given":"Ranjit"},{"family":"Shahid","given":"Mohammad"}],"issued":{"date-parts":[[2024]]},"DOI":"10.4018/979-8-3693-6834-3.ch002","URL":"https://doi.org/10.4018/979-8-3693-6834-3.ch002","source":"crossref"},{"id":"doi:10.4018/979-8-3693-1131-8.ch007","type":"article-journal","title":"Endometrial Cancer Detection Using Pipeline Biopsies Through Machine Learning Techniques","abstract":"Endometrial carcinoma (EC) is a common uterine cancer that leads to morbidity and death linked to cancer. Advanced EC diagnosis exhibits a subpar treatment response and requires a lot of time and money. Data scientists and oncologists focused on computational biology due to its explosive expansion and computer-aided cancer surveillance systems. Machine learning offers prospects for drug discovery, early cancer diagnosis, and efficient treatment. It may be pertinent to use ML techniques in EC diagnosis, treatments, and prognosis. Analysis of ML utility in EC may spur research in EC and help oncologists, molecular biologists, biomedical engineers, and bioinformaticians advance collaborative research in EC. It also leads to customised treatment and the growing trend of using ML approaches in cancer prediction and monitoring. An overview of EC, its risk factors, and diagnostic techniques are covered in this study. It concludes a thorough investigation of the prospective ML modalities for patient screening, diagnosis, prognosis, and the deep learning models, which gave the good accuracy.","author":[{"family":"Varshini","given":"Vemasani"},{"family":"Raja","given":"Maheswari"},{"family":"Jagannathan","given":"Sharath"}],"issued":{"date-parts":[[2024]]},"DOI":"10.4018/979-8-3693-1131-8.ch007","URL":"https://doi.org/10.4018/979-8-3693-1131-8.ch007","source":"crossref"},{"id":"doi:10.4018/979-8-3693-7076-6.ch004","type":"article-journal","title":"Quantum-Inspired Algorithms for AI and Machine Learning","abstract":"Quantum-inspired algorithms help handle complex optimization and inference challenges in AI and ML. Quantum-like algorithms based on quantum mechanics address challenges classical computer paradigms can't. These algorithms innovate data handling, solution finding, and pattern recognition. Quantum effects like superposition, entanglement, and tunneling influenced them. In particular, they work in parallel to explore large solution spaces and speed up processing for large datasets. Next, quantum-inspired algorithms including quantum annealing, optimization, and neural networks are discussed. Simulated annealing and the quantum approximate optimization algorithm (QAOA) make combinatorial optimization tasks in AI and ML easier. Quantum-inspired evolutionary and swarm intelligence algorithms tackle multi-modal optimization issues quickly. Furthermore, quantum-inspired neural networks (QNNs) have revolutionized deep learning. QNNs use quantum gates and quantum circuits to improve classification, regression, and generative modeling.","author":[{"family":"Pandurangan","given":"Kamaleswari"},{"family":"Priyadharshini","given":"A"},{"family":"Taseen","given":"Rakheeba"},{"family":"Galebathullah","given":"B"},{"family":"Yaseen","given":"Haseeba"},{"family":"Ravichandran","given":"P"}],"issued":{"date-parts":[[2024]]},"DOI":"10.4018/979-8-3693-7076-6.ch004","URL":"https://doi.org/10.4018/979-8-3693-7076-6.ch004","source":"crossref"},{"id":"doi:10.4018/979-8-3693-6303-4.ch016","type":"article-journal","title":"Neuromorphic Software Tools and Development Environments","abstract":"We are seeing a technological transformation now that was unthinkable ten years ago. Although introducing artificial intelligence (AI) in contemporary business theoretically permits unrestricted expansion, the dreaded power-wall issue in the parallel computing paradigm prevents us from fully using AI's potential. Because they are expected to operate at extremely low power, modern Neuromorphic accelerators provide a profitable substitute for conventional artificial neural network (ANN) accelerators for deep learning (DL). Neuromorphic accelerators are centred on Spiking Neural Networks (SNN), which seek to mimic the extremely energy-efficient mechanism operating in our brains. This chapter covers the general overview of Neuromorphic software tools and development environments, including platforms, frameworks, and best practices.","author":[{"family":"Sailaja","given":"D"},{"family":"Sharma","given":"Yogesh"},{"family":"Nune","given":"VLM"}],"issued":{"date-parts":[[2024]]},"DOI":"10.4018/979-8-3693-6303-4.ch016","URL":"https://doi.org/10.4018/979-8-3693-6303-4.ch016","source":"crossref"},{"id":"doi:10.1093/neuonc/noae165.0577","type":"article-journal","title":"DISP-03. HEALTH DISPARITIES IN NEURO-ONCOLOGY","abstract":"Abstract Significant disparities in diagnosis, research, and treatment exist in neuro-oncology. Patients who identify as racial or ethnic minorities, sexual or gender minorities, elderly, rural, or have low socioeconomic status are more likely to have adverse outcomes. Using PubMed and expert opinion, we summarized the existing literature on health disparities in neuro-oncology. We categorized disparities based on primary central nervous system tumors (CNS), secondary CNS tumors, pediatric neuro-oncology, clinical trial enrollment and diversity, and financial toxicity. For patients with primary CNS tumors, we found that marginalized groups faced delays in diagnosis, lower rates of resection, and lower rates of postoperative radiation. For secondary CNS tumors, marginalized groups had lower rates of indicated cancer screening, less access to genetic testing, less access to advanced imaging, and lower rates of stereotactic radiosurgery and targeted therapies. Pediatric neuro-oncology patients had differences in morbidity and mortality based on socioeconomic status and race and faced substantial delays in diagnosis. Financial toxicity, the side effects of the high financial cost of a cancer diagnosis, disproportionately affects marginalized populations. This may be further exacerbated by the use of long-term medications such as IDH inhibitors. There is also unequal representation of diverse populations in neuro-oncology clinical trials. Compared to other cancer subspecialties, there is less available data on disparities, however, the types of disparities such as delays in diagnosis and outcomes are similar. Notably, there are not many interventions focused on mitigating these disparities. There needs to be additional research on health disparities in neuro-oncology, including both descriptions of existing disparities, but also on methods and interventions to address these disparities.","author":[{"family":"Budhu","given":"Joshua"},{"family":"Michaelson","given":"Nara"},{"family":"Watsula","given":"Amanda"},{"family":"Bakare","given":"Anu"},{"family":"Mohamadpour","given":"Mali"},{"family":"Chukwueke","given":"Ugonma"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1093/neuonc/noae165.0577","URL":"https://doi.org/10.1093/neuonc/noae165.0577","source":"crossref"},{"id":"doi:10.1038/s41467-023-39070-8","type":"article-journal","title":"Eye accommodation-inspired neuro-metasurface focusing","abstract":"Abstract The human eye, which relies on a flexible and controllable lens to focus light onto the retina, has inspired many scientific researchers to understand better and imitate the biological vision system. However, real-time environmental adaptability presents an enormous challenge for artificial eye-like focusing systems. Inspired by the mechanism of eye accommodation, we propose a supervised-evolving learning algorithm and design a neuro-metasurface focusing system. Driven by on-site learning, the system exhibits a rapid response to ever-changing incident waves and surrounding environments without any human intervention. Adaptive focusing is achieved in several scenarios with multiple incident wave sources and scattering obstacles. Our work demonstrates the unprecedented potential for real-time, fast, and complex electromagnetic (EM) wave manipulation for various purposes, such as achromatic, beam shaping, 6 G communication, and intelligent imaging.","author":[{"family":"Lu","given":"Huan"},{"family":"Zhao","given":"Jiwei"},{"family":"Zheng","given":"Bin"},{"family":"Qian","given":"Chao"},{"family":"Cai","given":"Tong"},{"family":"Li","given":"Erping"},{"family":"Chen","given":"Hongsheng"}],"issued":{"date-parts":[[2023]]},"DOI":"10.1038/s41467-023-39070-8","URL":"https://doi.org/10.1038/s41467-023-39070-8","source":"crossref"},{"id":"doi:10.1049/cim2.70004","type":"article-journal","title":"Spiking neural network tactile classification method with faster and more accurate membrane potential representation","abstract":"Abstract Robot perception is an important topic in artificial intelligence field, and tactile recognition in particular is indispensable for human–computer interaction. Efficiently classifying data obtained by touch sensors has long been an issue. In recent years, spiking neural networks (SNNs) have been widely used in tactile data categorisation due to their temporal information processing benefits, low power consumption, and high biological dependability. However, traditional SNN classification methods often encounter under‐convergence when using membrane potential representation, decreasing their classification accuracy. Meanwhile, due to the time‐discrete nature of SNN models, classification requires a significant time overhead, which restricts their real‐time tactile sensing application potential. Considering these concerns, the authors propose a faster and more accurate SNN tactile classification approach using improved membrane potential representation. This method effectively overcomes model convergence problems by optimising the membrane potential expression and the relationship between the loss function and network parameters while significantly reducing the time overhead and enhancing the classification accuracy and robustness of the model. The experimental results show that the propose approach improves the classification accuracy by 4.16% and 2.71% and reduces the overall time by 8.00% and 8.14% on the EvTouch‐Containers dataset and EvTouch‐Objects dataset, respectively, when compared with existing models.","author":[{"family":"Yang","given":"Jing"},{"family":"Yu","given":"Zukun"},{"family":"Ji","given":"Xiaoyang"},{"family":"Su","given":"Zhidong"},{"family":"Li","given":"Shaobo"},{"family":"Cao","given":"Yang"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1049/cim2.70004","URL":"https://doi.org/10.1049/cim2.70004","source":"crossref"},{"id":"doi:10.3390/a17040156","type":"article-journal","title":"Spike-Weighted Spiking Neural Network with Spiking Long Short-Term Memory: A Biomimetic Approach to Decoding Brain Signals","abstract":"Background. Brain–machine interfaces (BMIs) offer users the ability to directly communicate with digital devices through neural signals decoded with machine learning (ML)-based algorithms. Spiking Neural Networks (SNNs) are a type of Artificial Neural Network (ANN) that operate on neural spikes instead of continuous scalar outputs. Compared to traditional ANNs, SNNs perform fewer computations, use less memory, and mimic biological neurons better. However, SNNs only retain information for short durations, limiting their ability to capture long-term dependencies in time-variant data. Here, we propose a novel spike-weighted SNN with spiking long short-term memory (swSNN-SLSTM) for a regression problem. Spike-weighting captures neuronal firing rate instead of membrane potential, and the SLSTM layer captures long-term dependencies. Methods. We compared the performance of various ML algorithms during decoding directional movements, using a dataset of microelectrode recordings from a macaque during a directional joystick task, and also an open-source dataset. We thus quantified how swSNN-SLSTM performed compared to existing ML models: an unscented Kalman filter, LSTM-based ANN, and membrane-based SNN techniques. Result. The proposed swSNN-SLSTM outperforms both the unscented Kalman filter, the LSTM-based ANN, and the membrane based SNN technique. This shows that incorporating SLSTM can better capture long-term dependencies within neural data. Also, our proposed swSNN-SLSTM algorithm shows promise in reducing power consumption and lowering heat dissipation in implanted BMIs.","author":[{"family":"Mcmillan","given":"Kyle"},{"family":"So","given":"Rosa"},{"family":"Libedinsky","given":"Camilo"},{"family":"Ang","given":"Kai"},{"family":"Premchand","given":"Brian"}],"issued":{"date-parts":[[2024]]},"DOI":"10.3390/a17040156","URL":"https://doi.org/10.3390/a17040156","source":"crossref"},{"id":"doi:10.1186/s40708-023-00192-w","type":"article-journal","title":"Prediction and detection of virtual reality induced cybersickness: a spiking neural network approach using spatiotemporal EEG brain data and heart rate variability","abstract":"Abstract Virtual Reality (VR) allows users to interact with 3D immersive environments and has the potential to be a key technology across many domain applications, including access to a future metaverse. Yet, consumer adoption of VR technology is limited by cybersickness (CS)—a debilitating sensation accompanied by a cluster of symptoms, including nausea, oculomotor issues and dizziness. A leading problem is the lack of automated objective tools to predict or detect CS in individuals, which can then be used for resistance training, timely warning systems or clinical intervention. This paper explores the spatiotemporal brain dynamics and heart rate variability involved in cybersickness and uses this information to both predict and detect CS episodes. The present study applies deep learning of EEG in a spiking neural network (SNN) architecture to predict CS prior to using VR (85.9%, F7) and detect it (76.6%, FP1, Cz). ECG-derived sympathetic heart rate variability (HRV) parameters can be used for both prediction (74.2%) and detection (72.6%) but at a lower accuracy than EEG. Multimodal data fusion of EEG and sympathetic HRV does not change this accuracy compared to ECG alone. The study found that Cz (premotor and supplementary motor cortex) and O2 (primary visual cortex) are key hubs in functionally connected networks associated with both CS events and susceptibility to CS. F7 is also suggested as a key area involved in integrating information and implementing responses to incongruent environments that induce cybersickness. Consequently, Cz, O2 and F7 are presented here as promising targets for intervention.","author":[{"family":"Yang","given":"Alexander"},{"family":"Kasabov","given":"Nikola"},{"family":"Cakmak","given":"Yusuf"}],"issued":{"date-parts":[[2023]]},"DOI":"10.1186/s40708-023-00192-w","URL":"https://doi.org/10.1186/s40708-023-00192-w","source":"crossref"},{"id":"doi:10.1101/2023.05.27.542589","type":"article-journal","title":"Slow ramping emerges from spontaneous fluctuations in spiking neural networks","abstract":"Abstract Highlights 1. We reveal a mechanism for slow-ramping signals before spontaneous voluntary movements. 2. Slow synapses stabilize spontaneous fluctuations in spiking neural network. 3. We validate model predictions in human frontal cortical single-neuron recordings. 4. The model recreates the readiness potential in an EEG proxy signal. 5. Neurons that ramp together had correlated activity before ramping onset. The capacity to initiate actions endogenously is critical for goal-directed behavior. Spontaneous voluntary actions are typically preceded by slow-ramping activity in medial frontal cortex that begins around two seconds before movement, which may reflect spontaneous fluctuations that influence action timing. However, the mechanisms by which these slow ramping signals emerge from single-neuron and network dynamics remain poorly understood. Here, we developed a spiking neural-network model that produces spontaneous slow ramping activity in single neurons and population activity with onsets ∼2 seconds before threshold crossings. A key prediction of our model is that neurons that ramp together have correlated firing patterns before ramping onset. We confirmed this model-derived hypothesis in a dataset of human single neuron recordings from medial frontal cortex. Our results suggest that slow ramping signals reflect bounded spontaneous fluctuations that emerge from quasi-winner-take-all dynamics in clustered networks that are temporally stabilized by slow-acting synapses.","author":[{"family":"Gavenas","given":"Jake"},{"family":"Rutishauser","given":"Ueli"},{"family":"Schurger","given":"Aaron"},{"family":"Maoz","given":"Uri"}],"issued":{"date-parts":[[2023]]},"DOI":"10.1101/2023.05.27.542589","URL":"https://doi.org/10.1101/2023.05.27.542589","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-2648993/v1","type":"article-journal","title":"Intrusion Detection for IoT Network Security with Deep Neural Network","abstract":"Abstract one of the most important challenges of the Internet of Things is security. Today, the Internet of Things has found an important place in information technology and human daily life. One of the main challenges of the Internet of Things is security. One of the common methods to intervene in Internet of Things services is Denial of Service (DoS) attacks and Distributed Denial of Service (DDoS) attacks. Therefore, intrusion detection systems or IDSs are currently the main and most complete parts of a network monitoring system. This paper uses the CICIDS 2017 data set to present an intrusion detection model in software-driven Internet of Things networks based on deep neural networks to detect distributed denial of service attacks and several other cyber attacks. In addition, we explored effective deep learning models to represent cyber security knowledge in Internet of Things networks, including CNN, DenseNet, CNN and LSTM hybrid models, and our proposed model.","author":[{"family":"Morshedi","given":"Roya"},{"family":"Matinkhah","given":"SM"},{"family":"Sadeghi","given":"Mohammad"}],"issued":{"date-parts":[[2023]]},"DOI":"10.21203/rs.3.rs-2648993/v1","URL":"https://doi.org/10.21203/rs.3.rs-2648993/v1","source":"crossref"},{"id":"doi:10.4018/979-8-3693-6303-4.ch017","type":"article-journal","title":"Neuromorphic Computing","abstract":"The exponential growth of data and information has stimulated technological progress in computing systems that utilize them to effectively discover patterns and produce important insights. Neural network algorithms have been applied to conventional silicon transistor-based hardware to do highly parallel computations, drawing inspiration from the structure and functions of biological synapses and neurons in the brain. Nevertheless, synapses composed of many transistors are limited to storing binary data, and the utilization of intricate silicon neuron circuits to handle these digital states poses challenges in achieving low-power and low-latency computing. This study examines the significance of developing memories and switches for synaptic and neural components in building Neuromorphic systems that can efficiently conduct cognitive tasks and recognition. This chapter closely examines and rates the latest progress in Neuromorphic computing, focusing on how these changes impact edge and Internet of Things technologies. It is also being thought about how to use tiny switches and short-term memory to copy the action of neurons. Once this is done, more Studies in many areas should be able to focus on the design, circuitry, and devices of Neuromorphic systems.","author":[{"family":"Pandey","given":"Devendra"},{"family":"Sharma","given":"Yogesh"},{"family":"Kumar","given":"Nimish"}],"issued":{"date-parts":[[2024]]},"DOI":"10.4018/979-8-3693-6303-4.ch017","URL":"https://doi.org/10.4018/979-8-3693-6303-4.ch017","source":"crossref"},{"id":"doi:10.4018/978-1-6684-6596-7.ch011","type":"article-journal","title":"Biologically Inspired SNN for Robot Control","abstract":"In recent years, there has been a trend towards more sophisticated robot control. This has been driven by advances in artificial intelligence (AI) and machine learning, which have enabled robots to become more autonomous and effective in completing tasks. One trend is towards using AI for robot control. This involves teaching robots how to carry out tasks by providing them with data and letting them learn from it. This approach can be used for tasks such as object recognition and navigation. Another trend is towards using machine learning for robot control. This involves using algorithms to learn from data and improve the performance of the robot. This approach can be used for tasks such as object recognition and navigation. A third trend is towards using more sophisticated sensors for robot control. This includes using sensors that can detect things such as temperature, humidity, and pressure.","author":[{"family":"Ganeshkumar","given":"S"},{"family":"Maniraj","given":"J"},{"family":"Gokul","given":"S"},{"family":"Ramaswamy","given":"Krishnaraj"}],"issued":{"date-parts":[[2023]]},"DOI":"10.4018/978-1-6684-6596-7.ch011","URL":"https://doi.org/10.4018/978-1-6684-6596-7.ch011","source":"crossref"},{"id":"doi:10.1093/neuonc/noae162","type":"article-journal","title":"Design and conduct of theranostic trials in neuro-oncology: Challenges and opportunities","abstract":"Abstract Theranostics is a new treatment modality integrating molecular imaging with targeted radionuclide therapy. Theranostic agents have received regulatory approval for some systemic cancers and have therapeutic potential in neuro-oncology. As clinical trials are developed to evaluate the efficacy of theranostic agents in brain tumors, specific considerations will have to be considered, taking into account lessons learned from previous studies examining other treatment modalities in neuro-oncology. These include the need for molecular imaging or surgical window-of-opportunity studies to confirm adequate passage across the blood-brain barrier, optimize eligibility criteria, and selection of the most appropriate response criteria and endpoints to address issues such as pseudoprogression. This review will discuss some of the issues that should be considered when designing clinical trials for theranostic agents.","author":[{"family":"Wen","given":"Patrick"},{"family":"Preusser","given":"Matthias"},{"family":"Albert","given":"Nathalie"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1093/neuonc/noae162","URL":"https://doi.org/10.1093/neuonc/noae162","source":"crossref"},{"id":"doi:10.1093/neuonc/noae064.721","type":"article-journal","title":"LMIC-04. BUILDING PEDIATRIC NEURO-ONCOLOGY CAPACITY IN LMICS THROUGH MULTIDISCIPLINARY EDUCATION AND COLLABORATION","abstract":"Abstract BACKGROUND The St. Jude Global Academy Neuro-Oncology Training Seminar (NOTS) is a course in pediatric neuro-oncology (PNO) designed for physicians from low-income and middle-income countries. The curriculum was systematically designed and includes 9 weeks of online learning and a 5-day in-person workshop. The NOTS has been run 3 times: 2019, 2022, and 2023. We sought to evaluate the outcomes of these courses. METHODS After the initial course where institutions were invited, applying institutions were selected based on the size of the neuro-oncology program, multidisciplinary representation, and perceived impact of the course. Data on course participants have been prospectively collected. A survey evaluating course impact was completed by graduates. RESULTS For 2022, 51 institutions from 29 countries applied and 15 were selected. In 2023, 60 institutions from 31 countries applied and 18 were selected. Overall, 41 institutions from 29 countries have participated in NOTS. Furthermore, 191 individuals, including 68 pediatric oncologists, 37 neurosurgeons, 22 radiation oncologists, 19 radiologists, 18 pathologists, and 27 other specialists have participated. Survey responses (n=39) describe that 47% did not have multidisciplinary tumor boards (MDTB) before NOTS. 55% of those institutions now have MDTB, all claiming to be influenced by their participation in NOTS. 89% of participants claim that multidisciplinary communication improved after the course. The participants of NOTS have continued to be engaged beyond the course through the creation of the Global Alliance in Pediatric Neuro-Oncology (GAPNO). GAPNO workstreams have included monthly cases discussions, a multisite retrospective review of medulloblastoma outcomes collecting &amp;gt;300 patients in 8 countries, and the design of a comprehensive tool to evaluate PNO services. CONCLUSIONS A multidisciplinary course focused on pediatric CNS tumors care for resource-limited settings can help expand international engagement in PNO. Furthermore, through multidisciplinary collaboration, the PNO community can be galvanized to expand activities in capacity building.","author":[{"family":"Moreira","given":"Daniel"},{"family":"Andujar","given":"Allyson"},{"family":"Qaddoumi","given":"Ibrahim"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1093/neuonc/noae064.721","URL":"https://doi.org/10.1093/neuonc/noae064.721","source":"crossref"},{"id":"doi:10.1093/neuonc/noae144.431","type":"article-journal","title":"P22.17.B THE NEURO-ONCOLOGICAL CAREGIVER ROLES AND RESPONSIBILITIES - A QUANTITATIVE SURVEY","abstract":"Abstract BACKGROUND In Denmark, cancer treatment is increasingly offered on an outpatient basis. This adds a responsibility on the caregivers, who often handle treatment agreements, provide 24/7 care, and offer practical support within the home environment. Especially, caregivers of individuals with brain cancer due to the disease’s unique characteristics, including cognitive and physical alterations. Qualitative exploration of the caregiver role has been undertaken, but there remains a notable absence of quantitative measurements regarding the burden faced by neuro-oncological caregivers. Therefore, this study aimed to quantitatively investigate the impact of caregiving on individuals caring for brain cancer patients. MATERIAL AND METHODS A cross-sectional study was conducted. The questionnaire used was the validated and standardized questionnaire “Caregiver roles and responsibilities scale” (CRRS). It consists of 51 questions grouped into 6 core areas: Support and Impact, Lifestyle, Emotional Health and Well-being, Self-care, Financial situation, and Work and Career. Responses are scored on a 5-point Likert scale ranging from 0 (“not at all”) to 4 (“very much”). The CRRS questionnaire has an official scoring guideline that calculates a score ranging from 0-152, in which 0 accounts for the most burdened, and 152 the least burdened. However, there isn’t a specific cut-off score to indicate when the CRRS score should raise concern. The questionnaire was distributed for two weeks to all caregivers attending the neuro-oncology outpatient clinic at Rigshospitalet. RESULTS Included were 27 primary caregivers (out of 30 eligible). 64% were female, and age was categorized in under 65 (younger) and over 65 (older); with a range from 39 to 80 years old. Data indicated that the younger caregivers (CRRS score = 87), those being in the initial 6 months of the disease (CRRS score = 85), and those receiving minimal to no practical assistance from family and friends (CRRS score = 90) were at a higher risk of burden, compared to those being older (CRRS score = 103), in the stage of 12-24 month post-diagnose (CRRS score = 106) and who received moderate to substantial support (CRRS score = 101). 18% of women rated their physical health as impaired to a large degree due to the patient’s disease, and none of the men noted this. CONCLUSION The CRRS questionnaire could serve as a valuable screening tool to identify the specific support required by individual caregivers. The study indicates that female caregivers under the age of 65, particularly within the initial 6 months of the disease onset, and those lacking practical support from friends and family, tend to experience elevated levels of burden. The study’s findings do not specify the frequency or timing for administering the CRRS questionnaire as a screening tool, initial steps could involve integrating the screening process within the first six months of caregiving.","author":[{"family":"Smith","given":"VR"},{"family":"Kjærgård","given":"SS"},{"family":"Piil","given":"K"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1093/neuonc/noae144.431","URL":"https://doi.org/10.1093/neuonc/noae144.431","source":"crossref"},{"id":"doi:10.1142/s1793292024500462","type":"article-journal","title":"Design of Nonlinear Autoregressive Neuro-Computing Structure for Bioconvective Micropolar Nanofluidic Model","abstract":"In the present arena, the tools of artificial intelligence (AI) play a significant role in research across various fields by enabling advanced data analysis, pattern recognition and decision-making. This research work presents the numerical investigation of bioconvective micropolar nanofluidic model (BCMNFM) by employing the knacks of AI-based nonlinear autoregressive (NAR) approach with a combination of backpropagated Levenberg–Marquardt neural networks (BLMNNs) represented as NAR-BLMNNs. This research work investigates the flow design to highlight the attributes of mass and heat exchange. A dataset for BCMNFM is created by applying the Adam numerical procedure by variation of unsteadiness parameter ([Formula: see text], magnetic field parameter ([Formula: see text] thermophoresis parameter (Nt), Brownian motion parameter (Nb), bioconvection Peclet number (Pe) and spin gradient viscosity parameter ([Formula: see text] The skills of AI-based NAR-BLMNNs technique is then utilized on the dataset created for BCMNFM to investigate the approximate solutions. The achieved and impactful values of performance consistently range between [Formula: see text] and [Formula: see text] across all scenarios of BCMNFM. The precision and the validation of predicted approach NAR-BLMNNs is exceptionally established by the graphical demonstration for all scenarios of MSE, regression metrics, error histograms and time series graphs. The numerical calculations attained through AI-based NAR-BLMNNs technique further rationalize the precision of the proposed methodology for solving the BCMNFM effectively and efficiently.","author":[{"family":"Shah","given":"Zahoor"},{"family":"Jamil","given":"Attika"},{"family":"Raja","given":"Muhammad"},{"family":"Shoaib","given":"Muhammad"},{"family":"Kiani","given":"Adiqa"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1142/s1793292024500462","URL":"https://doi.org/10.1142/s1793292024500462","source":"crossref"},{"id":"doi:10.2139/ssrn.4917843","type":"manuscript","title":"Intrusion detection in In-vehicle Networks using neuro computing","abstract":"A good way to detect illegitimate use is through motoring unusual activity in the In-vehicle Network (IVN). This paper proposes a detailed review&lt;br&gt;of the approaches used to detect intrusions in IVNs by applying methods inclusive of neuromorphic computing (NC). It explores Security threats to IVNs&lt;br&gt;and briefs security solutions using machine learning algorithms. NC offers effective intrusion detection systems to interpret intrusion attempts in the&lt;br&gt;incoming network traffic efficiently with minimal energy consumption. The review elaborates on IVN security solutions using NC. A multiscale histogram&lt;br&gt;method using a dataset based on a dictionary of Controller Area Network Identification (CANID) sequences subsequently applied with Deep Learning for&lt;br&gt;feature generation is explored. The results using Deep Convolutional Neural Network (DCNN) are proven to report very low error rates and false negatives.&lt;br&gt;The spiking neural network (SNN) based method incorporates presynaptic neuron modulation, which generates a spike in postsynaptic neurons when&lt;br&gt;membrane potential crosses the threshold. Another SNN-based method uses AWID and NSL-KDD intrusion detection datasets to demonstrate that nonleaky&lt;br&gt;neurons can be easily trained and show better performance. Supervised learning with indirect training is used to convert Deep Neural Networks (DNNs) to&lt;br&gt;SNNs with single-spike temporal coding, providing high-performance accuracy. A DNN is used to fetch statistical data from high-dimensional CAN packets&lt;br&gt;to extract corresponding features and provide an instant response to the intrusion. The Electronic Control Unit (ECU) embedded monitoring module identifies&lt;br&gt;the known attacks based on trained features. It uses a profiling module that records the attack and updates the system for upcoming data packets. Another&lt;br&gt;method uses KDD cup 99, which incorporates a mixture of symbolic and numeric values to detect intrusion. Intrusion Detection System (IDS) using the&lt;br&gt;Deep Learning (DL) method incorporates the Discrete vector factorization (DVF) method to generate synaptic weights, crossbar weight, and thresholds for&lt;br&gt;neurons and shows 90.12 % accuracy in network intrusion detection and shows 81.13 % recognition accuracy in detection of malicious and correct packets&lt;br&gt;with minimal power. From the various summarised experimental outcomes, it has been justified that NC delivers excellent classification and accuracy in&lt;br&gt;detection for IDS, and also consumes minimal power.","author":[{"family":"Parandkar","given":"Parag"},{"family":"Joshi","given":"Prashant"}],"issued":{"date-parts":[[2024]]},"DOI":"10.2139/ssrn.4917843","URL":"https://doi.org/10.2139/ssrn.4917843","source":"crossref"},{"id":"doi:10.4018/979-8-3693-6303-4.ch003","type":"article-journal","title":"A Systematic Review of Spiking Neural Networks and Their Applications","abstract":"These days, there is increasing curiosity regarding the topic of spiking neural networks (SNNs). Compared to artificial neural networks (ANNs), which are the subsequent equivalents, they bear a greater resemblance to the real neural networks found in the brain. SNNs are based on events such as neuromorphic factors; hardware based on SNNs may be less energy-intensive than ANNs. Since the energy usage would be far lower than that of typical deep learning models housed in the cloud today, this could result in a significant reduction in maintenance costs for neural network models. Such gear is still not readily accessible, however. This chapter presents a Systematic Review of Spiking Neural Networks and Their Applications. This study examines the benefits and drawbacks of various neural model types, coding techniques, methods for learning, and Neuromorphic platforms for computing. Based on these analyses, some anticipated developments are suggested, including balancing biological imitation and computing costs for neuron theories, the process of compounding coding techniques, unsupervised algorithms for learning in SNN, and digital-analog computation systems.","author":[{"family":"Singhal","given":"Tarun"},{"family":"Rani","given":"Ishta"},{"family":"Singh","given":"Divya"},{"family":"Kumar","given":"Bikram"},{"family":"Bhatia","given":"Vinay"},{"family":"Gupta","given":"Shubhi"}],"issued":{"date-parts":[[2024]]},"DOI":"10.4018/979-8-3693-6303-4.ch003","URL":"https://doi.org/10.4018/979-8-3693-6303-4.ch003","source":"crossref"},{"id":"doi:10.4018/979-8-3693-6834-3.ch008","type":"article-journal","title":"VM Placement in Cloud Computing Using Nature-Inspired Optimization Algorithms","abstract":"Cloud computing implements various techniques for the efficient utilization of computing resources. These resources are delivered over the internet, allowing users to access and manage them remotely. Often, these resources are provided to the users in the form of virtual machines (VMs). VMs are essentially software-based emulations of physical computers. In a cloud data center, numerous physical machines (PMs), also called servers, host various VMs. Placement of these VMs in Physical Machines is a critical task as there are many factors that need to be considered. Nature-inspired optimization algorithms such as Genetic Algorithms, Particle Swarm Optimization, and Ant Colony Optimization are inspired by natural phenomena and behaviour. These algorithms have been used in the past to generate near-optimal solutions in polynomial time for computationally intractable problems like VM Placement Problems (VPP). This chapter discusses how virtual machines are placed in cloud data centers using Nature-Inspired Optimization Algorithms.","author":[{"family":"Shah","given":"Monali"},{"family":"Rajwar","given":"Dipankar"},{"family":"Dehury","given":"Jitendra"},{"family":"Kumar","given":"Dinesh"}],"issued":{"date-parts":[[2024]]},"DOI":"10.4018/979-8-3693-6834-3.ch008","URL":"https://doi.org/10.4018/979-8-3693-6834-3.ch008","source":"crossref"},{"id":"doi:10.4018/979-8-3693-1131-8.ch001","type":"article-journal","title":"Bio-Inspired Algorithms Leveraging Blockchain Technology Enhancing Efficiency Security and Transparency","abstract":"Bio-inspired algorithms, which imitate the actions and procedures seen in biological systems, have proven to be incredibly effective at solving problems in a variety of fields. However, the combination of these algorithms with blockchain technology holds enormous promise for improving their potency while assuring effectiveness, efficiency, and security. This chapter gives a general overview of how bio-inspired algorithms are used effectively with blockchain technology, highlighting their main benefits and prospective uses. Bio-inspired algorithms can gain from improved security and confidence in their execution by utilizing the decentralized and immutable characteristics of blockchain. Blockchain technology offers a transparent and auditable platform that makes it easier to verify algorithmic operations and ensure the accuracy of data. Furthermore, distributed resource allocation and decision-making are made possible by blockchain's decentralized consensus mechanisms, promoting cooperation and collective intelligence.","author":[{"family":"Chitra","given":"P"},{"family":"Raja","given":"AS"},{"family":"Sivakumar","given":"V"}],"issued":{"date-parts":[[2024]]},"DOI":"10.4018/979-8-3693-1131-8.ch001","URL":"https://doi.org/10.4018/979-8-3693-1131-8.ch001","source":"crossref"},{"id":"doi:10.4018/979-8-3693-6303-4.ch001","type":"article-journal","title":"Introduction to Neuromorphic Computing","abstract":"Neuromorphic computing is an innovative and fast-evolving field poised to significantly impact human life in the coming decades. By emulating the brain's neural architecture using modern electronic systems, it aims to enhance computer architectures and revolutionize computing and machine learning tasks. Neuromorphic systems transcend traditional computing paradigms by integrating memory and computation in a massively parallel fashion, unlocking unprecedented computational efficiency and performance. This approach promises breakthroughs in pattern recognition, natural language processing, and autonomous decision-making. Additionally, neuromorphic computing could advance electronics and biology integration, leading to innovations like brain-computer interfaces and neuroprosthetic devices that revolutionize healthcare and human augmentation.","author":[{"family":"Londhe","given":"Supriya"},{"family":"Shivthare","given":"Sunayana"},{"family":"Sharma","given":"Yogeshkumar"}],"issued":{"date-parts":[[2024]]},"DOI":"10.4018/979-8-3693-6303-4.ch001","URL":"https://doi.org/10.4018/979-8-3693-6303-4.ch001","source":"crossref"},{"id":"doi:10.36227/techrxiv.170630368.82655420/v1","type":"article-journal","title":"Quantum-Inspired Differential Evolution with Decoding using Hashing for Efficient User Allocation in Edge Computing Environment","abstract":"Modern apps require high computing resources for real-time data processing, allowing app users (AUs) to access real-time information. Edge computing (EC) provides dynamic computing resources to AUs for real-time data processing. However, ESs in specific areas can only serve a limited number of AUs due to resource and coverage constraints. Hence, the app user allocation problem (AUAP) becomes challenging in the EC environment. In this paper, a quantum-inspired differential evolution algorithm (QDE-UA) is proposed for efficient user allocation in the EC environment. The quantum vector is designed to provide a complete solution to the AUAP. The fitness function considers factors such as minimum ES required, user allocation rate (UAR), energy consumption, and load balance. Extensive simulations are performed along with hypotheses-based statistical analyses (ANOVA, Friedman test) to show the significance of the proposed QDE-UA. The results indicate that QDE-UA outperforms existing strategies with an average UAR improvement of 116.63%, a 77.35% reduction in energy consumption, and 46.22% enhancement in load balance while utilizing 13.98% fewer ESs.","author":[{"family":"Bey","given":"Marlom"},{"family":"Kuila","given":"Pratyay"},{"family":"Naik","given":"Banavath"}],"issued":{"date-parts":[[2024]]},"DOI":"10.36227/techrxiv.170630368.82655420/v1","URL":"https://doi.org/10.36227/techrxiv.170630368.82655420/v1","source":"crossref"},{"id":"doi:10.5194/egusphere-egu24-8233","type":"article-journal","title":"Double Acoustic Emission events detection using U-net Neural Network","abstract":"In the past decade, the development of the Deep Neural Network formalism has emerged as a promising approach for addressing contemporary task in seismology, particularly in the effective and potentially automated processing of extensive datasets, such as seismograms. In this study, we introduce a 4D Neural Network (NN) based on the U-Net architecture, capable of simultaneously processing data from the entire seismic network. Our dataset comprises records/seismograms of Acoustic Emission (AE) events obtained during a laboratory loading experiment on a rock specimen. While AE event records share similarities with real seismograms, they exhibit simplifications in certain features.To assess the capability of the proposed NN in handling complex data, including occurrences of multiple events observed during experiments, we generated double-event seismograms through the augmentation of unambiguous single-event seismograms. These augmented datasets were employed for training, validation, and testing of the NN. Despite the individual station detection rate being approximately 30%, the simultaneous processing of multiple stations significantly increased efficiency, achieving an overall detection rate of 97%.In this work, we treat seismograms as \"images,\" adopting an approach that proves to be fruitful. The simultaneous processing of seismograms, coupled with this image-based treatment, demonstrates high potential for reliable automatic interpretation of seismic data. This approach (possibly combined with other methodologies), holds promise for seismogram processing.","author":[{"family":"Kolar","given":"Petr"},{"family":"Petružálek","given":"Matěj"},{"family":"Šílený","given":"Jan"},{"family":"Lokajíček","given":"Tomáš"}],"issued":{"date-parts":[[2024]]},"DOI":"10.5194/egusphere-egu24-8233","URL":"https://doi.org/10.5194/egusphere-egu24-8233","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-3172508/v1","type":"article-journal","title":"Heterogeneous 2D Memristor Array and Silicon Selector for Compute-in-Memory Hardware in Convolution Neural Networks","abstract":"Abstract Memristor crossbar arrays (CBAs) based on two-dimensional (2D) materials have emerged as a potential solution to overcome the limitations of energy consumption and latency associated with the conventional von Neumann architecture. However, current 2D memristor CBAs encounter specific challenges such as limited array size, high sneak path current, and lack of integration with peripheral circuits for hardware compute-in-memory (CIM) systems. In this work, we demonstrate a novel hardware CIM system that leverages the heterogeneous integration of scalable 2D hafnium diselenide (HfSe2) memristors and silicon (Si) selectors, as well as the integration between memristive CBAs and peripheral control-sensing circuits. The integrated 32 × 32 one-selector-one-memristor (1S1R) array effectively mitigates sneak current, exhibiting a high yield (89%) with notable uniformity. The integrated CBA demonstrates exceptional improvement of energy efficiency and response time comparable to state-of-the-art 2D materials-based memristors. To take advantage of low latency devices for achieving low energy systems, time-domain sensing circuits with the CBA are used, of which the power consumption surpasses that of analog-to-digital converters (ADCs) by 2.5 folds. Moreover, the implemented full-hardware binary convolution neural network (CNN) achieves remarkable accuracy (97.5%) in a pattern recognition task. Additionally, analog computing and in-built activation functions are demonstrated within the system, further augmenting energy efficiency. This silicon-compatible heterogeneous integration approach, along with the energy-efficient CIM system, presents a promising hardware solution for artificial intelligence (AI) applications.","author":[{"family":"Ang","given":"Kah"},{"family":"Li","given":"Sifan"},{"family":"Jain","given":"Samarth"},{"family":"Zheng","given":"Haofei"},{"family":"Li","given":"Lingqi"},{"family":"Fong","given":"Xuanyao"}],"issued":{"date-parts":[[2024]]},"DOI":"10.21203/rs.3.rs-3172508/v1","URL":"https://doi.org/10.21203/rs.3.rs-3172508/v1","source":"europepmc"},{"id":"doi:10.3934/mbe.2024147","type":"article-journal","title":"Synchronization of inertial complex-valued memristor-based neural networks with time-varying delays","abstract":"&lt;abstract&gt;&lt;p&gt;The synchronization of inertial complex-valued memristor-based neural networks (ICVMNNs) with time-varying delays was explored in the paper with the non-separation and non-reduced approach. Sufficient conditions required for the exponential synchronization of the ICVMNNs were identified with the construction of comprehensive Lyapunov functions and the design of a novel control scheme. The adaptive synchronization was also investigated based on the derived results, which is easier to implement in practice. What's more, a numerical example that verifies the obtained results was presented.&lt;/p&gt;&lt;/abstract&gt;","author":[{"family":"Wang","given":"Pan"},{"family":"Li","given":"Xuechen"},{"family":"Zheng","given":"Qianqian"}],"issued":{"date-parts":[[2024]]},"DOI":"10.3934/mbe.2024147","URL":"https://doi.org/10.3934/mbe.2024147","source":"crossref"},{"id":"doi:10.7498/aps.73.20231888","type":"article-journal","title":"A novel compound exponential locally active memristor coupled Hopfield neural network","abstract":"The neural network model coupled with memristors has been extensively studied due to its ability to more accurately represent the complex dynamic characteristics of the biological nervous system. Currently, the mathematical model of memristor used to couple neural networks mainly focuses on primary function, absolute value function, hyperbolic tangent function, etc. To further enrich the memristor-coupled neural network model and take into account the motion law of particles in some doped semiconductors, a new compound exponential local active memristor is proposed and used as a coupling synapse in the Hopfield neural network. Using the basic dynamic analysis method, the system’s dynamic behaviors are studied under different parameters and the coexistence of multiple bifurcation modes under different initial values. In addition, the influence of frequency change of external stimulation current on the system is also studied. The experimental results show that the internal parameters of memristor synapses regulate the system, and the system has a rich dynamic behavior, including symmetric attractor coexistence, asymmetric attractor coexistence, large-scale chaos as shown in attached figure, and bursting oscillation. Finally, the hardware of the system is realized by the STM32 microcontroller, and the experimental results verify the realization of the system.","author":[{"family":"Wang","given":"Meng"},{"family":"Yang","given":"Chen"},{"family":"He","given":"Shao"},{"family":"Li","given":"Zhi"}],"issued":{"date-parts":[[2024]]},"DOI":"10.7498/aps.73.20231888","URL":"https://doi.org/10.7498/aps.73.20231888","source":"crossref"},{"id":"doi:10.3934/era.2024156","type":"article-journal","title":"Synchronization analysis of delayed quaternion-valued memristor-based neural networks by a direct analytical approach","abstract":"&lt;abstract&gt;&lt;p&gt;This issue discusses the asymptotic synchronization and the exponential synchronization for memristor-based quaternion-valued neural networks under the time-varying delays. Some criteria for synchronization of the memristor-based quaternion-valued neural networks are given by exploiting the set-valued theory, the differential inclusion theory, some analytic techniques, as well as constructing novel controllers, It is worth noting that the synchronization problem about the memristor-based quaternion-valued neural networks were studied by the direct analysis method in this paper. Finally, the main theoretical results were verified by numerical simulations.&lt;/p&gt;&lt;/abstract&gt;","author":[{"family":"Guo","given":"Jun"},{"family":"Shi","given":"Yanchao"},{"family":"Wang","given":"Shengye"}],"issued":{"date-parts":[[2024]]},"DOI":"10.3934/era.2024156","URL":"https://doi.org/10.3934/era.2024156","source":"crossref"},{"id":"doi:10.3390/electronics12122715","type":"article-journal","title":"Embedding-Based Deep Neural Network and Convolutional Neural Network Graph Classifiers","abstract":"One of the most significant graph data analysis tasks is graph classification, as graphs are complex data structures used for illustrating relationships between entity pairs. Graphs are essential in many domains, such as the description of chemical molecules, biological networks, social relationships, etc. Real-world graphs are complicated and large. As a result, there is a need to find a way to represent or encode a graph’s structure so that it can be easily utilized by machine learning models. Therefore, graph embedding is considered one of the most powerful solutions for graph representation. Inspired by the Doc2Vec model in Natural Language Processing (NLP), this paper first investigates different ways of (sub)graph embedding to represent each graph or subgraph as a fixed-length feature vector, which is then used as input to any classifier. Thus, two supervised classifiers—a deep neural network (DNN) and a convolutional neural network (CNN)—are proposed to enhance graph classification. Experimental results on five benchmark datasets indicate that the proposed models obtain competitive results and are superior to some traditional classification methods and deep-learning-based approaches on three out of five benchmark datasets, with an impressive accuracy rate of 94% on the NCI1 dataset.","author":[{"family":"Elnaggar","given":"Sarah"},{"family":"Elsemman","given":"Ibrahim"},{"family":"Soliman","given":"Taysir"}],"issued":{"date-parts":[[2023]]},"DOI":"10.3390/electronics12122715","URL":"https://doi.org/10.3390/electronics12122715","source":"crossref"},{"id":"doi:10.17816/gc623428","type":"article-journal","title":"Design of a memristor-based neuron for spiking neural networks","abstract":"The primary objective of neuromorphic system design is to surpass limitations in energy efficiency and scaling of classical von Neumann computing systems, through the emulation of animals’ nervous systems. This is achieved by conducting calculations in memory and encoding information in impulse signals, ultimately leading to enhanced adaptability. Adhering to these principles allows for improved energy efficiency and computational speed when solving machine learning problems, encompassing biomedical applications, embedded systems, and cyber-physical systems. Functional blocks modeling the main elements of the central nervous system, namely neurons and synapses, offer an advantage in implementing learning on a chip. The use of memristive electronic components, capable of altering their resistance based on the charge flowing through them, opens new doors for hardware implementation of neuromorphic systems. These devices offer advantages over conventional transistor electronics with respect to power consumption, component density, and performance. To achieve optimal results, the architecture of neuromorphic systems should be optimized at the device level. Memristive components are utilized to create neurons and synapses. This thesis is specifically focused on producing memristive neuron-like spike signal generators. Previously, memristive neurons were crafted using a locally active element comprised of vanadium dioxide VO2, which incorporated a negative differential resistance section of the IV-curve. One of the recent advancements in this field is a spiking neuron with frequency adaptation [1]. Its drawbacks, however, involve separating the memristive and locally active elements physically, resulting in higher energy consumption and decreased integration quality. In [2], models of memristive neurons with minimal complexity are introduced, which incorporate the Leaky Integrate-and-Fire principle. However, the circuits presented require the application of negative voltage pulses to a DC battery to reset the memristor to its initial high-resistance state. This limitation restricts its sphere of application in neuromorphic systems. This paper proposes a model of a neuron that overcomes these limitations by using the negative differential resistance of the memristor to generate spikes, along with integrating supplementary circuit components to sustain the resistive switching cycles of the memristor. The neuron model under consideration is implemented using the NI Multisim 14.2 SPICE environment and has been verified in the NI LabVIEW 2022 tool environment. The equations of the modified model of the generalized mean metastable switch of the memristor with self-directed channel [3] represent the current in the memristor branch of the neuron equivalent circuit. The simplicity of the equivalent circuitry of the neuron is attained by merging all the nonlinear features necessary for spike generation into one memristor model. The experimental phase of the study employed obtainable memristors from Knowm Corporation and the laboratory prototyping platform NI ELVIS III. The investigation of the proposed neuron model was accomplished through the application of sinusoidal and rectangular input signals. The refractory time of the neuron model was calculated. The chosen stack of computer simulation and semi-natural modeling technologies is applied within the research-driven design concept of electronic devices. This approach considers the importance of refining the properties and identification of the design object or its components during the development cycle.","author":[{"family":"Ostrovskii","given":"VY"},{"family":"Druzhina","given":"OS"},{"family":"Kamal","given":"O"},{"family":"Karimov","given":"TI"},{"family":"Butusov","given":"DN"}],"issued":{"date-parts":[[2024]]},"DOI":"10.17816/gc623428","URL":"https://doi.org/10.17816/gc623428","source":"crossref"},{"id":"doi:10.1002/admt.202400965","type":"article-journal","title":"Au‐Nanodots Embedded Self‐Rectifying Analog Charge Trap Memristor with Modified Bias Voltage Application Method for Stable Multi‐Bit Hardware‐Based Neural Network","abstract":"Abstract The self‐rectifying memristor with a bilayer of trap‐rich HfO 2 and insulating Ta 2 O 5 oxide layers is considered one of the most promising candidates for the memristive crossbar array due to its superior switching performance, scalability with 3D stacking, and low operating power. However, the output current variation due to the electron detrapping from trap states can cause the failure of critical operations in neuromorphic applications. This work suggests two solutions to mitigate the switching variations and insufficient data retention time by embedding gold nanodots and modifying the bias voltage application methods for read, write, and erase operations in the crossbar array. The switching mechanism is studied by varying the embedded position of the gold nanodots across the thickness direction of the bilayered oxides, which helped to optimize device performance further. Combining the two solutions into the proposed self‐rectifying memristor enables a single device to have the 7‐possible, stable states by preventing interstate overlap and securing the retention. Consequently, the hardware neural network consisting of self‐rectifying memristors with gold nanodots with the modified bias voltage application methods demonstrates a high inference accuracy of 93.1% in MNIST handwritten digit classification, comparable to the software‐based accuracy of 93.4%, benefiting from the enhanced multi‐state uniformity.","author":[{"family":"Park","given":"Taegyun"},{"family":"Kim","given":"Jihun"},{"family":"Kwon","given":"Young"},{"family":"Kim","given":"Han"},{"family":"Yim","given":"Seong"},{"family":"Shin","given":"Dong"},{"family":"Kim","given":"Yeong"},{"family":"Kim","given":"Hae"},{"family":"Hwang","given":"Cheol"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1002/admt.202400965","URL":"https://doi.org/10.1002/admt.202400965","source":"crossref"},{"id":"doi:10.1002/rnc.7112","type":"article-journal","title":"Unified synchronization and fault‐tolerant anti‐disturbance control for synchronization of multiple memristor‐based neural networks","abstract":"Summary This work primarily concentrates on the design of fault‐tolerant anti‐disturbance control for synchronization of multiple memristor‐based neural networks subject to time delay, matched and mismatched disturbances. Moreover, in the addressed network model, we consider parameter uncertainties and actuator faults. Firstly in order to estimate the matched disturbances generated by the exogenous system, a disturbance observer is devised. Whereas, the mismatched part is tackled by employing the mixed and passivity performance indexes. Subsequently, a unified controller is designed by incorporating error feedback control and the disturbance estimate. Further, with the assistance of Lyapunov stability theory and linear matrix inequality technique, an adequate criteria is procured to ascertain the required synchronization criteria for the assayed network model with the mixed and passivity performance indexes. Following this, by basing on the established conditions, the explicit form of the controller and observer gain matrices is obtained. In the end, a numerical example with simulation results is shown to confirm the potential and usefulness of the conclusions acquired from the theoretical analysis.","author":[{"family":"Satheesh","given":"T"},{"family":"Sakthivel","given":"R"},{"family":"Aravinth","given":"N"},{"family":"Karimi","given":"HR"}],"issued":{"date-parts":[[2023]]},"DOI":"10.1002/rnc.7112","URL":"https://doi.org/10.1002/rnc.7112","source":"crossref"},{"id":"doi:10.1007/s11063-024-11466-7","type":"article-journal","title":"A Prototype-Based Neural Network for Image Anomaly Detection and Localization","abstract":"Abstract Image anomaly detection and localization perform not only image-level anomaly classification but also locate pixel-level anomaly regions. Recently, it has received much research attention due to its wide application in various fields. This paper proposes ProtoAD, a prototype-based neural network for image anomaly detection and localization. First, the patch features of normal images are extracted by a deep network pre-trained on nature images. Then, the prototypes of the normal patch features are learned by non-parametric clustering. Finally, we construct an image anomaly localization network (ProtoAD) by appending the feature extraction network with L 2 feature normalization, a $$1\\times 1$$ 1 × 1 convolutional layer, a channel max-pooling, and a subtraction operation. We use the prototypes as the kernels of the $$1\\times 1$$ 1 × 1 convolutional layer; therefore, our neural network does not need a training phase and can conduct anomaly detection and localization in an end-to-end manner. Extensive experiments on two challenging industrial anomaly detection datasets, MVTec AD and BTAD, demonstrate that ProtoAD achieves competitive performance compared to the state-of-the-art methods with a higher inference speed. The code and pre-trained models are publicly available at https://github.com/98chao/ProtoAD .","author":[{"family":"Huang","given":"Chao"},{"family":"Kang","given":"Zhao"},{"family":"Wu","given":"Hong"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1007/s11063-024-11466-7","URL":"https://doi.org/10.1007/s11063-024-11466-7","source":"crossref"},{"id":"doi:10.32493/jtsi.v7i1.34168","type":"article-journal","title":"Deteksi Leukemia Limfoblastik Akut menggunakan Convolutional Neural Network","abstract":"Leukemia limfoblastik akut merupakan jenis leukemia anak yang paling penting, dan menyumbang 25% dari kanker anak. Membedakan secara akurat prekursor sel normal dari sel kanker adalah kunci diagnosis leukemia limfoblastik akut. Namun, di bawah mikroskop, sel kanker sangat mirip dengan sel normal sehingga sulit untuk mengklasifikasikannya. Artikel ini menyajikan deteksi sel leukemia limfoblastik akut menggunakan Convolutional Neural Network (CNN). Dataset diperoleh dari ALL_IDB sejumlah 582 data citra berwarna yang dibagi menjadi 482 data citra latih dan 100 data citra uji. Data citra tersebut akan diubah ukurannya menjadi 128x128x3 sebelum akhirnya menjadi masukan dari model CNN. Model CNN yang digunakan adalah multi-scale CNN yang terdiri dari 3 lapisan konvolusi (ukuran filter 3x3, jumlah filter untuk masing-masing lapisan konvolusi yakni berurutan 32, 64, dan 128, dan fungsi aktivasi ReLU), 3 lapisan subsampling menggunakan maxpool dengan ukuran filter 2x2, 1 lapisan penggabungan yang digunakan untuk menggabungkan keluaran dari masing-masing lapisan subsampling, 1 lapisan fully-connected dengan fungsi aktivasi softmax dan fungsi galat cross-entropy, dan terakhir lapisan keluaran dengan jumlah kelas 2 yakni sel normal dan sel kanker. Model CNN akan dilatih menggunakan algoritma pelatihan Adam optimizer dengan laju pelatihan 0.0002 dan dilakukan iterasi sebanyak 20 kali. Berdasarkan hasil pelatihan setelah diiterasi sebanyak 20 kali, didapatkan nilai galat terkecil yakni 0,0001 dan nilai akurasi terbesar yakni 100% pada epoch ke-20. Model CNN kemudian diuji dengan 100 data citra uji dan menghasilkan tingkat akurasi 98% dan nilai galat 0,0482.","author":[{"family":"Akbar","given":"Mutaqin"},{"family":"Prasetyaningrum","given":"Putri"},{"family":"Setyaningsih","given":"Putry"},{"family":"Ahsan","given":"Moh"},{"family":"Budianto","given":"Alexius"}],"issued":{"date-parts":[[2024]]},"DOI":"10.32493/jtsi.v7i1.34168","URL":"https://doi.org/10.32493/jtsi.v7i1.34168","source":"crossref"},{"id":"doi:10.1007/s11063-024-11471-w","type":"article-journal","title":"A Vision Enhancement and Feature Fusion Multiscale Detection Network","abstract":"Abstract In the field of object detection, there is often a high level of occlusion in real scenes, which can very easily interfere with the accuracy of the detector. Currently, most detectors use a convolutional neural network (CNN) as a backbone network, but the robustness of CNNs for detection under cover is poor, and the absence of object pixels makes conventional convolution ineffective in extracting features, leading to a decrease in detection accuracy. To address these two problems, we propose VFN (A Vision Enhancement and Feature Fusion Multiscale Detection Network), which first builds a multiscale backbone network using different stages of the Swin Transformer, and then utilizes a vision enhancement module using dilated convolution to enhance the vision of feature points at different scales and address the problem of missing pixels. Finally, the feature guidance module enables features at each scale to be enhanced by fusing with each other. The total accuracy demonstrated by VFN on both the PASCAL VOC dataset and the CrowdHuman dataset is better than that of other methods, and its ability to find occluded objects is also better, demonstrating the effectiveness of our method.The code is available at https://github.com/qcw666/vfn .","author":[{"family":"Qian","given":"Chengwu"},{"family":"Qian","given":"Jiangbo"},{"family":"Wang","given":"Chong"},{"family":"Ye","given":"Xulun"},{"family":"Zhong","given":"Caiming"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1007/s11063-024-11471-w","URL":"https://doi.org/10.1007/s11063-024-11471-w","source":"crossref"},{"id":"doi:10.1007/s11063-024-11533-z","type":"article-journal","title":"FTUNet: A Feature-Enhanced Network for Medical Image Segmentation Based on the Combination of U-Shaped Network and Vision Transformer","abstract":"Abstract Semantic Segmentation has been widely used in a variety of clinical images, which greatly assists medical diagnosis and other work. To address the challenge of reduced semantic inference accuracy caused by feature weakening, a pioneering network called FTUNet (Feature-enhanced Transformer UNet) was introduced, leveraging the classical Encoder-Decoder architecture. Firstly, a dual-branch Encoder is proposed based on the U-shaped structure. In addition to employing convolution for feature extraction, a Layer Transformer structure (LTrans) is established to capture long-range dependencies and global context information. Then, an Inception structural module focusing on local features is proposed at the Bottleneck, which adopts the dilated convolution to amplify the receptive field to achieve deeper semantic mining based on the comprehensive information brought by the dual Encoder. Finally, in order to amplify feature differences, a lightweight attention mechanism of feature polarization is proposed at Skip Connection, which can strengthen or suppress feature channels by reallocating weights. The experiment is conducted on 3 different medical datasets. A comprehensive and detailed comparison was conducted with 6 non-U-shaped models, 5 U-shaped models, and 3 Transformer models in 8 categories of indicators. Meanwhile, 9 kinds of layer-by-layer ablation and 4 kinds of other embedding attempts are implemented to demonstrate the optimal structure of the current FTUNet.","author":[{"family":"Wang","given":"Yuefei"},{"family":"Yu","given":"Xi"},{"family":"Yang","given":"Yixi"},{"family":"Zeng","given":"Shijie"},{"family":"Xu","given":"Yuquan"},{"family":"Feng","given":"Ronghui"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1007/s11063-024-11533-z","URL":"https://doi.org/10.1007/s11063-024-11533-z","source":"crossref"},{"id":"doi:10.12688/f1000research.136097.2","type":"article-journal","title":"Graph neural network-based anomaly detection for river network systems","abstract":"Background Water is the lifeblood of river networks, and its quality plays a crucial role in sustaining both aquatic ecosystems and human societies. Real-time monitoring of water quality is increasingly reliant on in-situ sensor technology. Anomaly detection is crucial for identifying erroneous patterns in sensor data, but can be a challenging task due to the complexity and variability of the data, even under typical conditions. This paper presents a solution to the challenging task of anomaly detection for river network sensor data, which is essential for accurate and continuous monitoring. Methods We use a graph neural network model, the recently proposed Graph Deviation Network (GDN), which employs graph attention-based forecasting to capture the complex spatio-temporal relationships between sensors. We propose an alternate anomaly threshold criteria for the model, GDN+, based on the learned graph. To evaluate the model’s efficacy, we introduce new benchmarking simulation experiments with highly-sophisticated dependency structures and subsequence anomalies of various types. We also introduce software called gnnad. Results We further examine the strengths and weaknesses of this baseline approach, GDN, in comparison to other benchmarking methods on complex real-world river network data. Conclusions Findings suggest that GDN+ outperforms the baseline approach in high-dimensional data, while also providing improved interpretability.","author":[{"family":"Buchhorn","given":"Katie"},{"family":"Santos-Fernandez","given":"Edgar"},{"family":"Mengersen","given":"Kerrie"},{"family":"Salomone","given":"Robert"}],"issued":{"date-parts":[[2024]]},"DOI":"10.12688/f1000research.136097.2","URL":"https://doi.org/10.12688/f1000research.136097.2","source":"crossref"},{"id":"doi:10.1142/s0218127424501062","type":"article-journal","title":"Dynamics and Implementation of FPGA for Memristor-Coupled Fractional-Order Hopfield Neural Networks","abstract":"The coupling between neurons can lead to diverse neural network architectures, with the Hopfield neural network (HNN) being particularly noteworthy for its resemblance to human brain function and its potential in modeling chaotic systems. This paper introduces a novel approach: a fractional-order HNN coupled with a hyperbolic tangent-type memristor. Initially, we propose a new model for the hyperbolic tangent-type memristor and fingerprints. Subsequently, we construct a memristor-coupled fractional-order Hopfield neural network (mFOHNN) and explore its dynamic behavior using various analytical tools, including phase diagrams, bifurcation diagrams, Lyapunov exponent diagrams, Poincaré maps, and attractor basins. Our findings reveal rich coexisting bifurcation behavior in the neural network model, influenced by different initial values of coexisting attractors. Finally, we validate the model through analysis and implementation using Multisim circuit simulation software and FPGA hardware, respectively.","author":[{"family":"Yang","given":"Ningning"},{"family":"Liang","given":"Jiahao"},{"family":"Wu","given":"Chaojun"},{"family":"Guo","given":"Zhenshuo"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1142/s0218127424501062","URL":"https://doi.org/10.1142/s0218127424501062","source":"crossref"},{"id":"doi:10.1063/5.0190861","type":"article-journal","title":"A temperature sensing based Na0.5Bi0.5TiO3 ferroelectric memristor device for artificial neural systems","abstract":"With the development of artificial intelligence technology, it remains a challenge to improve the resistive switching performance of next-generation nonvolatile ferroelectric memristor device (FMD). Here, we report an epitaxial Na0.5Bi0.5TiO3 ferroelectric memristor device (NBT-FMD) with temperature sensing. The NBT epitaxial films with strong polarization strength and suitable oxygen vacancy concentration were obtained by temperature adjustment (700 °C). In addition, the function of the spiking-time-dependent plasticity and paired-pulse facilitation is simulated in ferroelectric memristor devices of Pt/NBT/SrRuO3 (SRO)/SrTiO3 (STO). More importantly, we have designed a neuronal circuit to confirm that NBT-FMD can serve as temperature receptors on the human skin, paving the way for bio-inspired application.","author":[{"family":"Zhou","given":"Lei"},{"family":"Pei","given":"Yifei"},{"family":"Li","given":"Changliang"},{"family":"He","given":"Hui"},{"family":"Liu","given":"Chao"},{"family":"Hou","given":"Yue"},{"family":"Tian","given":"Haoyuan"},{"family":"Guo","given":"Jianxin"},{"family":"Liu","given":"Baoting"},{"family":"Yan","given":"Xiaobing"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1063/5.0190861","URL":"https://doi.org/10.1063/5.0190861","source":"crossref"},{"id":"doi:10.32664/j-intech.v12i1.1273","type":"article-journal","title":"Identifikasi Tanda Tangan Dengan Menggunakan Metode Convolution Neural Network (CNN)","abstract":"This research aims to develop and evaluate a Convolutional Neural Network (CNN) model for signature identification. The CNN method is chosen for its capability to extract and analyze complex visual features from signature images. The data used in this study consists of a collection of signature images divided into training and testing sets. The proposed CNN model comprises several convolutional, pooling, and fully connected layers optimized for classification tasks. Evaluation results indicate that the CNN model achieves excellent performance with an accuracy of 0.97, demonstrating high accuracy and precision in signature recognition. With these results, CNN proves to be an effective and reliable method for signature identification, making a significant contribution to the field of biometric identity verification. These findings open opportunities for further applications in security and authentication systems requiring automatic signature recognition.","author":[{"family":"Indrianis","given":"Dechy"},{"family":"Sinaga","given":"Elya"},{"family":"Oktavia","given":"Grace"},{"family":"Syahputra","given":"Hermawan"},{"family":"Ramadhani","given":"Fanny"}],"issued":{"date-parts":[[2024]]},"DOI":"10.32664/j-intech.v12i1.1273","URL":"https://doi.org/10.32664/j-intech.v12i1.1273","source":"crossref"},{"id":"doi:10.1007/s10462-024-10826-y","type":"article-journal","title":"Heterogeneous wireless network selection using feed forward double hierarchy linguistic neural network","abstract":"Abstract Network selection in heterogeneous wireless networks (HWNs) is a complex issue that requires a thorough understanding of service features and user preferences. This is because the various wireless access technologies have varying capabilities and limitations, and the best network for a voice, video, and data service depends on a variety of factors. For selecting the optimal network in HWNs, varying factors such as the user’s position, accessible network resources, service quality requirements, and user preferences must be considered. The classical decision making procedure is very difficult and uncertain to select the desirable HWNs for voice, video, and data. Therefore, we develop a novel decision making model based on feed-forward neural networks under the double hierarchy linguistic information for the selection of the best HWNs for voice, video, and data. In this article, we introduce a novel feed-forward double hierarchy linguistic neural network using the Hamacher t-norm and t-conorm. Further, the feed-forward double hierarchy linguistic neural network applies to the decision making model for the selection of the best HWNs for voice, video, and data. In this novel decision making model, we first take the given data about HWNs and use the converting function to convert the given data into a double hierarchy linguistic term set. We calculate the hidden layer and output layer information by using Hamacher aggregation operations. Finally, we use the sigmoid activation function on the output layer information to decide on the best HWNs for voice, video, and data according to ranking. The proposed approach is compared with other existing models of decision making and the results of the comparison show that the proposed technique is applicable and reliable for the decision support model.","author":[{"family":"Abdullah","given":"Saleem"},{"family":"Ullah","given":"Ihsan"},{"family":"Ghani","given":"Fazal"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1007/s10462-024-10826-y","URL":"https://doi.org/10.1007/s10462-024-10826-y","source":"crossref"},{"id":"doi:10.4018/ijisscm.345654","type":"article-journal","title":"Physical Delivery Network Optimization Based on Ant Colony Optimization Neural Network Algorithm","abstract":"The development of modern logistics chains is not just simple cargo transportation, it has become a cross-integrated industry that integrates many emerging technologies such as IoT technology, intelligent transportation, cloud computing and mobile Internet. Based on the ant colony algorithm (ACA), this paper optimizes the physical delivery network of the optimized neural network algorithm, establishes a mathematical model for the constraints and optimization objectives in the optimization of the physical delivery path, and proposes some improvements to the ACA to improve the convergence of the algorithm. speed and global search ability, so as to use the improved algorithm to solve the physical delivery path optimization problem. Experiments show that the optimal distance of physical delivery path planning calculated by traditional ACA is 207.8544km, while the optimal distance of improved ACA path planning is 197.9879km. The performance of the improved ACA is improved by analyzing the results of solving typical examples.","author":[{"family":"Wu","given":"Shujuan"},{"family":"Cheng","given":"Hanlie"},{"family":"Qin","given":"Qiang"}],"issued":{"date-parts":[[2024]]},"DOI":"10.4018/ijisscm.345654","URL":"https://doi.org/10.4018/ijisscm.345654","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-3957770/v1","type":"article-journal","title":"D-GAN: An Automatic Acne Detection, Severity, and Assessment Framework using Generative Adversarial Network with Deep Neural Network","abstract":"Abstract With the robust development of Artificial Intelligence (AI), especially image processing has made information technology more efficient and effective in the sense to evaluate facial features, even though there has been masked on the face. However, the accuracy of acne detection and related severity analysis is becoming a significant prospect for the precise treatment of patients. Due to this, close severity is one of the features that need to be added first, while it is considered a highly challenging aspect for dermatologists because the similar appearance of acne in the face reduces the rate of accuracy when examining. It poses a serious problem in the domain of biomedical processing and controls. In this paper, we contribute to four different folds. Initially, this paper presents a novel framework that provides a platform in order to measure localization and segmentation. In this process, consultants receive better accuracy and efficiency during the process of acne detection and severity analysis. Second, this paper utilizes deep neural networks (DNNs) as a backend process to lightning the extraction of multi-scale features through a multi-hierarchy neural net for regionalized facial features to investigate distinction and localization. Third, a class-based segmentation approach customizes and integrates with the proposed framework to examine the background and facial skin separation to distinguish different classes to obtain severity marking. Fourth, the facial skin segmentation classes are built as a cluster segment using Generative Adversarial Network (GAN). It provides a high-resolution network that enhances severity masking awareness and related attentions, including shuffle and conditional channels like an update in weight block. Although, the simulation illustrates that the technological collaboration achieves promising results and elaborates on the accuracy and efficiency while the detection of acne as compared to other state-of-the-art baseline methods. In addition, the proposed framework illustrates good results in severity analysis for acne detection comparably far better than the previously published articles with a performance rate of 3.112% (accuracy), 1.131% (segmentation), 2.317% (localization), and 1.573% (GAN-based high-resolution network management), respectively.","author":[{"family":"Khalid","given":"Umara"},{"family":"Chen","given":"Li"},{"family":"Khan","given":"Abdullah"},{"family":"Mehmood","given":"Faisal"}],"issued":{"date-parts":[[2024]]},"DOI":"10.21203/rs.3.rs-3957770/v1","URL":"https://doi.org/10.21203/rs.3.rs-3957770/v1","source":"crossref"},{"id":"doi:10.3390/su16083474","type":"article-journal","title":"A Time Series Prediction Model for Wind Power Based on the Empirical Mode Decomposition–Convolutional Neural Network–Three-Dimensional Gated Neural Network","abstract":"In response to the global challenge of climate change and the shift away from fossil fuels, the accurate prediction of wind power generation is crucial for optimizing grid operations and managing energy storage. This study introduces a novel approach by integrating the proportional–integral–derivative (PID) control theory into wind power forecasting, employing a three-dimensional gated neural (TGN) unit designed to enhance error feedback mechanisms. The proposed empirical mode decomposition (EMD)–convolutional neural network (CNN)–three-dimensional gated neural network (TGNN) framework starts with the pre-processing of wind data using EMD, followed by feature extraction via a CNN, and time series forecasting using the TGN unit. This setup leverages proportional, integral, and differential control within its architecture to improve adaptability and response to dynamic wind patterns. The experimental results show significant improvements in forecasting accuracy; the EMD–CNN–TGNN model outperforms both traditional models like autoregressive integrated moving average (ARIMA) and support vector regression (SVR), and similar neural network approaches, such as EMD–CNN–GRU and EMD–CNN–LSTM, across several metrics including mean absolute error (MAE), mean squared error (MSE), root mean squared error (RMSE), and coefficient of determination (R2). These advancements substantiate the model’s effectiveness in enhancing the precision of wind power predictions, offering substantial implications for future renewable energy management and storage solutions.","author":[{"family":"Guo","given":"Zhiyong"},{"family":"Wei","given":"Fangzheng"},{"family":"Qi","given":"Wenkai"},{"family":"Han","given":"Qiaoli"},{"family":"Liu","given":"Huiyuan"},{"family":"Feng","given":"Xiaomei"},{"family":"Zhang","given":"Minghui"}],"issued":{"date-parts":[[2024]]},"DOI":"10.3390/su16083474","URL":"https://doi.org/10.3390/su16083474","source":"crossref"},{"id":"doi:10.1115/gt2024-127933","type":"article-journal","title":"The Parameter Augmentation Pretraining Neural Network Model of Aero-Engine Thrust Prediction","abstract":"Abstract Accurately predicting the thrust of aero-engine is important for engine health management and engine performance control. Thrust prediction neural network models have been widely applied in recent years. This paper proposes a parameter augmentation pretraining method and designs an aero-engine neural network model with physical significance. It is known that ground test system measures more engine parameters compared to onboard system. Due to the limitations of using onboard models, typically only the parameters measured by both systems are used as training parameters, which may lead to incomplete training. In this study, a 0-1 encoded vector is utilized to distinguish ground test system data and onboard system data. The model is first pretrained using the measurable parameters from the ground test system, following by retraining using parameters measured by both the ground test system and the onboard system. The parameter augmentation pretrain method allows the model to learn more mapping relationships of aero-engine and resolves the issue of parameter dimension matching. Furthermore, this study investigates the impact of the weight loss function on the prediction accuracy of the model. This research provides guidance for engine performance evaluation and further studies.","author":[{"family":"You","given":"Rui"},{"family":"Xiao","given":"Hong"},{"family":"Chen","given":"Guo"},{"family":"Liang","given":"Yufeng"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1115/gt2024-127933","URL":"https://doi.org/10.1115/gt2024-127933","source":"crossref"},{"id":"doi:10.5121/csit.2024.140601","type":"article-journal","title":"Knowledge-Augmented Dynamic Neural Network Model and its Application in Credit Evaluation","abstract":"Credit risk is the most significant risk faced by credit businesses. Currently, various approaches are widely used in credit evaluation. However, methods based on expert knowledge exhibit obvious subjective cognitive bias, while both statistical and machine learning methods require a substantial amount of historical data. In cases with limited data, the machine-learning effect is poor. Inspired by the structural similarity between neural networks (NN) and the Analytic Hierarchy Process (AHP), we propose a knowledge-augmented dynamic neural network model called KADNN to construct an effective credit evaluation model. This composite architecture will help effectively utilize existing data to alleviate the initial low-data dilemma and can be further utilized for training neural networks. Subsequent data updates can be dynamically incorporated to improve model accuracy. Additionally, this approach improves the comprehensibility and premature convergence issues of the NN model. The proposed approach is validated and evaluated through credit evaluation simulation.","author":[{"family":"Xiong","given":"Qingyue"},{"family":"Zhang","given":"Liwei"},{"family":"Lan","given":"Qiujun"}],"issued":{"date-parts":[[2024]]},"DOI":"10.5121/csit.2024.140601","URL":"https://doi.org/10.5121/csit.2024.140601","source":"crossref"},{"id":"doi:10.59188/eduvest.v2i8.514","type":"article-journal","title":"Covid-19 Sentiment Analysis Using Convolutional Neural Network / Reccurent Neural Network Method","abstract":"Social media is a very important tool in this modern era , one of which is namely twitter. Twitter allows user for give opinion / opinion to various issues and topics hot / viral trending . Trending on twitter is so so fast in the process of spreading so that Twitter becomes a medium of information that often become a media issue conspiracy . Covid-19 is a moderate epidemic / disease _ experienced the whole world when this . Issues circulating in the population , they believe that Covid-19 is a real pandemic and a conspiracy , issue this make population confused differentiate Among second issue that . Because of that required a fast and accurate analysis _ for produce valid results , that Covid-19 a real thing _ or conspiracy seen from opinion population and corner views written on Twitter. CNN/RNN or combined from RNN(LSTM) and CNN methods are method used _ for classify opinion population about Covid-19 issues . Study this also done with compare is correct RNN/CNN accuracy same like deep RNN even more fast for in the process . Research results state that accuracy from combined RNN/CNN no different remote , even RNN/CNN in the process more fast than deep RNNs. Research results about opinion / opinion residents on twitter who believe about Covid-19 is conspiracy more low than residents who have confidence about Covid-19 is something the real thing . Percentage classification opinion / opinion from sentiment positive by 63.15% and opinion / opinion sentiment negative by 28.60%, this is results calculation use RNN/CNN method , with accuracy reached 58%. Accuracy from method used _ make Covid-19 issues that exist in the population no Becomes hoax news so population more alert against the ongoing Covid-19 pandemic happen.","author":[{"family":"Matatula","given":"Ravensca"},{"family":"Manongga","given":"Danny"},{"family":"Hendry","given":"Hendry"}],"issued":{"date-parts":[[2023]]},"DOI":"10.59188/eduvest.v2i8.514","URL":"https://doi.org/10.59188/eduvest.v2i8.514","source":"crossref"},{"id":"doi:10.33395/sinkron.v8i1.11944","type":"article-journal","title":"Comparison of Convolutional Neural Network and Artificial Neural Network for Rice Detection","abstract":"Rice is a staple food for people in tropical countries. Indonesia is a country that needs a lot of rice for its people in providing food. This country has implemented various ways to plant rice properly. Many agricultural fields have implemented harvests up to three times a year, due to the role of technology which has helped a lot in agriculture. Planting to harvest already uses advanced technology and tools. A good rice harvest can improve the welfare of the surrounding community. Meanwhile with lots of rice products because many rice plants produce with lots of rice. The type of rice from different regions of origin, the yield of rice is also different from other regions of origin. But with advances in technology, it is possible to plant rice whose types of plants come from other regions. The rice sold to the public varies, so that people who are unfamiliar with the types of rice find it difficult to detect the types of rice. Machine learning is present in detecting various kinds of rice. Machine learning, especially deep learning can make better detection, because one of the deep learning methods works similar to the human brain. In the human brain there are millions or even billions of neurons. This research uses neural networks in experiments using public datasets. Experiments using Artificial Neural Networks achieve an training accuracy of 98.2%, loss: 0.2351. It takes about 10 minutes of training. Testing accuracy reaches accuracy: 96%, loss: 0.6641. By conducting experiments using the Convolution Neural Network, it achieves an accuracy of 99.3% and the training time requires around 18 hours. The purpose of this research is to classify the rice image dataset and detect the rice image.","author":[{"family":"Suherman","given":"Endang"},{"family":"Hindarto","given":"Djarot"},{"family":"Makmur","given":"Amelia"},{"family":"Santoso","given":"Handri"}],"issued":{"date-parts":[[2023]]},"DOI":"10.33395/sinkron.v8i1.11944","URL":"https://doi.org/10.33395/sinkron.v8i1.11944","source":"crossref"},{"id":"doi:10.14311/nnw.2024.34.012","type":"article-journal","title":"Understanding Travel Behavior: A Deep Neural Network and SHAP Approach to Mode Choice Determinants","abstract":"Understanding individual travel behavior is crucial for developing effective travel demand management strategies and informed transportation policies. This study investigates the factors influencing individuals’ mode choices by analyzing data from a comprehensive travel survey. We employ a deep neural network model to explore the relationships between survey variables and respondents’ transportation mode preferences, focusing on both observable and latent factors. The SHAP method is applied to interpret the model’s outputs, providing global and local explanations that offer detailed insights into the contribution of each variable to mode choice decisions. By identifying the key determinants of mode selection and uncovering the complex interactions between these factors, this research provides valuable insights for designing targeted policies that can better address transportation needs and influence sustainable travel behavior.","author":[{"family":"Çevik","given":"Halil"},{"family":"Přibyl","given":"Ondřej"},{"family":"Samandar","given":"Shoaib"}],"issued":{"date-parts":[[2024]]},"DOI":"10.14311/nnw.2024.34.012","URL":"https://doi.org/10.14311/nnw.2024.34.012","source":"crossref"},{"id":"doi:10.22541/au.171309702.22057241/v1","type":"article-journal","title":"Improvement the classification of a nanocomposite using nanoparticules based on a meta-analysis study, Recurrent Neural Network and Recurrent Neural Network Monte-Carlo algorithms","abstract":"This paper may be the first meta-analysis that presents a comprehensive synthesis of scientific works spanning the last five years, focusing on methodologies and results related to the analysis of nanocomposite using nanoparticules. The primary objective is to identify the optimal algorithm using software information and leading to better classification methodology. Specifically, this study come up with the advantages and the drawbacks of the most used algorithms and proposes an enhancement and performance of Recurrent Neural Networks based Long Short Term Memory (LSTM) neurons. Besides, a comparaison of Deep Learning methods for the classification of polymeric nanoparticles, with polypropylene serving as a case study will be implemented. Experiment comparison were conducted to assess with one physical property, later expanded to four properties and finally to eight properties. Neural networks, including Artificial Neural Networks (ANN), Recurrent Neural Networks (RNN), and Recurrent Neural Networks-Monte Carlo, were employed for simulations. The evaluation criteria encompassed accuracy, calculation time, mean square error (MSE) and other metrics. The findings contribute to the selection of an optimal algorithm for the analysis of polymeric nanoparticles, emphasizing the potential of Deep Learning methodologies, particularly Recurrent Neural Networks Monte Carlo, in advancing classification accuracy and efficiency.","author":[{"family":"Loukil","given":"Rania"},{"family":"Gazehi","given":"Wejdene"},{"family":"Besbes","given":"Mongi"}],"issued":{"date-parts":[[2024]]},"DOI":"10.22541/au.171309702.22057241/v1","URL":"https://doi.org/10.22541/au.171309702.22057241/v1","source":"crossref"},{"id":"doi:10.1088/1361-6528/acebf5","type":"article-journal","title":"Unsupervised learning in hexagonal boron nitride memristor-based spiking neural networks","abstract":"Abstract Resistive random access memory (RRAM) is an emerging non-volatile memory technology that can be used in neuromorphic computing hardware to exceed the limitations of traditional von Neumann architectures by merging processing and memory units. Two-dimensional (2D) materials with non-volatile switching behavior can be used as the switching layer of RRAMs, exhibiting superior behavior compared to conventional oxide-based devices. In this study, we investigate the electrical performance of 2D hexagonal boron nitride (h-BN) memristors towards their implementation in spiking neural networks (SNN). Based on experimental behavior of the h-BN memristors as artificial synapses, we simulate the implementation of unsupervised learning in SNN for image classification on the Modified National Institute of Standards and Technology dataset. Additionally, we propose a simple spike-timing-dependent-plasticity (STDP)-based dropout technique to enhance the recognition rate in h-BN memristor-based SNN. Our results demonstrate the viability of using 2D-material-based memristors as artificial synapses to perform unsupervised learning in SNN using hardware-friendly methods for online learning.","author":[{"family":"Afshari","given":"Sahra"},{"family":"Xie","given":"Jing"},{"family":"Musisi-Nkambwe","given":"Mirembe"},{"family":"Radhakrishnan","given":"Sritharini"},{"family":"Esqueda","given":"Ivan"}],"issued":{"date-parts":[[2023]]},"DOI":"10.1088/1361-6528/acebf5","URL":"https://doi.org/10.1088/1361-6528/acebf5","source":"crossref"},{"id":"doi:10.36227/techrxiv.171436020.02962844/v1","type":"article-journal","title":"Hierarchical Physics-Informed Neural Network Framework for 3D Magnetic Modeling of Medium Frequency Transformers","abstract":"Neural Network (NN) technology is revolutionizing the modeling paradigm in the engineering arena, enhancing simultaneously model precision and process speed. Nevertheless, the deficiency of physical interpretability in current NN models causes an unattainable demand for the quantity of training data, especially when dealing with high-dimension physical behaviours. In this work, a Hierarchical Physics-Informed Neural Network (HPINN) framework is proposed to address the 3D magnetic modeling issue of Medium Frequency Transformers (MFTs). By establishing the knowledge transfer channel between the cross section and the equivalent space, the inferior 2D model parameter search space, which is restricted by Dowell’s equation and accessible 2D magnetic data information, can be mapped into the superior 3D search space. Accordingly, with limited 3D training data, the performance of the HPINN framework is still guaranteed around the optimal state. The structural embedding of physics knowledge reduces the model's reliance on 3D training data, thereby relieving the strict computing requirement at the data generation stage. Finally, according to the training results and experiment verifications, with only one-third computational burden used in the classic NN model, the average error of this proposed HPINN framework is reduced to 1%, and the maximum error does not exceed 10%.","author":[{"family":"Yang","given":"Xiao"},{"family":"Shu","given":"Liangcai"},{"family":"Yang","given":"Dongsheng"}],"issued":{"date-parts":[[2024]]},"DOI":"10.36227/techrxiv.171436020.02962844/v1","URL":"https://doi.org/10.36227/techrxiv.171436020.02962844/v1","source":"crossref"},{"id":"doi:10.3365/kjmm.2024.62.3.212","type":"article-journal","title":"Simulation Study of Interfacial Switching Memristor Structure and Neural Network Performance","abstract":"In this study, the architecture of an interfacial switching memristor, which has a metal-insulatormetal structure of Pt/SrTiO&lt;sub&gt;3&lt;/sub&gt;/Nb-SrTiO&lt;sub&gt;3&lt;/sub&gt; was investigated. The performance of a neural network that uses memristors as its synapse components was also examined with system-level simulations. A finite element solver, COMSOL Multiphysics, was used to simulate synaptic device characteristics, specifically, the conductance change, using a series of pulses for a given architecture. An open-source software, NeuroSim, was used to simulate the ability of the neural network to recognize and identify handwritten digits. Electrostatics, mass transport, and thermionic emission equations were numerically solved in a fully coupled manner to model the Schottky barrier height modulation at the Pt/SrTiO&lt;sub&gt;3&lt;/sub&gt; contact using the applied bias. The barrier height is a function of the oxygen vacancy concentration in the SrTiO&lt;sub&gt;3&lt;/sub&gt; near the contact. The gradual change of the oxygen vacancy concentration profile caused by successive pulses results in the gradual change of conductance. Utilizing the simulations, the influences of device structure modification, and more specifically, changing the size of the Schottky contact, on long-term potentiation and depression were analyzed for planar devices. The results show that a smaller Schottky contact yields a higher digit recognition rate. Based on this finding, a three-dimensional device architecture that is vertically stackable was designed.","author":[{"family":"Song","given":"Yun"},{"family":"Lim","given":"Ji"},{"family":"Khot","given":"Sagar"},{"family":"Jung","given":"Dongmyung"},{"family":"Kwon","given":"Yongwoo"}],"issued":{"date-parts":[[2024]]},"DOI":"10.3365/kjmm.2024.62.3.212","URL":"https://doi.org/10.3365/kjmm.2024.62.3.212","source":"crossref"},{"id":"doi:10.20944/preprints202411.0183.v2","type":"manuscript","title":"Neural Network for Enhancing Robot Assisted Rehabilitation: A Systematic Review","abstract":"The integration of neural networks into robotic exoskeletons for physical rehabilitation has become popular due to their ability to interpret complex physiological signals. Surface electromyography (sEMG), electromyography (EMG), electroencephalography (EEG), and other physiological signals enable communication between the human body and robotic systems. Utilizing physiological signals for communicating with robots plays a crucial role in robot assisted neurorehabilitation. This systematic review synthesizes 44 peer-reviewed studies, exploring how neural networks can improve exoskeleton robot assisted rehabilitation for individuals with impaired upper limbs. By categorizing the studies based on robot assisted joints, sensor systems, and control methodologies, we offer a comprehensive overview of neural network applications in this field. Our findings demonstrate that neural networks, such as Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and Radial Basis Function Neural Networks (RBFNN), and other forms of neural network significantly contribute to patient specific rehabilitation by enabling adaptive learning and personalized therapy. CNNs improve motion intention estimation and control accuracy, while LSTM networks capture temporal muscle activity patterns for real-time rehabilitation. RBFNNs improve human-robot interaction by adapting to individual movement patterns, leading to more personalized and efficient therapy. This review highlights the potential of neural networks to revolutionize upper limb rehabilitation, improving motor recovery and patient outcomes in both clinical and home-based settings. It also recommends the future direction to customize existing neural networks for robot assisted rehabilitation applications.","author":[{"family":"Hasan","given":"Sk"},{"family":"Alam","given":"Nafizul"},{"family":"Mashud","given":"Gazi"},{"family":"Bhujel","given":"Subodh"}],"issued":{"date-parts":[[2024]]},"DOI":"10.20944/preprints202411.0183.v2","URL":"https://doi.org/10.20944/preprints202411.0183.v2","source":"crossref"},{"id":"doi:10.1101/2024.05.24.595714","type":"article-journal","title":"Electroencephalogram (EEG) Classification using a bio-inspired Deep Oscillatory Neural Network","abstract":"Abstract In this paper, we propose two models of oscillatory neural networks - the Deep Oscillatory Neural Network (DONN) and a convolutional variation of it named Oscillatory Convolutional Neural Network (OCNN) – and apply the models to a variety of problems involving the classification and prediction of Electroencephalogram (EEG) signals. Deep neural networks applied to signal processing problems will have to incorporate various architectural features to remember the history of the input signals e.g., loops between the layers, “gated” neurons, and tapped delay lines. But real brains have rich dynamics expressed in terms of frequency bands like alpha, beta, gamma, delta, etc. To incorporate this aspect of brain dynamics in a Recurrent Neural Network (RNN) we propose to use nonlinear oscillators as dynamic neuron models in the hidden layers. The two oscillatory deep neural networks proposed are applied to the following EEG classification and prediction problems: Prediction of nearby EEG channels, classification of single-channel EEG data (healthy vs. epileptic, different stages of sleep stage classification), and multi-channel EEG data (Epileptic vs. Normal, Left vs. right-hand Motor imagery movement, and healthy vs. Claustrophobic EEG).","author":[{"family":"Ghosh","given":"Sayan"},{"family":"Vigneswaran","given":"C"},{"family":"Rohan","given":"Nr"},{"family":"Chakravarthy","given":"VS"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1101/2024.05.24.595714","URL":"https://doi.org/10.1101/2024.05.24.595714","source":"crossref"},{"id":"doi:10.20944/preprints202401.1285.v1","type":"manuscript","title":"Predictive Neural Network Modeling for Almond Harvest Dust Control","abstract":"This study introduces a neural network-based approach to predict dust emission, specifically PM2.5 particles, during almond harvesting in California. Using a feedforward neural network (FNN), the research predicts PM2.5 emissions by analyzing key operational parameters of an advanced almond harvester. The model is trained on extensive field data from the almond pickup system, including variables like brush speed, angular velocity, and harvester forward speed. The results demonstrate a notable predictive accuracy of the FNN model, with a Mean Squared Error (MSE) of 0.02 and a Mean Absolute Error (MAE) of 0.01, indicating a high degree of precision in forecasting PM2.5 levels. The study also finds a strong correlation between certain operational parameters and PM2.5 emissions, highlighting specific areas for optimization in harvesting techniques. By integrating machine learning with agricultural practices, this research provides a significant tool for environmental management in almond production, offering a method to reduce harmful emissions while maintaining operational efficiency. This model presents a solution for the almond industry and sets a precedent for applying predictive analytics in sustainable agriculture.","author":[{"family":"Serajian","given":"Reza"},{"family":"Sun","given":"Jian"},{"family":"Cobian-Iñiguez","given":"Jeanette"},{"family":"Ehsani","given":"Reza"}],"issued":{"date-parts":[[2024]]},"DOI":"10.20944/preprints202401.1285.v1","URL":"https://doi.org/10.20944/preprints202401.1285.v1","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-3825788/v1","type":"article-journal","title":"Physics-informed two-tier neural network for non-linear model order reduction","abstract":"Abstract In recent years, machine learning (ML) has had a great impact in the area of non-intrusive, non-linear model order reduction (MOR). However, the offline training phase often still entails high computational costs since it requires numerous, expensive, full-order solutions as the training data. Furthermore, in state-of-the-art methods, neural networks trained by a small amount of training data cannot be expected to generalize sufficiently well, and the training phase often ignores the underlying physical information when it is applied with MOR. Moreover, state-of-the-art MOR techniques that ensure an efficient online stage, such as hyper reduction techniques, are either intrusive or entail high offline computational costs. To resolve these challenges, inspired by recent developments in physics-informed and physics-reinforced neural networks, we propose a non-intrusive, physics-informed, two-tier deep network (TTDN) method. The proposed network, in which the first tier achieves the regression of the unknown quantity of interest and the second tier rebuilds the physical constitutive law between the unknown quantities of interest and derived quantities, is trained using pretraining and semi-supervised learning strategies. To illustrate the efficiency of the proposed approach, we perform numerical experiments on challenging non-linear and non-affine problems, including multi-scale mechanics problems.","author":[{"family":"Hong","given":"Yankun"},{"family":"Bansal","given":"Harshit"},{"family":"Veroy","given":"Karen"}],"issued":{"date-parts":[[2024]]},"DOI":"10.21203/rs.3.rs-3825788/v1","URL":"https://doi.org/10.21203/rs.3.rs-3825788/v1","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-4199827/v1","type":"article-journal","title":"Constitutive Artificial Neural Network espoused Plant Leaf Disease Detection","abstract":"Abstract The emergence of new diseases on plant leaves poses a substantial threat to global food safety and agricultural productivity. To mitigate this risk, accurate and swift detection of plant illnesses is crucial, reducing unnecessary expenses and minimizing financial losses and environmental damage. This study proposes a method called Plant Leaf Disease Detection with a Constitutive Artificial Neural Network (PLDD-CANN) to provide advancements in deep learning. The approach begins by gathering data from the Plant Village dataset and subjecting it to pre-processing techniques. This includes noise removal and image enhancement using a Variational Marginalized Particle Filter (AVMPF). Next, an Adaptive Convex Clustering (ACC) method is employed for image segmentation, followed by feature extraction using Fast Fourier and Continuous Wavelet (FFCWT) transforms. Finally, a Constitutive Artificial Neural Network (CANN) is utilized to categorize the input image to one of several categories, including healthy and various disease types like Yellow Leaf Curl Virus, Septoria Leaf Spot, Two-Spotted Spider Mite, Bacterial Spot, Target Spot, Leaf Mold, Mosaic Virus, Early Blight, and Late Blight. Then, the proposed technique is simulated using Python under several performance metrics including precision, f1-score, error rate accuracy, sensitivity, specificity and ROC. The proposed PLDD-CANN method provides 26.75%, 25.83% and 27.46% higher accuracy comparing with existing methods an enhanced CNN technique for plant leaves disease diagnosis in tomato (CNN-PLDD), A Novel Approach for Plant Leaf Disease Predictions with Recurrent Neural Network RNN Classification Method (RNN-PLDD), Detection of tomato leaf diseases for agro-based industries (FRCNN-PLDD) respectively.","author":[{"family":"Kanagaraj","given":"Kaavya"},{"family":"Kulandaivel","given":"Madhumitha"},{"family":"Shajin","given":"FH"},{"family":"Prabhakaran","given":"Salini"}],"issued":{"date-parts":[[2024]]},"DOI":"10.21203/rs.3.rs-4199827/v1","URL":"https://doi.org/10.21203/rs.3.rs-4199827/v1","source":"crossref"},{"id":"doi:10.20944/preprints202407.0392.v1","type":"manuscript","title":"Enhanced Brain-to-Brain Communication Security via Adversarial Neural Network Training","abstract":"Brain-to-brain communication (B2B-C) is rapidly expanding, integrating communication technology and neuroscience to enable direct neural data transfer between people. Nonetheless, because neural data is susceptible to noise, interference, and hostile attacks, guaranteeing the security and resilience of B2B-C systems continues to be difficult. This work aims to use Adversarial Neural Network Training (ANNT) to improve the security of B2B-C systems by using Steady-State Visually Evoked Potentials (SSVEP) EEG data. We use two large SSVEP datasets for a thorough analysis: Lee2019_SSVEP and Nakanishi2015. We use the Fast Gradient Sign Method (FGSM) to create adversarial instances and ANNT to train the model on clean and adversarially perturbed data. The system&amp;#039;s accuracy and resilience are significantly improved by ANNT, as seen by the up to 17% increase in adversarial accuracy and the average 0.03 point improvement in the Area Under the Curve (AUC). This work demonstrates how ANNT may strengthen B2B-C systems against advanced cyberattacks, opening the door for dependable and safe neural communication technologies.","author":[{"family":"Ahmadi","given":"Hossein"},{"family":"Kuhestani","given":"Ali"},{"family":"Keshavarzi","given":"Mohammadreza"},{"family":"Mesin","given":"Luca"}],"issued":{"date-parts":[[2024]]},"DOI":"10.20944/preprints202407.0392.v1","URL":"https://doi.org/10.20944/preprints202407.0392.v1","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-5556309/v1","type":"article-journal","title":"A Novel Multi-Input Neural Network Model for MicroRNA Target-Site Detection","abstract":"Abstract Purpose: MicroRNAs are tiny non-coding RNA sequences that regulate gene expressions and play crucial roles in controlling cell activities. They function by binding to their target messenger RNAs (mRNAs) and suppressing protein synthesis. Detecting the binding location of a microRNA is an important step towards discovering microRNA functions. Given the many possibilities that a microRNA can bind to an mRNA, microRNA target-site detection is a challenging task in vivo and in vitro; from time and cost limitations of experimental methods to high false positive rates of computational methods. Methods: In this paper, we propose a multi-input neural network-based algorithm that yields high recall and precision results simultaneously. Moreover, we designed a Dynamic Programming(DP) algorithm to predict the duplex structure of microRNA target-site, specific to the context of these sequences. The predicted duplex structures, substructures resulting from the DP algorithm, minimum free energy (MFE) of the substructures, and a probabilistic image of possible base pairs in the duplex are all fed in parallel into our algorithm to learn a comprehensive and precise model. Results and Conclusions: Our method on an experimentally validated test set, detects target-sites with AUPRC of 0.9373, Precision of 0.8725, Recall of 0.8703, and outperforms several commonly used computational methods of microRNA target predictions. In addition, using the duplex structure, MFE, and binding probabilities, rather than the nucleotide sequences of the microRNA and targetsites, enables our model to generalize beyond specific sequence contexts and perform well on sequentially distant samples. The source code for our algorithm and the accompanying datasets are freely accessible on Github at: https://github.com/mohebbimg/minn.git","author":[{"family":"Mohebbi","given":"Mohammad"},{"family":"Bennett","given":"Ethan"},{"family":"Williams","given":"Phillip"}],"issued":{"date-parts":[[2024]]},"DOI":"10.21203/rs.3.rs-5556309/v1","URL":"https://doi.org/10.21203/rs.3.rs-5556309/v1","source":"crossref"},{"id":"doi:10.1101/2024.02.15.580574","type":"article-journal","title":"A Neural Network Approach to Identify Left-Right Orientation of Anatomical Brain MRI","abstract":"Abstract Left-right orientation misidentification in brain MRIs presents significant challenges due to several factors, including metadata loss or ambiguity, which often occurs during the de-identification of medical images for research, conversion between image formats, software operations that strip or overwrite metadata, and the use of older imaging systems that stored orientation differently. This study presents a novel application of deep-learning to enhance the accuracy of left-right orientation identification in anatomical brain MRI scans. A three-dimensional Convolutional Neural Network model was trained using 350 MRIs and evaluated on eight distinct brain MRI databases, totaling 3,384 MRIs, to assess its performance across various conditions, including neurodegenerative diseases. The proposed deep-learning framework demonstrated a 99.6% accuracy in identifying the left-right orientation, thus addressing challenges associated with the loss of orientation metadata. GradCAM was used to visualize areas of the brain where the model focused, demonstrating the importance of the right planum temporale and surrounding areas in judging left-right orientation. The planum temporale is known to exhibit notable left-right asymmetry related to language functions, underscoring the biological validity of the model. More than half of the ten left-right misidentified MRIs involved notable brain feature variations, such as severe temporal lobe atrophy, arachnoidal cysts adjacent to the temporal lobe, or unusual cerebral torque, indicating areas for further investigation. This approach offers a potential solution to the persistent issue of left-right misorientation in brain MRIs and supports the reliability of neuroscientific research by ensuring accurate data interpretation.","author":[{"family":"Nishimaki","given":"Kei"},{"family":"Iyatomi","given":"Hitoshi"},{"family":"Oishi","given":"Kenichi"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1101/2024.02.15.580574","URL":"https://doi.org/10.1101/2024.02.15.580574","source":"crossref"},{"id":"doi:10.20944/preprints202411.0183.v1","type":"manuscript","title":"Neural Network for Enhancing Robot Assisted Rehabilitation: A Systematic Review","abstract":"Recently, the integration of neural networks into robotic exoskeletons for physical rehabilitation has become popular due to their ability to interpret complex physiological signals. Surface electromyography (sEMG), electromyography (EMG), electroencephalography (EEG), and other physiological signals enable communication between the human body and robotic systems. Utilizing physiological signals for communicating with robots plays a crucial role in robot assisted neurorehabilitation. This systematic review synthesizes 44 peer-reviewed studies, exploring how neural networks can improve exoskeleton robot assisted rehabilitation for individuals with impaired upper limbs. By categorizing the studies based on robot assisted joints, sensor systems, and control methodologies, we offer a comprehensive overview of neural network applications in this field. Our findings demonstrate that neural networks, such as Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and Radial Basis Function Neural Networks (RBFNN), and other forms of neural network significantly contribute to patient specific rehabilitation by enabling adaptive learning and personalized therapy. CNNs improve motion intention estimation and control accuracy, while LSTM networks capture temporal muscle activity patterns for real-time rehabilitation. RBFNNs improve human-robot interaction by adapting to individual movement patterns, leading to more personalized and efficient therapy. This review highlights the potential of neural networks to revolutionize upper limb rehabilitation, improving motor recovery and patient outcomes in both clinical and home-based settings. It also recommends the future direction to customize existing neural networks for robot assisted rehabilitation applications.","author":[{"family":"Hasan","given":"Sk"},{"family":"Alam","given":"Nafizul"},{"family":"Mashud","given":"Gazi"},{"family":"Bhujel","given":"Subodh"}],"issued":{"date-parts":[[2024]]},"DOI":"10.20944/preprints202411.0183.v1","URL":"https://doi.org/10.20944/preprints202411.0183.v1","source":"crossref"},{"id":"doi:10.1101/2024.04.02.587669","type":"article-journal","title":"Recurrent issues with deep neural network models of visual recognition","abstract":"Abstract Object recognition requires flexible and robust information processing, especially in view of the challenges posed by naturalistic visual settings. The ventral stream in visual cortex is provided with this robustness by its recurrent connectivity. Recurrent deep neural networks (DNNs) have recently emerged as promising models of the ventral stream, surpassing feedforward DNNs in the ability to account for brain representations. In this study, we asked whether recurrent DNNs could also better account for human behaviour during visual recognition. We assembled a stimulus set that included manipulations that are often associated with recurrent processing in the literature, like occlusion, partial viewing, clutter, and spatial phase scrambling. We obtained a benchmark dataset from human participants performing a categorisation task on this stimulus set. By applying a wide range of model architectures to the same task, we uncovered a nuanced relationship between recurrence, model size, and performance. While recurrent models reach higher performance than their feedforward counterpart, we could not dissociate this improvement from that obtained by increasing model size. We found consistency between humans and models patterns of difficulty across the visual manipulations, but this was not modulated in an obvious way by the specific type of recurrence or size added to the model. Finally, depth/size rather than recurrence makes model confusion patterns more human-like. Contrary to previous assumptions, our findings challenge the notion that recurrent models are better models of human recognition behaviour than feedforward models, and emphasise the complexity of incorporating recurrence into computational models.","author":[{"family":"Maniquet","given":"Tim"},{"family":"Beeck","given":"Hans"},{"family":"Costantino","given":"Andrea"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1101/2024.04.02.587669","URL":"https://doi.org/10.1101/2024.04.02.587669","source":"crossref"},{"id":"doi:10.22541/au.172114776.66144810/v1","type":"article-journal","title":"Machine tool operating vibration prediction based on multi-sensor fusion and LSTM neural network","abstract":"In this study, a machine tool operating vibration prediction method based on multi-sensor fusion and long short-term memory (LSTM) network is proposed. Machine tool vibration has a significant impact on machining quality, workpiece surface roughness, dimensional accuracy, and tool’s wear. This study combines deep learning technology with industrial applications to achieve high-precision machine tool vibration prediction by fusing multiple sensor data. The real-time data is input into the LSTM model to predict the vibration situation at the next moment. The experimental results show that the method has strong prediction ability for the periodic vibration of the machine tool and the vibration error specific to the machining action. And it can effectively predict machine vibration and improve machining accuracy.","author":[{"family":"Shi","given":"Zhonglou"},{"family":"Duan","given":"Jinjie"},{"family":"Li","given":"Faquan"}],"issued":{"date-parts":[[2024]]},"DOI":"10.22541/au.172114776.66144810/v1","URL":"https://doi.org/10.22541/au.172114776.66144810/v1","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-4618247/v1","type":"article-journal","title":"Multi-path Hybrid Attention Deep Neural Network for Valve Detection","abstract":"Abstract Valves Detection is a basic function of rescue robots in various disaster situations. However, due to the small differences between similar valves, rescue robots suffer great challenges in valve detection in complex environments. To address this challenge, this paper proposes a multi-path hybrid attention deep neural network (MHADNN). By weighting features at different scales and spatial positions, the MHADNN can help valve detection models focus on more discriminative subtle features, thereby enhancing the ability to distinguish similar valves. This paper combines the MHADNN with the YOLOv5n to develop a valve detection model. The comparative experiments are conducted on a similar valve dataset collected in the simulated environment of a chemical industrial park. The experimental results show that compared with YOLOv5n, the proposed valve detection model has an average precision improvement of 4.20%. It has excellent performance in distinguishing similar valves.","author":[{"family":"Zhang","given":"First"},{"family":"Huang","given":"Second"},{"family":"Qu","given":"Xiwen"}],"issued":{"date-parts":[[2024]]},"DOI":"10.21203/rs.3.rs-4618247/v1","URL":"https://doi.org/10.21203/rs.3.rs-4618247/v1","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-3824277/v1","type":"article-journal","title":"CryoNeFEN: High-resolution reconstruction of cryo-EM structures using neural field network","abstract":"Abstract The elucidation of three-dimensional (3D) structures is crucial to unraveling protein function and illuminating mechanisms in structural biology. Cryogenic electron microscopy (cryo-EM) single-particle analysis provides direct measurements to determine the structures of macromolecules. However, the main challenge is reconstructing high-resolution 3D structures from extremely noisy and randomly oriented 2D projection images. Most existing methods primarily involve the optimization of multiple 2D slices in the Fourier domain but ignore the anisotropy among these slices, thus limiting the reconstruction of high-frequency structures. In this paper, we propose a cryo-EM neural field reconstruction network (cryoNeFEN), a new 3D spatial domain optimization paradigm, that learns a directional isotropic representation of the cryo-EM structure by mapping spatial coordinates to corresponding density values. We qualitatively and quantitatively evaluate cryoNeFEN on four experimental datasets. The results demonstrate the improved directional isotropy and 3D density resolution beyond the limits of existing algorithms in homogeneous reconstruction and resolve the missing elements of SARS-CoV-2 distinctly in heterogeneous reconstruction.","author":[{"family":"Liu","given":"Manhua"},{"family":"Huang","given":"Yue"},{"family":"Chengguang","given":"Zhu"}],"issued":{"date-parts":[[2024]]},"DOI":"10.21203/rs.3.rs-3824277/v1","URL":"https://doi.org/10.21203/rs.3.rs-3824277/v1","source":"crossref"},{"id":"doi:10.1063/5.0211178","type":"article-journal","title":"Effect of neural firing pattern on NbOx/Al2O3 memristor-based reservoir computing system","abstract":"The implementation of reservoir computing using resistive random-access memory as a physical reservoir has attracted attention due to its low training cost and high energy efficiency during parallel data processing. In this work, a NbOx/Al2O3-based memristor device was fabricated through a sputter and atomic layer deposition process to realize reservoir computing. The proposed device exhibits favorable resistive switching properties (&amp;gt;103 cycle endurance) and demonstrates short-term memory characteristics with current decay. Utilizing the controllability of the resistance state and its variability during cycle repetition, electrical pulses are applied to investigate the synapse-emulating properties of the device. The results showcase the functions of potentiation and depression, the coexistence of short-term and long-term plasticity, excitatory post-synaptic current, and spike-rate dependent plasticity. Building upon the functionalities of an artificial synapse, pulse spikes are categorized into three distinct neural firing patterns (normal, adapt, and boost) to implement 4-bit reservoir computing, enabling a significant distinction between “0” and “1.”","author":[{"family":"Ju","given":"Dongyeol"},{"family":"Ji","given":"Hyeonseung"},{"family":"Lee","given":"Jungwoo"},{"family":"Kim","given":"Sungjun"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1063/5.0211178","URL":"https://doi.org/10.1063/5.0211178","source":"crossref"},{"id":"doi:10.14311/nnw.2023.33.024","type":"article-journal","title":"Upgrading the JANET neural network by introducing a new storage buffer of working memory","abstract":"Recurrent neural networks (RNNs), along with long short-term memory networks (LSTMs), have been successfully used on a wide range of sequential data problems and have been entitled as extraordinarily powerful tools for learning and processing such data. However, the search for a new or derived architecture that would model very long-term dependencies is still an active area of research. In this paper, a relatively psychologically plausible architecture named event buffering JANET (EB-JANET) is proposed. The architecture is derived from the forgetgate- only version of the LSTM, which is also called just another network (JANET). The new architecture implements a new working memory mechanism that operates on information represented as dynamic events. The event buffer, as a container of events, is a reference to the state of the relevant pre-activation values on the basis of which historical candidate values were generated relative to the current timestep. The buffer is emptied as needed and depending on the context of information. The proposed architecture has achieved world-class results and it outperforms JANET on multiple benchmark datasets. Moreover, the new architecture is applicable to a wider class of problems and showed superior resilience when processing longer sequences, as opposed to JANET which experienced catastrophic failures on certain tasks.","author":[{"family":"Tolic","given":"Antonio"},{"family":"Boshkoska","given":"Biljana"},{"family":"Skansi","given":"Sandro"}],"issued":{"date-parts":[[2024]]},"DOI":"10.14311/nnw.2023.33.024","URL":"https://doi.org/10.14311/nnw.2023.33.024","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-4273139/v1","type":"article-journal","title":"Prediction Model For Digital Image Tampering Using Customized Deep Neural Network Techniques","abstract":"Abstract Image tampering detection is a critical area of research, given the widespread use of manipulated images for deceptive purposes. Convolutional Neural Networks (CNNs) have shown significant potential in automating the identification of tampered images. This paper presents customized deep learning model to detect tampering class with comparative analysis of CNN architectures - ResNet50V2, InceptionNetV3, MobileNetV2, and the proposed CNN, for image tampering detection. The proposed approach encompasses a dataset comprising four distinct classes: copy-move, inpaint, splicing, and normal images. This study sheds light on the comparative strengths and weaknesses of these CNN architectures. The dataset encompasses the key tampered classes, offering a holistic assessment of each model's ability to identify various tampering techniques. The custom CNN architecture is specifically tailored for this task, aiming to evaluate its efficiency compared to the established CNNs. Metrics for training and evaluation are standardized to generate equitable comparisons, encompassing performance indicators such as accuracy, precision, recall, and F1-score.This research contributes the knowledge in the field of image tampering detection, offering a comprehensive evaluation of multiple CNN architectures. Additionally, the effectiveness of separable convolutional layers is explored in deep neural networks, showcasing their potential to enhance scalability and effectiveness across various tasks in machine learning and computer vision. The proposed model, designed with separable convolution layers, exhibits superior validation accuracy and training accuracy compared to the other models under evaluation. Notably, The proposed customized model achieved an impressive F1 score of 96%, highlighting its proficiency in accurately detecting tampered regions within images while minimizing false positives.","author":[{"family":"Saxena","given":"Sachin"},{"family":"Singh","given":"Archana"},{"family":"Tiwari","given":"Shailesh"}],"issued":{"date-parts":[[2024]]},"DOI":"10.21203/rs.3.rs-4273139/v1","URL":"https://doi.org/10.21203/rs.3.rs-4273139/v1","source":"crossref"},{"id":"doi:10.2139/ssrn.4864049","type":"manuscript","title":"NERD: Neural Network for Edict of Risky Data Streams","abstract":"Cyber incidents can have a wide range of cause from a simple connection loss to an insistent attack. Once a potential cyber security incidents and system failures have been identified, deciding how to proceed is often complex. Especially, if the real cause is not directly in detail determinable. Therefore, we developed the concept of a Cyber Incident Handling Support System. The developed system is enriched with information by multiple sources such as intrusion detection systems and monitoring tools. It uses over twenty key attributes like sync-package ratio to identify potential security incidents and to classify the data into different priority categories. Afterwards, the system uses artificial intelligence to support the further decision-making process and to generate corresponding reports to brief the Board of Directors. Originating from this information, appropriate and detailed suggestions are made regarding the causes and troubleshooting measures. Feedback from users regarding the problem solutions are included into future decision-making by using labelled flow data as input for the learning process. The prototype shows that the decision making can be sustainably improved and the Cyber Incident Handling process becomes much more effective.","author":[{"family":"Hillmann","given":"Peter"},{"family":"Passarelli","given":"Sandro"},{"family":"Gündogan","given":"Cem"},{"family":"Stiemert","given":"Lars"},{"family":"Schopp","given":"Matthias"}],"issued":{"date-parts":[[2024]]},"DOI":"10.2139/ssrn.4864049","URL":"https://doi.org/10.2139/ssrn.4864049","source":"crossref"},{"id":"doi:10.1145/3665138","type":"article-journal","title":"Efficient Automation of Neural Network Design: A Survey on Differentiable Neural Architecture Search","abstract":"In the past few years, Differentiable Neural Architecture Search (DNAS) rapidly imposed itself as the trending approach to automate the discovery of deep neural network architectures. This rise is mainly due to the popularity of DARTS (Differentiable ARchitecTure Search), one of the first major DNAS methods. In contrast with previous works based on Reinforcement Learning or Evolutionary Algorithms, DNAS is faster by several orders of magnitude and uses fewer computational resources. In this comprehensive survey, we focused specifically on DNAS and reviewed recent approaches in this field. Furthermore, we proposed a novel challenge-based taxonomy to classify DNAS methods. We also discussed the contributions brought to DNAS in the past few years and its impact on the global NAS field. Finally, we concluded by giving some insights into future research directions for the DNAS field.","author":[{"family":"Heuillet","given":"Alexandre"},{"family":"Nasser","given":"Ahmad"},{"family":"Arioui","given":"Hichem"},{"family":"Tabia","given":"Hedi"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1145/3665138","URL":"https://doi.org/10.1145/3665138","source":"crossref"},{"id":"doi:10.1088/1741-2552/ad5404","type":"article-journal","title":"Blindly separated spontaneous network-level oscillations predict corticospinal excitability","abstract":"Abstract Objective. The corticospinal responses of the motor network to transcranial magnetic stimulation (TMS) are highly variable. While often regarded as noise, this variability provides a way of probing dynamic brain states related to excitability. We aimed to uncover spontaneously occurring cortical states that alter corticospinal excitability. Approach. Electroencephalography (EEG) recorded during TMS registers fast neural dynamics—unfortunately, at the cost of anatomical precision. We employed analytic Common Spatial Patterns technique to derive excitability-related cortical activity from pre-TMS EEG signals while overcoming spatial specificity issues. Main results. High corticospinal excitability was predicted by alpha-band activity, localized adjacent to the stimulated left motor cortex, and suggesting a travelling wave-like phenomenon towards frontal regions. Low excitability was predicted by alpha-band activity localized in the medial parietal–occipital and frontal cortical regions. Significance. We established a data-driven approach for uncovering network-level neural activity that modulates TMS effects. It requires no prior anatomical assumptions, while being physiologically interpretable, and can be employed in both exploratory investigation and brain state-dependent stimulation.","author":[{"family":"Ermolova","given":"Maria"},{"family":"Metsomaa","given":"Johanna"},{"family":"Belardinelli","given":"Paolo"},{"family":"Zrenner","given":"Christoph"},{"family":"Ziemann","given":"Ulf"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1088/1741-2552/ad5404","URL":"https://doi.org/10.1088/1741-2552/ad5404","source":"crossref"},{"id":"doi:10.1101/2024.08.13.607720","type":"article-journal","title":"Deep graph convolutional neural network for one-dimensional hepatic vascular haemodynamic prediction","abstract":"Abstract Hepatic vascular hemodynamics is an important reference indicator in the diagnosis and treatment of hepatic diseases. However, Method based on Computational Fluid Dynamics(CFD) are difficult to promote in clinical applications due to their computational complexity. To this end, this study proposed a deep graph neural network model to simulate the one-dimensional hemodynamic results of hepatic vessels. By connecting residuals between edges and nodes, this framework effectively enhances network prediction accuracy and efficiently avoids over-smoothing phenomena. The graph structure constructed from the centerline and boundary conditions of the hepatic vasculature can serve as the network input, yielding velocity and pressure information corresponding to the centerline. Experimental results indicate that our proposed method achieves higher accuracy on a hepatic vasculature dataset with significant individual variations and can be extended to applications involving other blood vessels. Following training, errors in both the velocity and pressure fields are maintained below 1.5%. The trained network model can be easily deployed on low-performance devices and, compared to CFD-based methods, can output velocity and pressure along the hepatic vessel centerline at a speed three orders of magnitude faster. Author summary When using deep learning methods for hemodynamic analysis, simple point cloud data cannot express the real geometric structure of the blood vessels, and it is necessary for the network to have additional geometric information extraction capability. In this paper, we use graph structure to express the structure of hepatic blood vessels, and deep graph neural network to predict the corresponding hemodynamic parameters. The graph structure can effectively express the geometric information of hepatic blood vessels and the topology of branch blood vessels, which can effectively improve the prediction accuracy with strong geometric generalisation ability. The results show that the method achieves the highest prediction accuracy in the one-dimensional hepatic vessel blood flow simulation dataset, and the experimental results on the human aorta also show that our method can be effectively applied to the blood flow simulation of other vascular organs.","author":[{"family":"Zhang","given":"Weiqng"},{"family":"Shi","given":"Shuaifeng"},{"family":"Qi","given":"Quan"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1101/2024.08.13.607720","URL":"https://doi.org/10.1101/2024.08.13.607720","source":"crossref"},{"id":"doi:10.22541/essoar.172202798.82044672/v1","type":"article-journal","title":"PEGSGraph: a Graph Neural Network for fast earthquake characterization based on Prompt ElastoGravity Signals","abstract":"State-of-the-art earthquake early warning systems use the early records of seismic waves to estimate the magnitude and location of the seismic source before the shaking and the tsunami strike. Because of the inherent properties of early seismic records, those systems systematically underestimate the magnitude of large events, which results in catastrophic underestimation of the subsequent tsunamis. Prompt elastogravity signals (PEGS) are low-amplitude, light-speed signals emitted by earthquakes, which are highly sensitive to both their magnitude and focal mechanism. Detected before traditional seismic waves, PEGS have the potential to produce unsaturated magnitude estimates faster than state-of-the-art systems. Accurate instantaneous tracking of large earthquake magnitude using PEGS has been proven possible through the use of a Convolutional Neural Network (CNN). However, the CNN architecture is sub-optimal as it does not allow to capture the geometry of the problem. To address this limitation, we design PEGSGraph, a novel deep learning model relying on a Graph Neural Network (GNN) architecture. PEGSGraph accurately estimates the magnitude of synthetic earthquakes down to Mw 7.6-7.7 and determines their focal mechanisms (thrust, strike-slip or normal faulting) within 70 seconds of the event’s onset, offering crucial information for predicting potential tsunami wave amplitudes. Our comparative analysis on Alaska and Western Canada data shows that the GNN outperforms the CNN, especially on test samples with low signal-to-noise ratios, providing more reliable rapid magnitude estimates and enhancing tsunami warning reliability.","author":[{"family":"Hourcade","given":"Céline"},{"family":"Juhel","given":"Kévin"},{"family":"Bletery","given":"Quentin"}],"issued":{"date-parts":[[2024]]},"DOI":"10.22541/essoar.172202798.82044672/v1","URL":"https://doi.org/10.22541/essoar.172202798.82044672/v1","source":"crossref"},{"id":"doi:10.2139/ssrn.4927691","type":"manuscript","title":"Neural Network Compression Using Binarization and Few Full-Precision Weights","abstract":"Quantization and pruning are two effective Deep Neural Networks model compression methods.In this paper, we propose Automatic Prune Binarization (APB), a novel compression technique combining quantization with pruning. APB enhances the representational capability of binary networks using a few full-precision weights.Our technique jointly maximizes the accuracy of the network while minimizing its memory impact by deciding whether each weight should be binarized or kept in full precision.We show how to efficiently perform a forward pass through layers compressed using APB by decomposing it into a binary and a sparse-dense matrix multiplication. Moreover, we design two novel efficient algorithms for extremely quantized matrix multiplication on CPU, leveraging highly efficient bitwise operations. The proposed algorithms are 6.9x and 1.5x faster than available state-of-the-art solutions. We extensively evaluate APB on two widely adopted model compression datasets, namely Cifar-10 and ImageNet. APB shows to deliver better accuracy/memory trade-off compared to state-of-the-art methods based on i) quantization, ii) pruning, and iii) a combination of pruning and quantization.APB also outperforms quantization in the accuracy/efficiency trade-off, being up to 2x faster than the 2-bits quantized model with no loss in accuracy.","author":[{"family":"Rulli","given":"Cosimo"},{"family":"Nardini","given":"Franco"},{"family":"Trani","given":"Salvatore"},{"family":"Venturini","given":"Rossano"}],"issued":{"date-parts":[[2024]]},"DOI":"10.2139/ssrn.4927691","URL":"https://doi.org/10.2139/ssrn.4927691","source":"crossref"},{"id":"doi:10.1115/isfa2024-141304","type":"article-journal","title":"A Recurrent Neural Network Enhanced Unscented Kalman Filter for Human Motion Prediction","abstract":"Abstract This paper presents a deep learning enhanced adaptive unscented Kalman filter (UKF) for predicting human arm motion in the context of manufacturing. Unlike previous network-based methods that solely rely on captured human motion data, which is represented as bone vectors in this paper, we incorporate a human arm dynamic model into the motion prediction algorithm and use the UKF to iteratively forecast human arm motions. Specifically, a Lagrangian-mechanics-based physical model is employed to correlate arm motions with associated muscle forces. Then a Recurrent Neural Network (RNN) is integrated into the framework to predict future muscle forces, which are transferred back to future arm motions based on the dynamic model. Given the absence of measurement data for future human motions that can be input into the UKF to update the state, we integrate another RNN to directly predict human future motions and treat the prediction as surrogate measurement data fed into the UKF. A noteworthy aspect of this study involves the quantification of uncertainties associated with both the data-driven and physical models in one unified framework. These quantified uncertainties are used to dynamically adapt the measurement and process noises of the UKF over time. This adaption, driven by the uncertainties of the RNN models, addresses inaccuracies stemming from the data-driven model and mitigates discrepancies between the assumed and true physical models, ultimately enhancing the accuracy and robustness of our predictions. One unique point of our method is that it integrates a dynamic model of human arms and two RNN models, and uses Monte Carlo dropout sampling to quantify the uncertainties inherent in our RNN prediction models and transforms them into the covariances of the UKF’s measurement and process noises respectively. Compared to the traditional RNN-based prediction, our method demonstrates improved accuracy and robustness in extensive experimental validations of various types of human motions.","author":[{"family":"Liu","given":"Wansong"},{"family":"Tian","given":"Sibo"},{"family":"Hu","given":"Boyi"},{"family":"Liang","given":"Xiao"},{"family":"Zheng","given":"Minghui"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1115/isfa2024-141304","URL":"https://doi.org/10.1115/isfa2024-141304","source":"crossref"},{"id":"doi:10.5121/ijnsa.2024.16602","type":"article-journal","title":"Improving Intrusion Detection System using the Combination of Neural Network and Genetic Algorithm","abstract":"One of the essential issues in network-based systems is a fault attack which is caused by intrusion. It is the responsibility of intrusion detection to provide capabilities such as adaptation, fault tolerance, high computational speed, and error resilience in the face of noisy information. Thus, the construction of an efficient intrusion detection model is highly appreciated to increase the detection rates as well as to decrease false detection. Currently, researchers are more focusing on abnormal behaviour of network as this system can easily recognize new attacks without updating the daily recognized databases. However, the capability of current developing machine learning algorithms suffers from inefficient use of intrusion detection particularly once it involved some huge datasets of irrelevant and redundant features. The main objective of this thesis is to achieve the higher detection rate with lower false detection for attack recognition in order to support efficient application of intrusion detection. To achieve this goal, a new machine learning model was designed and developed to provide intelligent recognition with new attack patterns. New proposed and improved algorithm GA-ANN is constructed to support the proof of concept. For evaluation, five datasets namely KDD CUP 99 from the online data repositories are used in the experiment. In the above scenarios, GA-ANN provides the highest detection rate for pattern recognition which was 98.98% based on 18 selected features. This means that the proposed IDS model is significant and increases the network security.","author":[{"family":"Dastanpour","given":"Amin"},{"family":"Farizani","given":"Amirabbas"},{"family":"Mahmood","given":"Raja"}],"issued":{"date-parts":[[2024]]},"DOI":"10.5121/ijnsa.2024.16602","URL":"https://doi.org/10.5121/ijnsa.2024.16602","source":"crossref"},{"id":"doi:10.1177/01423312231200514","type":"article-journal","title":"Delay-independent control for synchronization of memristor-based BAM neural networks with parameter perturbation and strong mismatch via finite-time technology","abstract":"This paper mainly studies the synchronization problem of memristor-based bidirectional associative memory neural networks (MBAMNNs) via finite-time technology. Different from the existing neural network dynamic models, the given model in this paper is focused on the impact of parameter perturbation and strong mismatch, where strong mismatch includes parameter mismatch and time-varying delay mismatch. These characteristics can make the model be closer to the actual situation. A delay-independent feedback control scheme, which can stabilize the error system within finite-time regardless of whether the past state is known or not, is designed. It is worth noting that the constant is replaced by a function with the exponential term in the delay-independent controller, which can save the control cost to a certain extent. Based on the integral inequality technique, some sufficient conditions for MBAMNNs to converge to the equilibrium point within finite-time are provided. The validity and correctness of the theoretical results are finally confirmed by numerical simulation.","author":[{"family":"Zhou","given":"Lili"},{"family":"Zhang","given":"Huiying"},{"family":"Tan","given":"Fei"},{"family":"Liu","given":"Kaiyue"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1177/01423312231200514","URL":"https://doi.org/10.1177/01423312231200514","source":"crossref"},{"id":"doi:10.1038/s41467-023-44620-1","type":"article-journal","title":"Purely self-rectifying memristor-based passive crossbar array for artificial neural network accelerators","abstract":"Abstract Memristor-integrated passive crossbar arrays (CAs) could potentially accelerate neural network (NN) computations, but studies on these devices are limited to software-based simulations owing to their poor reliability. Herein, we propose a self-rectifying memristor-based 1 kb CA as a hardware accelerator for NN computations. We conducted fully hardware-based single-layer NN classification tasks involving the Modified National Institute of Standards and Technology database using the developed passive CA, and achieved 100% classification accuracy for 1500 test sets. We also investigated the influences of the defect-tolerance capability of the CA, impact of the conductance range of the integrated memristors, and presence or absence of selection functionality in the integrated memristors on the image classification tasks. We offer valuable insights into the behavior and performance of CA devices under various conditions and provide evidence of the practicality of memristor-integrated passive CAs as hardware accelerators for NN applications.","author":[{"family":"Jeon","given":"Kanghyeok"},{"family":"Ryu","given":"Jin"},{"family":"Im","given":"Seongil"},{"family":"Seo","given":"Hyun"},{"family":"Eom","given":"Taeyong"},{"family":"Ju","given":"Hyunsu"},{"family":"Yang","given":"Min"},{"family":"Jeong","given":"Doo"},{"family":"Kim","given":"Gun"}],"issued":{"date-parts":[[2024]]},"DOI":"10.1038/s41467-023-44620-1","URL":"https://doi.org/10.1038/s41467-023-44620-1","source":"crossref"},{"id":"doi:10.5194/egusphere-egu24-11928","type":"article-journal","title":"The dynamics of field soil water retention curves predicted by autoencoder neural network","abstract":"The matric potential plays a pivotal role in understanding of water movement, plant water availability, and mechanical stability. In lack of direct measurements, the matric potential dynamics must be deduced from soil water content values, using the soil water retention curve. This approach is of particular importance at larger scales where only the water content (but not the potential) can be deduced from satellite data. However, because the relationship between water content and matric potential in natural field soils is highly ambiguous, not unique and dynamic, the prediction of matric potential from water content data is a big challenge. This ambiguity is related to different structures controlling drainage and wetting, dynamic effects, and seasonal changes of structures controlling the water distribution. In this study we present an autoencoder neural network as a new approach to analyze the soil moisture dynamics and to predict matric potential from water content data. The autoencoder compresses the water content time series into a site-specific feature (denoted as autoencoder value, AUV) that is representative of the underlying soil moisture dynamics. The AUV can then be used as predictor of the matric potential and the highly hysteretic soil water retention curve. The approach was tested successfully for nine soil profiles in the region of Solothurn (Switzerland). Three sites were chosen to establish the connection between AUV and the ambiguous soil water retention curve using a deep neural network, that was then applied to predict the matric potential dynamics of the other six sites. This method offers the potential to (i) deduce matric potential dynamics by relying solely on soil water content measurements (including satellite data), even when strong seasonal effects challenge standard methods, and (ii) serves as a warning system for changes in soil properties and in the intricate relationship between soil water content and matric potential dynamics.","author":[{"family":"Aqel","given":"Nedal"},{"family":"Carminati","given":"Andrea"},{"family":"Lehmann","given":"Peter"}],"issued":{"date-parts":[[2024]]},"DOI":"10.5194/egusphere-egu24-11928","URL":"https://doi.org/10.5194/egusphere-egu24-11928","source":"crossref"},{"id":"doi:10.1088/2634-4386/acb2f0","type":"article-journal","title":"Text classification in memristor-based spiking neural networks","abstract":"Abstract Memristors, emerging non-volatile memory devices, have shown promising potential in neuromorphic hardware designs, especially in spiking neural network (SNN) hardware implementation. Memristor-based SNNs have been successfully applied in a wide range of applications, including image classification and pattern recognition. However, implementing memristor-based SNNs in text classification is still under exploration. One of the main reasons is that training memristor-based SNNs for text classification is costly due to the lack of efficient learning rules and memristor non-idealities. To address these issues and accelerate the research of exploring memristor-based SNNs in text classification applications, we develop a simulation framework with a virtual memristor array using an empirical memristor model. We use this framework to demonstrate a sentiment analysis task in the IMDB movie reviews dataset. We take two approaches to obtain trained SNNs with memristor models: (1) by converting a pre-trained artificial neural network (ANN) to a memristor-based SNN, or (2) by training a memristor-based SNN directly. These two approaches can be applied in two scenarios: offline classification and online training. We achieve the classification accuracy of 85.88% by converting a pre-trained ANN to a memristor-based SNN and 84.86% by training the memristor-based SNN directly, given that the baseline training accuracy of the equivalent ANN is 86.02%. We conclude that it is possible to achieve similar classification accuracy in simulation from ANNs to SNNs and from non-memristive synapses to data-driven memristive synapses. We also investigate how global parameters such as spike train length, the read noise, and the weight updating stop conditions affect the neural networks in both approaches. This investigation further indicates that the simulation using statistic memristor models in the two approaches presented by this paper can assist the exploration of memristor-based SNNs in natural language processing tasks.","author":[{"family":"Huang","given":"Jinqi"},{"family":"Serb","given":"Alexantrou"},{"family":"Stathopoulos","given":"Spyros"},{"family":"Prodromakis","given":"Themis"}],"issued":{"date-parts":[[2023]]},"DOI":"10.1088/2634-4386/acb2f0","URL":"https://doi.org/10.1088/2634-4386/acb2f0","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-2601995/v1","type":"article-journal","title":"Convolutional neural network classifier incorporating misclassification information","abstract":"Abstract In this paper we introduce misclassification information for the improved training of convolutional neural network classifiers (CNNCs) for image recognition. We construct a additional autoencoder neural network, called tutor, that forces the CNNCs to learn the difference between the misclassified picture and the picture corresponding to the misclassified category. Making full use of the classification results to guide the CNNCs for purposeful learning is expected to improve the learning efficiency and classification performance. We integrate the proposed tutor into several state-of-the-art CNNCs architectures and demonstrate improvement in their recognition performance on CIFAR-10/100 and MNIST datasets. Our results suggest that making the most of misclassification information to guide the training of the model can lead to significant performance improvement.","author":[{"family":"Hu","given":"Junying"},{"family":"Fei","given":"Rongrong"},{"family":"Du","given":"Fang"},{"family":"Chang","given":"Peiju"},{"family":"Zhang","given":"Jiangshe"}],"issued":{"date-parts":[[2023]]},"DOI":"10.21203/rs.3.rs-2601995/v1","URL":"https://doi.org/10.21203/rs.3.rs-2601995/v1","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-4171645/v1","type":"article-journal","title":"DCNN: A Novel Binary and Multi-Class Network Intrusion Detection Model via Deep Convolutional Neural Network","abstract":"Abstract Network security has become imperative in the context of our interconnected networks and everyday communications. Recently, many deep learning models have been proposed to tackle the problem of predicting intrusions and malicious activities in interconnected systems. However, they solely focus on binary classification and lack reporting on individual class performance in case of multi-class classification. Therefore, the need for an efficient and accurate network intrusion detection system has reached a pivotal point. In this paper, we propose a novel intelligent detection system based on convolutional neural network, namely DCNN. The proposed model can be utilized to analyze and detect attacks and intrusions in intelligent network systems. The DCNN model is applied against two benchmark datasets and compared with state-of-the-art models. Experimental results show that the proposed model improved resilience to intrusions and malicious activities for binary as well as multi-class classification. Furthermore, our DCNN model outperforms similar intrusion detection systems in terms of positive predicted value, true positive rate, F1 measure, and accuracy. The scores obtained for binary and multi-class classifications on the CICIoT2023 dataset are 99.50% and 99.25%, respectively. Additionally, for the CICIDS-2017 dataset, DCNN attains a score of 99.96% for both binary and multi-class classifications.","author":[{"family":"Shebl","given":"Ahmed"},{"family":"Elsedimy","given":"Sayed"},{"family":"Ismail","given":"Amr"},{"family":"Salama","given":"Ahmed"},{"family":"Herajy","given":"Mostafa"}],"issued":{"date-parts":[[2024]]},"DOI":"10.21203/rs.3.rs-4171645/v1","URL":"https://doi.org/10.21203/rs.3.rs-4171645/v1","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-2508012/v1","type":"article-journal","title":"N-Net: A Convolutional Neural Network for Medical Image Segmentation","abstract":"Abstract This paper presents a novel supervised convolutional neural network architecture, \"N-Net\", capable of effectively learning and generalizing from small amounts of medical images to perform accurate segmentation tasks. Our model utilizes an encoder-decoder structure with a residual downsampling mechanism and a custom convolutional block to capture and process image information at multiple resolutions in the encoder segment. We employ data augmentation techniques to enrich the training set, thus increasing our model's performance. While our architecture is versatile and applicable to various segmentation tasks, in this study, we demonstrate its capabilities specifically for polyp segmentation in colonoscopy images. We evaluate the performance of our method on several popular benchmark datasets for polyp segmentation, Kvasir-SEG, CVC-ClinicDB, CVC-ColonDB, and ETIS-LARIBPOLYPDB showing that it achieves state-of-the-art results in terms of mean Dice coefficient, Jaccard index, Precision, Recall, and Accuracy. Our approach demonstrates strong generalization capabilities, achieving excellent performance even with limited training data.","author":[{"family":"Dumitru","given":"Razvan"},{"family":"Peteleaza","given":"Darius"},{"family":"Craciun","given":"Catalin"}],"issued":{"date-parts":[[2023]]},"DOI":"10.21203/rs.3.rs-2508012/v1","URL":"https://doi.org/10.21203/rs.3.rs-2508012/v1","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-3935900/v1","type":"article-journal","title":"Step Network: A Neural Network that Takes Into Account Spatial Structure and Texture Features for Human Pose Transfer","abstract":"Abstract Portrait synthesis guided by pose presents a challenging frontier in image generation. In our latest research, we have introduced an innovative network called the Step network, specifically designed to overcome the limitations identified in previous works. Our approach differs from traditional methods by honing in on the spatial structure of the pose, enabling a gradual migration of the pose while minimizing the loss of spatial structure information at each step. Taking inspiration from triplet loss, we have incorporated Style Discriminator to elevate texture generation. Moreover, in contrast to prior research, we have placed greater emphasis on refining the generation of facial areas. To achieve this, we employed a specialized loss function that combines triplet loss and L1 loss to optimize facial features, resulting in images that are more aligned with human perception. Additionally, we have also extended our method to include the ability to replace clothing on the body. To evaluate the quality of the images generated, we utilized evaluation metrics wihch are PSNR, SSIM, FID, and LPIPS. Through both qualitative and quantitative experiments comparing our approach with state-of-the-art models, we have demonstrated significant improvements across these metrics, confirming its superiority. Specifically, our method achieved a PSNR of 18.0376, SSIM of 0.7686, FID of 10.8102, and LPIPS of 0.1665. Our experimental results and open source code in the following url: https://github.com/FineURRight/Step-Network/tree/main.","author":[{"family":"Mo","given":"Han"},{"family":"Xu","given":"Yang"},{"family":"Zhang","given":"Caideng"},{"family":"Zhang","given":"Yongdan"}],"issued":{"date-parts":[[2024]]},"DOI":"10.21203/rs.3.rs-3935900/v1","URL":"https://doi.org/10.21203/rs.3.rs-3935900/v1","source":"crossref"},{"id":"oa:W4406428599","type":"article-journal","title":"Advanced Materials Research at CUHK: From Biomedicine to Electronics and Beyond","abstract":"This special issue spans a diverse array of topics, including nanomedicine, tissue engineering, regenerative medicine, organs-on-chips, biosensing, soft robotics, smart devices, nanofabrication, energy saving and storage, catalysis, spintronics, soft electronics, and neuromorphic computing. It showcases the breadth and depth of advanced materials research at the Chinese University of Hong Kong (CUHK), highlighting the innovation, collaboration, and excellence of CUHK's materials scientists.","author":[{"family":"Mao","given":"Chuanbin"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1002/adma.202418618","URL":"https://doi.org/10.1002/adma.202418618","source":"openalex"},{"id":"oa:W4296780357","type":"article-journal","title":"Statics and dynamics of skyrmions interacting with disorder and nanostructures","abstract":"The study of skyrmions has grown as they have been realized in topologically stable spin-textured magnetic excitations. This overview treats magnetic skyrmions and skyrmion assemblies using a particle-based approach to describe a system in which the energy scales of the skyrmion-skyrmion interactions, sample disorder, temperature, and drive magnitude compete. Pinning and dynamic effects of skyrmions interacting with disordered or ordered substrates are of technological importance for skyrmion applications. Analogs to vortices in type-II superconductors, charge density waves, and colloidal particles are presented.","author":[{"family":"Reichhardt","given":"CJO"},{"family":"Reichhardt","given":"CJO"},{"family":"Miloševıć","given":"MV"}],"issued":{"date-parts":[[2022]]},"DOI":"10.1103/revmodphys.94.035005","URL":"https://doi.org/10.1103/revmodphys.94.035005","source":"openalex"},{"id":"doi:10.15151/esrf-es-2419625002","type":"article-journal","title":"Strain in Reconfigurable Transistors for SiGe based neuromorphic computing","abstract":"We propose to apply Scanning X-ray Diffraction Microscopy (SXDM) at beamline ID01 to acquire nanoscopic maps of the full strain tensor in reconfigurable transistors (RFETs) based on Ge/Si nanosheet heterostructures on silicon on insulator (SOI) substrates. The RFETs contain Schottky barriers formed through an Al-Ge/Si exchange mechanism. Evaluating spatial material compositions and strain fields that are influenced by the device fabrication process is crucial for understanding and modelling the Schottky barriers, as well as improving the carrier mobility in the channel. Thus, we will measure the spatial strain distribution and composition in the top-down-defined Ge/Si nanowires at different stages of device fabrication and evaluate the crystallographic relation between the Al contacts and the Si/Ge layers.","author":[{"family":"Aberl","given":"Johannes"},{"family":"Brehm","given":"Moritz"},{"family":"Capellini","given":"Giovanni"},{"family":"Knaller","given":"Nikolas"},{"family":"Nazzari","given":"Daniele"},{"family":"Prado Navarrete","given":"Enrique"},{"family":"Sistani","given":"Masiar"},{"family":"Weber","given":"Walter"}],"DOI":"10.15151/esrf-es-2419625002","URL":"https://doi.org/10.15151/esrf-es-2419625002","source":"datacite"},{"id":"doi:10.26083/tuprints-00026641","type":"article-journal","title":"Neuromorphic Perception using Time-of-Flight-based Encoding of Lidar Data : A Potential and Feasibility Study","abstract":"This master thesis is dedicated to the research of neuromorphic perception in the domain of automated driving systems. In view of the rapid progress in the automotive industry and the increasing realization of automated driving, questions regarding the energy efficiency and performance of such systems are coming into focus. This work discusses the possibility of processing lidar data using spiking neural networks to reduce energy consumption in object detection while increasing perceptual capabilities. Automated vehicles not only promise a radical change in the transport sector, but also numerous benefits beyond pure mobility. These include reducing road accidents, reducing congestion, improving access to mobility and saving time in traffic. However, the implementation of automated driving systems faces technical and legal challenges, including the allocation of liability in the event of an accident and the need for intensive research into technical implementation. A key challenge is the perception of the vehicle, which is essential in order to maneuver safely and react to changing traffic conditions. Automated vehicles rely on various sensor systems and often use artificial intelligence methods to recognize objects and perform other tasks for Automated driving. However, these methods require considerable computing power and consume a lot of energy, which is particularly challenging in the context of electromobility, where energy is often limited by battery capacity. This thesis investigates a promising solution to this problem by integrating neuromorphic technology into the perception system. This technology is inspired by the neurobiological information processing of the human brain and aims to develop hardware and software that are similar in function and structure to biological neurons. Because the brain uses discrete impulses to process information, it consumes very little energy compared to conventional computers for comparable tasks. Energy is only consumed when these pulses occur and information is processed. This work focuses on the use of lidar sensors, which emit infrared light and measure the reflected laser beams to determine the distance to objects in the environment. An important parameter is the time of flight of the laser beams. The ToF can be used to convert the lidar data into discrete pulses that the neuromorphic perception system communicates with. The integration of neuromorphic perception systems into automated vehicles has the potential to significantly increase energy efficiency and improve the ability to perceive the environment. In order to analyze the potential and feasibility of a neuromorphic perception system using the ToF of lidar data, a morphological analysis is performed. This analysis takes into account different approaches and aspects that have been elaborated in an extensive literature review. The perception pipeline is divided into subsystems and subfunctions, leading to two concepts that fulfill different requirements for a neuromorphic perception system. These two concepts serve as a starting point for further research and development of this promising technology. The thesis is divided into several chapters that provide a comprehensive overview of the topic of neuromorphic perception systems for automated driving. It covers the basics of automated driving systems, artificial neural networks and spiking neural networks. Lidar sensors, different coding schemes, point cloud object detection, spiking neuron models, SNN hardware implementations and existing approaches are also discussed. The results of the literature review are evaluated based on a morphological analysis. A baseline and an advanced concept of a neuromorphic perception system are presented. In conclusion, this thesis shows the enormous potential of neuromorphic perception systems for automated driving and provides an outlook on possible future developments and research directions. Reducing energy consumption is crucial to ach","author":[{"family":"Schulte","given":"Jonas"}],"issued":{"date-parts":[[2024]]},"DOI":"10.26083/tuprints-00026641","URL":"https://doi.org/10.26083/tuprints-00026641","source":"datacite"},{"id":"doi:10.26083/tuprints-00024217","type":"article-journal","title":"Substoichiometric Phases of Hafnium Oxide with Semiconducting Properties","abstract":"Since the dawn of the information age, all developments that provided a significant improvement in information processing and data transmission have been considered as key technologies. The impact of ever new data processing innovations on the economy and almost all areas of our daily lives is unprecedented and a departure from this trend is unimaginable in the near future. Even though the end of Moore's Law has been predicted all too often, the steady exponential growth of computing capacity remains unaffected to this day, due to tremendous commercial pressure. While the minimum physical size of the transistor architecture is a serious constraint, the steady evolution of computing effectiveness is not limited in the predictable future. However, the focus of development will have to expand more strongly to other technological aspects of information processing. For example, the development of new computer paradigms which mark a departure from the digitally dominated van Neumann architecture will play an increasingly significant role. The category of so-called next-generation non-volatile memory technologies, based on various physical principles such as phase transformation, magnetic or ferroelectric properties or ion diffusion, could play a central role here. These memory technologies promise in part strongly pronounced multi-bit properties up to quasi-analog switching behavior. These attributes are of fundamental importance especially for new promising concepts of information processing like in-memory computing and neuromorphic processing. In addition, many next-generation non-volatile memory technologies already show advantages over conventional media such as Flash memory. For example, their application promises significantly reduced energy consumption and their write and especially read speeds are in some cases far superior to conventional technology and could therefore already contribute significant technological improvements to the existing memory hierarchy. However, these alternative concepts are currently still limited in terms of their statistical reliability, among other things. Even though phase change memory in the form of the 3D XPoint, for example, has already been commercialized, the developments have not yet been able to compete due to the enormous commercial pressure in Flash memory research. Nevertheless, the further development of alternative concepts for the next and beyond memory generations is essential and the in-depth research on next-generation non-volatile memory technologies is therefore a hot and extremely important scientific topic. This work focuses on hafnium oxide, a key material in next-generation non-volatile memory research. Hafnium oxide is very well known in the semiconductor industry, as it generated a lot of attention in the course of high-k research due to its excellent dielectric properties and established CMOS compatibility. However, since the growing interest in so-called memristive memory, research efforts have primarily focused on the value of hafnium oxide in the form of resistive random-access memory (RRAM) and, with the discovery of ferroelectricity in HfO₂, ferroelectric resistive random-access memory (FeRAM). RRAM is a next-generation non-volatile memory technology that features a simple metal-insulator-metal (MIM) structure, excellent scalability, and potential 3D integration. In particular, the aforementioned gradual to quasi-continuous switching behavior has been demonstrated on a variety of RRAM systems. A significant change of the switching properties is achievable, for example, by the choice of top and bottom electrodes, the introduction of doping elements, or by designated oxygen deficiency. In particular, the last point is based on the basic physical principle of the hafnium oxide-based RRAM mechanism, in which local oxygen ions are stimulated to diffuse by applying an electrical potential, and a so-called conducting filament is formed by the remaining vacancies, which el","author":[{"family":"Kaiser","given":"Nico"}],"issued":{"date-parts":[[2023]]},"DOI":"10.26083/tuprints-00024217","URL":"https://doi.org/10.26083/tuprints-00024217","source":"datacite"},{"id":"doi:10.6084/m9.figshare.26577538","type":"article-journal","title":"A Fully Configurable Open-Source Software-Defined Digital Quantized Spiking Neural Core Architecture","abstract":"Neuromorphic computing systems, which mimic biological neurons and synapses, can implement machine-learning (ML) tasks in power-efficient fashion. Major challenges for neuromorphic computing lie in its adoption by users and from a system developer's perspective, to cope with faster time-to-market pressure for new chip designs. Our project is developing a software infrastructure called NeuroXplorer which helps both end-users as well as developers of neuromorphic systems. It allows for ML tasks to be efficiently mapped onto neuromorphic architectures in the form of Spiking Neural Networks (SNNs). As part of this effort, we will introduce the QUANTIzed Spiking-Enabled Neural Core (QUANTISENC), a fully configurable layer-based design to improve the performance of neuromorphic systems. The parameterized Verilog-based design is compatible with ASIC and FPGA. QUANTISENC’s software-defined hardware design methodology allows the user to train an SNN model using Python and evaluate performance of its hardware implementation, such as area, power, latency, and throughput. QUANTISENC is made available at https://github.com/drexel-DISCO/quantisenc-public.git under the MIT license to allow academia and industry to access the design without restriction.","author":[{"family":"Kandasamy","given":"Nagarajan"},{"family":"Das","given":"Anup"}],"issued":{"date-parts":[[2024]]},"DOI":"10.6084/m9.figshare.26577538","URL":"https://doi.org/10.6084/m9.figshare.26577538","source":"datacite"},{"id":"doi:10.6084/m9.figshare.24221317","type":"article-journal","title":"Elements: Software infrastructure for programming and architectural exploration of neuromorphic computing systems","abstract":"Neuromorphic computing systems, which mimic biological neurons and synapses, can implement machine-learning (ML) tasks in power-efficient fashion. Major challenges for neuromorphic computing lie in its adoption by users and from a system developer's perspective, to cope with faster time-to-market pressure for new chip designs. Our project is developing a software infrastructure called NeuroXplorer, which helps both end-users as well as developers of neuromorphic systems. It allows for ML tasks to be mapped onto neuromorphic architectures in the most efficient way possible; and provides analysis and synthesis tools to explore new chip designs to meet the needs of machine-learning workloads. NeuroXplorer supports code generation for neuromorphic chips from a high-level specification of the ML task; provides synthesis tools to map ML tasks on novel neuromorphic architectures built using FPGAs; and supports hardware/software design-space exploration of new architectures. NeuroXplorer is distributed under an open-source license to promote adoption of neuromorphic computing.","author":[{"family":"Kandasamy","given":"Nagarajan"},{"family":"Das","given":"Anup"}],"issued":{"date-parts":[[2023]]},"DOI":"10.6084/m9.figshare.24221317","URL":"https://doi.org/10.6084/m9.figshare.24221317","source":"datacite"},{"id":"doi:10.6084/m9.figshare.12807224","type":"article-journal","title":"How Spike based Neuromorphic Computing can help monitor Social Distancing Efficiently","abstract":"This poster was presented at IEEE Brain Sponsored Neuromatch 2.0 Conference. Brief Description: As countries decide to lift lockdowns (due to the recent Covid-19 pandemic), it leaves no doubt, that our lifestyles will not be the same anymore. The new normal for several months will be to maintain social distance in everything we do. For most places, social distancing is likely to be monitored using cameras mounted on drones. However, artificial neural networks powering these devices require substantial computational and energy costs, thus limiting their use in mobile applications. Limitations include possibility of only short-term deployment, requirement of connectivity to remote servers for inference which adds to latency, and inability to take decisions by itself. Though, recent advent of edge computing devices like the Jetson Nano solve these limitations to some extent, still energy efficiency remains a big issue. Spiking Neural Networks have shown great potential as a solution for realizing ultra-low power consumption using spike based neuromorphic hardware. However practical deployment of such devices may take time. In this work, we compare edge computing and neuromorphic computing approaches in terms of energy efficiency. In addition to this, we present a possible approach for monitoring of social distancing in public places.","author":[{"family":"Srivastava","given":"Varad"},{"family":"Singh","given":"Manoj"}],"issued":{"date-parts":[[2020]]},"DOI":"10.6084/m9.figshare.12807224","URL":"https://doi.org/10.6084/m9.figshare.12807224","source":"datacite"},{"id":"doi:10.25349/d9031z","type":"article-journal","title":"Extracellular Recordings from Human Brain Organoids Using High-density CMOS Arrays","abstract":"Human brain organoids replicate much of the cellular diversity and developmental anatomy of the human brain. However, the physiological behavior of neuronal circuits within organoids remains relatively under-explored. With high-density CMOS microelectrode arrays (26,400 electrodes) and shank electrodes (960 electrodes), we probed broadband and three-dimensional extracellular field recordings generated by spontaneous activity of human brain organoids. These recordings simultaneously captured local field potentials (LFPs) and single-unit activity extracted through spike sorting. From spiking activity, we estimated a directed functional connectivity graph of synchronous neural network activity, which showed a large number of weak functional connections enmeshed within a network skeleton of significantly fewer strong connections. Treatment of the organoid with a benzodiazepine induced a reproducible signature response that shortened the inter-burst intervals, increased the uniformity of the firing pattern within each burst and decreased the population of weakly connected edges. Simultaneously examining the spontaneous LFPs and their phase alignment to spiking showed that spike bursts were coherent with theta oscillations in the LFPs. Our results demonstrate that human brain organoids have self-organized neuronal assemblies of sufficient size, cellular orientation, and functional connectivity to co-activate and generate field potentials from their collective transmembrane currents that phase-lock to spiking activity. These results point to the potential of brain organoids for the study of neuropsychiatric diseases, drug mechanisms, and the effects of external stimuli upon neuronal networks.","author":[{"family":"Sharf","given":"Tal"}],"issued":{"date-parts":[[2022]]},"DOI":"10.25349/d9031z","URL":"https://doi.org/10.25349/d9031z","source":"datacite"},{"id":"doi:10.5075/epfl-thesis-9739","type":"article-journal","title":"Taming neuronal noise with large networks","abstract":"How does reliable computation emerge from networks of noisy neurons? While individual neurons are intrinsically noisy, the collective dynamics of populations of neurons taken as a whole can be almost deterministic, supporting the hypothesis that, in the brain, computation takes place at the level of neuronal populations. Mathematical models of networks of noisy spiking neurons allow us to study the effects of neuronal noise on the dynamics of large networks. Classical mean-field models, i.e., models where all neurons are identical and where each neuron receives the average spike activity of the other neurons, offer toy examples where neuronal noise is absorbed in large networks, that is, large networks behave like deterministic systems. In particular, the dynamics of these large networks can be described by deterministic neuronal population equations. In this thesis, I first generalize classical mean-field limit proofs to a broad class of spiking neuron models that can exhibit spike-frequency adaptation and short-term synaptic plasticity, in addition to refractoriness. The mean-field limit can be exactly described by a multidimensional partial differential equation; the long time behavior of which can be rigorously studied using deterministic methods. Then, we show that there is a conceptual link between mean-field models for networks of spiking neurons and latent variable models used for the analysis of multi-neuronal recordings. More specifically, we use a recently proposed finite-size neuronal population equation, which we first mathematically clarify, to design a tractable Expectation-Maximization-type algorithm capable of inferring the latent population activities of multi-population spiking neural networks from the spike activity of a few visible neurons only, illustrating the idea that latent variable models can be seen as partially observed mean-field models. In classical mean-field models, neurons in large networks behave like independent, identically distributed processes driven by the average population activity -- a deterministic quantity, by the law of large numbers. The fact the neurons are identically distributed processes implies a form of redundancy that has not been observed in the cortex and which seems biologically implausible. To show, numerically, that the redundancy present in classical mean-field models is unnecessary for neuronal noise absorption in large networks, I construct a disordered network model where networks of spiking neurons behave like deterministic rate networks, despite the absence of redundancy. This last result suggests that the concentration of measure phenomenon, which generalizes the ``law of large numbers'' of classical mean-field models, might be an instrumental principle for understanding the emergence of noise-robust population dynamics in large networks of noisy neurons.","author":[{"family":"Schmutz","given":"Valentin"}],"issued":{"date-parts":[[2022]]},"DOI":"10.5075/epfl-thesis-9739","URL":"https://doi.org/10.5075/epfl-thesis-9739","source":"datacite"},{"id":"doi:10.5075/epfl-thesis-7546","type":"article-journal","title":"Multi-memristive synaptic architectures for training neural networks","abstract":"Highly data-centric AI workloads require that new computing paradigms be adopted because the performance of traditional CPU- and GPU-based systems are limited by data access and transfer. Training deep neural networks with millions of tunable parameters takes days or even weeks using relatively powerful heterogeneous systems, and it consumes more than hundreds of kilowatts of power. In-memory computing with memristive devices is a very promising avenue to accelerate deep learning because computations take place within the memory itself, eliminating the need to move the data around. The synaptic weights can be represented by the analog conductance states of memristive devices organized in crossbar arrays. The computationally expensive operations associated with deep learning training can be performed in place by exploiting the physical attributes and state dynamics of memristive devices and circuit laws. Memristive cores are also of particular interest due to the non-volatility, scalability, CMOS-compatibility and fast access time of the constituent devices. In addition, the multi-level storage capability of certain memristive technologies is especially attractive for increasing the information storage capacity of such cores. Large-scale demonstrations that combine memristive synapses with digital or analog CMOS circuitry indicate the potential of in-memory computing to accelerate deep learning. However, these implementations are highly vulnerable to non-ideal memristive device behavior. In particular, the limited weight representation capability, intra-device and array-level variability and temporal variations of conductance states pose significant challenges to achieving training accuracies that are comparable to conventional von Neumann implementations. Design solutions that can address these non-idealities without introducing significant implementation complexities will be critical for future memristive systems. This thesis proposes a novel synaptic architecture that can overcome a multitude of the aforementioned device non-idealities. In particular, it investigates the use of multiple memristive devices as a single computational primitive to represent a neural network weight and examines experimentally how such a compute primitive can improve non-desired memristive behavior. We propose a novel technique to arbitrate between the constituent devices of the synapse that can easily be implemented in hardware and adds only minimal energy overhead. We explore the proposed concept over various networks such as conventional non-spiking and spiking neural networks. The efficacy of this synaptic architecture is demonstrated for different training approaches including fully memristive and mixed-precision in-memory training by means of experiments using more than 1 million phase-change memory devices. Furthermore, we show that the proposed concept can be a key enabler to exploit binary memristive devices for deep-learning training.","author":[{"family":"Boybat Kara","given":"Irem"}],"issued":{"date-parts":[[2020]]},"DOI":"10.5075/epfl-thesis-7546","URL":"https://doi.org/10.5075/epfl-thesis-7546","source":"datacite"},{"id":"doi:10.26092/elib/2182","type":"article-journal","title":"Spontaneous synchronization in recurrent neural networks: From mathematical analysis to flexible information processing in spiking networks","abstract":"The brain is a highly complex distributed system that not only continuously performs demanding computational tasks but also flexibly adapts its processing to changing requirements on short time scales. A prime example of this flexibility is provided by selective visual attention. Neurons in the visual cortex have been shown to respond to simultaneous presentation of multiple stimuli in their receptive field as if only the stimulus receiving attention was present. This constitutes an attention mediated change of the functional configuration of information processing resulting in selective routing of the behaviorally relevant signals. It is not fully understood which neural mechanisms enable flexible information processing in the brain. This thesis uses theoretical analysis, model building and experimental investigations to generate novel insights into the key aspects coordination, configuration and control of flexible information processing. On a fundamental level, neural activity consists of spatio-temporal spike patterns which can contain transient bursts of synchronous activity of an assembly of neurons in form of spike avalanches. Theoretical studies suggest that a critical dynamics exhibiting power law distributed avalanche sizes maximizes computational capabilities, but usually assume homogeneously coupled populations in the large network size limit. Neural populations in the brain are instead finite, highly structured and strongly driven. To help bridge this gap, we develop a theory of spike pattern formation for finite and arbitrarily coupled networks of spiking neurons resulting in closed form avalanche assembly distributions. This closed form contains the networks graph Laplacian and thus provides explicit insights into how the network structure coordinates spontaneously synchronized activity.","author":[{"family":"Schünemann","given":"Maik"}],"issued":{"date-parts":[[2023]]},"DOI":"10.26092/elib/2182","URL":"https://doi.org/10.26092/elib/2182","source":"datacite"},{"id":"doi:10.82419/234","type":"article-journal","title":"RespiroDynamics Unveiled: A Groundbreaking Multi-Modal Deep Learning and Spiking Neural Network Framework for Revolutionizing Non-Invasive Lung Health Assessment","abstract":"This thesis investigates non-invasive lung health assessment using deep learning and Spiking Neural Networks (SNNs) to analyze thermal and RGB video data. Traditional respiratory diagnostics often require direct physical interaction, which can cause patient discomfort. This research aims to develop a non-contact, robust, precise, and energy-efficient lung health model using thermal or mobile video recordings and personal data, eliminating the need for traditional spirometry. The study collected a unique dataset from 60 male participants of various demographics and health backgrounds, including thermal and RGB videos, heart rate, ECG data, and detailed metadata. The methodology involved creating and testing various neural network models, including Convolutional Neural Networks (CNNs) for classification and regression tasks, and SNNs that process temporal respiratory patterns. Innovations include data augmentation, the Adaptive Precision-Tuned Regression (APTR) loss function, multimodal data integration, attention mechanisms, and ensemble learning to enhance model performance. Results revealed high efficacy in both classification and regression tasks. In the FVC Normal vs. Abnormal classification, the thermal model achieved a perfect score of 100%, and the RGB model scored 99.7%. In the Peak Expiratory Flow (PEF) classification, the thermal model outperformed with 97.14% accuracy compared to 96% for RGB. SNNs showed an accuracy improvement from 91.99% to 99.5% after data aggregation for thermal videos, and from 81.17% to 99% for RGB. In regression tasks, ensemble learning significantly boosted performance; the thermal model reported a Relative Root Mean Square Error of 0.11, a Relative Mean Absolute Error of 0.09, and a Pearson Correlation of 0.93. Comparatively, the RGB model showed poorer performance with respective values of 0.26, 0.21, and 0.79. These findings highlight the superior performance of thermal imaging over RGB in detecting respiratory patterns and the beneficial impact of integrating metadata into the models, setting new standards in the field.","author":[{"family":"Sharshar","given":"Ahmed"}],"issued":{"date-parts":[[2024]]},"DOI":"10.82419/234","URL":"https://doi.org/10.82419/234","source":"datacite"},{"id":"oa:W4283332145","type":"article-journal","title":"Heusler alloys for metal spintronics","abstract":"Abstract Heusler alloys have been theoretically predicted and experimentally demonstrated to be an ideal spin source due to their half-metallicity at room temperature. The half-metallicity also offers low Gilbert damping constants for fast magnetization reversal with low switching current density. These intrinsic properties can offer better operationability in spin-transfer-torque-based devices. In addition spin–orbit torque can be exerted using Heusler alloys for spin Hall and caloritronic effects. These properties can be precisely controlled by substituting the constituent elements in a Heusler alloy. We review the recent development on these spintronic devices and summarize their future perspectives. Graphical abstract","author":[{"family":"Hirohata","given":"Atsufumi"},{"family":"Lloyd","given":"David"}],"issued":{"date-parts":[[2022]]},"DOI":"10.1557/s43577-022-00350-1","URL":"https://doi.org/10.1557/s43577-022-00350-1","source":"openalex"},{"id":"doi:10.1063/5.0257074","type":"article-journal","title":"Inversion-sensing SiO2-based MOS capacitive synapse for neuromorphic computing","abstract":"In this work, an inversion-sensing SiO2-based capacitive synapse device was introduced by using the lateral coupling effect in a concentric metal–oxide–semiconductor structure. The device achieved a CHCS/CLCS ratio of 24 with a low programming voltage of VPGM = −2.5 V. Technology Computer Aided Design (TCAD) simulations confirmed the device's high sensitivity to changes in external charges. For oxide with an effective positive charge density (Neff) exceeding 2.8×1011 cm−2, a small variation of 5×109 cm−2 could influence a lot in the capacitance value of the device in the inversion region. This sensitivity enabled multi-state capacitance modulation by adjusting the number of pulses and operating voltages. Additionally, the scalability of the device was validated through simulations. The on/off ratio could be further improved by substituting the gate dielectric material. Overall, the lateral coupling effect not only enhances the performance of charge-trapping-based devices but also provides a viable strategy for expanding memory windows across various types of capacitive memory technologies.","author":[{"family":"Kao","given":"Chi"},{"family":"Hwu","given":"Jenn"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1063/5.0257074","URL":"https://doi.org/10.1063/5.0257074","source":"crossref"},{"id":"doi:10.53941/ldm.2026.100002","type":"article-journal","title":"Spatio-Temporal Confinement in Two-Dimensional Channels for Neuromorphic Computing","abstract":"Graphene oxide (GO), with its unique two-dimensional structure, adjustable functional groups, and tunable nanofluidic channels, has emerged as a promising platform for bio-inspired neuromorphic computing. This perspective explores the structural and functional analogies between GO membranes and biological ion channels, emphasizing GO’s ability to support selective ion transport, stimuli-responsive behavior, and synaptic plasticity. Recent advances in material engineering and device integration have enabled GO-based artificial synapses, including memristors and ion-gated transistors, to emulate key neuronal features such as excitatory postsynaptic currents, paired-pulse facilitation, and spike-timing-dependent plasticity with sub-millisecond response times and picojoule-level energy consumption. Moreover, the incorporation of GO with polymers, quantum dots, and semiconductors has facilitated multimodal control via electric, optical, and chemical inputs. Together, these developments position GO as a powerful material system for future neuromorphic devices that operate in aqueous and dynamic biological environments, paving the way toward brain-inspired hardware, neuroprosthetics, and intelligent biointerfaces.","author":[{"family":"Zhu","given":"Hongwei"},{"family":"Wang","given":"Honglin"}],"issued":{"date-parts":[[2026]]},"DOI":"10.53941/ldm.2026.100002","URL":"https://doi.org/10.53941/ldm.2026.100002","source":"crossref"},{"id":"doi:10.3390/photonics13050431","type":"article-journal","title":"Memristors for the Post-Von Neumann Era: Hardware Paradigms, Neuromorphic Perception, and Computing Systems","abstract":"Memristors, as transformative electronic devices designed to transcend the von Neumann architecture, enable the physical unification of information storage and computation, thereby offering a foundational hardware pathway toward energy-efficient, brain-inspired computing. Their intrinsic analog resistive switching, non-volatility, and history-dependent learning capabilities allow them to natively implement in-memory computing and emulate synaptic plasticity, addressing the critical bottlenecks of energy and speed in conventional systems. Notably, the evolution from electrically controlled memristors to optoelectronic memristors marks a paradigm shift from pure computing to integrated sensing-processing, opening new dimensions for high-speed, parallel, and adaptive signal processing. In recent years, significant progress has been made in the development of memristor-based neuromorphic vision and tactile systems, on-chip signal processors, and dynamic trajectory trackers, demonstrating their potential in edge intelligence, adaptive robotics, and real-time perceptual tasks. This review systematically summarizes the latest advances in memristor technology, providing a comprehensive analysis of their operating mechanisms, material and structural innovations, and cutting-edge applications in neuromorphic perception and computing. Furthermore, it discusses the key challenges and future directions for the development and integration of memristor-based systems in the post-von Neumann era.","author":[{"family":"Fu","given":"Kerui"},{"family":"Qin","given":"Tianling"}],"issued":{"date-parts":[[2026]]},"DOI":"10.3390/photonics13050431","URL":"https://doi.org/10.3390/photonics13050431","source":"crossref"},{"id":"doi:10.20944/preprints202511.1462.v1","type":"manuscript","title":"The Neuromorphic Conductor: A Speculative Framework for Brain-Chip Interfaces to Restore Bodily Function","abstract":"This report is a deep dive into the complex world of using neuromorphic chips to help people with severe brain damage regain control of their bodies. We’ll look at the fundamental science behind neuromorphic computing, explore the current landscape of brain-computer interfaces (BCIs), and confront the biological and ethical challenges of making such a technology a reality. The main idea is to create a kind of “digital nervous system” that could bypass damaged parts of the brain to restore basic functions like movement and breathing. This isn't just a technical paper; it’s a detailed exploration of the immense hurdles and profound questions that must be answered before we can truly build a bridge between mind and machine. This document is a starting point for anyone looking to understand this fascinating and difficult field.","author":[{"family":"Rawat","given":"Anand"},{"family":"Yadav","given":"Anamika"}],"issued":{"date-parts":[[2025]]},"DOI":"10.20944/preprints202511.1462.v1","URL":"https://doi.org/10.20944/preprints202511.1462.v1","source":"europepmc"},{"id":"doi:10.31219/osf.io/gtv6q","type":"article-journal","title":"Extremely high bandwidth optical neuromorphic processing, microwave photonics and data transmission with Kerr microcombs","abstract":"We report ultrahigh bandwidth applications of Kerr microcombs to optical neural networks and to optical data transmission, at data rates from 44 Terabits/s (Tb/s) to approaching 100 Tb/s. Convolutional neural networks (CNNs) are a powerful category of artificial neural networks that can extract the hierarchical features of raw data to greatly reduce the network complexity and enhance the accuracy for machine learning tasks such as computer vision, speech recognition, playing board games and medical diagnosis [1-7]. Optical neural networks can dramatically accelerate the computing speed to overcome the inherent bandwidth bottleneck of electronics. We use a new and powerful class of micro-comb called soliton crystals that exhibit robust operation and stable generation as well as a high intrinsic efficiency with an extremely low spacing of 48.9 GHz. We demonstrate a universal optical vector convolutional accelerator operating at 11 Tera-OPS/s (TOPS) on 250,000 pixel images for 10 kernels simultaneously — enough for facial image recognition. We use the same hardware to sequentially form a deep optical CNN with ten output neurons, achieving successful recognition of full 10 digits with 900 pixel handwritten digit images. We also report world record high data transmission over standard optical fiber from a single optical source, at 44.2 Terabits/s over the C-band, with a spectral efficiency of 10.4 bits/s/Hz, with a coherent data modulation format of 64 QAM. We achieve error free transmission across 75 km of standard optical fiber in the lab and over a field trial with a metropolitan optical fiber network. Our work demonstrates the ability of optical soliton crystal micro-combs to exceed other approaches in performance for the most demanding practical optical communications applications.","author":[{"family":"Moss","given":"David"}],"issued":{"date-parts":[[2025]]},"DOI":"10.31219/osf.io/gtv6q","URL":"https://doi.org/10.31219/osf.io/gtv6q","source":"crossref"},{"id":"doi:10.1142/s0217979226300112","type":"article-journal","title":"Emerging Quantum Materials for Neuromorphic Computing: From Fundamental Physics to Device Architectures","abstract":"The explosive growth of artificial intelligence (AI), edge computing, and brain-inspired algorithms has spurred the development of neuromorphic systems that mimic biological information processing. Achieving such functionality at the hardware level requires materials that exhibit neuron- and synapse-like behaviors in a scalable, low-power, and CMOS-compatible manner. In this review, we provide a comprehensive assessment of emerging quantum materials including Mott insulators, phase change materials (PCMs), topological insulators (TIs), twodimensional (2D) materials, and ferroelectrics highlighting their unique physical mechanisms and their relevance to neuromorphic device operation. Each material class is examined in terms of its electronic properties, switching dynamics, and compatibility with spiking neural networks (SNNs) and in-memory computing architectures. We compare their performance across key metrics such as energy efficiency, analog programmability, synaptic plasticity, and integration scalability. Furthermore, we engage in a comprehensive discussion regarding device prototypes that are predicated upon quantum materials, delineate the contemporary challenges associated with integration, and provide insights into hardware–algorithm co-design methodologies. This review additionally recognizes nascent trends such as hybrid material heterostructures, quantumclassical neuromorphic frameworks, and bioinspired learning paradigms that leverage intrinsic material dynamics. Our examination underscores that the amalgamation of the distinctive functionalities inherent in quantum materials with neuromorphic hardware presents a promising trajectory towards mitigating the constraints imposed by traditional computing paradigms. This synthesis facilitates the development of quantum neuromorphic platforms capable of real-time learning, operating with minimal energy consumption, and possessing adaptive learning architectures that dynamically adjust in accordance with cognitive requirements. Such platforms are poised to usher in the subsequent generation of computing systems that can rival or potentially surpass the performance of conventional von Neumann architectures.","author":[{"family":"Alharbi","given":"Abdullah"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1142/s0217979226300112","URL":"https://doi.org/10.1142/s0217979226300112","source":"crossref"},{"id":"doi:10.1088/2634-4386/ae84f7","type":"article-journal","title":"TEMPO: a stochastic benchmarking protocol for evaluating temporal robustness in spiking neural networks","abstract":"Abstract Spiking neural networks are set apart from conventional architectures by their leaky-integrator dynamics and temporal memory, yet the benchmarks most often used to evaluate them do not exercise these properties. In datasets such as N-MNIST and DVS-Gesture the discriminative information lies in spatial features, so collapsing all spike timing into a single bin leaves classification accuracy largely intact. Morse code is a natural antidote, since its meaning resides entirely in the duration of marks and the silences between them. Existing Morse benchmarks, however, assume deterministic machine-perfect timing and ignore the rhythmic variability of real human operators. We introduce the Temporal Encoding Morse Protocol for Operators, a benchmark protocol for 26-letter Morse classification that places human operator variability at the center of its design. Each Morse element produces a single spike at the moment its mark completes, encoded on one of two channels at a 1 ms time base compatible with ITU standards and neuromorphic hardware. Three stochastic parameters, the dash-to-dot ratio, word-level speed, and per-element jitter, model an operator’s characteristic ‘fist,’ and their distributions are fitted to measurements from 44 real recordings spanning diverse speeds and keying styles. A proof-of-concept multi-timescale network reaches 75.7% accuracy. Randomizing spike order while preserving counts collapses accuracy to a count-based baseline far below the intact result, confirming that the task cannot be solved without timing. A network trained only on clean signals still performs well above chance under severe noise, revealing an implicit temporal robustness that spatial benchmarks cannot expose. The benchmark and its generation code are openly available, with 75.7% standing as an open baseline for future architectures.","author":[{"family":"Hindman","given":"Lucas"},{"family":"Cantley","given":"Kurtis"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/2634-4386/ae84f7","URL":"https://doi.org/10.1088/2634-4386/ae84f7","source":"crossref"},{"id":"doi:10.36227/techrxiv.174742729.99297491/v1","type":"article-journal","title":"The Neuromorphic Cyber-Twin: A Conceptual Architecture for Cognitive Defense in Digital Twin Ecosystems","abstract":"As cyber-physical systems become increasingly virtualized, Digital Twins are emerging as critical components for real-time system monitoring, simulation, and control. However, their growing complexity and exposure to dynamic network environments render them susceptible to sophisticated cyber threats. Traditional cybersecurity models, often based on rules or based on machine learning, struggle to adapt in real time to evolving attack patterns, especially within decentralized and resourceconstrained settings. In this conceptual paper, we introduce the Neuromorphic Cyber-Twin (NCT), a brain-inspired architectural framework that leverages spiking neural networks (SNNs) and event-driven cognition to endow digital twins with adaptive, self-evolving cyber defense capabilities. The NCT framework is grounded in neuromorphic principles such as sparse coding, temporal encoding, and synaptic plasticity, e.g., Spike-Timing-Dependent Plasticity (STDP), enabling it to process telemetry data from the digital twin layer as spike-based sensory input. We present a layered architecture wherein the cognitive layer continuously monitors behavioral deviations, performs anomaly inference, and autonomously adapts its defense responses in alignment with system dynamics. This biologically inspired paradigm offers low-latency detection, contextual awareness, and energy efficiency, essential for scalable security in virtualized ecosystems. We discuss theoretical underpinnings, architectural components, and application scenarios across smart infrastructure, autonomous transport, and industrial control systems. The paper concludes with key research challenges and a roadmap for future implementation of neuromorphic cybersecurity within digital twin environments.","author":[{"family":"Nasir","given":"Nida"},{"family":"Hamadi","given":"Hussam"}],"issued":{"date-parts":[[2025]]},"DOI":"10.36227/techrxiv.174742729.99297491/v1","URL":"https://doi.org/10.36227/techrxiv.174742729.99297491/v1","source":"crossref"},{"id":"doi:10.62311/nesx/rp3jy-30072026","type":"article-journal","title":"Topological Photonics for Neuromorphic Computing, Quantum Communications and Ultralow-Power Edge Intelligence","abstract":"Abstract: Topological photonics offers a route to guide, process and generate optical states through band structures whose global invariants can suppress selected forms of backscattering and disorder sensitivity. Neuromorphic photonics, quantum communications and edge intelligence, however, impose requirements that exceed protected transport alone: nonlinear computation, trainability, quantum-state fidelity, nanosecond-scale latency, low conversion overhead, fabrication tolerance and accountable system governance must be addressed jointly. This paper develops a model-based interdisciplinary framework for integrating these requirements. It introduces the Topological Photonic Neuromorphic–Quantum Edge Intelligence (TPNQ-EI) Framework, which couples Maxwell eigenproblems and Berry-topological invariants with tight-binding matrices, nonlinear photonic-neuron dynamics, quantum-channel fidelity, energy-per-inference accounting and constrained multi-objective optimization. The methodology combines critical literature synthesis, formal systems modelling, variable architecture, scenario comparison and an explicitly illustrative normalized numerical model. The numerical demonstration produces a gross capability score of 8.05/10 and a risk-adjusted score of 7.65/10 after accounting for optical loss, thermal drift, fabrication disorder and quantum decoherence. Analytical results indicate that topological design can improve resilience when disorder and routing defects are material, but cannot compensate automatically for absorption, out-of-plane scattering, calibration energy, detector noise or poorly matched multiphoton spectra. The paper contributes a unified vocabulary, mathematical architecture, validation protocol and governance perspective for evaluating topological photonic platforms across neuromorphic inference, quantum-network interfaces and ultralow-power edge applications. The proposed results are conceptual and illustrative; empirical claims require device-level measurement, cross-platform benchmarking and lifecycle assessment. Keywords Topological photonics; photonic neural networks; neuromorphic computing; quantum communications; topological quantum optics; edge intelligence; Chern number; Berry curvature; non-Hermitian photonics; photonic integrated circuits; ultralow-power inference; quantum-channel fidelity; disorder robustness; physics-aware training; multi-objective photonic optimization.","author":[{"family":"Pasupuleti","given":"Murali"}],"issued":{"date-parts":[[2026]]},"DOI":"10.62311/nesx/rp3jy-30072026","URL":"https://doi.org/10.62311/nesx/rp3jy-30072026","source":"crossref"},{"id":"doi:10.1088/2634-4386/ae6728","type":"article-journal","title":"An energy-efficient spiking neural network with continuous learning for self-adaptive brain–machine interface","abstract":"Abstract The number of simultaneously recorded neurons follows an exponentially increasing trend in implantable brain–machine interfaces (iBMIs). Integrating the neural decoder in the implant is an effective data compression method for future wireless iBMIs. However, the non-stationarity of the system makes the performance of the decoder unreliable. To avoid frequent retraining of the decoder and to ensure the safety and comfort of the iBMI user, continuous learning is essential for real-life applications. Since deep spiking neural networks (DSNNs) are being recognized as a promising approach for developing a resource-efficient neural decoder, we propose continuous learning approaches with reinforcement learning (RL) algorithms adapted for DSNNs. Banditron and AGREL are chosen as the two candidate RL algorithms since they can be trained with limited computational resources, effectively addressing the non-stationary problem and fitting the energy constraints of implantable devices. To assess the effectiveness of the proposed methods, we conducted both open-loop and closed-loop experiments. The accuracy of open-loop experiments conducted with DSNN_Banditron and DSNN_AGREL remains stable over extended periods. Meanwhile, the time-to-target in the closed-loop experiment with perturbations, DSNN_Banditron performed comparably to that of DSNN_AGREL while achieving reductions of 98% in memory access usage and 99% in the requirements for multiply-and-accumulate operations during training. Compared to previous continuous learning SNN decoders, DSNN_Banditron requires 98% less computes making it a prime candidate for future wireless iBMI systems.","author":[{"family":"Biyan","given":"Zhou"},{"family":"Basu","given":"Arindam"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/2634-4386/ae6728","URL":"https://doi.org/10.1088/2634-4386/ae6728","source":"crossref"},{"id":"doi:10.1002/admt.202500588","type":"article-journal","title":"Recent Progress in Memristor Array‐Based Neuromorphic Computing for on‐Chip Vector‐Matrix Multiplication","abstract":"Abstract Recently, the global issue of energy consumption has become a critical concern for the development of artificial intelligence (AI), which demands extensive computational resources and large‐scale data processing as network architectures and system algorithms grow increasingly complex. The conventional von Neumann digital computing architecture faces inherent limitations in handling the continuously growing big data, primarily due to its sequential data processing nature in vector–matrix multiplication (VMM) and the bottlenecks between processor and memory units. To address this challenge, brain‐inspired neuromorphic computing has emerged, emulating the human nervous system, particularly through memristor crossbar array architectures. These arrays function as cumulative operation units, enabling efficient parallel data processing. This paper discusses recent progress in hardware implementations of neuromorphic computing using memristor array devices, with a focus on circuit integration and on‐chip applications of AI algorithms rather than synaptic plasticity or unit‐cell switching mechanisms. The principle of parallel VMM operations is briefly reviewed, followed by hardware‐based VMM demonstrations, including convolutional transformations, neural network perceptrons, learning rule optimization, and on‐chip operations. The review also provides perspectives on future research directions, highlighting key challenges. Thus, memristor array‐based neuromorphic computing holds significant promise for scalable, energy‐efficient, and application‐ready AI hardware.","author":[{"family":"Jang","given":"Jingon"},{"family":"Gi","given":"Sang‐gyun"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1002/admt.202500588","URL":"https://doi.org/10.1002/admt.202500588","source":"crossref"},{"id":"doi:10.1149/ma2026-01341564mtgabs","type":"article-journal","title":"(\n                    <i>Invited</i>\n                    ) Ion–Electron Interactions In Ionically Gated 2D Transistors for Neuromorphic and Energy-Efficient Computing","abstract":"Two-dimensional (2D) materials offer a compelling platform for next-generation electronic and iontronic devices due to their nanoscale thickness, high carrier mobility, and strong coupling to electrochemical environments. When interfaced with solid ion conductors, these crystals can exhibit both electrostatic electric double layer (EDL) gating and electrochemical ion intercalation, providing complementary modes of modulation that are useful for low-power and neuromorphic device operation [1]. Yet, despite significant progress, the boundary between purely electrostatic operation and Faradaic ion insertion remains difficult to control, and understanding this interplay is essential for achieving stable, low-power, and high-performance devices [2]. In this talk, I will first discuss the unique opportunities and challenges of using 2D materials as channels in electrochemical RAM (ECRAM) devices, where their nanoscale thickness and strong coupling to solid ion conductors create both advantages and constraints for stable analog operation. I will then share recent efforts in our group to improve the stability and reversibility of ion intercalation in 2D channels through materials engineering. Because 2D-based ECRAM offers analog programmability suitable for on-chip training of artificial neural networks [3], we also examine how key device parameters such as on/off ratio, accessible conductance states, and weight-update linearity jointly determine neural-network performance. Through combined device measurements and simulation, we quantify trade-offs between the number of programmable states and weight-update linearity. Using a predictive ionic-weight-update model informed by measured device behavior [4], we demonstrate that linear and symmetric conductance modulation can be achieved while preserving multilevel operation, which in turn improves ANN accuracy and accelerates the training process. Finally, I will outline emerging directions in ionically controlled 2D devices for real-time and edge computing, where understanding and tailoring ion transport at electrolyte–2D crystal interfaces remain a central scientific challenge. By integrating insights from both the electronic device and electrochemical communities, this work aims to help establish design principles for stable, energy-efficient, and programmable iontronic systems based on 2D materials. [1] Xu, K.; Fullerton-Shirey, S. K. J. Phys. Mater. 2020 , 3 (3), 032001. [2] Xu, K.; Fullerton-Shirey, S. K. 2D Mater. 2025 , 12 (2), 023003. [3] Han, H.; Inman, J.; Maczynski, S.; Indovina, M.; Ganguly, A.; Das, T.; Xu, K. Proc. IEEE Microelectron. Des. Test Symp. (MDTS) 2025 , 1–6. [4] Manimaran, N. H.; Sutton, C. L. M.; Streamer, J. W.; Merkel, C.; Xu, K. J. Phys. Mater. 2024 , 8 (1), 015008.","author":[{"family":"Xu","given":"Ke"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1149/ma2026-01341564mtgabs","URL":"https://doi.org/10.1149/ma2026-01341564mtgabs","source":"crossref"},{"id":"doi:10.1088/2634-4386/ae5fc6","type":"article-journal","title":"Unsupervised feature learning in spiking neural networks using nonlinear interface dipole modulation-based synaptic devices","abstract":"Abstract Recently, a three-terminal interface dipole modulation field-effect transistor (IDM FET) memory device has been proposed that leverages electric-field-induced dipole modulation at oxide/oxide interfaces. This device has been reported to exhibit a double-pulse-induced response analogous to the spike-timing-dependent plasticity (STDP) observed in biological synapses. Although the STDP behavior of the IDM FET exhibits pronounced nonlinearity, previous simulation studies have suggested that it can still be applied to unsupervised feature learning in spiking neural networks (SNNs) when combined with an additional frequency-independent (FI) depression operation. In this study, we first briefly review the nonlinear IDM response based on experimental observations and clarify that the nonlinearity is intrinsic to the IDM interface, originating from changes in the interface dipole states. We then present the synaptic weight-update model of IDM FETs employed in our SNN simulations and analyze the weight-update dynamics during feature learning using a simple single-layer SNN. Based on this analysis, we examine the optimal update conditions in terms of the balance between potentiation and depression rates. Furthermore, we evaluate feature learning on the MNIST handwritten-digit dataset using a two-layer network. Based on frequency-dependent rate-equilibrium considerations, we propose a switching FI depression/potentiation algorithm to improve feature‐learning performance, demonstrating enhanced robustness, improved classification accuracy, and reasonable tolerance to device-to-device variation.","author":[{"family":"Miyata","given":"Noriyuki"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/2634-4386/ae5fc6","URL":"https://doi.org/10.1088/2634-4386/ae5fc6","source":"crossref"},{"id":"doi:10.1073/pnas.2528654122","type":"article-journal","title":"Can neuromorphic computing help reduce AI's high energy cost?","abstract":"Artificial intelligence systems, and large language models (LLMs) in particular, have become almost ubiquitous.The implications are many.Among them: serious concerns about massive amounts of energy usage.The initial training of GPT-3 required as much energy as powering 120 houses for a year.That's before adding the energy required for the chatbot to respond to users' prompts (1-3).GPT-4 required an estimated 50 times as much energy as its predecessor to train, and although OpenAI has not divulged official figures, GPT-5 likely required much more.To date, most efforts to improve the energy consumption of AI systems have focused on developing more efficient algorithms, increasing the use of energy from sustainable sources, or even building smaller language models.To build DeepSeek, which debuted last January, researchers in China built a model that only activates a fraction of the total model for each query.Other efforts focus on packing more transistors onto smaller chips, which results in a shorter distance for the data to travel and enables more parallel computing.But many researchers are developing a different kind of computing architecture, one inspired by the efficient mechanisms of the brain.Dubbed neuromorphic computing, it was first proposed decades ago but has recently seen a resurgence of Some researchers believe that a computing architecture inspired by the mechanisms of the brain could offer a path to innovation that helps tackle rampant AI energy consumption.","author":[{"family":"Ornes","given":"Stephen"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1073/pnas.2528654122","URL":"https://doi.org/10.1073/pnas.2528654122","source":"europepmc"},{"id":"doi:10.20944/preprints202512.2404.v1","type":"manuscript","title":"HyperFabric Interconnect (HFI): A Unified, Scalable Communication Fabric for HPC, AI, Quantum, and Neuromorphic Workloads","abstract":"The evolution of high-performance computing (HPC) interconnects has produced specialized fabrics such as InfiniBand[1], Intel Omni-Path, and NVIDIA NVLink[2], each optimized for distinct workloads. However, the increasing convergence of HPC, AI/ML, quantum, and neuromorphic computing requires a unified communication substrate capable of supporting diverse requirements including ultra-low latency, high bandwidth, collective operations, and adaptive routing. We present HyperFabric Interconnect (HFI), a novel design that combines the strengths of existing interconnects while addressing their scalability and workload-fragmentation limitations. Our evaluation on simulated clusters demonstrates HFI’s ability to reduce job completion time (JCT) by up to 30%, improve tail latency consistency by 45% under mixed loads and 4× better jitter control in latency-sensitive applications., and sustain efficient scaling across heterogeneous workloads. Beyond simulation, we provide an analytical model and deployment roadmap that highlight HFI’s role as a converged interconnect for the exascale and post-exascale era.","author":[{"family":"Bajpai","given":"Krishna"}],"issued":{"date-parts":[[2025]]},"DOI":"10.20944/preprints202512.2404.v1","URL":"https://doi.org/10.20944/preprints202512.2404.v1","source":"europepmc"},{"id":"doi:10.22541/au.175580538.80500974/v1","type":"article-journal","title":"Radiation-Aware Meta-Plasticity (RAMP): A Bio-Inspired Learning Rule for Neuromorphic Computing in Radiation-Prone Environments","abstract":"Neuromorphic computing systems, particularly those based on spiking neural networks (SNNs), provide exceptional energy efficiency and real-time processing, making them attractive for space, nuclear, and high-energy physics (HEP) applications. Their deployment, however, is challenged by ionizing radiation, which induces stochastic degradation, weight drift, and functional failure in CMOS-based synaptic components. To address this, we propose Radiation-Aware Meta-Plasticity (RAMP), a bio-inspired learning rule that dynamically modulates spike-timing-dependent plasticity (STDP) in response to neuronal activity and radiation exposure. RAMP introduces a dose- and time-dependent learning rate, integrating cumulative radiation dose, a membrane-voltage-based environmental regularizer, and explicit modeling of radiation-induced effects including noise, single-event transients, and total ionizing dose (TID) drift. MATLAB simulations of Leaky Integrate-and-Fire neurons under Poisson spike trains and mixed radiation profiles demonstrate that RAMP reduces synaptic weight variance by 32.8% compared to standard STDP while preserving Hebbian learning and convergence. Statistical analysis confirms that RAMP selectively suppresses noise during bursts without halting long-term adaptation. Furthermore, the effective learning rate decays exponentially with dose, mimicking biological synaptic downscaling under chronic stress. These results establish RAMP as a foundation for radiation-resilient neuromorphic intelligence, with direct implications for aerospace autonomy, nuclear safety, and HEP experiments.","author":[{"family":"Banteywalu","given":"Solomon"},{"family":"Leroux","given":"Paul"}],"issued":{"date-parts":[[2025]]},"DOI":"10.22541/au.175580538.80500974/v1","URL":"https://doi.org/10.22541/au.175580538.80500974/v1","source":"europepmc"},{"id":"doi:10.1101/2025.07.25.666748","type":"article-journal","title":"Neuromodulation enhances the capability and efficiency of spiking neural networks","abstract":"Abstract Spiking neurons underlie the brain’s extreme energy efficiency, and therefore have great potential in neuromorphic computing, although realising this efficiency in practice has proven challenging. We use neuromodulation, a biological mechanism that lets the network dynamically and contextually modify its own parameters. We find it substantially increases performance across a range of sensory processing tasks, including a challenging new speech-in-noise dataset we introduce, with very few additional resources (neurons, energy, parameters). Neuromodulatory networks are space and energy efficient thanks to two mechanisms that are directly relevant to neuromorphic computing and biology: firstly, they allow for an order-of-magnitude reduction in the number of neurons required; and secondly, they enable very sparse firing, achieving better results using orders of magnitude fewer spikes. Together, these properties may throw light on the computational role of neuromodulation in biology, and make neuromodulation an ideal mechanism to improve performance and efficiency for neuromorphic devices.","author":[{"family":"Alkilany","given":"Abdelqader"},{"family":"Goodman","given":"Dan"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1101/2025.07.25.666748","URL":"https://doi.org/10.1101/2025.07.25.666748","source":"preprints"},{"id":"doi:10.22541/au.175638505.58202572/v1","type":"article-journal","title":"Toward Trustworthy Neuromorphic AI: A Bayesian Framework for Uncertainty-Aware Spiking Neural Networks","abstract":"Spiking Neural Networks (SNNs) offer a biologically plausible and energy-efficient paradigm for processing temporal data, particularly suited for neuromorphic computing platforms. However, their deterministic nature limits their deployment in safety-critical applications where reliable uncertainty quantification is essential. This paper introduces a novel Bayesian Spiking Neural Network (BSNN) framework that integrates variational inference with surrogate gradient learning to enable robust uncertainty estimation while preserving the efficiency of SNNs. We formulate a scalable Bayesian framework using mean-field variational approximation over network weights and biases, enabling full predictive uncertainty quantification through Monte Carlo sampling. To address the non-differentiability of spiking dynamics, we employ a smooth surrogate gradient method based on sigmoidal derivatives during backpropagation through time. A tailored Poisson encoding scheme ensures rich temporal input representation, while a KL annealing strategy stabilizes training by gradually increasing the regularization pressure from prior distributions. Comprehensive experiments on a synthetic dataset demonstrate competitive classification accuracy (77.78%) alongside well-calibrated uncertainty estimates, as evidenced by strong calibration curves and high AUROC scores (&gt; 0.9). The model exhibits stable training dynamics, with controlled gradients and balanced loss components. Our approach bridges the gap between efficient neuromorphic computation and trustworthy AI, offering a pathway toward deployable, uncertainty-aware edge intelligence systems.","author":[{"family":"Banteywalu","given":"Solomon"},{"family":"Leroux","given":"Paul"}],"issued":{"date-parts":[[2025]]},"DOI":"10.22541/au.175638505.58202572/v1","URL":"https://doi.org/10.22541/au.175638505.58202572/v1","source":"europepmc"},{"id":"doi:10.1149/ma2025-01361728mtgabs","type":"article-journal","title":"Synaptic Capacitance Modulation in MOS Capacitors via Lateral Coupling Effect for Neuromorphic Computing","abstract":"Recently, capacitive memory devices, such as MFM capacitors [1], FeFETs [2], and SONOS capacitors [3], have been explored for capacitive crossbar arrays in compute-in-memory (CIM) applications. Compared to memristors, capacitors reduce static energy costs and provide a wide memory window (C HCS /C LCS ) that can be read at 0 V. This work proposed a simple MOS capacitor (MOSCap) structure (Al/SiO₂/Si/Al) utilizing the lateral coupling effect to achieve a maximum C HCS /C LCS ≥ 36 at a read voltage of 0.5 V, with low program/erase voltages (-2.5 V/+1 V). Figure 1(a) depicts the schematic of the device, with the SiO₂ thickness measured as 4.3 nm. Figure 1(b) shows the C-V curves under four operating frequencies. Frequency dispersion was evident in both the depletion and inversion regions. In the depletion region, this dispersion is attributed to possible interface states at the SiO₂/Si interface. In the inversion region, capacitance variations across different frequencies were observed. It is notice that, the frequency at 1 kHz is still too high for minority carriers to effectively response in silicon. However, it is observed the inversion capacitance at 1 kHz existed low frequency behavior. Consequently, the measured capacitance might include contributions from the outer region of the device, this phenomenon was known as the lateral coupling effect [4]. As illustrated in Figure 2(a), positive oxide charges existed in the SiO₂ layer deplete majority carriers, forming an n-type inversion layer and a conductive channel. The gate voltage controls the coupling extent between the MOSCap and the outer region. Trapped charges at the device edge could effectively modulate the inversion capacitance (C inv ). Figures 2(b) and (c) illustrate the mechanism: electrons trapped in SiO₂ after a negative stress, reducing C inv by disconnecting the device from the outer region, while positive stress removes these electrons, restoring C inv . TCAD simulations (Figure 3) confirmed this, showing significant reductions in C inv and the supply of lateral electron current density (J e-,x ) outside the device as oxide charge density (N eff ) decreased below 2.6×10 11 cm −2 . Figure 1(c) demonstrates variations of C-V curves after different programming times (t PRG ) at V PRG = -2.5 V, while Figure 1(d) shows retention characteristics of C inv at 0 V. Continuous stress cycles also modulated C inv , as shown in Figures 4(a)-(d). Longer t PRG amplified the reduction of C inv , while a opposite erase voltage (V ERS = +1 V) applied for a shorter duration (t ERS = 0.1 s) restored it, as shown in Figure 4(d). Figures 5(a)–5(c) show the variation of C inv after single potentiation/depression (continuous erase/program stress) cycle read at different voltages (V read ), with a maximum C HCS /C LCS ratio was 36 happened as V read = 0.5 V after 20 continuous erase/program stress cycles. Figure 5(d) demonstrate multi-cycle potentiation and depression at V read = 0.5 V. Stress voltage modulation also impacts C inv , as shown in Figures 6(a) and 6(b). A series of positive (amplitudes: 0.1 V to 1.2 V in 0.1 V steps; width: 0.1 s) and negative (amplitudes: -1.5 V to -2.7 V in 0.1 V steps; width: 5 s) pulses resulted in more linear C inv changes (Figure 6(c)) compared to stress cycle modulation. Figure 6(d) displays multi-cycle potentiation and depression characteristics at V read = 0 V. In conclusion, the proposed MOSCap structure exhibited synaptic modulation behavior through the lateral coupling effect, offering an energy-efficient approach with lower operating voltages for possible neuromorphic computing applications. [1] N. Liu et al., IEEE Electron Device Letters, vol. 45, no. 7, pp. 1357-1360, 2024. [2] T. -H. Kim et al., IEEE Electron Device Letters, vol. 44, no. 10, pp. 1628-1631, 2023. [3] D. Kwon and I. -Y. Chung, IEEE Electron Device Letters, vol. 41, no. 3, pp. 493-496, 2020. [4] E. H. Nicollian and A. Goetzberger, IEEE Transactions on Electron Devices, vol. ","author":[{"family":"Kao","given":"Chi"},{"family":"Hwu","given":"Jenn"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1149/ma2025-01361728mtgabs","URL":"https://doi.org/10.1149/ma2025-01361728mtgabs","source":"crossref"},{"id":"doi:10.3390/nano15050348","type":"article-journal","title":"Electrolyte Gated Transistors for Brain Inspired Neuromorphic Computing and Perception Applications: A Review","abstract":"Emerging neuromorphic computing offers a promising and energy-efficient approach to developing advanced intelligent systems by mimicking the information processing modes of the human brain. Moreover, inspired by the high parallelism, fault tolerance, adaptability, and low power consumption of brain perceptual systems, replicating these efficient and intelligent systems at a hardware level will endow artificial intelligence (AI) and neuromorphic engineering with unparalleled appeal. Therefore, construction of neuromorphic devices that can simulate neural and synaptic behaviors are crucial for achieving intelligent perception and neuromorphic computing. As novel memristive devices, electrolyte-gated transistors (EGTs) stand out among numerous neuromorphic devices due to their unique interfacial ion coupling effects. Thus, the present review discusses the applications of the EGTs in neuromorphic electronics. First, operational modes of EGTs are discussed briefly. Second, the advancements of EGTs in mimicking biological synapses/neurons and neuromorphic computing functions are introduced. Next, applications of artificial perceptual systems utilizing EGTs are discussed. Finally, a brief outlook on future developments and challenges is presented.","author":[{"family":"Wang","given":"Weisheng"},{"family":"Zhu","given":"Liqiang"}],"issued":{"date-parts":[[2025]]},"DOI":"10.3390/nano15050348","URL":"https://doi.org/10.3390/nano15050348","source":"europepmc"},{"id":"doi:10.63382/jni.v1i1.8","type":"article-journal","title":"Machine Learning for Soft Robotics","abstract":"Soft Robotics refers to the design, fabrication, and control of robots made from highly deformable materials that mimic biological organisms. They can perform tasks that require flexibility and soft dexterity. With the application of Machine Learning (ML)， soft robots are now able to learn from data, redesign their shape, and handle various tasks with high precision. This literature review explores the integration of ML techniques in soft robotics, examining various data-driven strategies, applications, knowledge gaps, and future directions. Soft robots often require complex, nonlinear dynamics, and autonomous decision-making capability without explicit programming in order to fit in unpredictable environments.Therefore, despite the significant advances in this field, there are still a few technical challenges topical to various implementations of soft robots. In this literature review, five representative ML methods and their diverse applications for soft robots are analyzed. Lastly, the emerging bio-inspired energy-efficient neuromorphic learning is introduced, which uses far less power than traditional computing technologies and allows for a rapid adaption of soft robots to complex environmental changes. Therefore, neuromorphic learning is regarded as a promising tool for efficient event-driven sensing and adaptive locomotion control in the field of soft robotics.","author":[{"family":"Zhang","given":"Baiyu"},{"family":"Qiu","given":"Jingjing"}],"issued":{"date-parts":[[2025]]},"DOI":"10.63382/jni.v1i1.8","URL":"https://doi.org/10.63382/jni.v1i1.8","source":"crossref"},{"id":"doi:10.22214/ijraset.2025.66411","type":"article-journal","title":"Advancements and Challenges in Neuromorphic Computing: Bridging Neuroscience and Artificial Intelligence","abstract":"Neuromorphic computing represents a paradigm shift in computational design, aiming to emulate the neural structures and functionalities of the human brain. This approach seeks to enhance efficiency and adaptability in artificial intelligence (AI) systems. This paper provides a comprehensive review of recent advancements in neuromorphic hardware and software, highlighting their potential to revolutionize AI by enabling real-time processing and energy-efficient computations. Additionally, it examines the challenges inherent in replicating complex neural processes, including issues related to scalability, material limitations, and the integration of neuromorphic systems with existing technologies. By bridging the disciplines of neuroscience and AI, neuromorphic computing offers promising avenues for the development of intelligent systems that closely mirror human cognitive functions.","author":[{"family":"Elfighi","given":"Melad"}],"issued":{"date-parts":[[2025]]},"DOI":"10.22214/ijraset.2025.66411","URL":"https://doi.org/10.22214/ijraset.2025.66411","source":"crossref"},{"id":"doi:10.1002/est2.70272","type":"article-journal","title":"A Comprehensive Review of Phase Change Memory for Neuromorphic Computing: Advancements, Challenges, and Future Directions","abstract":"ABSTRACT The human brain functions as a highly efficient control center, inspiring the field of neuromorphic computing, which seeks to replicate its structure and behavior through hardware systems. Neuromorphic computing integrates processing and memory functions using artificial neurons and synapses designed with electronic circuits, enabling parallel, energy‐efficient data handling. One of the leading technologies supporting this paradigm is phase change memory (PCM), a non‐volatile memory that stores data through reversible transitions between amorphous (high resistance) and crystalline (low resistance) states of chalcogenide materials, particularly Ge 2 Sb 2 Te 5 (GST225). PCM exhibits fast read/write speeds, excellent data retention, and scalability, making it ideal for neuromorphic architectures. This review highlights recent advancements in PCM for neuromorphic computing, including innovations in doping strategies and device engineering. Notable developments include arsenic‐doped ovonic threshold switches (OTS) for enhanced selector performance, monolayer Sb 2 Te 3 for atomic‐scale devices, and heater‐all‐around (HAA) 3D architectures for reduced energy consumption. Integration with machine learning tools enables precise atomistic modeling, accelerating material and device optimization. Furthermore, emerging variants like ovonic unified memory (OUM) and interfacial PCM (IPCM) offer unique performance advantages. While PCM promises significant benefits, key challenges such as resistance drift, endurance limits, and thermal crosstalk must be addressed. The global neuromorphic computing market is poised for exponential growth, driven by innovations in materials, algorithms, and architectures. The PCM and neuromorphic computing represent a transformative leap toward intelligent, adaptive, and energy‐efficient computing systems.","author":[{"family":"Bhatnagar","given":"Vikas"},{"family":"Kumar","given":"Adesh"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1002/est2.70272","URL":"https://doi.org/10.1002/est2.70272","source":"crossref"},{"id":"doi:10.31224/5651","type":"article-journal","title":"JAQA: Building a Joint Alliance of Quantum Assurance for Future-Proof Cybersecurity, Energy Optimization, and Precision Healthcare Leveraging Post-Quantum Cryptography, Neuromorphic Computing, 6G, and XR Technologies in Kerala and India","abstract":"The accelerating convergence of disruptive technologies necessitates innovative frameworks for national security, sustainable development, and advanced public services. This paper details the strategic design and implementation of JAQA, a Joint Alliance of Quantum Assurance, as an integrated multi-stakeholder model for Kerala and India. The alliance aims to establish quantum-safe digital infrastructure, optimized energy systems, and next-generation precision healthcare by leveraging post-quantum cryptography (PQC), neuromorphic computing, 6G networks, extended reality (XR), and autonomous systems. Through a comprehensive methodological approach involving systematic literature review and thematic analysis, this work synthesizes existing research and policy contexts to address the central inquiry: how Kerala and India can pioneer global leadership in secure, intelligent, and sustainable governance by 2030 through JAQA. Qualitative findings reveal critical themes concerning governance structures, technology adoption pathways, ethical considerations, and scalability. Policy implications underscore the imperative for harmonized regulatory frameworks, robust capacity building, and strategic international collaborations. An evaluation plan emphasizes key performance indicators related to cybersecurity resilience, energy efficiency gains, and healthcare outcome improvements, ensuring an iterative refinement process. The proposed JAQA framework provides a blueprint for integrating cutting-edge technologies within a coherent governance ecosystem, fostering trust, agility, and sustainability across vital sectors.","author":[{"family":"Suresh","given":"Amit"}],"issued":{"date-parts":[[2025]]},"DOI":"10.31224/5651","URL":"https://doi.org/10.31224/5651","source":"crossref"},{"id":"doi:10.1002/smll.202511256","type":"article-journal","title":"Self-Powered Neuromorphic Touch Sensors Based on Triboelectric Devices: Current Approaches and Open Challenges.","abstract":"ABSTRACT Advanced neuromorphic systems mimicking the human sensory and nervous system will enable artificial perception for intelligent robotics and human machine interfaces. Among sensing modalities, tactile perception is crucial for replicating human somatosensory and motor functions, with significant potential to restore impaired tactile capabilities. Artificial neuromorphic sensors can directly sense, store and process various stimuli information and implement computation functions such as perception, learning, and memory. However, computational energy efficiency must be achieved with novel neuromorphic systems capable of environmental energy harvesting enabling self‐powered sensing, and real‐time edge data processing. Here, we focus on the integration of tactile self‐powered sensors based on triboelectric nanogenerators (TENGs) with neuromorphic devices. We systematically discuss current approaches for coupling TENGs with artificial synapses and neurons, covering the main integration architectures (ex situ, discrete circuit, direct gating, monolithic), the primary operational modes (displacement‐driven, pulse‐driven), and neuromorphic functions as short‐ and long‐term plasticity, memory, and logic‐in‐memory computing. We also highlight the mechanisms of signal generation and transduction, and the strategies used to enhance performance and energy efficiency. The review concludes with a discussion on key challenges and future directions for developing sustainable, low‐power, and multifunctional neuromorphic tactile systems, paving the way toward fully integrated self‐powered artificial somatosensory platforms.","author":[{"family":"Torricelli","given":"Fabrizio"},{"family":"Pace","given":"Giuseppina"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1002/smll.202511256","URL":"https://doi.org/10.1002/smll.202511256","source":"europepmc"},{"id":"doi:10.1088/2631-8695/adfbbb","type":"article-journal","title":"Towards brain-inspired edge AI: a review of memristor-based neuromorphic computing and learning algorithms","abstract":"Abstract With the advent of innovative technologies and the emerging demands for in situ edge decision-making, the weaknesses of the conventional von Neumann computing paradigm have become increasingly evident. While such architectures were considered reliable until recently, they have struggled to meet the growing requirements for greater processing speed, lower power consumption, and enhanced scalability. To highlight these challenges, several studies since the early 2010s, have focused on neuromorphic computing based on memristors. Unlike traditional computing, this approach relies more on how the human brain processes information, offering a promising alternative for future edge AI applications. Memristor-based systems, particularly those employing crossbar architectures like 1T1R and 1R arrays, enable analog computation and efficient matrix-vector multiplication, laying the foundation for energy-efficient, scalable hardware. This comprehensive review paper explores the fundamentals of memristor, neuromorphic computing with memristor, learning algorithms focusing on Artificial Neural Networks (ANN), Binary Neural Networks (BNN), and Spiking Neural Networks (SNN), while exploring other learning models as well by comparing the accuracy percentage of each model and edge AI tasks such as image classification, emotion recognition, and multimodal learning. Furthermore, it addresses the intrinsic advantages of computation-in-memory architecture regarding scalability, latency reduction, and parallelism. It also offers insights into the challenges such as device-to-device variability, low endurance, non-ideal switching properties, and compatibility with complementary metal–oxide–semiconductor (CMOS) technologies. This study flow will help understand each technique’s central concept to highlight future development challenges, discussing the status and prospects of neuromorphic and brain-inspired computing technologies.","author":[{"family":"Hussein","given":"Salma"},{"family":"Ho","given":"Patrick"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1088/2631-8695/adfbbb","URL":"https://doi.org/10.1088/2631-8695/adfbbb","source":"crossref"},{"id":"doi:10.1088/2634-4386/aea01b","type":"article-journal","title":"A closer-to-brain heterosynaptic learning rule for spatiotemporal spike pattern detection with low-resolution synapse","abstract":"Abstract The brain is believed to process information efficiently in a different manner from deep learning-based artificial intelligence (AI). Brain-like next-generation AI is gaining attention owing to its potential to perform human-like, highly adaptive, robust, and power-efficient computation. To realize such AI, one crucial approach is the bottom-up implementation of the neuronal systems, capturing their electrophysiological characteristics in electronic circuits. However, this neuromorphic approach generally focuses on simplified neuronal models that do not refer to many biological findings. Developing closer-to-brain models is a natural direction that serve as a fundamental computing model for next-generation AI. One of the constraints of neuromorphic circuits is the bit resolution of synaptic efficacy memory, as the memory footprint scales with its precision. Although low-resolution synaptic efficacy is essential for minimizing memory circuit footprint and energy consumption, it generally leads to performance degradation in many tasks such as the spatiotemporal spike pattern detection. This study proposed a closer-to-brain learning rule that incorporates heterosynaptic plasticity (HP) induced by glutamate spillover. It is demonstrated that our model mitigates the performance degradation associated with low-bit resolution synaptic efficacy, achieving the pattern detection success rate with 3-bit resolution synaptic efficacy, which is comparable to 64-bit floating-point precision. Furthermore, the findings of the study indicate that HP based model accelerates the convergence of the synaptic efficacy and effectively potentiates the synapses relevant to the pattern detection while suppressing irrelevant ones, thereby promoting a bimodal distribution of synaptic efficacies. These findings may provide a basic framework for constructing an energy-efficient, brain-like next-generation AI that maintains high performance under hardware constraints.","author":[{"family":"Furuichi","given":"Shunta"},{"family":"Kohno","given":"Takashi"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/2634-4386/aea01b","URL":"https://doi.org/10.1088/2634-4386/aea01b","source":"crossref"},{"id":"doi:10.38124/ijisrt/25aug550","type":"article-journal","title":"Energetic Signatures and Quantum States: Toward a Consciousness-Driven Architecture for  Neuromorphic Computing","abstract":"We introduce a neuromorphic approach to representing the energetic signature of a quantum state. In this scheme, a quantum Hamiltonian is decomposed into a linear combination of energies that we term eigenergy components. A more widely understood concept of eigenspectra emerges when the quantum state itself is an eigenstate of the Hamiltonian. Eigenenergy components of a general state are then obtained by factoring the corresponding energy eigenvalue with the associated eigenstate probability. The complex behavior of the components collectively defines the energetic signature of the state. From such a signature, the original quantum state can be recovered when energetic constraints are lifted. A solid-state mixed-signal implementation is described that leverages properties of the co-integrated analog neuromorphic–digital platform BrainScaleS-2, regarded as world-leading in neuromorphic hardware. Here, spiking neurons realize the linear coupling between components and quantum states and convert a quantum Hamiltonian to a temporal eigenergy distribution through an address-event representation. From this standard communication protocol, the eigenergy components—encoded temporally in the resulting spike pattern—are extracted again by downstream cores. In the neuromorphic output, an energetic signature is mapped to a population of leaky integrate-and-fire neurons that asynchronously evokes a corresponding spiking probability. The functionality of the architecture is demonstrated for Hamiltonians—sparse, dense, and low rank—that arise from models of bilayer graphene relationships [1]. Quantum states remain at the core of various means of information encoding and processing in domains such as communication, sensing, and computing. Algorithms in these domains are usually expressed in mathematical frameworks based on the axioms of quantum theory. Encoded data and subsequent manipulations are regarded as determined by a wholly different probabilistic rule than those found in classical digital computers [2]. Using standard digital hardware, access to states and associated operations therefore poses disproportionate challenges. Co-integrated digital-analog neuromorphic computing architectures present an alternative, reminiscent of single photons transmitting information through an array of gates on a linear-optical circuit. They qualify— as physically motivated data structures—for engineerable representations of quantum states. Mapping quantum systems directly to spiking activity and its propagation through dedicated emulation circuits offers a distinctively novel platform for processing quantum information that operates in a natural synchronous-to-asynchronous mode.","author":[{"family":"Thethali","given":"Aruna"},{"family":"Mandava","given":"Kranthi"}],"issued":{"date-parts":[[2025]]},"DOI":"10.38124/ijisrt/25aug550","URL":"https://doi.org/10.38124/ijisrt/25aug550","source":"crossref"},{"id":"doi:10.1088/2634-4386/adebaa","type":"article-journal","title":"Training and synchronizing oscillator networks with Equilibrium Propagation","abstract":"Abstract Oscillator networks represent a promising technology for unconventional computing and artificial intelligence. Thus far, these systems have primarily been demonstrated in small-scale implementations, such as Ising machines for solving combinatorial problems and associative memories for image recognition, typically trained without state-of-the-art gradient-based algorithms. Scaling up oscillator-based systems requires advanced gradient-based training methods that also ensure robustness against frequency dispersion between individual oscillators. Here, we demonstrate through simulations that the Equilibrium Propagation algorithm enables effective gradient-based training of oscillator networks, facilitating synchronization even when initial oscillator frequencies are significantly dispersed. We specifically investigate two oscillator models: purely phase-coupled oscillators and oscillators coupled via both amplitude and phase interactions. Our results show that these oscillator networks can scale successfully to standard image recognition benchmarks, such as achieving nearly 98% test accuracy on the Modified National Institute of Standards and Technology (MNIST) dataset, despite noise introduced by imperfect synchronization. This work thus paves the way for practical hardware implementations of large-scale oscillator networks, such as those based on spintronic devices.","author":[{"family":"Rageau","given":"Théophile"},{"family":"Grollier","given":"Julie"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1088/2634-4386/adebaa","URL":"https://doi.org/10.1088/2634-4386/adebaa","source":"crossref"},{"id":"doi:10.5267/j.ijdns.2026.37","type":"article-journal","title":"Neuromorphic computing for healthcare engineering: A bibliometric analysis of materials, devices, architectures, and biomedical applications based on 200 highly cited publications (2009–2025)","abstract":"This bibliometric survey examines 200 highly cited publications spanning from 2009 to 2025, retrieved from the Scopus database using search terms targeting neuromorphic computing and its applications in healthcare engineering and biomedical systems. The analysis reveals a rapidly maturing interdisciplinary field at the convergence of materials science, device physics, circuit design, computer architecture, and biomedical engineering. Key findings indicate that memristive devices, particularly those based on metal oxides and two-dimensional materials, constitute the dominant hardware platform for neuromorphic computing, with phase-change materials, spintronic devices, and ferroelectric tunnel junctions representing significant alternative approaches. The United States, China, Germany, South Korea, and Switzerland emerge as the most productive nations, with extensive international collaboration networks reflecting the field's multidisciplinary nature. Thematic clustering identifies five major research domains: (1) memristive devices and resistive switching mechanisms, (2) two-dimensional materials for neuromorphic applications, (3) spiking neural networks and neuromorphic architectures, (4) optical and photonic neuromorphic computing, and (5) biomedical and healthcare applications including electronic skin, neural prosthetics, and biosensing. Emerging trends include the integration of neuromorphic computing with edge artificial intelligence for wearable healthcare devices, the development of reservoir computing for real-time biomedical signal processing, and the convergence of neuromorphic systems with biohybrid interfaces. This survey provides a comprehensive mapping of the intellectual landscape, identifies persistent challenges including device variability, scalability, and biocompatibility, and proposes future directions for healthcare-oriented neuromorphic engineering.","author":[{"family":"Jenab","given":"Kouroush"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5267/j.ijdns.2026.37","URL":"https://doi.org/10.5267/j.ijdns.2026.37","source":"crossref"},{"id":"doi:10.5281/zenodo.21028926","type":"article-journal","title":"Neural Topological Variational Dynamics (NTVD): A Self-Organizing Microstate Scheduling Theory for Next-Generation Operating Systems","abstract":"Conventional operating system scheduling theories rely heavily on discrete, heuristic algorithms to allocate computing resources from an external, centralized controller. While effective for classical architectures, this paradigm faces a fundamental scalability wall when confronted with the non-linear coherency and extreme heterogeneity of exascale and neuromorphic computing systems. In this paper, we introduce Neural Topological Variational Dynamics (NTVD), a fundamentally new scheduling paradigm that abandons discrete task-switching in favor of continuous self-organization. We redefine the entire operating system as a non-equilibrium statistical dynamical system evolving on a high-dimensional Riemannian manifold, termed the Topological Resource Manifold. Computing tasks are no longer represented as static threads, but as continuous spatio-temporal probability density functions. By formulating a global free energy functional—which incorporates intrinsic task urgency, mutual resource conflicts, and computational temperature—scheduling is mathematically transformed into a variational minimization problem. We derive the Neural Topological Variational Flow Equation, proving that tasks can autonomously navigate the manifold via gradient descent, naturally segregating from resource conflicts without external intervention. Furthermore, we establish strict mathematical stability proofs and demonstrate dead-lock immunity leveraging topological invariants and phase transitions. Finally, we integrate deep functional learning directly into the manifold's dissipating structures, enabling tasks to dynamically reshape the resource landscape through physical feedback fields. This work establishes a rigorous, purely continuous mathematical framework for autonomous resource orchestration, paving the way for next-generation distributed and non-von Neumann operating systems.","author":[{"family":"Zhang","given":"Jincheng"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.21028926","URL":"https://doi.org/10.5281/zenodo.21028926","source":"datacite"},{"id":"doi:10.5281/zenodo.21028600","type":"article-journal","title":"Generative Causal Flow Dynamics: A Continuous Non-Equilibrium Computing Paradigm for High-Dimensional Concurrency Theory","abstract":"Traditional formalisms of concurrency theory—such as the Pi-calculus, Actor models, and Petri nets—have long been anchored in the discrete paradigms of state-transition systems and combinatorial graph theory. While highly successful for foundational verification, these approaches suffer from catastrophic state-space explosion when confronting modern, ultra-large-scale, heterogeneous, and dynamically evolving computational topologies. This paper breaks away from the discrete assumption and introduces Generative Causal Flow Dynamics (GCFD), a radically new concurrency theory that redefines concurrent computation as the evolution of non-equilibrium probability density flows over high-dimensional differentiable manifolds. Under this framework, individual process executions are modeled as projections of continuous coordinate dimensions, driven by the coupling of gradient flows and stochastic diffusion within a potential energy field constrained by a novel asymmetric causal metric tensor. We introduce the concept of \"causal geodesics\" to characterize the temporal flow of concurrent events, formulate synchronization mechanisms as topological phase transitions where probability flows collapse onto compact hyperplanes, and define classic concurrent anomalies such as deadlocks strictly as geometric singularities or energy sinks on the manifold. By extending functional variational principles and generalizing Generative Flow Networks (GFlowNets) into continuous domains, we construct a mathematically rigorous, fully composable continuous concurrent algebra. We analytically prove the global consistency, mass conservation, and monotonic convergence of the system without relying on discrete state exploration. This theory establishes a new mathematical foundation for analyzing ultra-large scale parallel systems and paves the way for future native continuous neuromorphic computing architectures.","author":[{"family":"Zhang","given":"Jincheng"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.21028600","URL":"https://doi.org/10.5281/zenodo.21028600","source":"datacite"},{"id":"doi:10.5281/zenodo.21028599","type":"article-journal","title":"Generative Causal Flow Dynamics: A Continuous Non-Equilibrium Computing Paradigm for High-Dimensional Concurrency Theory","abstract":"Traditional formalisms of concurrency theory—such as the Pi-calculus, Actor models, and Petri nets—have long been anchored in the discrete paradigms of state-transition systems and combinatorial graph theory. While highly successful for foundational verification, these approaches suffer from catastrophic state-space explosion when confronting modern, ultra-large-scale, heterogeneous, and dynamically evolving computational topologies. This paper breaks away from the discrete assumption and introduces Generative Causal Flow Dynamics (GCFD), a radically new concurrency theory that redefines concurrent computation as the evolution of non-equilibrium probability density flows over high-dimensional differentiable manifolds. Under this framework, individual process executions are modeled as projections of continuous coordinate dimensions, driven by the coupling of gradient flows and stochastic diffusion within a potential energy field constrained by a novel asymmetric causal metric tensor. We introduce the concept of \"causal geodesics\" to characterize the temporal flow of concurrent events, formulate synchronization mechanisms as topological phase transitions where probability flows collapse onto compact hyperplanes, and define classic concurrent anomalies such as deadlocks strictly as geometric singularities or energy sinks on the manifold. By extending functional variational principles and generalizing Generative Flow Networks (GFlowNets) into continuous domains, we construct a mathematically rigorous, fully composable continuous concurrent algebra. We analytically prove the global consistency, mass conservation, and monotonic convergence of the system without relying on discrete state exploration. This theory establishes a new mathematical foundation for analyzing ultra-large scale parallel systems and paves the way for future native continuous neuromorphic computing architectures.","author":[{"family":"Zhang","given":"Jincheng"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.21028599","URL":"https://doi.org/10.5281/zenodo.21028599","source":"datacite"},{"id":"doi:10.5281/zenodo.20843912","type":"article-journal","title":"A Neodymium-Inspired Fractal State-Space Generator for Neuromorphic Control","abstract":"Modern software typically represents memory as a data structure: arrays, buffers, queues, key-value stores, or attention contexts. In time-critical control, however, expanding the memory structure directly increases memory traffic, latency, and engineering complexity. This preprint proposes NdFractal, a neodymium-inspired fractal state-space generator for neuromorphic control. Instead of storing past inputs as an ever-growing sequence, NdFractal lets past inputs deform a compact dynamical state space. New sensory inputs then flow through this deformed space, where attractor-like dynamics separate control-relevant regimes. In this view, memory is not only stored in data; it is formed into the geometry of the state space itself. NdFractal is motivated by three converging ideas: reservoir computing, fractional or power-law memory, and the spin/hysteresis/remanence metaphor of neodymium-like magnetic materials. The proposed role of NdFractal is not to replace the neuromorphic controller, but to act as a front-end state-space generator for spiking liquid neural networks and related adaptive control systems. Conceptually, the distinction can be summarized as follows: Transformers unfold memory into addressable context; state-space models such as Mamba fold memory into recurrent state; NdFractal folds and unfolds the memory space itself as a function. Preliminary internal observations suggest that neodymium-inspired fractal preprocessing can improve already-optimized neuromorphic control systems by single-digit relative margins in selected domains, while also improving state separation and temporal basis richness. These observations are presented here only as motivation. This manuscript does not provide a reproducible implementation.","author":[{"family":"Lee","given":"Kyuchul"},{"family":"Cordingai"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20843912","URL":"https://doi.org/10.5281/zenodo.20843912","source":"datacite"},{"id":"doi:10.5281/zenodo.20844346","type":"article-journal","title":"A Neodymium-Inspired Fractal State-Space Generator for Neuromorphic Control","abstract":"Modern software typically represents memory as a data structure: arrays, buffers, queues, key-value stores, or attention contexts. In time-critical control, however, expanding the memory structure directly increases memory traffic, latency, and engineering complexity. This preprint proposes NdFractal, a neodymium-inspired fractal state-space generator for neuromorphic control. Instead of storing past inputs as an ever-growing sequence, NdFractal lets past inputs deform a compact dynamical state space. New sensory inputs then flow through this deformed space, where attractor-like dynamics separate control-relevant regimes. In this view, memory is not only stored in data; it is formed into the geometry of the state space itself. NdFractal is motivated by three converging ideas: reservoir computing, fractional or power-law memory, and the spin/hysteresis/remanence metaphor of neodymium-like magnetic materials. The proposed role of NdFractal is not to replace the neuromorphic controller, but to act as a front-end state-space generator for spiking liquid neural networks and related adaptive control systems. Conceptually, the distinction can be summarized as follows: Transformers unfold memory into addressable context; state-space models such as Mamba fold memory into recurrent state; NdFractal folds and unfolds the memory space itself as a function. Preliminary internal observations suggest that neodymium-inspired fractal preprocessing can improve already-optimized neuromorphic control systems by single-digit relative margins in selected domains, while also improving state separation and temporal basis richness. These observations are presented here only as motivation. This manuscript does not provide a reproducible implementation.","author":[{"family":"Lee","given":"Kyuchul"},{"family":"Cordingai"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20844346","URL":"https://doi.org/10.5281/zenodo.20844346","source":"datacite"},{"id":"doi:10.5281/zenodo.20707853","type":"article-journal","title":"The Coherence Cost: 1/√Depth as the Universal Law of Compiled Existence","abstract":"The Coherence Cost: 1/√Depth as the Universal Law of Compiled Existence Driven By Dean A. Kulik June 2026 Abstract Five independent systems — a deep residual neural network, a neural optimizer, a graph clustering algorithm, an online learning protocol, and a neuromorphic hardware architecture — all obey the same law: the cost of maintaining a coherent compiled state against a running background decays as the inverse square root of depth. These systems were developed by different researchers, in different decades, for different purposes. They share no surface-level connection. The convergence is not coincidence. It is a measurement. This paper asks what geometry forces that measurement, derives the answer, and shows that the derivation demands a single conclusion: logic is downstream of shape, change is identical to computation, and existence is what survives the running against the holding. The ontology is not a philosophical claim imposed on the data. It is what the data requires. I. The Law First Every theory should begin where it is hardest to deny. Here is the data: Domain System The Law NEXUS Engine 18 Deep Residual Networks α* ≈ 2.5/√L Neural Optimization RLD vs AdamW Diffusion scales as D_t^{−1/2} Graph Theory Laplacian Clustering ψ_j(i) = u_j(i)/√λ_j Online Learning Confidence-Budget Matching Query when confidence > 1/√budget Neuromorphic Hardware Synaptic Memory Memory strength decays as 1/√age In each case a system is trying to maintain structured state across time, depth, or budget. In each case the cost of maintaining that structure — the gain required, the confidence threshold, the decay rate — scales with the inverse square root of the accumulation variable. This is the starting point. Not a metaphysics. Not a framework assertion. Five measurements of the same number in five places that have never communicated. The question this paper answers is: what geometry forces this? II. The Shape That Forces 1/√Depth The geometry of accumulation A random walk of N independent steps produces displacement proportional to √N. This is not a discovered empirical fact about random walks — it follows directly from the geometry of independence. At each step a variance σ² is added. Variances add when steps are independent. After N steps, total variance is N·σ². Standard deviation is √N·σ. This is why diffusion scales as √N. Not because of any property of the diffusing thing. Because independence plus accumulation equals square root growth. The square root is the signature of uncorrelated steps summing over depth. Now introduce a constraint. A compiled state is not a random walk — it is a path through a space that must satisfy structural requirements at every step. At each step, the constraint filters out proposals that violate structure and admits only those that satisfy it. The filtered proposals form the next step. Here is the critical point: filtering does not eliminate accumulation. Among the surviving proposals at each step, there is residual variance — the constraint space has internal width. That residual variance accumulates across steps. If the constraints at successive steps are independent — if what is permitted at step k does not fully determine what is permitted at step k+1 — then the residual variance sums, and total accumulated variance after D steps is proportional to D. Standard deviation of the accumulated drift is proportional to √D. To correct this — to maintain a signal-to-noise ratio of order 1 at depth D — you must apply a corrective gain proportional to 1/√D. Not because of the domain. Not because of the specific system. Because the geometry of independent accumulation always produces √D drift, and correcting √D drift always costs 1/√D. This is the geometry that forces the law. What the constraint does and does not do A constraint system reduces the width of the proposal space at each step. If the constraint were total — if only one proposal were permitted at each step — there would be no variance and no acc","author":[{"family":"Kulik","given":"Dean"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20707853","URL":"https://doi.org/10.5281/zenodo.20707853","source":"datacite"},{"id":"doi:10.5281/zenodo.20707854","type":"article-journal","title":"The Coherence Cost: 1/√Depth as the Universal Law of Compiled Existence","abstract":"The Coherence Cost: 1/√Depth as the Universal Law of Compiled Existence Driven By Dean A. Kulik June 2026 Abstract Five independent systems — a deep residual neural network, a neural optimizer, a graph clustering algorithm, an online learning protocol, and a neuromorphic hardware architecture — all obey the same law: the cost of maintaining a coherent compiled state against a running background decays as the inverse square root of depth. These systems were developed by different researchers, in different decades, for different purposes. They share no surface-level connection. The convergence is not coincidence. It is a measurement. This paper asks what geometry forces that measurement, derives the answer, and shows that the derivation demands a single conclusion: logic is downstream of shape, change is identical to computation, and existence is what survives the running against the holding. The ontology is not a philosophical claim imposed on the data. It is what the data requires. I. The Law First Every theory should begin where it is hardest to deny. Here is the data: Domain System The Law NEXUS Engine 18 Deep Residual Networks α* ≈ 2.5/√L Neural Optimization RLD vs AdamW Diffusion scales as D_t^{−1/2} Graph Theory Laplacian Clustering ψ_j(i) = u_j(i)/√λ_j Online Learning Confidence-Budget Matching Query when confidence > 1/√budget Neuromorphic Hardware Synaptic Memory Memory strength decays as 1/√age In each case a system is trying to maintain structured state across time, depth, or budget. In each case the cost of maintaining that structure — the gain required, the confidence threshold, the decay rate — scales with the inverse square root of the accumulation variable. This is the starting point. Not a metaphysics. Not a framework assertion. Five measurements of the same number in five places that have never communicated. The question this paper answers is: what geometry forces this? II. The Shape That Forces 1/√Depth The geometry of accumulation A random walk of N independent steps produces displacement proportional to √N. This is not a discovered empirical fact about random walks — it follows directly from the geometry of independence. At each step a variance σ² is added. Variances add when steps are independent. After N steps, total variance is N·σ². Standard deviation is √N·σ. This is why diffusion scales as √N. Not because of any property of the diffusing thing. Because independence plus accumulation equals square root growth. The square root is the signature of uncorrelated steps summing over depth. Now introduce a constraint. A compiled state is not a random walk — it is a path through a space that must satisfy structural requirements at every step. At each step, the constraint filters out proposals that violate structure and admits only those that satisfy it. The filtered proposals form the next step. Here is the critical point: filtering does not eliminate accumulation. Among the surviving proposals at each step, there is residual variance — the constraint space has internal width. That residual variance accumulates across steps. If the constraints at successive steps are independent — if what is permitted at step k does not fully determine what is permitted at step k+1 — then the residual variance sums, and total accumulated variance after D steps is proportional to D. Standard deviation of the accumulated drift is proportional to √D. To correct this — to maintain a signal-to-noise ratio of order 1 at depth D — you must apply a corrective gain proportional to 1/√D. Not because of the domain. Not because of the specific system. Because the geometry of independent accumulation always produces √D drift, and correcting √D drift always costs 1/√D. This is the geometry that forces the law. What the constraint does and does not do A constraint system reduces the width of the proposal space at each step. If the constraint were total — if only one proposal were permitted at each step — there would be no variance and no acc","author":[{"family":"Kulik","given":"Dean"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20707854","URL":"https://doi.org/10.5281/zenodo.20707854","source":"datacite"},{"id":"doi:10.21203/rs.3.rs-9949762/v1","type":"article-journal","title":"Dual-Mode Deep-Ultraviolet Photodetection and neuromorphic vision sensor via Thermal Engineering of Oxygen Vacancies in Ga2O3","abstract":"Abstract The integration of neuromorphic vision sensors and photodetectors has attracted considerable interest. Here, a facile single-step thermal oxidation strategy is employed to fabricate Ga 2 O 3 thin films, realizing dual functionalities within a single material through differentiated oxygen vacancy (V O ) concentrations. Interestingly, this concurrent process of Ga 2 O 3 formation and V O modulation achieves a broad tunability of the V O concentration from 13.46% to 44.06%, solely through annealing temperature variation. Low-temperature (500°C) devices contain a high density of V O , and the energy barriers formed by V O photoionization give rise to a pronounced persistent photoconductivity effect. It enables the devices diverse synaptic plasticity, emulation of artificial visual memory, image filtering as preprocessing for recognition, and physical reservoir computing with 96.33% handwritten digit recognition accuracy under 10% noise. High-temperature (900°C) devices, with substantially reduced V O concentration, exhibit ultralow dark current (1.76 × 10 − 10 A), a photo-to-dark current ratio of 387.85, and fast response (rise/decay times of 0.19/0.20 s), making them excellent solar-blind photodetectors for optical communication and imaging. By vacancy tailoring, this work merges photodetection and neuromorphic computation within a single Ga 2 O 3 platform, providing a streamlined fabrication paradigm and mechanistic insights for advanced multifunctional solar-blind optoelectronics.","author":[{"family":"Cao","given":"Yanqiang"}],"issued":{"date-parts":[[2026]]},"DOI":"10.21203/rs.3.rs-9949762/v1","URL":"https://doi.org/10.21203/rs.3.rs-9949762/v1","source":"europepmc"},{"id":"doi:10.21203/rs.3.rs-10051531/v1","type":"article-journal","title":"Neuromorphic Satellite Intelligence: A Formal Framework for Event-Driven Architectures, Spike-Based Algorithms, and Quantitative Energy-Accuracy Trade-offs","abstract":"Abstract Satellite onboard computing faces an irreducible tension: mission intelligence requirements grow monotonically while power, mass, and radiation budgets remain fixed. This paper advances the State of the art on three fronts. First, we develop a formal mathematical model of spiking neural computation for satellite workloads, deriving closed-form expressions for energy consumption as a function of network sparsity, firing rate, and synaptic fan-out. Second, we propose three original algorithms — a Spiking Kalman-Complementary Filter (SKCF) for attitude estimation, a Temporal Contrast Anomaly Detector (TCAD) for spacecraft health monitoring, and a Spike-Based Adaptive Modulation and Coding (SAMC) scheme for cognitive satellite links — each with pseudocode and analytical complexity bounds. Third, we introduce a quantitative Energy-Accuracy-Latency (EAL) trade-off surface and populate it with numerical estimates derived from published silicon measurements of Loihi and SpiNNaker, enabling direct comparison with FPGA and GPU baselines. A taxonomy of four radiation-aware neuromorphic architecture classes is refined with fault-rate models. Continual on-orbit learning is formalized through a constrained Spike-Timing Dependent Plasticity (STDP) rule that provably bounds weight drIft. Case studies for a 12U CubeSat Earth observation mission, a GEO communications payload, and an autonomous deep-space probe provide concrete, quantIfied design points. Our analysis shows that neuromorphic implementations can achieve 8–47× energy reduction over GPU baselines while maintaining inference accuracy within 2.1\\% for representative satellite workloads.","author":[{"family":"Tp","given":"Suresh"},{"family":"Agnihotri","given":"Vikas"}],"issued":{"date-parts":[[2026]]},"DOI":"10.21203/rs.3.rs-10051531/v1","URL":"https://doi.org/10.21203/rs.3.rs-10051531/v1","source":"europepmc"},{"id":"doi:10.54254/2755-2721/2025.19928","type":"article-journal","title":"Research Progress of Neuromorphic Chips","abstract":"The increasing amount of data in the era of artificial intelligence imposes higher demands on the computational power of neural networks, and in order to fulfill this demand, there is a pressing need to overcome the limitations imposed by the von Neumann architecture's memory wall. Memristors, with their characteristics, are considered the optimal electronic devices for implementing neuromorphic computing. Therefore, in order to better utilize memristors for the design and research of neuromorphic chips, this paper summarizes and comparatively analyzes the memristor characteristics, the RRAM basic principles, memristor array research, crossbar array designs based on memristors, and the study of memristor-based neuromorphic computing chips through the review. The paper emphasizes the challenges that memristor-based neuromorphic computing chips still face in the future, such as non-linear resistance variation. In addition, potential future research directions for amnesia-based neuromorphic computing chips, including amnesia architecture, programming techniques, and instruction set development, are discussed and investigated","author":[{"family":"Fan","given":"Luwei"}],"issued":{"date-parts":[[2025]]},"DOI":"10.54254/2755-2721/2025.19928","URL":"https://doi.org/10.54254/2755-2721/2025.19928","source":"crossref"},{"id":"doi:10.1149/ma2025-02341691mtgabs","type":"article-journal","title":"<i>(Invited)</i>\n                    Low-Dimensional Neuromorphic Electronic Materials and Applications","abstract":"The exponentially improving performance of digital computers has recently slowed due to power consumption issues resulting from the von Neumann bottleneck. In contrast, neuromorphic computing circumvents these limitations by spatially co-locating logic and memory in a manner analogous to biological neuronal networks [1]. This talk will explore how low-dimensional nanoelectronic materials enable gate-tunable neuromorphic devices. For example, by utilizing self-aligned, atomically thin heterojunctions, dual-gated Gaussian transistors have been realized, which show tunable anti-ambipolarity for artificial spiking neurons and mixed-kernel support vector machines [2]. Charge transfer in van der Waals interfacial junction transistors provides further opportunities for tailoring doping profiles to enable hardware-efficient reconfigurable fuzzy logic and fuzzy neural networks [3]. In addition, field-driven defect motion in polycrystalline monolayer MoS 2 yields gate-tunable memristive phenomena for memtransistors that concurrently achieve logic and memory functions. The planar geometry of memtransistors further allows multiple contacts that mimic the behavior of biological systems such as heterosynaptic responses. Moreover, control over polycrystalline grain structure enhances the tunability of potentiation and depression for unsupervised continuous learning in spiking neural networks [4]. Finally, the moiré potential in asymmetric twisted bilayer graphene/hexagonal boron nitride heterostructures gives rise to robust room-temperature electronic ratchet states that underlie diverse bio-realistic neuromorphic functionalities [5]. By extending these strategies to magnetically ordered low-dimensional semiconductors, additional opto-spintronic degrees of freedom can be exploited for neuromorphic function [6]. [1] S. Hadke, et al., Chemical Reviews, 125, 835 (2025). [2] X. Yan, et al., Nature Electronics, 6, 862 (2023). [3] H. Liu, et al., Nature Electronics, 7, 876 (2024). [4] X. Yan, et al., Advanced Materials, 34, 2108025 (2022). [5] X. Yan, et al., Nature, 624, 551 (2023). [6] J. T. Gish, et al., Nature Electronics, 7, 336 (2024).","author":[{"family":"Hersam","given":"Mark"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1149/ma2025-02341691mtgabs","URL":"https://doi.org/10.1149/ma2025-02341691mtgabs","source":"crossref"},{"id":"doi:10.63363/aijfr.2025.v06i06.2067","type":"article-journal","title":"Neuromorphic Adaptation and Cognitive Parallelism — A VerbaTerra Project Study","abstract":"This study, Neuromorphic Adaptation and Cognitive Parallelism, is a core output of the VerbaTerra Project, which seeks to unify neuroscience, linguistics, anthropology, and artificial intelligence under the principle of resonant coherence — the alignment of feedback rhythms across cognition, culture, and computation. Using only secondary academic data and a simulation-based framework, the research develops the vSION Neuromorphic Engine, a four-layer adaptive system (Perceptual, Linguistic, Cognitive, Energetic) capable of learning through rhythmic modulation rather than static optimisation. The simulated data follow the entropy–energy–ethics balance equations described in Annex A and the resonance logic detailed in Annex C. Results show that: 1. Rhythmic learning cycles reduce informational entropy while improving coherence; 2. Creativity peaks at energetic equilibrium; 3. Cross-modal transfer between perception and language enhances adaptability; and 4. Ethical energy regulation functions as a physical limit sustaining long-term cognition. These findings, while conceptual, demonstrate the theoretical viability of resonant ethics as a measurable component of intelligence. By providing executable code in open form (Annex D–F), the paper transforms theory into an interactive experiment—an embodiment of VerbaTerra’s belief that knowledge must remain participatory, transparent, and rhythmic.","author":[{"family":"Gupta","given":"Harshit"}],"issued":{"date-parts":[[2025]]},"DOI":"10.63363/aijfr.2025.v06i06.2067","URL":"https://doi.org/10.63363/aijfr.2025.v06i06.2067","source":"crossref"},{"id":"doi:10.1002/admt.71088","type":"article-journal","title":"Self‐Rectifying Second Order Memristive Behavior in WO\n                    <sub>3</sub>\n                    Films for Neuromorphic Computing Applications","abstract":"ABSTRACT Brain‐inspired neuromorphic computing paves an alternative to the Von Neumann bottleneck on implementing highly efficient and parallel processing hardware. However, in a crossbar array architecture, crosstalk occurs between devices. To address this issue, in this study, a Pt/WO 3 /FTO self‐rectifying memristor was fabricated using pulsed laser deposition, exhibiting a rectification ratio ≈10 3 , endurance of 5 × 10 3 cycles, negligible device‐to‐device variability, highly stable up to 25 cycles, energy consumption of 1.1 pJ/µm 2 in writing operation, and excellent synaptic behavior without any additional material engineering. This study comprehensively investigates the neuromorphic computing characteristics of the fabricated Pt/WO 3 /FTO structure. The conduction mechanism was explained using charge trapping/detrapping and Schottky barrier formation at the Pt/WO 3 interface, resulting in excellent rectification behavior. Interestingly, the fabricated structure exhibited second‐order memristive behavior, attributed to internal ion dynamics. The neuromorphic computing characteristics were comprehensively assessed using artificial neural network simulations, and fabricated devices achieved 90% accuracy on the MNIST handwritten digits dataset and 75% accuracy on the Fashion‐MNIST dataset, thus emulating synaptic functionality and showing excellent potential of the fabricated devices in real‐world artificial intelligence applications.","author":[{"family":"Kanthaswamy","given":"Agesthian"},{"family":"Thakre","given":"Atul"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1002/admt.71088","URL":"https://doi.org/10.1002/admt.71088","source":"crossref"},{"id":"doi:10.5281/zenodo.20647975","type":"article-journal","title":"NeuroSilk - Bio Neuromorphic Humanoid Robotics: Integration of Proton Donor–Acceptor (PrDA) Artificial Spider Silk Hydrogel Fibers for Soft Ionic Signal Transmission, HASEL Artificial Muscles, and Carbon Fiber Endoskeleton toward Embodied AI Systems","abstract":"This thesis introduces the NeuroSilk Hybrid Robotics Architecture (NSS), a biomimetic framework for humanoid robots that emulates the human musculoskeletal and nervous systems through the coordinated integration of three principal technologies: neuron-inspired Proton Donor–Acceptor (PrDA) artificial spider silk hydrogel fibres serving as a soft, adhesive, and ionically conductive nervous system; Hydraulically Amplified Self-healing Electrostatic (HASEL) actuators serving as compliant artificial muscles; and a lightweight carbon fibre composite endoskeleton providing structural support. A four-layer architecture (endoskeleton → muscles → nerves → silicone skin with evaporative cooling) is proposed, fabricated at the component level, and characterised mechanically, electrically, and thermally. The integrated system addresses well-documented limitations of rigid robotics — high energy consumption, poor compliance, limited distributed sensing, and unsafe human interaction — while advancing the emerging field of embodied neuromorphic AI through physical, brain-like ionic signal processing and morphological computation. The work is conducted independently and shared in alignment with NASA's Transform to Open Science (TOPS) initiative. Keywords: NeuroSilk; PrDA hydrogel; HASEL actuators; soft robotics; embodied AI; hybrid robotics; neuromorphic computing; humanoid robotics; aerospace engineering; mechatronics.","author":[{"family":"Vatulia","given":"Leona"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20647975","URL":"https://doi.org/10.5281/zenodo.20647975","source":"datacite"},{"id":"doi:10.5281/zenodo.20647974","type":"article-journal","title":"NeuroSilk - Bio Neuromorphic Humanoid Robotics: Integration of Proton Donor–Acceptor (PrDA) Artificial Spider Silk Hydrogel Fibers for Soft Ionic Signal Transmission, HASEL Artificial Muscles, and Carbon Fiber Endoskeleton toward Embodied AI Systems","abstract":"This thesis introduces the NeuroSilk Hybrid Robotics Architecture (NSS), a biomimetic framework for humanoid robots that emulates the human musculoskeletal and nervous systems through the coordinated integration of three principal technologies: neuron-inspired Proton Donor–Acceptor (PrDA) artificial spider silk hydrogel fibres serving as a soft, adhesive, and ionically conductive nervous system; Hydraulically Amplified Self-healing Electrostatic (HASEL) actuators serving as compliant artificial muscles; and a lightweight carbon fibre composite endoskeleton providing structural support. A four-layer architecture (endoskeleton → muscles → nerves → silicone skin with evaporative cooling) is proposed, fabricated at the component level, and characterised mechanically, electrically, and thermally. The integrated system addresses well-documented limitations of rigid robotics — high energy consumption, poor compliance, limited distributed sensing, and unsafe human interaction — while advancing the emerging field of embodied neuromorphic AI through physical, brain-like ionic signal processing and morphological computation. The work is conducted independently and shared in alignment with NASA's Transform to Open Science (TOPS) initiative. Keywords: NeuroSilk; PrDA hydrogel; HASEL actuators; soft robotics; embodied AI; hybrid robotics; neuromorphic computing; humanoid robotics; aerospace engineering; mechatronics.","author":[{"family":"Vatulia","given":"Leona"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20647974","URL":"https://doi.org/10.5281/zenodo.20647974","source":"datacite"},{"id":"doi:10.5281/zenodo.20644744","type":"article-journal","title":"Beyond Transformers: Emerging Neural Architectures, Energy-Efficient AI, and Multimodal Fusion for Next-Generation Artificial Intelligence","abstract":"It analyzes the limitations of traditional Transformer models (such as quadratic computational complexity and high energy consumption) and explores promising alternative architectures like State Space Models (SSMs) and Mamba. The paper also investigates techniques for energy-efficient AI (including quantization, pruning, knowledge distillation, and neuromorphic computing) alongside methods for multimodal fusion—integrating text, image, audio, and sensor data into unified AI systems. Overall, it argues that the future of AI lies in developing scalable, sustainable, and efficient models that move beyond a single dominant architecture","author":[{"family":"Akazou","given":"Ibtissam"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20644744","URL":"https://doi.org/10.5281/zenodo.20644744","source":"datacite"},{"id":"doi:10.5281/zenodo.20644743","type":"article-journal","title":"Beyond Transformers: Emerging Neural Architectures, Energy-Efficient AI, and Multimodal Fusion for Next-Generation Artificial Intelligence","abstract":"It analyzes the limitations of traditional Transformer models (such as quadratic computational complexity and high energy consumption) and explores promising alternative architectures like State Space Models (SSMs) and Mamba. The paper also investigates techniques for energy-efficient AI (including quantization, pruning, knowledge distillation, and neuromorphic computing) alongside methods for multimodal fusion—integrating text, image, audio, and sensor data into unified AI systems. Overall, it argues that the future of AI lies in developing scalable, sustainable, and efficient models that move beyond a single dominant architecture","author":[{"family":"Akazou","given":"Ibtissam"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20644743","URL":"https://doi.org/10.5281/zenodo.20644743","source":"datacite"},{"id":"doi:10.26183/pn9n-3t62","type":"article-journal","title":"Hardware architectures for event-driven feature extraction algorithms","abstract":"Over the last decade, Neuromorphic Engineering (NE) has evolved into a prominent scientific discipline, advancing biologically inspired sensors, algorithms, hardware architectures, and processing techniques that emulate key operational principles of the human brain. Among these, the most notable innovation is the silicon retinas, commonly known as Dynamic Vision Sensor (DVS), which produce a stream of events in response to changes in luminosity in the light. Conventional frame-based processing techniques pose limitations in handling these streams of events, leading to the emergence of Event-Based (EB) processing techniques designed to manage the asynchronous stream of events generated by DVS sensors. In addition, the scientific community has witnessed a tremendous growth in EB vision solutions, where DVS have been deployed across diverse domains, including recognition, tracking, motion estimation, space do- main awareness, and space imaging. The first part of the thesis proposes a reconfigurable hardware accelerator implemented on the Field Programmable Gate Array (FPGA) for the FEAST network. The novelty of the design lies in splitting the FEAST learning rule into two different sets of tasks and executing them independently in a time-multiplexed fashion, using a minimum number of hardware resources. The proposed hardware architecture, operated at 200 megahertz (MHz) clock frequency, can process 183⇥103 events/sec in training mode and 196⇥103 events/sec in inference mode. The second part of the thesis examines the FPGA implementation of a reconfigurable hardware accelerator designed to support flexible Event Context (E C) computation of the FEAST algorithm. The architecture introduces a novel 1-D computing core, which operates as a broadcast structure utilising a loop unrolling mechanism that supports the T S and FEAST learning rule execution with minimum hardware utilisation. Additionally, the proposed flexible hardware offers configurability across a wide range of E C sizes, with radii ranging from 1 to 7. Compared to the first FEAST implementation, this architecture processes 3.95⇥ events. The final part of the thesis explores the design of the first-ever multi-layer FEAST network and its FPGA implementation. The novelty of the design lies in the introduction of a 1-D computing core, which interleaves the FEAST layer computations without incurring additional hardware overhead. The proposed architecture achieves an event processing rate of 247⇥103 events/sec while the FEAST layers are configured with an E C size of 15 ⇥ 15.","author":[{"family":"Jose","given":"Philip"}],"issued":{"date-parts":[[2026]]},"DOI":"10.26183/pn9n-3t62","URL":"https://doi.org/10.26183/pn9n-3t62","source":"datacite"},{"id":"doi:10.5281/zenodo.20202223","type":"article-journal","title":"ExSpike: A General Full-Event Neuromorphic Architecture for Exploiting Irregular Sparsity with Event Compression","abstract":"Spiking neural networks (SNNs) promise energy-efficient computing due to their sparse spatio-temporal activity. However, effectively translating such irregular sparsity into practical performance and energy gains remains challenging, as full-event computing architectures are still underexplored. This paper proposes ExSpike, a general full-event neuromorphic architecture that fully exploits irregular sparsity in SNNs. To realize pure event-driven execution, we first propose a set of dataflow optimizations to ensure that the inputs to each SNN layer remain spike-based, thereby enabling full-event execution throughout the network. We then design a hardware-efficient full-event architecture, named ExSpike, which supports the optimized pure event-driven dataflow and an additional Attention Core for spike-driven self-attention. To further improve computing efficiency, we introduce adjacent-position event compression to reduce redundant accumulations across spatially adjacent spike sequences. ExSpike is implemented on an AMD Xilinx Virtex-7 FPGA and evaluated on both classification and segmentation workloads. Experimental results show that ExSpike achieves high normalized energy efficiency across diverse SNN models while maintaining competitive accuracy, delivering up to 479.15 GOPS, 281.85 GOPS/W, and 0.80 GOPS/W/PE. In particular, ExSpike achieves up to 10x higher normalized energy efficiency than prior FPGA-based accelerators.","author":[{"family":"Chen","given":"Yuehai"},{"family":"Merchant","given":"Farhad"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20202223","URL":"https://doi.org/10.5281/zenodo.20202223","source":"datacite"},{"id":"doi:10.5281/zenodo.20600138","type":"article-journal","title":"ExSpike: A General Full-Event Neuromorphic Architecture for Exploiting Irregular Sparsity with Event Compression","abstract":"Spiking neural networks (SNNs) promise energy-efficient computing due to their sparse spatio-temporal activity. However, effectively translating such irregular sparsity into practical performance and energy gains remains challenging, as full-event computing architectures are still underexplored. This paper proposes ExSpike, a general full-event neuromorphic architecture that fully exploits irregular sparsity in SNNs. To realize pure event-driven execution, we first propose a set of dataflow optimizations to ensure that the inputs to each SNN layer remain spike-based, thereby enabling full-event execution throughout the network. We then design a hardware-efficient full-event architecture, named ExSpike, which supports the optimized pure event-driven dataflow and an additional Attention Core for spike-driven self-attention. To further improve computing efficiency, we introduce adjacent-position event compression to reduce redundant accumulations across spatially adjacent spike sequences. ExSpike is implemented on an AMD Xilinx Virtex-7 FPGA and evaluated on both classification and segmentation workloads. Experimental results show that ExSpike achieves high normalized energy efficiency across diverse SNN models while maintaining competitive accuracy, delivering up to 479.15 GOPS, 281.85 GOPS/W, and 0.80 GOPS/W/PE. In particular, ExSpike achieves up to 10x higher normalized energy efficiency than prior FPGA-based accelerators.","author":[{"family":"Chen","given":"Yuehai"},{"family":"Merchant","given":"Farhad"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20600138","URL":"https://doi.org/10.5281/zenodo.20600138","source":"datacite"},{"id":"doi:10.5281/zenodo.20576755","type":"article-journal","title":"Variational-Spectral-GKSL Closed Dynamical Theory of a Quantum-Information Field","abstract":"Version:v0.1-formalization Author:Vitaliy Bazarov(VIT-BAZ) ORCID:0009-0007-9967-5807 Status:Preprint / theoretical specification / formal research archive Description:This archive contains the current formalization package for the UAU–ESVG–SAAU–Σ* framework: a layered variational, spectral, categorical, operator-algebraic, and hardware-realizable theory of dissipative adaptive dynamical systems. The package includes a unified formalization synthesis, contradiction report, working mathematical instrument, categorical omega-closure formalization, GKSL operator-field formalization, and physical Sigma-star hardware/runtime layer. The framework should be read as a conditional mathematical and engineering specification. Claims of reconstruction, universality, and physical realization are valid only under the explicit assumptions stated in the documents: well-posed dissipative dynamics, admissible Koopman/GKSL spectral resolution, separating observables, finite effective sector or Σ*_ε approximation, and controlled physical implementation. Scientific Status Statement:This release is a theoretical and formal specification package. It contains conditional mathematical frameworks, proof programs, hardware abstraction layers, and contradiction audits. It is not a claim of completed silicon validation, medical validation, universal physical implementation, or experimentally verified cognition theory. Core Technical Statement:The central structure of the framework is: E(U) → b = -G∇E → Φ_t → K_t / L_K → Σ / Σ*_ε → Σ_agent → GKSL / τ_t / A* → EEG/FPGA/SAAU runtime → admissible categorical closure. Exact finite Σ* reconstruction requires finite Koopman closure and separating observables. In general implementations, Σ*_ε should be used as an approximate effective spectral kernel. Contents:1. UAU_Full_Formalization_Synthesis.md Unified synthesis of the current UAU–ESVG–SAAU–Σ* formalization corpus. 2. UAU_Formalization_Contradictions_Report.md Contradictions, risks, overclaims, and repair rules found in the source corpus. 3. UAU_Sigma_Working_Math_Instrument.md Working mathematical instrument and research protocol for future development. 4. Categorical_Omega_Closure_Formalization.md Formal categorical omega-closure and spectral reconstruction theorem. 5. GKSL_Operator_Field_Formalization.md Operator-algebraic GKSL extension of the framework. 6. Sigma_Physical_Hardware_Formalization.md Formal physical Sigma-star hardware/runtime layer. Keywords:Koopman operator theory; Wasserstein gradient flow; spectral reconstruction; neuromorphic computing; FPGA spectral runtime; GKSL dynamics; Lindblad generator; category theory; dissipative dynamical systems; operator algebra; adaptive systems; Sigma-star; UAU; SAAU; ESVG. Recommended Citation:Bazarov, V. (2026). UAU–ESVG–SAAU–Σ*: Variational–Spectral–Categorical Formalization Package (v0.1-formalization). Zenodo. DOI: [insert DOI after publication] Release Notes:- First public formalization package.- Includes unified synthesis and contradiction audit.- Separates state vector field b = -G∇E from Koopman generator L_K.- Replaces unconditional exact Σ* claims with Σ*_ε where finite Koopman closure is not proved.- Adds formal GKSL/operator-algebraic layer under CP semigroup and faithful invariant-state assumptions.- Adds physical Sigma-star hardware layer with domain, semigroup, spectral, noise, retention, and approximate universality assumptions.- Hardware layer is formal/architectural and not yet HDL-validated.- EEG/FPGA/SAAU layers are treated as measurable or implementable finite effective spectral sectors, not as proofs of consciousness, emergent gravity, or arbitrary universal physical computation. Bazarov, Vitaliy. (VIT-BAZ) (2026). UAU–ESVG–SAAU–Σ*: Variational–Spectral–Categorical Formalization Package (v0.1-formalization). Zenodo. https://doi.org/10.5281/zenodo.20576756","author":[{"family":"Bazarov","given":"Vitaly"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20576755","URL":"https://doi.org/10.5281/zenodo.20576755","source":"datacite"},{"id":"doi:10.5281/zenodo.20576756","type":"article-journal","title":"Variational-Spectral-GKSL Closed Dynamical Theory of a Quantum-Information Field","abstract":"Version:v0.1-formalization Author:Vitaliy Bazarov(VIT-BAZ) ORCID:0009-0007-9967-5807 Status:Preprint / theoretical specification / formal research archive Description:This archive contains the current formalization package for the UAU–ESVG–SAAU–Σ* framework: a layered variational, spectral, categorical, operator-algebraic, and hardware-realizable theory of dissipative adaptive dynamical systems. The package includes a unified formalization synthesis, contradiction report, working mathematical instrument, categorical omega-closure formalization, GKSL operator-field formalization, and physical Sigma-star hardware/runtime layer. The framework should be read as a conditional mathematical and engineering specification. Claims of reconstruction, universality, and physical realization are valid only under the explicit assumptions stated in the documents: well-posed dissipative dynamics, admissible Koopman/GKSL spectral resolution, separating observables, finite effective sector or Σ*_ε approximation, and controlled physical implementation. Scientific Status Statement:This release is a theoretical and formal specification package. It contains conditional mathematical frameworks, proof programs, hardware abstraction layers, and contradiction audits. It is not a claim of completed silicon validation, medical validation, universal physical implementation, or experimentally verified cognition theory. Core Technical Statement:The central structure of the framework is: E(U) → b = -G∇E → Φ_t → K_t / L_K → Σ / Σ*_ε → Σ_agent → GKSL / τ_t / A* → EEG/FPGA/SAAU runtime → admissible categorical closure. Exact finite Σ* reconstruction requires finite Koopman closure and separating observables. In general implementations, Σ*_ε should be used as an approximate effective spectral kernel. Contents:1. UAU_Full_Formalization_Synthesis.md Unified synthesis of the current UAU–ESVG–SAAU–Σ* formalization corpus. 2. UAU_Formalization_Contradictions_Report.md Contradictions, risks, overclaims, and repair rules found in the source corpus. 3. UAU_Sigma_Working_Math_Instrument.md Working mathematical instrument and research protocol for future development. 4. Categorical_Omega_Closure_Formalization.md Formal categorical omega-closure and spectral reconstruction theorem. 5. GKSL_Operator_Field_Formalization.md Operator-algebraic GKSL extension of the framework. 6. Sigma_Physical_Hardware_Formalization.md Formal physical Sigma-star hardware/runtime layer. Keywords:Koopman operator theory; Wasserstein gradient flow; spectral reconstruction; neuromorphic computing; FPGA spectral runtime; GKSL dynamics; Lindblad generator; category theory; dissipative dynamical systems; operator algebra; adaptive systems; Sigma-star; UAU; SAAU; ESVG. Recommended Citation:Bazarov, V. (2026). UAU–ESVG–SAAU–Σ*: Variational–Spectral–Categorical Formalization Package (v0.1-formalization). Zenodo. DOI: [insert DOI after publication] Release Notes:- First public formalization package.- Includes unified synthesis and contradiction audit.- Separates state vector field b = -G∇E from Koopman generator L_K.- Replaces unconditional exact Σ* claims with Σ*_ε where finite Koopman closure is not proved.- Adds formal GKSL/operator-algebraic layer under CP semigroup and faithful invariant-state assumptions.- Adds physical Sigma-star hardware layer with domain, semigroup, spectral, noise, retention, and approximate universality assumptions.- Hardware layer is formal/architectural and not yet HDL-validated.- EEG/FPGA/SAAU layers are treated as measurable or implementable finite effective spectral sectors, not as proofs of consciousness, emergent gravity, or arbitrary universal physical computation. Bazarov, Vitaliy. (VIT-BAZ) (2026). UAU–ESVG–SAAU–Σ*: Variational–Spectral–Categorical Formalization Package (v0.1-formalization). Zenodo. https://doi.org/10.5281/zenodo.20576756","author":[{"family":"Bazarov","given":"Vitaly"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20576756","URL":"https://doi.org/10.5281/zenodo.20576756","source":"datacite"},{"id":"doi:10.5281/zenodo.20555674","type":"article-journal","title":"An Exact-Fixed-Point Reference Benchmark and Validation Methodology for Simulators of Signed k-State Voter Dynamics on Networks","abstract":"Open technical disclosure · v1.0 · 5 June 2026. This document specifies, in fully reproducible and implementation-ready detail, a tiered reference benchmark and an associated validation methodology for simulators of signed k-state voter dynamics on networks, anchored on an exact, seed-independent, machine-precision fixed point established by the companion empirical study (P02_TOPOLOGY_PHASE_MAP, v0.5). It is released for unrestricted public use, study, implementation, modification, and redistribution. The reference quantities, test definitions, data schemas, and reference code given below are placed in the public record so that any implementer, on any substrate, may verify their software against a common, freely available standard. No registration, credential, fee, or grant of permission is required to use anything described herein. Preprint · Reproducibility & Reference Benchmarking in the Statistical Physics of Networks An Exact-Fixed-Point Reference Benchmark and Validation Methodology for Simulators of Signed k-State Voter Dynamics on Networks A freely reusable, implementation-agnostic conformance suite built on a machine-precision analytic invariant, a topology-invariant stress baseline, and a degree-attribution mechanism signature János Gábor Melegh1 1Independent researcher. 5 June 2026· Manuscript ID: P02_BENCHMARK_SUITE· v1.0 · companion to P02_TOPOLOGY_PHASE_MAP v0.5 Abstract Computational studies of stochastic dynamics on networks are difficult to validate because the quantities they produce are usually distributional: a result that is “close enough” can mask a defect in the update rule, the random number stream, the graph generator, the floating-point reduction, or the parallel decomposition. We observe that the companion empirical program establishes a quantity that is not merely close but exact: for a per-node local stress Ti and a per-node local admissibility ψi, the across-node Spearman rank correlation H3 = ρS(T, ψ) equals exactly −1 to machine precision on every degree-regular ensemble, independent of size, negative-edge fraction, and random seed (sample standard deviation of order 10−16). An exact, seed-independent invariant is the ideal anchor for a software conformance test, because a correct implementation must reproduce it to floating-point tolerance while almost any category of defect breaks it. Around this anchor we define a four-tier benchmark suite: Tier 0 exact/analytic tests (machine precision, seed-free), Tier 1 distributional baselines with calibrated tolerance bands (including the topology-invariant mean stress ⟨T⟩ = 0.324, Kruskal–Wallis p = 1.00), Tier 2 mechanism-attribution signatures (mean-degree variance dominance: random-forest importance 0.94, cross-validated R2 = 0.99, collapse classifier AUC 0.998), and Tier 3 perturbation-response signatures (matched-degree residual ≈ 0.19; a sharp degree–collapse boundary; and a sign-and-order-of-magnitude asymmetric bistability under degree-targeted perturbation). We give the precise observable contract, canonical generator specifications, a portable on-disk manifest schema, reference pseudocode, and acceptance criteria; we then enumerate, in deliberate breadth, the practical contexts in which the suite can be used — continuous-integration regression testing, cross-language and cross-substrate conformance, floating-point and reduction-order auditing, random-stream verification, GPU/distributed/out-of-core reproducibility, differentiable and surrogate-model validation, hardware/accelerator/neuromorphic/probabilistic substrate qualification, mean-field and closure-solver validation, teaching and assessment, peer-review artifact evaluation and conformance certification, and applied multi-agent/swarm/IoT consensus-controller qualification — and we describe how the same construction generalizes to other dynamics, observables, alphabet sizes, and graph families. The entire specification is placed in the public record for free use. Keywords — reference benc","author":[{"family":"Melegh","given":"Janos"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20555674","URL":"https://doi.org/10.5281/zenodo.20555674","source":"datacite"},{"id":"doi:10.5281/zenodo.20555673","type":"article-journal","title":"An Exact-Fixed-Point Reference Benchmark and Validation Methodology for Simulators of Signed k-State Voter Dynamics on Networks","abstract":"Open technical disclosure · v1.0 · 5 June 2026. This document specifies, in fully reproducible and implementation-ready detail, a tiered reference benchmark and an associated validation methodology for simulators of signed k-state voter dynamics on networks, anchored on an exact, seed-independent, machine-precision fixed point established by the companion empirical study (P02_TOPOLOGY_PHASE_MAP, v0.5). It is released for unrestricted public use, study, implementation, modification, and redistribution. The reference quantities, test definitions, data schemas, and reference code given below are placed in the public record so that any implementer, on any substrate, may verify their software against a common, freely available standard. No registration, credential, fee, or grant of permission is required to use anything described herein. Preprint · Reproducibility & Reference Benchmarking in the Statistical Physics of Networks An Exact-Fixed-Point Reference Benchmark and Validation Methodology for Simulators of Signed k-State Voter Dynamics on Networks A freely reusable, implementation-agnostic conformance suite built on a machine-precision analytic invariant, a topology-invariant stress baseline, and a degree-attribution mechanism signature János Gábor Melegh1 1Independent researcher. 5 June 2026· Manuscript ID: P02_BENCHMARK_SUITE· v1.0 · companion to P02_TOPOLOGY_PHASE_MAP v0.5 Abstract Computational studies of stochastic dynamics on networks are difficult to validate because the quantities they produce are usually distributional: a result that is “close enough” can mask a defect in the update rule, the random number stream, the graph generator, the floating-point reduction, or the parallel decomposition. We observe that the companion empirical program establishes a quantity that is not merely close but exact: for a per-node local stress Ti and a per-node local admissibility ψi, the across-node Spearman rank correlation H3 = ρS(T, ψ) equals exactly −1 to machine precision on every degree-regular ensemble, independent of size, negative-edge fraction, and random seed (sample standard deviation of order 10−16). An exact, seed-independent invariant is the ideal anchor for a software conformance test, because a correct implementation must reproduce it to floating-point tolerance while almost any category of defect breaks it. Around this anchor we define a four-tier benchmark suite: Tier 0 exact/analytic tests (machine precision, seed-free), Tier 1 distributional baselines with calibrated tolerance bands (including the topology-invariant mean stress ⟨T⟩ = 0.324, Kruskal–Wallis p = 1.00), Tier 2 mechanism-attribution signatures (mean-degree variance dominance: random-forest importance 0.94, cross-validated R2 = 0.99, collapse classifier AUC 0.998), and Tier 3 perturbation-response signatures (matched-degree residual ≈ 0.19; a sharp degree–collapse boundary; and a sign-and-order-of-magnitude asymmetric bistability under degree-targeted perturbation). We give the precise observable contract, canonical generator specifications, a portable on-disk manifest schema, reference pseudocode, and acceptance criteria; we then enumerate, in deliberate breadth, the practical contexts in which the suite can be used — continuous-integration regression testing, cross-language and cross-substrate conformance, floating-point and reduction-order auditing, random-stream verification, GPU/distributed/out-of-core reproducibility, differentiable and surrogate-model validation, hardware/accelerator/neuromorphic/probabilistic substrate qualification, mean-field and closure-solver validation, teaching and assessment, peer-review artifact evaluation and conformance certification, and applied multi-agent/swarm/IoT consensus-controller qualification — and we describe how the same construction generalizes to other dynamics, observables, alphabet sizes, and graph families. The entire specification is placed in the public record for free use. Keywords — reference benc","author":[{"family":"Melegh","given":"Janos"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20555673","URL":"https://doi.org/10.5281/zenodo.20555673","source":"datacite"},{"id":"doi:10.25560/128198","type":"article-journal","title":"Novel scanning probe methods for manipulating and characterising pentacene thin films and crystals","abstract":"Molecular semiconductors are rapidly emerging as a key enabling technology in fields such as quantum sensing, neuromorphic computing and energy generation. Their flexibility, tunability and low-cost fabrication make them promising materials for the highly-integrated electronic architectures of future technology. This thesis devises strategies to control and optimise the functional properties of molecular thin films and crystals for organic electronic applications. With charge carrier mobility comparable to amorphous silicon, pentacene is one of the most well-researched molecular semiconductors, both as a model system and for its own technological utility. Like other small-molecule semiconductors, the functional properties of pentacene such are highly orientation-dependent, which currently limits its applications in devices such as photovoltaics. Strongly-interacting substrates can be used to template the molecular orientation of thin films and improve charge transport and light-matter interactions. This thesis identifies poled ferroelectric polymer thin films as effective templating layers for pentacene. Molecular orientation is correlated to optoelectronic properties and a higher out-of-plane mobility is demonstrated for the flat-lying (edge-on) pentacene compared to the upright (end-on) orientation. In the pursuit of high-performance nanoscale devices, the efficacy of oxidation lithography is demonstrated for defining nanoscale conductive channels in pentacene thin films and crystals. Scanning probe methods based on atomic force microscopy provide the nanoscale imaging capability, sensitivity and versatility to shed light on the properties of these functional materials. This thesis applies advanced modes such as piezoresponse force microscopy, Kelvin probe force microscopy, conductive AFM and local anodic oxidation to investigate both pentacene and the ferroelectric polymer poly(vinylidene fluoride–trifluoroethylene). Developing reliable methods to compare properties on the nanoscale, microscale and macroscale is of paramount importance for the optimisation of organic electronic device materials. A robust multi-modal approach is developed including scanning probe microscopy, spectroscopy and in-situ x-ray scattering to explore orientation, order and performance of molecular semiconductors.","author":[{"family":"Bryan","given":"Emma"}],"issued":{"date-parts":[[2025]]},"DOI":"10.25560/128198","URL":"https://doi.org/10.25560/128198","source":"datacite"},{"id":"doi:10.11591/ijece.v15i6.pp5173-5182","type":"article-journal","title":"Improving time-domain winner-take-all circuit for neuromorphic computing systems","abstract":"With the rapid advancements of information processing systems, winner- take-all (WTA) circuits have emerged as essential components in a wide range of cognitive functions and decision-making applications. Neuromorphic computing systems, inspired by the biological brain, utilize WTA circuits as selective mechanisms that identify and retain the strongest signal while suppressing all others. In this study, we present an effective time-domain WTA circuit with optimized multiple-input NOT AND (NAND) gate and delay circuit for neuromorphic computing applications. The circuit is evaluated using sinusoidal current inputs with varying phase delays, which successfully demonstrating precise winner selection. When applied to neuromorphic image recognition task, the enhanced time-domain WTA achieves an improvement of 0.2% in precision while significantly reducing power consumption, yielding a low figure of merit (FoM) of 0.03 µW/MHz, compared to the previous study with FoM of 0.25 µW/MHz. The optimized WTA circuit is highly promising for large-scale neuromorphic applications.","author":[{"family":"Truong","given":"Son"},{"family":"Ngo","given":"Tu"}],"issued":{"date-parts":[[2025]]},"DOI":"10.11591/ijece.v15i6.pp5173-5182","URL":"https://doi.org/10.11591/ijece.v15i6.pp5173-5182","source":"crossref"},{"id":"doi:10.31613/ceramist.2025.00038","type":"article-journal","title":"Recent Advances in Next-Generation Electrochemical Ionic Synapse for Neuromorphic Computing","abstract":"Neuromorphic computing, inspired by the structure and function of biological neural networks, offers a promising alternative to traditional von Neumann architectures by integrating memory and computation for enhanced energy efficiency and processing speed. Electrochemical ionic synapses (EIS) have emerged as a next-generation memory technology, leveraging the reversible insertion and extraction of mobile ions to modulate conductance. Among various implementations, lithium-ion and proton-based EIS devices have gained attention due to their high tunability, fast response, and energy efficiency. This review delves into recent advancements in EIS technology. Lithium-ion-based EIS devices, employing transition metal oxides such as LiCoO2 and LixTiO2, demonstrate stable and scalable synaptic behavior. Meanwhile, proton-based EIS, utilizing materials like WO3 and MoO3, offer rapid ion transport and complementary metal-oxide-semiconductor compatibility, paving the way for practical neuromorphic hardware. We discuss the critical challenges related to material selection, retention stability, and large-scale integration, while exploring future directions for advancing EIS-based computing. By providing insights into the underlying mechanisms and engineering strategies, this review aims to support the development of energy-efficient and high-performance neuromorphic systems, bringing artificial intelligence closer to the capabilities of the human brain.","author":[{"family":"Kim","given":"Hanju"},{"family":"Jung","given":"Woo"}],"issued":{"date-parts":[[2025]]},"DOI":"10.31613/ceramist.2025.00038","URL":"https://doi.org/10.31613/ceramist.2025.00038","source":"crossref"},{"id":"doi:10.71465/csb199","type":"article-journal","title":"Energy-Efficient Neuromorphic Computing with Spiking Neural Networks on Edge Devices: A Dynamic Programming Approach","abstract":"The proliferation of Internet of Things devices has necessitated a paradigm shift from centralized cloud computing to edge computing, primarily to mitigate latency and bandwidth constraints. However, deploying complex deep learning models on resource-constrained edge devices remains a significant challenge due to power limitations. Neuromorphic computing, specifically utilizing Spiking Neural Networks, offers a promising solution by mimicking the event-driven, sparse processing capabilities of the biological brain. Despite their theoretical energy efficiency, the optimal mapping of Spiking Neural Networks onto hardware architectures with limited cores and memory is a non-trivial problem, often resulting in suboptimal energy utilization when heuristic methods are employed. This paper proposes a novel framework that leverages Dynamic Programming to optimize the placement and scheduling of spiking neurons on multicoreneuromorphic processors. By formulating the mapping problem as a multi-stage decision process, our approach systematically balances the trade-offs between inter-core communication costs and computational load balancing. We demonstrate that the proposed method significantly reduces energy consumption compared to standard greedy and random mapping strategies while maintaining high inference accuracy. The results underscore the viability of Dynamic Programming as a deterministic tool for enhancing the operational efficiency of next-generation edge intelligence.","author":[{"family":"Miller","given":"Robert"},{"family":"Bennett","given":"Sarah"}],"issued":{"date-parts":[[2026]]},"DOI":"10.71465/csb199","URL":"https://doi.org/10.71465/csb199","source":"crossref"},{"id":"doi:10.64229/q5g7dh32","type":"article-journal","title":"Neuromorphic Computing-Enabled Digital Twin Framework for Sustainable IT Supply Chain Integration in Smart Urban Ecosystems","abstract":"The convergence of neuromorphic computing, digital twin technologies, and sustainable supply chain management presents unprecedented opportunities for transforming urban cyber-physical systems. This viewpoint paper introduces a novel conceptual framework that integrates neuromorphic computing architectures with digital twin methodologies to optimize sustainable IT supply chain operations within smart urban ecosystems. The proposed framework addresses critical issues related to real-time data processing, energy-efficient computation, and adaptive decision-making for green technology adoption across complex urban supply networks. Through theoretical analysis and conceptual modeling, we demonstrate how brain-inspired computing paradigms can enhance digital twin capabilities for monitoring, predicting, and optimizing supply chain sustainability metrics. The framework incorporates spatiotemporal knowledge graph embeddings, hybrid intelligence systems, and circular economy principles to create responsive, self-adapting supply chain networks. Our approach offers significant implications for urban planners, supply chain managers, and technology implementers seeking to advance computational sustainability science. The integration of neuromorphic processing units with digital twin architectures enables unprecedented energy efficiency improvements of up to 1000x compared to traditional computing approaches while maintaining real-time responsiveness for critical supply chain decisions. This research plays role in emerging urban sector computational sustainability by contributing a foundational framework for next-generation smart city supply chain management systems.","author":[{"family":"Tathavadekar","given":"Viraj"},{"family":"Mahankale","given":"Nitin"}],"issued":{"date-parts":[[2025]]},"DOI":"10.64229/q5g7dh32","URL":"https://doi.org/10.64229/q5g7dh32","source":"crossref"},{"id":"doi:10.62311/nesx/rp2-30042026","type":"article-journal","title":"Algebraic Quantum-AI Co-Design of Energy-Adaptive Semiconductor Architectures for Edge Intelligence and Neuromorphic Computing","abstract":"Abstract This study investigates the algebraic co-design of quantum-AI abstractions, energy-adaptive semiconductor architectures, and neuromorphic computing pathways for edge intelligence. The core problem addressed is the growing mismatch between the computational intensity of modern inference workloads and the strict power, thermal, and memory limits of edge devices that must operate continuously in industrial, mobile, and cyber-physical settings. Existing edge-AI platforms provide high throughput, yet they often rely on dense data movement, externally managed memory hierarchies, and power envelopes that complicate always-on deployment. In parallel, neuromorphic platforms show promising event-driven sparsity and very low inference power, but the design space remains fragmented across device physics, architecture, workload shape, and deployment objective. The present paper develops an algebraic quantum-AI co-design framework that formalises this design space and links architecture selection to measurable energy-performance relations. Real data were collected from official NVIDIA Jetson specifications, the Hailo-8 product brief, the MLPerf Tiny benchmark suite, the Intel Loihi 2 technology brief, and published neuromorphic benchmark evidence including NeuroBench and a BrainChip comparative report. No synthetic hardware values were introduced. The analysis combines explicit equations for dynamic power, event-driven energy, energy-delay utility, and a constrained architecture-selection objective with tables and figures derived from the collected corpus. The results show a clear division of labour across edge semiconductors: high-throughput Jetson-class modules dominate dense multi-stream inference, Hailo-8 offers a strong low-power dataflow alternative, and neuromorphic designs offer the strongest always-on inference advantage when sparse event streams and adaptive signalling dominate. The study concludes that future edge intelligence systems will benefit from heterogeneous co-designed stacks in which dense accelerators, memory-local dataflow engines, and neuromorphic coprocessors are orchestrated through algebraic runtime policies rather than treated as isolated hardware categories. Keywords: quantum-AI co-design; energy-adaptive semiconductors; edge intelligence; neuromorphic computing; Loihi 2; Akida; Hailo-8; Jetson Orin; MLPerf Tiny; algebraic architecture optimisation","author":[{"family":"Pasupuleti","given":"Murali"}],"issued":{"date-parts":[[2026]]},"DOI":"10.62311/nesx/rp2-30042026","URL":"https://doi.org/10.62311/nesx/rp2-30042026","source":"crossref"},{"id":"doi:10.70675/399bee3cz14e7z4cbfz94feze902eb9ecb6c","type":"article-journal","title":"Neuromorphic vision combining events and frames","abstract":"Vision neuromorphique combinant événements et images L'intégration des machines avancées dans la vie quotidienne devient progressivement indispensable. Un élément clé de cette intégration réside dans la capacité de ces systèmes à percevoir, comprendre et naviguer de manière autonome. Parmi les modalités sensorielles, la vision occupe une place centrale, offrant une représentation riche et détaillée du monde environnant. Déployer la perception visuelle dans des systèmes autonomes tels que robots, drones ou véhicules impose cependant un compromis entre précision, consommation énergétique et latence. Les caméras classiques à trames, combinées à l'apprentissage profond, atteignent de bonnes performances mais présentent des limitations : faible résolution temporelle, flou de mouvement, saturation en forte luminosité et forte consommation de ressources. Cette thèse propose de surmonter ces limites en combinant deux concepts inspirés biologiquement : i) les réseaux de neurones à impulsions (SNN) et ii) les caméras à événements. Les SNN utilisent des impulsions asynchrones, offrant une alternative économe en énergie aux réseaux classiques. Les caméras à événements, inspirées de la rétine biologique, capturent les variations de luminosité de manière asynchrone avec précision microseconde, grande plage dynamique et faible consommation. Cependant, en plus du manque d'informations de texture, essentielles pour l'identification des objets, les méthodes de vision par ordinateur restent optimisées pour des données denses basées sur des trames. Pour combler cette lacune, nous proposons de fusionner les deux caméras, afin de tirer parti de leurs complémentarités. Nous explorons cette fusion pour la segmentation sémantique. Les flux d'événements sont convertis en pseudo-trames traitées par un SNN. Cette approche dépasse les méthodes existantes mais reste limitée. Pour l'améliorer, nous introduisons la distillation des connaissances : un réseau neuronal conventionnel entraîné sur des images fusionnées avec des événements transfère les caractéristiques apprises au SNN traitant les événements. Cette stratégie résout efficacement le flou et l'éblouissement inhérents aux caméras à trames. Ensuite, nous exploitons la densité d'événements comme une forme d'attention visuelle, en concentrant le calcul sur les régions de trames présentant une forte activité événementielle pour améliorer l'efficacité et réduire la latence. Enfin, nous proposons un cadre de segmentation inter-trames qui tire parti de la haute résolution temporelle des événements. En exploitant les événements déclenchés entre les trames, nous capturons le mouvement sur différents intervalles temporels et l'interpolons avec les segmentations basées sur les trames. Cette thèse démontre que la fusion de la vision basée sur les événements et sur les trames avec des architectures SNN permet une perception performante, efficace et à faible latence, ouvrant la voie à des pipelines neuromorphiques pratiques pour les systèmes autonomes du futur.","author":[{"family":"Hareb","given":"Dalia"}],"issued":{"date-parts":[[2026]]},"DOI":"10.70675/399bee3cz14e7z4cbfz94feze902eb9ecb6c","URL":"https://doi.org/10.70675/399bee3cz14e7z4cbfz94feze902eb9ecb6c","source":"crossref"},{"id":"doi:10.70675/16013f38zb45dz4205za1d3zc1e15e7f5ff0","type":"article-journal","title":"Neuromorphic photonic systems for information processing","abstract":"Systèmes photoniques neuro-morphiques pour le traitement de l'information Par une utilisation performante de nombreux algorithmes dont les réseaux neuronaux, l'intelligence artificielle révolutionne le développement de la société numérique. Néanmoins, la tendance actuelle dépasse les limites prédites par la loi de Moore et celle de Koomey, ce qui implique des limitations éventuelles des implémentations numériques de ces systèmes. Pour répondre plus efficacement aux besoins calculatoires spécifiques de cette révolution, des systèmes physiques innovants tentent en amont d'apporter des solutions, nommées \"neuro-morphiques\" puisqu'elles imitent le fonctionnement des cerveaux biologiques. Les systèmes existants sont basés sur des techniques dites de \"Reservoir Computing\" ou \"coherent Ising Machine.\" Leurs versions photoniques, ont permis de démontrer l'intérêt de ces techniques notamment pour la reconnaissance vocale avec un état de l'art en 2017 attestant de bonnes performances en termes de reconnaissance à un rythme d'1 million de mots par seconde. Nous proposons dans un premier temps une technique d'ajustement automatique des hyperparamètres pour le \"Reservoir Computing\", accompagnée d'une étude théorique de convergence. Nous proposons ensuite une solution au problème de la détection précoce de la maladie d'Alzheimer de type \"Reservoir Computing\" optoélectronique. En plus des taux de classifications obtenus meilleurs que l'état de l'art, une étude complète du compromis coût énergétique performance démontre la validité de cette approche. Enfin, le problème de la restauration d'image par maximum de vraisemblance est abordé à l'aide d'une implémentation optoélectronique appropriée de type \"coherent Ising Machine\".","author":[{"family":"Mwamsojo","given":"Nickson"}],"issued":{"date-parts":[[2026]]},"DOI":"10.70675/16013f38zb45dz4205za1d3zc1e15e7f5ff0","URL":"https://doi.org/10.70675/16013f38zb45dz4205za1d3zc1e15e7f5ff0","source":"crossref"},{"id":"doi:10.1088/1361-6463/ae2c9c","type":"article-journal","title":"Nanoscale resistive switching in mixed-valence CuO\n                    <i>\n                      <sub>x</sub>\n                    </i>\n                    memristors for neuromorphic sensing–computing applications","abstract":"Abstract Memristive devices are pivotal for the next generation of neuromorphic computing systems due to their ability to mimic biological synapses. This study investigates the nanoscale resistive switching (RS) characteristics of copper oxide (CuO x ) memristors, employing conducting atomic force microscopy (cAFM) to probe the fundamental properties of a single device junction. Our research reveals that these devices exhibit robust, counter-eight-wise bipolar RS with a high on/off ratio of ∼100 at low operating voltages (±2 V) for more than 500 cycles. Beyond memory, nanoscale memristors successfully replicate key synaptic functions essential for brain-inspired computing, including potentiation, depression, spike-rate-dependent plasticity, and paired-pulse facilitation − enabled by their analog RS. Furthermore, nociceptor-like responses such as threshold detection, relaxation, and sensitization are also realized, highlighting their multifunctional neuromorphic capabilities. The devices demonstrate a very low power consumption under a single synaptic event (∼20 pJ for potentiation and ∼8 pJ for depression). The observed bipolar resistance modulation is attributed to the modulation of the Schottky barrier at the cAFM tip/CuO x nanoscale junction, driven by ionic migration under electrical bias. These findings establish CuO x as a promising material platform for energy-efficient memory and multifunctional neuromorphic devices. Furthermore, the emulation of synaptic and nociceptive behaviors within a single device provides versatile building blocks for next-generation sensing–computing architectures.","author":[{"family":"Mandal","given":"Rupam"},{"family":"Som","given":"Tapobrata"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1088/1361-6463/ae2c9c","URL":"https://doi.org/10.1088/1361-6463/ae2c9c","source":"crossref"},{"id":"doi:10.1038/s42256-025-01143-2","type":"article-journal","title":"Solving sparse finite element problems on neuromorphic hardware","abstract":"Abstract The finite element method (FEM) is one of the most important and ubiquitous numerical methods for solving partial differential equations (PDEs) on computers for scientific and engineering discovery. Applying the FEM to larger and more detailed scientific models has driven advances in high-performance computing for decades. Here we demonstrate that scalable spiking neuromorphic hardware can directly implement the FEM by constructing a spiking neural network that solves the large, sparse, linear systems of equations at the core of the FEM. We show that for the Poisson equation, a fundamental PDE in science and engineering, our neural circuit achieves meaningful levels of numerical accuracy and close to ideal scaling on modern, inherently parallel and energy-efficient neuromorphic hardware, specifically Intel’s Loihi 2 neuromorphic platform. We illustrate extensions to irregular mesh geometries in both two and three dimensions as well as other PDEs such as linear elasticity. Our spiking neural network is constructed from a recurrent network model of the brain’s motor cortex and, in contrast to black-box deep artificial neural network-based methods for PDEs, directly translates the well-understood and trusted mathematics of the FEM to a natively spiking neuromorphic algorithm.","author":[{"family":"Theilman","given":"Bradley"},{"family":"Aimone","given":"James"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1038/s42256-025-01143-2","URL":"https://doi.org/10.1038/s42256-025-01143-2","source":"crossref"},{"id":"doi:10.55041/ijsmt.v2i6.053","type":"article-journal","title":"Neuromorphic Computing-Based Real-Time EEG Epileptic Seizure Detection Using Spiking Neural Networks","abstract":"Epilepsy is a long-term neurological condition marked by recurring seizures that affect a large population across the globe. Detecting seizures in real time is particularly challenging because electroencephalogram (EEG) signals exhibit highly dynamic, non-linear, and patient-specific behavior. Conventional machine learning and deep learning approaches, although effective in certain scenarios, often demand significant computational resources and power, which limits their applicability in continuous monitoring systems. This study proposes a framework based on neuromorphic computing to overcome these challenges, enabling real-time detection of epileptic seizures through the use of spiking neural networks (SNNs). The proposed approach adopts an event-driven processing mechanism inspired by biological neural systems, allowing efficient handling of temporal EEG data. A spike encoding method is utilized to transform continuous EEG signals into distinct spike sequences, allowing them to be compatible with neuromorphic systems. The proposed system is designed to enhance detection performance while reducing latency and energy consumption. By combining temporal signal representation with low-power computation, the framework offers a promising approach for applications in real-time healthcare. This study demonstrates how neuromorphic computing can contribute to the development of next-generation intelligent monitoring systems for neurological disorders.","author":[{"family":"Raju","given":"Machiraju"},{"family":"Babu","given":"GABG"}],"issued":{"date-parts":[[2026]]},"DOI":"10.55041/ijsmt.v2i6.053","URL":"https://doi.org/10.55041/ijsmt.v2i6.053","source":"crossref"},{"id":"doi:10.70675/78b51dcaze575z4535zad33zaa56bebca7eb","type":"article-journal","title":"MRAM based neuromorphic cell for Artificial Intelligence","abstract":"Cellule MRAM neuromorphique pour l'Intelligence Artificielle La constante évolution de la société en termes de besoin computationnel tend vers les applications d'intelligence artificiel qui étaient autrefois réservés exclusivement à l'Homme. Le développement d'algorithme reposant sur l'apprentissage profond a rendu possible le dépassement des capacités du cerveau humain dans ces domaines de prédilections comme la reconnaissance d'images et de paroles, la prise de décision et les problèmes d'optimisation. Pourtant, les réseaux de neurones artificiels capable de résoudre ces problèmes repose de manière intensive sur d'énorme quantité de donnés et requiert de nombreuses opérations en parallèle. Quand ces algorithmes tournent sur des architectures classiques de type Van Neumann où l'unité de calcul est séparée de l'unité mémoire, la latence du réseau et son énergie consommée augmente exponentiellement avec sa taille. De ce constat, la communauté scientifique commença a développé des schémas informatiques inspiré du cerveau afin de surpasser les limitations observées. En particulier, les réseaux de neurones impulsionnels ont été prédit par W. Maass en 1997 comme étant un candidat adapté à la maximisation du dispersement des opérations dans le réseau tout en démontrant des performances égales, sinon meilleures, aux premières générations de réseaux de neurones. Jusqu'à maintenant, des circuits intégrés orienté vers une application donnée (ASICs) ont été développé par des industriels tel que Intel ou IBM pour émuler des réseaux de neurones impulsionnels, se basant sur des technologies courantes CMOS. Le nombre de transistors nécessaire pour accomplir certaines fonctionnalités critique propre aux réseaux de neurones, comme les neurones impulsionnels, sont toujours très important, ce qui n'est pas souhaitable dans une stratégie de réduction de miniaturisation poussées.Dans ce contexte, de nouvelles solutions ont été proposé pour reproduire les caractéristiques synaptiques et neuronales tout essayant de réduire un maximum l'empreinte physique et la consommation énergétique. En outre, différent type de nano-synapses reposant sur des mémoires émergeantes non-volatile ont exploré des poids synaptiques multi-niveaux ainsi que la reproduction des fonctions mémoires court-terme et long-terme. Parmi elles, les solutions basées sur une technologie spintronique sont celle les plus matures d'un point de vue industriel puisque les mémoires à accès direct magnétique (MRAM), pouvant servir de synapses binaires, ont déjà atteint le marché international depuis une dizaine année. Pourtant, l'existence d'un neurone impulsionnel compatible avec une synapses spintroniques est toujours manquant dans la littérature actuelle.Cette thèse prend donc sa place dans ce contexte de développement de nouvelles solutions spintroniques afin d'émuler un neurone impulsionnel. Les jonctions tunnels magnétiques (JTMs) largement utilisées en spintronique pour leur endurance en tant que mémoire tout en démontrant des énergies d'écriture et de lecture très faible ainsi qu'une compatibilité \"back-end of line\" et CMOS. La solution élaborée dans ce présent manuscrit prends l'avantage des JTMs pour concevoir un neurone impulsionnel basé sur la dynamique \"windmill\". Le concept de JTM à double couche libre est modélisé, nano fabriqué et caractérisé électriquement pour donner une perspective constructive sur son potentiel utilisation en tant que neurone impulsionnel.","author":[{"family":"Farcis","given":"Louis"}],"issued":{"date-parts":[[2026]]},"DOI":"10.70675/78b51dcaze575z4535zad33zaa56bebca7eb","URL":"https://doi.org/10.70675/78b51dcaze575z4535zad33zaa56bebca7eb","source":"crossref"},{"id":"doi:10.58723/ijsei.v2i1.151","type":"article-journal","title":"Neuromorphic Computing Chips for Edge AI: A Comprehensive Analysis of Brain-Inspired Hardware Architecture for Real-Time Intelligent Systems","abstract":"Background of study: Edge computing devices like autonomous robots and IoT sensors need sophisticated AI for real-time decisions, but conventional processors consume 15-300 watts during inference, creating critical limitations for battery-powered deployments. GPU-based accelerators face memory bottlenecks and high energy costs from data movement, making sustained autonomous operation impractical. Aims of paper: This research compares neuromorphic platforms (Intel Loihi 2, IBM TrueNorth, BrainChip Akida) against conventional accelerators (NVIDIA Jetson, Google Coral) to evaluate if neuromorphic architectures can solve edge AI energy efficiency challenges across five representative workloads. Methods: Using an experimental design with hardware benchmarking and power analysis, we evaluated five edge AI workloads. ANOVA and regression modeling were then applied to rigorously compare computing paradigms while controlling for variables. Result: Neuromorphic platforms demonstrated 15-50× improved energy efficiency versus conventional GPU accelerators for event-driven workloads. Intel Loihi 2 achieved 2,400 inferences/joule at 1.8 watts versus 180 inferences/joule at 18.5 watts for NVIDIA Jetson. IBM TrueNorth operated at 70 milliwatts for pattern recognition. BrainChip Akida achieved 94.6% accuracy on keyword spotting at 0.8 watts. Event-driven processing exhibited 0.4ms latency versus 5.1ms for frame-based systems. Neuromorphic chips maintained stable performance without active cooling below 65°C, while conventional accelerators required thermal management above 85°C. Conclusion: Neuromorphic processors (0.6-5W) excel in power-efficient edge AI for event-driven data. While hybrid architectures optimize performance, adoption is hindered by immature software ecosystems, limited training frameworks, and a 2-4% accuracy gap compared to conventional methods.","author":[{"family":"Sathio","given":"Anwar"},{"family":"Maheshwari","given":"Chiragh"}],"issued":{"date-parts":[[2026]]},"DOI":"10.58723/ijsei.v2i1.151","URL":"https://doi.org/10.58723/ijsei.v2i1.151","source":"crossref"},{"id":"doi:10.5281/zenodo.20313736","type":"article-journal","title":"Adaptive Dissipative Attractor Dynamics (ADAD)","abstract":"Adaptive Dissipative Attractor Dynamics (ADAD)Adaptive Dissipative Attractor Dynamics (ADAD) is a mathematical and computational framework describing adaptive neural-field dynamics with dissipative homeostatic regulation, stochastic activity control, and energy-driven attractor stabilization. The framework models collective spiking activity as a coupled dynamical system in which a global adaptive scale parameter R(t)R(t)R(t) evolves in response to population activity fluctuations. The system combines threshold-based spiking dynamics, mean-field activity feedback, dissipative second-order attractor evolution, and stochastic perturbations into a unified variational structure. At the core of the model is an effective attractor energy functional: E(R,ρ)=12R2−(ρ2+c)log⁡R,E(R,\\rho) = \\frac12R^2 - (\\rho^2+c)\\log R,E(R,ρ)=21R2−(ρ2+c)logR, where: RRR is an adaptive attractor radius or gain-scale parameter, ρ\\rhoρ is collective neural activity, c>0c>0c>0 is a regularization constant. The adaptive variable R(t)R(t)R(t) evolves according to a damped second-order gradient-flow system: R¨+γR˙+ϵ∂RE(R,ρ)=0,\\ddot R + \\gamma \\dot R + \\epsilon \\partial_R E(R,\\rho) = 0,R¨+γR˙+ϵ∂RE(R,ρ)=0, which yields stable dissipative convergence toward equilibrium attractor states: R∗(ρ)=ρ2+c.R_*(\\rho)=\\sqrt{\\rho^2+c}.R∗(ρ)=ρ2+c. The framework admits: Lyapunov stability structure, stochastic extensions, Fokker–Planck formulations, McKean–Vlasov mean-field limits, fixed-point neuromorphic implementation. The model is designed for low-power adaptive neuromorphic systems and supports integer/fixed-point computation compatible with embedded hardware and spiking neural accelerators. ADAD does not model quantum systems, cosmology, or consciousness. The framework should be interpreted strictly as a class of adaptive stochastic dynamical systems for homeostatic neural regulation and attractor-based computation. Main Features Adaptive attractor stabilization Dissipative energy minimization Mean-field activity regulation Stochastic adaptive dynamics Homeostatic feedback control Fixed-point neuromorphic implementation Hardware-safe saturation regulation Activity fluctuation adaptation Lyapunov-stable evolution McKean–Vlasov-compatible formulation Mathematical Foundations The framework combines methods from: dissipative dynamical systems, stochastic differential equations, McKean–Vlasov mean-field theory, Wasserstein and variational dynamics, adaptive control theory, neuromorphic computation, nonlinear attractor systems. Potential Applications Neuromorphic Computing Adaptive spiking control for low-power neuromorphic processors and edge AI systems. Embedded AI Energy-efficient adaptive regulation in constrained hardware environments. Adaptive Control Systems Real-time feedback stabilization under stochastic conditions. Spiking Neural Networks Dynamic threshold adaptation and attractor stabilization in recurrent neural architectures. Mean-Field Neural Modeling Population-level adaptive dynamics and collective activity regulation. Stochastic Optimization Noise-regulated adaptive parameter evolution in nonlinear systems. Research Status The framework currently represents: a formal mathematical specification, a computational dynamical model, a research-stage adaptive systems architecture. The following components are mathematically established: dissipative stability structure, Lyapunov energy decay, bounded adaptive dynamics, stochastic extension formulation, fixed-point implementation scheme. The following remain open research problems: rigorous propagation-of-chaos proof, global mean-field convergence, ergodicity analysis, spectral stability theory, large-scale PDE limits, hardware convergence guarantees. Scientific Positioning ADAD should be classified as: adaptive stochastic dynamics, dissipative attractor theory, mean-field neural dynamics, neuromorphic adaptive control, variational adaptive systems. It should not be interpreted as: a physical theory, a cosmological fram","author":[{"family":"Bazarov","given":"Vitaly"},{"family":"Bazarov","given":"Vitaliy"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20313736","URL":"https://doi.org/10.5281/zenodo.20313736","source":"datacite"},{"id":"doi:10.5281/zenodo.20313737","type":"article-journal","title":"Adaptive Dissipative Attractor Dynamics (ADAD)","abstract":"Adaptive Dissipative Attractor Dynamics (ADAD)Adaptive Dissipative Attractor Dynamics (ADAD) is a mathematical and computational framework describing adaptive neural-field dynamics with dissipative homeostatic regulation, stochastic activity control, and energy-driven attractor stabilization. The framework models collective spiking activity as a coupled dynamical system in which a global adaptive scale parameter R(t)R(t)R(t) evolves in response to population activity fluctuations. The system combines threshold-based spiking dynamics, mean-field activity feedback, dissipative second-order attractor evolution, and stochastic perturbations into a unified variational structure. At the core of the model is an effective attractor energy functional: E(R,ρ)=12R2−(ρ2+c)log⁡R,E(R,\\rho) = \\frac12R^2 - (\\rho^2+c)\\log R,E(R,ρ)=21R2−(ρ2+c)logR, where: RRR is an adaptive attractor radius or gain-scale parameter, ρ\\rhoρ is collective neural activity, c>0c>0c>0 is a regularization constant. The adaptive variable R(t)R(t)R(t) evolves according to a damped second-order gradient-flow system: R¨+γR˙+ϵ∂RE(R,ρ)=0,\\ddot R + \\gamma \\dot R + \\epsilon \\partial_R E(R,\\rho) = 0,R¨+γR˙+ϵ∂RE(R,ρ)=0, which yields stable dissipative convergence toward equilibrium attractor states: R∗(ρ)=ρ2+c.R_*(\\rho)=\\sqrt{\\rho^2+c}.R∗(ρ)=ρ2+c. The framework admits: Lyapunov stability structure, stochastic extensions, Fokker–Planck formulations, McKean–Vlasov mean-field limits, fixed-point neuromorphic implementation. The model is designed for low-power adaptive neuromorphic systems and supports integer/fixed-point computation compatible with embedded hardware and spiking neural accelerators. ADAD does not model quantum systems, cosmology, or consciousness. The framework should be interpreted strictly as a class of adaptive stochastic dynamical systems for homeostatic neural regulation and attractor-based computation. Main Features Adaptive attractor stabilization Dissipative energy minimization Mean-field activity regulation Stochastic adaptive dynamics Homeostatic feedback control Fixed-point neuromorphic implementation Hardware-safe saturation regulation Activity fluctuation adaptation Lyapunov-stable evolution McKean–Vlasov-compatible formulation Mathematical Foundations The framework combines methods from: dissipative dynamical systems, stochastic differential equations, McKean–Vlasov mean-field theory, Wasserstein and variational dynamics, adaptive control theory, neuromorphic computation, nonlinear attractor systems. Potential Applications Neuromorphic Computing Adaptive spiking control for low-power neuromorphic processors and edge AI systems. Embedded AI Energy-efficient adaptive regulation in constrained hardware environments. Adaptive Control Systems Real-time feedback stabilization under stochastic conditions. Spiking Neural Networks Dynamic threshold adaptation and attractor stabilization in recurrent neural architectures. Mean-Field Neural Modeling Population-level adaptive dynamics and collective activity regulation. Stochastic Optimization Noise-regulated adaptive parameter evolution in nonlinear systems. Research Status The framework currently represents: a formal mathematical specification, a computational dynamical model, a research-stage adaptive systems architecture. The following components are mathematically established: dissipative stability structure, Lyapunov energy decay, bounded adaptive dynamics, stochastic extension formulation, fixed-point implementation scheme. The following remain open research problems: rigorous propagation-of-chaos proof, global mean-field convergence, ergodicity analysis, spectral stability theory, large-scale PDE limits, hardware convergence guarantees. Scientific Positioning ADAD should be classified as: adaptive stochastic dynamics, dissipative attractor theory, mean-field neural dynamics, neuromorphic adaptive control, variational adaptive systems. It should not be interpreted as: a physical theory, a cosmological fram","author":[{"family":"Bazarov","given":"Vitaly"},{"family":"Bazarov","given":"Vitaliy"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20313737","URL":"https://doi.org/10.5281/zenodo.20313737","source":"datacite"},{"id":"doi:10.5281/zenodo.20191325","type":"article-journal","title":"Emergent Intelligence for Chaotic Supply Chain Resilience in Contested Environments","abstract":"This paper introduces Vacuum Intelligence, a computational framework in which coherent, adaptive decision structure emerges from a high-entropy generative substrate rather than being imposed through explicit parameterization. The framework draws on reservoir computing principles, extended with synchronicity detection, decay-rebirth cycling, and a hypothesis-generation output layer that produces actionable logistics recommendations rather than binary threshold alerts. A prototype system is exercised against synthetic data streams modeled on the 2026 Strait of Hormuz supply chain crisis, the largest energy supply disruption in the history of the global oil market, to evaluate emergent attractor formation, early tipping-point detection, and antifragile recovery across simulated coherence-decay cycles. Results demonstrate promising behavior across all three dimensions, including progressive resilience improvement across repeated stress cycles consistent with antifragile system dynamics. Five defense application domains are developed in depth: early warning and anomaly detection from multi-source weak signals including AIS vessel tracking, satellite imagery, and open-source intelligence; resilient logistics planning under deep uncertainty; counter-sanctions and shadow-fleet disruption modeling with adaptive countermeasure hypothesis generation; antifragile command system architecture for degraded and jammed environments; and automated wargaming through large-scale internal scenario exploration cycles. The framework is positioned as a nonlinear hypothesis-generation frontend for existing human decision loops, targeted for eventual neuromorphic edge deployment on platforms including Intel Loihi 2. All scenario data is derived from publicly available reporting as of May 2026. This is exploratory conceptual research. No claims of operational readiness are advanced. External validation on sanitized or cleared logistics datasets is identified as the critical next step.","author":[{"family":"John","given":"Damon"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20191325","URL":"https://doi.org/10.5281/zenodo.20191325","source":"datacite"},{"id":"doi:10.5281/zenodo.20288933","type":"article-journal","title":"Interference-Field-Grown Bio-Synaptic Semiconductor Organisms: A Conceptual Framework for Wave-Written Living Electronic Morphogenesis","abstract":"This paper proposes a conceptual framework for interference-field-grown bio-synaptic semiconductor organisms: living or life-like material systems that form, repair, and adapt semiconductor-like electronic networks through biological growth, material deposition, electron transport, and externally imposed wave fields. The central idea is that optical, electrical, or acoustic interference patterns can act not only as fabrication masks, but as spatial growth commands for living electronic morphogenesis. In this framework, an interference field writes an initial circuit-like landscape, while living or biohybrid material processes complete, repair, and adapt the resulting conductive or semiconductor-like network. The paper introduces a minimal mathematical model in which interference intensity, frequency-selective biological response, conductive growth, metabolic repair, current-dependent reinforcement, and material memory jointly define a time-evolving graph of electronic connectivity. The proposed system is not intended as an immediate replacement for silicon logic. Instead, it targets slowly adaptive, self-healing, environmentally coupled bioelectronic skins, sensor films, analog computing substrates, pollution-recording membranes, biohybrid photoreactive surfaces, and living neuromorphic materials. The work connects engineered living materials, living electronics, interference lithography, optogenetic biofilm patterning, and biohybrid semiconductor systems into a single conceptual architecture. Its core claim is that circuits may be understood not only as static manufactured artifacts, but as adaptive morphologies that can be grown, trained, damaged, repaired, and partially remembered by the material itself.","author":[{"family":"Labo","given":"Trinity"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20288933","URL":"https://doi.org/10.5281/zenodo.20288933","source":"datacite"},{"id":"doi:10.5281/zenodo.20288932","type":"article-journal","title":"Interference-Field-Grown Bio-Synaptic Semiconductor Organisms: A Conceptual Framework for Wave-Written Living Electronic Morphogenesis","abstract":"This paper proposes a conceptual framework for interference-field-grown bio-synaptic semiconductor organisms: living or life-like material systems that form, repair, and adapt semiconductor-like electronic networks through biological growth, material deposition, electron transport, and externally imposed wave fields. The central idea is that optical, electrical, or acoustic interference patterns can act not only as fabrication masks, but as spatial growth commands for living electronic morphogenesis. In this framework, an interference field writes an initial circuit-like landscape, while living or biohybrid material processes complete, repair, and adapt the resulting conductive or semiconductor-like network. The paper introduces a minimal mathematical model in which interference intensity, frequency-selective biological response, conductive growth, metabolic repair, current-dependent reinforcement, and material memory jointly define a time-evolving graph of electronic connectivity. The proposed system is not intended as an immediate replacement for silicon logic. Instead, it targets slowly adaptive, self-healing, environmentally coupled bioelectronic skins, sensor films, analog computing substrates, pollution-recording membranes, biohybrid photoreactive surfaces, and living neuromorphic materials. The work connects engineered living materials, living electronics, interference lithography, optogenetic biofilm patterning, and biohybrid semiconductor systems into a single conceptual architecture. Its core claim is that circuits may be understood not only as static manufactured artifacts, but as adaptive morphologies that can be grown, trained, damaged, repaired, and partially remembered by the material itself.","author":[{"family":"Labo","given":"Trinity"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20288932","URL":"https://doi.org/10.5281/zenodo.20288932","source":"datacite"},{"id":"doi:10.5281/zenodo.20248382","type":"article-journal","title":"The Hexad Methodology: Toward an Environmentally Sustainable Pathway to AGI","abstract":"Abstract The pursuit of Artificial General Intelligence (AGI) has historically been characterized by escalating computational and energetic demands, raising profound concerns about ecological sustainability. This paper introduces the Hexad Methodology—a six■principle framework for developing AGI within planetary boundaries. Drawing on thermodynamics, resource economics, embodied cognition, and ecological governance, the Hexad proposes that environmental constraint is not an impediment to AGI but a necessary condition for its ethical and practical realization. We argue that the most resource■efficient pathway—leveraging neuromorphic computing, sparse architectures, renewable integration, and circular hardware economies—is also epistemologically superior, as it forces a disciplined alignment between intelligence and its material substrate. Unlike existing “Green AI” efforts that treat efficiency as a post■hoc optimization, the Hexad embeds ecological constraint as a first■order architectural and governance principle. We conclude that indefinite exponential scaling is ecologically unsustainable, and that constraint■driven design may yield more general, robust, and legitimate forms of intelligence.","author":[{"family":"Kain","given":"Adam"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20248382","URL":"https://doi.org/10.5281/zenodo.20248382","source":"datacite"},{"id":"doi:10.5281/zenodo.20248383","type":"article-journal","title":"The Hexad Methodology: Toward an Environmentally Sustainable Pathway to AGI","abstract":"Abstract The pursuit of Artificial General Intelligence (AGI) has historically been characterized by escalating computational and energetic demands, raising profound concerns about ecological sustainability. This paper introduces the Hexad Methodology—a six■principle framework for developing AGI within planetary boundaries. Drawing on thermodynamics, resource economics, embodied cognition, and ecological governance, the Hexad proposes that environmental constraint is not an impediment to AGI but a necessary condition for its ethical and practical realization. We argue that the most resource■efficient pathway—leveraging neuromorphic computing, sparse architectures, renewable integration, and circular hardware economies—is also epistemologically superior, as it forces a disciplined alignment between intelligence and its material substrate. Unlike existing “Green AI” efforts that treat efficiency as a post■hoc optimization, the Hexad embeds ecological constraint as a first■order architectural and governance principle. We conclude that indefinite exponential scaling is ecologically unsustainable, and that constraint■driven design may yield more general, robust, and legitimate forms of intelligence.","author":[{"family":"Kain","given":"Adam"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20248383","URL":"https://doi.org/10.5281/zenodo.20248383","source":"datacite"},{"id":"doi:10.5281/zenodo.20124584","type":"article-journal","title":"Evaluating Container Orchestration for Neuromorphic Workloads in Virtual Edge Environments","abstract":"The growing adoption of edge computing has created an increasing need for workloads capable of operating under strict resource and energy constraints. Neuromorphic computing, and spiking neural networks (SNNs) in particular, offers an energy-efficient alternative to conventional machine learning through event-driven computation. However, how SNN workloads behave when deployed within modern container orchestration frameworks, especially in edge environments, remains largely unexplored. This paper investigates the feasibility of deploying and orchestrating SNN workloads in a virtual edge environment using Kubernetes, focusing on end-to-end latency, throughput, classification accuracy, infrastructure overhead, and runtime behavior under concurrent load. Experiments were conducted on a single-node K3d cluster running on a Windows 11 host with WSL2 and Docker Desktop. The results show that SNN workloads are highly sensitive to resource availability. Restricting CPU to 0.5 cores increased median latency by 47.6 times and reduced throughput by 49 times, while the most constrained configuration failed entirely due to insufficient memory. Classification accuracy remained stable across all working configurations. From an orchestration perspective, K3d successfully deployed and scaled SNN workloads, though its default round-robin routing policy introduced severe tail latency under replica scaling, highlighting a mismatch between stateless load balancing assumptions and long-running inference workloads. Overall, the study provides a practical baseline for deploying neuromorphic workloads in containerized edge environments and highlights the importance of resource provisioning and orchestration configuration. Future work should explore improved routing strategies, memory optimization, and validation on physical edge hardware.","author":[{"family":"Pham","given":"Huyen"},{"family":"Silverajan","given":"Bilhanan"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20124584","URL":"https://doi.org/10.5281/zenodo.20124584","source":"datacite"},{"id":"doi:10.5281/zenodo.20124585","type":"article-journal","title":"Evaluating Container Orchestration for Neuromorphic Workloads in Virtual Edge Environments","abstract":"The growing adoption of edge computing has created an increasing need for workloads capable of operating under strict resource and energy constraints. Neuromorphic computing, and spiking neural networks (SNNs) in particular, offers an energy-efficient alternative to conventional machine learning through event-driven computation. However, how SNN workloads behave when deployed within modern container orchestration frameworks, especially in edge environments, remains largely unexplored. This paper investigates the feasibility of deploying and orchestrating SNN workloads in a virtual edge environment using Kubernetes, focusing on end-to-end latency, throughput, classification accuracy, infrastructure overhead, and runtime behavior under concurrent load. Experiments were conducted on a single-node K3d cluster running on a Windows 11 host with WSL2 and Docker Desktop. The results show that SNN workloads are highly sensitive to resource availability. Restricting CPU to 0.5 cores increased median latency by 47.6 times and reduced throughput by 49 times, while the most constrained configuration failed entirely due to insufficient memory. Classification accuracy remained stable across all working configurations. From an orchestration perspective, K3d successfully deployed and scaled SNN workloads, though its default round-robin routing policy introduced severe tail latency under replica scaling, highlighting a mismatch between stateless load balancing assumptions and long-running inference workloads. Overall, the study provides a practical baseline for deploying neuromorphic workloads in containerized edge environments and highlights the importance of resource provisioning and orchestration configuration. Future work should explore improved routing strategies, memory optimization, and validation on physical edge hardware.","author":[{"family":"Pham","given":"Huyen"},{"family":"Silverajan","given":"Bilhanan"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.20124585","URL":"https://doi.org/10.5281/zenodo.20124585","source":"datacite"},{"id":"doi:10.1038/s41377-024-01719-4","type":"article-journal","title":"Polariton lattices as binarized neuromorphic networks","abstract":"Abstract We introduce a novel neuromorphic network architecture based on a lattice of exciton-polariton condensates, intricately interconnected and energized through nonresonant optical pumping. The network employs a binary framework, where each neuron, facilitated by the spatial coherence of pairwise coupled condensates, performs binary operations. This coherence, emerging from the ballistic propagation of polaritons, ensures efficient, network-wide communication. The binary neuron switching mechanism, driven by the nonlinear repulsion through the excitonic component of polaritons, offers computational efficiency and scalability advantages over continuous weight neural networks. Our network enables parallel processing, enhancing computational speed compared to sequential or pulse-coded binary systems. The system’s performance was evaluated using diverse datasets, including the MNIST dataset for image recognition and the Speech Commands dataset for voice recognition tasks. In both scenarios, the proposed system demonstrates the potential to outperform existing polaritonic neuromorphic systems. For image recognition, this is evidenced by an impressive predicted classification accuracy of up to 97.5%. In voice recognition, the system achieved a classification accuracy of about 68% for the ten-class subset, surpassing the performance of conventional benchmark, the Hidden Markov Model with Gaussian Mixture Model.","author":[{"family":"Sedov","given":"Evgeny"},{"family":"Kavokin","given":"Alexey"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1038/s41377-024-01719-4","URL":"https://doi.org/10.1038/s41377-024-01719-4","source":"crossref"},{"id":"doi:10.54254/2755-2721/2026.ka26804","type":"article-journal","title":"Impact of Device Materials on Neuromorphic Circuits","abstract":"Facing the rapid development and energy-efficiency demands of artificial intelligence and neuromorphic computing, traditional von Neumann architectures struggle with storage and power consumption bottlenecks. Neuromorphic devices, as next-generation computing technologies, have been validated using machine-learning software and CMOS hardware, but still face issues in power consumption and learning speed. This paper introduces four distinctive device mechanisms. Zein-based memristive synapses combine biocompatibility with mechanical flexibility, offering a naturally sustainable solution for wearable neuromorphic systems. Low-dimensional nanomaterials exploit quantum confinement and van der Waals heterojunctions to realize ultra-low-energy plasticity and gate-tunable multistate storage, outperforming conventional CMOS in integration density and energy efficiency. Y₂O₃ memristors, analyzed via coupled electro-thermal modeling, demonstrate that surface roughness and yttrium oxygen-reservoir layers strongly modulate conductive-filament stability and switching voltages, providing quantitative guidelines for nanoscale interface engineering. WO₃₋ₓ-based electrochemical random-access memory leverages oxygen-rich/oxygen-poor phase separation to achieve non-volatile retention under short-circuit conditions, satisfying the requirements of low-power inference chips. These material pathways collectively drive neuromorphic hardware toward brain-like, scalable, and ultra-low-power computation.","author":[{"family":"Feng","given":"Yiming"}],"issued":{"date-parts":[[2025]]},"DOI":"10.54254/2755-2721/2026.ka26804","URL":"https://doi.org/10.54254/2755-2721/2026.ka26804","source":"crossref"},{"id":"doi:10.64038/cel.0220248","type":"article-journal","title":"EXPLORING THE POTENTIAL OF NEUROMORPHIC COMPUTING TO ENHANCE EFFICIENCY AND PERFORMANCE IN AI WORKLOADS","abstract":"The increasing processing needs of artificial intelligence (AI) has inspired utilizing the architecture of the human brain as a viable way to address this challenge and helped neuromorphic computing become a reality. In contrast to von Neumann architecture-based computer systems, neuromorphic systems offer event driven parallel processing which can make functionally compute thousands more times in parallel and with much lower energy consumption compared to today's technologies. The scope of this work is to study how neuromorphic computing may enable the performance and effectiveness of AI tasks that rely on demanding resources, such as in image recognition, natural language processing and real-time data processing. On the basis of a proper examination of arrangements of the present in neuromorphic equipment, for example, IBM's True North, Brain Chip’s Akida and Intel's Loihi, it tends to be contended that the framework is thoroughly greater than when contrasted and customary GPUs and CPUs as far as power utilization and fundamentally increasingly proficient. The integration of spiking neural networks (SNNs) into neuromorphic systems has been demonstrated to improve real time job performance and offer important advantages for future edge AI applications where energy efficiency and delay are essential. In addition, to facilitate data interpretation and decision making in an edge database system, the work studies the possibility for synergy between neuromorphic computing and a hybrid AI model that unifies deep learning with symbolic reasoning. For instance, the research showed that neuromorphic computing and hybrid AI models have a range of practical applications in smart cities, medical diagnostics, self driving vehicles and smart agriculture among others. Although neuromorphic computing offers various advantages, there are some things to be worked out, like scaling, interfacing with existing AI pipelines, and developing a standardized programming framework. In the near future, efforts for future research should focus on establishing standardized framework s and evaluation measures to close the gap between neuromorphic systems and conventional AI architectures. Finally, neuromorphic computing and hybrid AI model allow us to take on these computational issues in modern AI workloads in a revolutionary way. With continued advancement of these technologies, the possibility of smarter, more responsive, energy efficient systems are possible: systems that can scale to support the needs of real-time applications in a number of fields.","author":[{"family":"Shariff","given":"Mohammad"},{"family":"Khan","given":"Sara"}],"issued":{"date-parts":[[2026]]},"DOI":"10.64038/cel.0220248","URL":"https://doi.org/10.64038/cel.0220248","source":"crossref"},{"id":"doi:10.48175/ijarsct-36820","type":"article-journal","title":"Neuromorphic Computing: From Device Physics to Scalable Systems – A Comprehensive Review","abstract":"Neuromorphic computing has emerged as a promising paradigm for addressing the growing energy demands of modern computing by incorporating brain-inspired principles into physical hardware. However, a major challenge lies in scaling individual synaptic devices from laboratory demonstrations to integrated systems capable of real-world applications. This review examines recent progress from device-level innovations to system-level architectures and standardized benchmarking frameworks. In particular, it discusses emerging implementations of artificial synapses based on spintronic and photonic technologies, including photoresponsiveheterostructures and optical scattering techniques that exploit intrinsic physical phenomena for information processing. At the device scale, mechanisms such as magnetic switching, optical response, and wave interference enable efficient and parallel computation. Translating these devices into practical neuromorphic systems requires overcoming integration challenges related to device interconnectivity, power management, and compatibility across heterogeneous platforms, including photonic and electronic circuits. Furthermore, the development of standardized benchmarking frameworks is crucial for evaluating performance, comparing architectures, and identifying system limitations. This review consolidates recent advances in device physics, neuromorphic architectures, and evaluation methodologies, highlighting key directions for the development of scalable and energy-efficient neuromorphic computing systems beyond conventional architectures.","author":[{"family":"Anushree A Rajput","given":"Suresh"}],"issued":{"date-parts":[[2026]]},"DOI":"10.48175/ijarsct-36820","URL":"https://doi.org/10.48175/ijarsct-36820","source":"crossref"},{"id":"doi:10.7868/s3034508125050034","type":"article-journal","title":"NANOSCALE STRANTRONIC MAGNETOELETRIC CELL FOR NEUROMORPHIC SYSTEMS","abstract":"Results are reported on numerical-analytic modelling of functional characteristics of a nanoscale neuron-like magnetoelectric cell. Nonlinear transfer activation functions of a composite cell and conditions of their formation in spin-reorientation processes in the magnet sub-system were determined. As applied to the transformation of random pulsed signals, threshold modes of generation of inverse polarity spikes were demonstrated as well as phenomena of potential accumulation followed by an abrupt change of the system’s state of the Integrate-and-Fire type. Magnitude of signals at input and output of the nanoscale cell values few millivolts. The activation function type and the threshold values of input signals are controlled by magnetizing field, which permits to expand the functional capabilities of components for analog neuromorphic systems.","author":[{"family":"Krutyansky","given":"LM"},{"family":"Preobrazhensky","given":"VL"}],"issued":{"date-parts":[[2026]]},"DOI":"10.7868/s3034508125050034","URL":"https://doi.org/10.7868/s3034508125050034","source":"crossref"},{"id":"doi:10.3390/jlpea16030032","type":"article-journal","title":"Hybrid Filamentary Switching in Fe2O3—Incorporated TiO2 Memristors for Memory and Neuromorphic Computing Application","abstract":"Memristive devices have emerged as promising candidates for next-generation non-volatile memory and neuromorphic computing systems owing to their simple device architecture, low power consumption, and capability for analog conductance modulation. In this work, Fe2O3-incorporated TiO2 thin films were employed as the active switching layer to investigate their resistive switching and artificial synaptic characteristics. The fabricated Ag/TiO2 + Fe2O3/FTO device exhibits stable switching performance with a low operating voltage of approximately ±0.6 V, a large ON/OFF ratio of ~103, retention of nearly 104 s, and endurance over 5000 pulse cycles, representing a significant improvement compared with pristine TiO2 devices. Electrical transport analysis demonstrates Schottky emission-dominated conduction in the high-resistance state and Ohmic conduction in the low-resistance state. Furthermore, analog switching behavior was achieved by operating the device under a higher compliance current, demonstrating gradual conductance modulation and synaptic weight update characteristics, indicating the potential of the device for neuromorphic applications. These results demonstrate that Fe2O3 incorporation provides an effective approach for simultaneously enhancing memory performance and synaptic functionality in TiO2-based memristors, highlighting their potential for future memory and neuromorphic computing applications.","author":[{"family":"Sahu","given":"Dwipak"}],"issued":{"date-parts":[[2026]]},"DOI":"10.3390/jlpea16030032","URL":"https://doi.org/10.3390/jlpea16030032","source":"crossref"},{"id":"doi:10.1088/2053-1583/adb8c3","type":"article-journal","title":"Ionically gated transistors based on two-dimensional materials for neuromorphic computing","abstract":"Abstract Neuromorphic computing is a low-power and energy efficient alternative to von Neumann computing that demands new materials and computing architectures. Two-dimensional (2D) van der Waals materials and ions are a particularly favorable pair for neuromorphic computing. The large surface to volume ratio of 2D layered materials makes them sensitive to the presence of ions, detected as orders of magnitude change in electrical resistance. Quantum confinement of 2D crystals limits carrier scattering and enhances mobility, which decreases power consumption. Moreover, the 2D crystal-ion pair can provide volatile and non-volatile responses in the same device, as well as dynamic synaptic properties, such as spike-timing dependent plasticity. These dynamic properties are particularly relevant because they mirror the mechanisms involved in biological learning and memory. In this perspective, we first summarize recent progress in the field, categorize 2D crystal-ion devices in terms of their mechanisms (either electrostatic or electrochemical), and highlight key synaptic functionalities these devices can replicate. We underscore the differences between artificial and biological synapses, and between devices meant to emulate biological functions versus those optimized for compatibility with digital artificial neural networks (ANNs). We note that the use of ionically gated transistors based on 2D crystals (2D IGTs) in ANNs has primarily focused on their non-volatile memory functions, rather than fully exploiting their dynamic synaptic properties. We assert that the energy-efficient operation of 2D IGTs, enabled by their high capacitance density and tunable ion dynamics, makes them particularly suited for low-power edge computing applications. Finally, our perspective is that realizing the full potential of 2D crystals and ions in neuromorphic systems will require bridging the gap between demonstrated synaptic functionalities and their practical implementations in neural networks.","author":[{"family":"Xu","given":"Ke"},{"family":"Fullerton-Shirey","given":"Susan"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1088/2053-1583/adb8c3","URL":"https://doi.org/10.1088/2053-1583/adb8c3","source":"crossref"},{"id":"doi:10.60023/qvsndc03","type":"article-journal","title":"Neuromorphic Computing Model Based on Spiking Neural Network for an Efficient and Resilient Tsunami Early Warning System in Indonesia’s Small Islands","abstract":"This study aims to develop a fault-tolerant neuromorphic computing system for tsunami early detection in Indonesia’s small islands, which face significant limitations in energy and network infrastructure. The research was conducted over a three-month period (January–March 2025) using a simulated experimental approach with ocean wave data obtained from BMKG and NOAA. The system model was designed using a Spiking Neural Network (SNN) that mimics biological neuron activity to adaptively recognize ocean wave anomaly patterns. Simulation results show a detection accuracy rate of 94%, maintaining stable performance above 85% even under 25% signal interference. Furthermore, the system’s power consumption was recorded at only 0.42 watts—approximately 40–60% more efficient than conventional CNN-based models. The implications of this study include scientific contributions to the development of adaptive and energy-efficient artificial intelligence, as well as practical benefits for agencies such as BMKG and BNPB in designing autonomous and resilient tsunami early warning systems for remote and underdeveloped regions. In the future, this system has the potential to serve as a prototype for edge computing–based disaster mitigation solutions powered by artificial intelligence, particularly relevant for archipelagic nations.","author":[{"family":"Jayaun","given":"Jayaun"}],"issued":{"date-parts":[[2026]]},"DOI":"10.60023/qvsndc03","URL":"https://doi.org/10.60023/qvsndc03","source":"crossref"},{"id":"doi:10.5121/ijsc.2025.16101","type":"article-journal","title":"Dynamic Cognitive Ontology Networks: Advanced Integration of Neuromorphic Event Processing and Tropical Hyper Dimensional Representations","abstract":"This journal article introduces a significantly expanded framework for integrating dynamic neuromorphic event-based processing with tropical hyperdimensional computing in cognitive ontology networks. Enhancements include the exploration of practical applications, ethical implications, and scalability to real-world datasets. This approach employs advanced dynamic adaptive learning rates, hierarchical relationship encoding, and novel binding mechanisms to achieve substantial improvements in clustering and classification tasks. A comparative study with state-of-the-art models highlights the framework's robustness, scalability, and biological plausibility. Additionally, the paper discusses new real-time hardware implementation potentials and multimodal integration, paving the way for future advancements.","author":[{"family":"Menemy","given":"Robert"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5121/ijsc.2025.16101","URL":"https://doi.org/10.5121/ijsc.2025.16101","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-10021251/v1","type":"article-journal","title":"A Physio-Informatic Paradigm for Quantum Annealing and Neuromorphic AI: Extending PIRA for Sub-Exponential Energy Optimization","abstract":"Abstract This paper introduces PIRA-X, extending the Physio-Informatic Reduction Algorithm (PIRA) paradigm from discrete classical boundaries to continuous-state physical architectures, notably Quantum Annealing (QA) substrates and Neuromorphic crossbar arrays. Combinatorial NP-hard mappings and deep neural network optimization suffer from exponential energy scaling on classical von Neumann architectures. By integrating Landauer's thermodynamic erasure principle with non-equilibrium physical state variables, we model optimization as a native thermodynamic relaxation lifecycle. We present a multi-dimensional tensor governing equation that bounds physical information pressure gradients, enforcing a strict sub-exponential, logarithmic convergence trajectory of T(d)=O(d \\ln d) across rugged, non-convex energy landscapes. Simulated benchmarks across extensive TSPLIB instances demonstrate significant execution gains against conventional stochastic baselines. Rigorous statistical validation via Wilcoxon signed rank testing (p=0.00012) and an exceptional Cohen's d effect size (4.72) confirm the computational viability of hardware-native, physio-informatic acceleration.","author":[{"family":"Ali","given":"Md"}],"issued":{"date-parts":[[2026]]},"DOI":"10.21203/rs.3.rs-10021251/v1","URL":"https://doi.org/10.21203/rs.3.rs-10021251/v1","source":"crossref"},{"id":"doi:10.1149/ma2026-01341576mtgabs","type":"article-journal","title":"(\n                    <i>Invited</i>\n                    ) Solution-Processed Neuromorphic Transistors with Tunable Temporal Dynamics for Wearable Sensing","abstract":"Wearable biological sensing devices promise continuous, low-power monitoring of physiological and biochemical signals. However, these emerging platforms often require increasingly complex signal computation. One common solution is to continuously offload data via wireless transmission for cloud-based processing, but this approach incurs significant power and communication overhead that fundamentally limits wearable operation. Neuromorphic computing architectures offer an alternative by enabling low-power, on-board computation directly at the sensing node. Critically, such devices must exhibit temporal dynamics that are tuned to the intrinsic timescales of the biological signals being measured. In this talk, I will discuss how we engineer solution-processed materials that maintain high performance across biologically relevant and tunable temporal regimes. Ultimately, this work establishes a materials design framework for developing sensor-specific computing architectures for next-generation wearable technologies. The devices are based on hybrid semiconducting channels composed of carbon nanotube (CNT) networks and the ion-permeable polymer poly(3-hexylthiophene) (P3HT), fabricated using low-temperature, scalable processes compatible with flexible substrates. By controlling the polymer crystallization pathway, we engineer distinct microstructures that regulate ionic penetration and trapping, yielding synaptic transistors spanning fast, surface-dominated responses to slow, long-term ionic memory. Electrical and synaptic characterization reveals that these material-defined temporal dynamics directly govern device behavior, including short- and long-term plasticity, recovery time constants, and energy efficiency. Using experimentally measured device responses, we demonstrate through neuromorphic simulations that fast CNT-dominant devices are well-suited for high-frequency spiking operations, while hybrid CNT/P3HT devices with extended memory enable efficient temporal integration and data compression for slowly varying inputs. These regimes closely mirror the signal characteristics encountered in wearable sensing, such as motion, strain, and sweat-based biomarker monitoring. Together, this work establishes materials-controlled temporal dynamics as a key design parameter for solid-state electronics in wearable and biomedical systems, enabling flexible sensor platforms that perform signal processing directly at the body interface without increasing circuit or algorithmic complexity.","author":[{"family":"Andrews","given":"Joseph"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1149/ma2026-01341576mtgabs","URL":"https://doi.org/10.1149/ma2026-01341576mtgabs","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-10010940/v1","type":"article-journal","title":"Neuromorphic control of a simulated shape-memory-alloy-driven multi-legged robot using a spiking neural network","abstract":"Abstract We propose a spiking-neural-network (SNN)-based deep reinforcement learning (DRL) method for efficient autonomous locomotion control of a starfish-inspired multi-legged soft robot driven by shape memory alloy (SMA) actuators. Conventional soft-robot controllers have predominantly relied on open-loop, pre-designed periodic signals; however, manually designing and tuning coordinated motion in systems with many degrees of freedom and strong nonlinearities remains difficult. Here, we train an SNN controller for SMAs, whose nonlinear contractile responses resemble key aspects of biological muscle. Specifically, we propose a Spiking Actor Network (SAN), in which the actor network of Twin Delayed Deep Deterministic Policy Gradient (TD3), an algorithm well suited to continuous action spaces, is replaced by an SNN. To account for the membrane potential, an internal state of SNNs, we extend the replay buffer to store and reuse membrane-potential information during training. To improve inference-time energy efficiency, we also add a regularization term that minimizes the squared membrane potentials, thereby encouraging action generation with fewer spike firings. In physics-simulation experiments on autonomous locomotion, the proposed method enabled the agent to acquire the leg coordination required to move toward a target. Analysis of the output spike trains showed that the learned firing patterns developed periodicity matched to the physical characteristics of the SMA, forming a coordinated gait that generated continuous propulsion while avoiding excessive firing. These results demonstrate the potential of autonomous neuromorphic control for energy-efficient soft-robot locomotion.","author":[{"family":"Miki","given":"Daisuke"},{"family":"Takigasaki","given":"Hiroto"}],"issued":{"date-parts":[[2026]]},"DOI":"10.21203/rs.3.rs-10010940/v1","URL":"https://doi.org/10.21203/rs.3.rs-10010940/v1","source":"europepmc"},{"id":"doi:10.54254/2753-8818/2025.ad26488","type":"article-journal","title":"Energy-Efficient Neuromorphic Chips for Real-Time Robotic Control: A Review","abstract":"Neuromorphic computing has gained increasing attention as a bio-inspired solution to the limitations of traditional computing systems in power and real-time constraints. In robotic control tasks, real-time processing and energy efficiency are crucial, especially for autonomous and mobile systems. Neuromorphic chips mimic the structure and operation of biological neurons, enabling low-latency and low-power processing. This review examines the architectures, performance benchmarks, and application cases of mainstream neuromorphic hardware platforms used in robotic perception and control. It also compare their energy consumption and latency with conventional control platforms. Furthermore, this paper identifies key challenges in system integration and suggests future directions including improved scalability, online learning, and the combination with edge AI frameworks. The paper finds that neuromorphic chips can significantly reduce energy consumption and improve real-time performance in robotic control. However, challenges in algorithm adaptation, standardization, and hardware integration remain. This study aims to provide researchers with insights into the practical implementation of neuromorphic control systems in energy-sensitive robotic applications.","author":[{"family":"Liu","given":"Shuming"}],"issued":{"date-parts":[[2025]]},"DOI":"10.54254/2753-8818/2025.ad26488","URL":"https://doi.org/10.54254/2753-8818/2025.ad26488","source":"crossref"},{"id":"doi:10.56147/aaiet.2.2.117","type":"article-journal","title":"Neuromorphic Modular Intelligence Architecture: Overcoming Transformer Limits Toward AGI Beyond Monolithic LLM Limits","abstract":"Transformer-based Large Language Models (LLMs) achieve remarkable pattern matching but struggle with deep reasoning due to fixed computation depth, bounded inference complexityandlimited token memory. These structural constraints (e.g.,fixed-layer depth yields only O(1)-depth circuit expressivityandcontext windows cap memory) hinder tasks requiring iterative, algorithmic reasoning. I propose the Neuromorphic Modular Intelligence Architecture (NMIA), a brain inspired system combining multiple specialized LLM modules with persistent memory, external symbolic verificationanda non-linguistic control layer.I analyze asymptotic complexity: Amonolithic transformer incurs O(n2d)cost per inference (with ntokens and dimension d), while NMIA with kmodules of size n/k yields approximately O(n2d/k + O(nd)), a factor-k speedup. In a simulated experiment (n=5000), NMIA (with 5 modules) significantly outperforms a monolithic LLM on a realistic reasoning task: Modular inference is faster (mean 48.6ms vs.104.4ms, d=6.63, 95% CI (51.5, 60.2)) and more accurate (accuracy 85.4% vs.69.4%, d=2.04, ∆=0.16, 95% CI (0.12, 0.20)). These results support NMIA as a viable AGI path: It demonstrates that careful modular decomposition and memory can circumvent fundamental transformer limitations without invoking purely linguistic reasoning. The analysis and simulation together argue that while monolithic LLMs have formal limits on computation, a neuromorphic, multi-agent design can extend reasoning capacity and is a promising route toward general intelligence.","author":[{"family":"Mishra","given":"Anindya"}],"issued":{"date-parts":[[2026]]},"DOI":"10.56147/aaiet.2.2.117","URL":"https://doi.org/10.56147/aaiet.2.2.117","source":"crossref"},{"id":"doi:10.33774/coe-2026-7rm70","type":"article-journal","title":"SuperNeuronBench - A Computational Benchmark for Comparing Time-Multiplexed Neuromorphic Architectures on Open-Source FPGAs to Parallel Biological Software Models","abstract":"The real-time simulation of biological neurons is a constraint in computational neuroscience. One human brain cortical column includes tens of thousands of neurons and hundreds of millions to billions of synaptic connections that have to update in the same clock cycle. Time multiplexing is the scientific method used to mitigate this problem. One processing element, named a \"superneuron\", cycles through the states of simulated neurons in a sequence. This method is used by systems such as SpiNNaker, Intel Loihi, and IBM TrueNorth. But this sequential process creates a time gap that does not exist in the parallel process of biological neurons. No study isolates the effect of this timing gap on neural computation types, and this problem is not addressed for open-source neuromorphic hardware such as NeuroCoreX. This proposal introduces SuperNeuronBench, a benchmark that measures the reduction in biological accuracy caused by time multiplexing across task types. Those task types are pattern storage and retrieval, spike-timing-dependent plasticity(STDP), temporal gating, and coincidence detection. The benchmark uses the van Rossum spike-train distance against a parallel NEST or Brian2 software-biological reference to process vand calculate a normalized biological accuracy value, across superneuron configurations. Those configurations are x, the number of neurons per superneuron; z, the number of time-block subdivisions, and C, the number of superneurons in parallel. Testing will be done using open-source models on Intel Altera Cyclone IV EP4CE15 and Cyclone V SoC hardware. The main hypothesis are that time-dependent tasks reduce the accuracy at a lower multiplexing density than rate-coded tasks.","author":[{"family":"Desilva","given":"Neksha"}],"issued":{"date-parts":[[2026]]},"DOI":"10.33774/coe-2026-7rm70","URL":"https://doi.org/10.33774/coe-2026-7rm70","source":"crossref"},{"id":"doi:10.31026/j.eng.2026.03.01","type":"article-journal","title":"Design of Low-Power Neuromorphic Architectures for IoT Applications","abstract":"The rapid growth of the Internet of Things (IoT) demands computing systems that remain highly intelligent while adhering to tight energy constraints. For always-on edge applications, traditional processors are too power-hungry. This neuromorphic system is developed for IoT to achieve computing integrity and has remarkable efficiency. We propose computer-memory architecture which essentially is event-driven processing and temporal-spike coding. Architectural breakthroughs include clockless processing and adaptive precision and therefore exploit temporality with hierarchical event encoding. While current hardware solutions show a power consumption of 70 μW to 680 μW, our system shows 10 – 100× improvement in overall efficiency. Testing proved that with radar gesture recognition, audio pattern matching and visual event detection, we have more than 96 % accuracy. Inference energy is 1.38 nJ, and molecular operation cost is 9.9 pJ of the architecture with these efficient metrics, a new family of autonomous IoT applications can be developed, from battery-free sensor networks to implantable devices that run for years off a single charge.","author":[{"family":"Ismael","given":"Enji"}],"issued":{"date-parts":[[2026]]},"DOI":"10.31026/j.eng.2026.03.01","URL":"https://doi.org/10.31026/j.eng.2026.03.01","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-10081852/v1","type":"article-journal","title":"Energy–Accuracy Trade-offs in Spiking Neural Networks: A Pareto Analysis on Fashion-MNIST","abstract":"Abstract We tested a directly-trained, rate-coded SNN for Fashion-MNIST [28] (60,000 training images / 10,000 test images, each 28×28 greyscale, 10 classes) for varying numbers of simulation timesteps T = {1, 2, 4, 8, 16, 32, 64} over three randomly seeded runs. The overall energy cost of the full ANN on test was 21,361.7 nJ, the best SNN (T = 8) was 89.69% and 3,910.1 nJ, 5.5× less while introducing a 3.28% accuracy cost. For every doubling of timestep there is a corresponding loss of accuracy for the marginal accuracy gain dropped from + 4.17% per timestep doubling out to + 1.39% per timestep doubling at T = 8, after which it plummeted below + 0.5%. T = 8 was taken as the onset of the practical plateau. The findings of a capacity-constrained SmallANN ablation clearly demonstrates that the SNN efficiency advantage at T = 4–8 is not due to parameter reductions alone. SNNs do not match the ANN at full accuracy, but rather operate in a regime of diminishing energy dissipation at a Pareto-efficient position. Results showed timestep budget to be a key design parameter for SNN efficiency and indicate that by using a modest temporal resolution, a high level of accuracy is attained with greatly reduced computational costs.","author":[{"family":"Farooq","given":"Hassan"}],"issued":{"date-parts":[[2026]]},"DOI":"10.21203/rs.3.rs-10081852/v1","URL":"https://doi.org/10.21203/rs.3.rs-10081852/v1","source":"crossref"},{"id":"doi:10.5821/dissertation-2117-427834","type":"article-journal","title":"Highly scalable hardware architecture for real-time execution of spiking neural networks applied to neural cognitive applications","abstract":"(English) This thesis contributes to the field of neuromorphic hardware. In particular, to the significant improvement of a scalable hardware architecture, named Hardware Emulator of Evolvable Neural Spiking Systems (HEENS), for the real-time execution of spiking neural networks (SNNs) in cognitive applications. SNNs are neural networks inspired by the biological activity of the brain, designed to process discrete temporal events, offering improved capabilities for handling temporal data as well as local plasticity, thus exhibiting greater energy efficiency compared to traditional neural networks.The HEENS architecture is implemented on several AMD (Xilinx) Zynq hardware platforms, which combine ARM processing cores with programmable logic (FPGAs), known for their high flexibility and parallelism in the execution of different neural models. One of the key achievements of this thesis is the optimization of the architecture to reduce latency, minimize resource usage, increase processing capacity, overcome previous architectural issues, and improve adaptability to various cognitive applications, such as sensory processing and pattern recognition.The results obtained can be divided into architectural contributions and experimental results in real applications. In the first part, this work improves the synaptic mapping capability, develops new system configuration mechanisms, and introduces interaction with external sensors.The enhanced system integrates an HDMI interface, demonstrating real-time visualization of neural activity and the neural and synaptic parameters of the SNN allows continuous monitoring without affecting system performance. This real-time monitoring capability has been tested in multiple experiments, where a precise representation of neural activity was observed on a 1 ms time scale, considered to be real-time, with support to other time scales, both faster or slower. The use of HEENS in physical sensor processing is highlighted, where it was proven that the architecture can adapt in real-time to changes in input signals, making it an ideal platform for applications in robotics and autonomous systems.Regarding testbenches and applications, significant results are presented in the implementation of cognitive applications using HEENS. In handwritten digit recognition, the architecture showed high accuracy using SNN models with synaptic plasticity, achieving good performance in terms of processing time and energy consumption compared to other existing solutions. Another key result is the hardware emulation of neuronal cultures, comparing the behavior of simulated neural networks in vitro, in silico, and in duris silico (physical hardware). The experiments showed that HEENS is capable of faithfully replicating the behavior and statistical properties observed in real neuronal cultures, opening new possibilities for research in neuroscience and biomedicine.Finally, the thesis concludes that the HEENS architecture not only offers a flexible and efficient environment for the research and development of SNN but also enables its implementation in real-world applications demanding real-time processing with good energy performance. The advances achieved in this work represent an important step toward the creation of next-generation scalable neuromorphic systems capable of efficiently emulating complex cognitive functions. (Català) Aquesta tesi contribueix al camp del maquinari neuromòrfic. En particular, a la millora significativa d'una arquitectura de maquinari escalable, anomenada Hardware Emulator of Evolvable Neural Spiking Systems (HEENS), per a l'execució en temps real de xarxes neuronals de polsos (SNNs) en aplicacions cognitives. Les SNNs són xarxes neuronals inspirades en l'activitat biològica del cervell, dissenyades per processar esdeveniments temporals discrets, oferint capacitats millorades per manejar dades temporals així com emprar plasticitat local, mostrant així una major eficiència energètica en comparació","author":[{"family":"Mancero","given":"Bernardo"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5821/dissertation-2117-427834","URL":"https://doi.org/10.5821/dissertation-2117-427834","source":"crossref"},{"id":"doi:10.36227/techrxiv.177220005.52986713/v1","type":"article-journal","title":"Sleep-Mediated Replay Prevents Catastrophic Forgetting in Spiking Neural Networks Trained on Sequential Tasks","abstract":"Catastrophic forgetting is a fundamental challenge in sequential learning for artificial neural networks, particularly spiking neural networks (SNNs), which aim to emulate biologically plausible neuronal dynamics. In this study, we investigate the role of sleep-mediated replay in mitigating catastrophic forgetting during sequential task learning. We trained a multilayer SNN on two sequentially presented image classification tasks (Task A and Task B) using spike-timing-dependent plasticity (STDP) learning rules. During offline periods, we implemented spontaneous replay of Task A patterns to emulate biological sleep consolidation. Our results demonstrate that sleep-mediated replay keeps Task A performance (0.589 before Task B, 0.206 after sleep) while allowing effective acquisition of Task B (0.672 before sleep). In contrast, networks trained sequentially without sleep exhibit severe forgetting (Task A drops to 0.011 after Task B). These findings suggest that sleep-like replay mechanisms can enable continual learning in SNNs, providing a computationally and biologically grounded approach to memory consolidation. We further analyze synaptic weight distributions and firing rate dynamics, revealing that sleep stabilizes task-specific synaptic patterns without impeding new learning. This work bridges computational neuroscience and artificial intelligence, suggesting that sleep-like replay could be an effective approach for continual learning in SNNs.","author":[{"family":"Sabbineni","given":"Yeteesh"},{"family":"Qiu","given":"Ethan"}],"issued":{"date-parts":[[2026]]},"DOI":"10.36227/techrxiv.177220005.52986713/v1","URL":"https://doi.org/10.36227/techrxiv.177220005.52986713/v1","source":"crossref"},{"id":"doi:10.1162/neco_a_01752","type":"article-journal","title":"Elucidating the Theoretical Underpinnings of Surrogate Gradient Learning in Spiking Neural Networks","abstract":"Abstract Training spiking neural networks to approximate universal functions is essential for studying information processing in the brain and for neuromorphic computing. Yet the binary nature of spikes poses a challenge for direct gradient-based training. Surrogate gradients have been empirically successful in circumventing this problem, but their theoretical foundation remains elusive. Here, we investigate the relation of surrogate gradients to two theoretically well-founded approaches. On the one hand, we consider smoothed probabilistic models, which, due to the lack of support for automatic differentiation, are impractical for training multilayer spiking neural networks but provide derivatives equivalent to surrogate gradients for single neurons. On the other hand, we investigate stochastic automatic differentiation, which is compatible with discrete randomness but has not yet been used to train spiking neural networks. We find that the latter gives surrogate gradients a theoretical basis in stochastic spiking neural networks, where the surrogate derivative matches the derivative of the neuronal escape noise function. This finding supports the effectiveness of surrogate gradients in practice and suggests their suitability for stochastic spiking neural networks. However, surrogate gradients are generally not gradients of a surrogate loss despite their relation to stochastic automatic differentiation. Nevertheless, we empirically confirm the effectiveness of surrogate gradients in stochastic multilayer spiking neural networks and discuss their relation to deterministic networks as a special case. Our work gives theoretical support to surrogate gradients and the choice of a suitable surrogate derivative in stochastic spiking neural networks.","author":[{"family":"Gygax","given":"Julia"},{"family":"Zenke","given":"Friedemann"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1162/neco_a_01752","URL":"https://doi.org/10.1162/neco_a_01752","source":"crossref"},{"id":"doi:10.20944/preprints202605.2023.v1","type":"manuscript","title":"Convolutive Kernel-Guarded Spiking Neural P Systems for Local Feature Computation","abstract":"Spiking Neural P systems provide a rule-based model of distributed computation inspired by membrane computing, while kernel P systems use guarded transformations and structured control of rule applicability. This paper introduces Convolutive Kernel-Guarded Spiking Neural P systems (CKSNP systems), a formal and trainable framework in which spike-rule applicability may depend on local kernel responses computed over ordered neighborhoods of spike multiplicities. The proposed model provides a general mechanism for local feature computation, combining explicit operational semantics with kernel-based predicates that can be fixed, selected, or embedded in trainable realizations. We define the syntax and transition semantics of the model, relate the construction to delay-free extended Spiking Neural P systems and kernel P systems under stated assumptions, and present a reproducible instantiation for electrocardiographic beat classification under a patient-independent protocol. The empirical study illustrates how CK-SN P local responses can be combined with RR, Gaussian, and Fourier descriptors and evaluated with classical and neural classifiers. Overall, the study clarifies both the formal role of guarded local computation and its practical use as an interpretable feature-generation mechanism.","author":[{"family":"Constantin","given":"Doru"},{"family":"Bălcău","given":"Costel"}],"issued":{"date-parts":[[2026]]},"DOI":"10.20944/preprints202605.2023.v1","URL":"https://doi.org/10.20944/preprints202605.2023.v1","source":"crossref"},{"id":"doi:10.25958/tfkv-5t55","type":"article-journal","title":"Towards neuromorphic visual SLAM: A spiking neural network for efficient pose estimation and loop closure based on event camera data","abstract":"The need for effective Simultaneous Localisation and Mapping (SLAM) solutions has been pivotal across a wide range of applications, including autonomous vehicles, industrial robotics, and mobile service platforms where accurate localisation and environmental perception are essential. Visual SLAM (VSLAM) has emerged as a popular approach due to its cost effectiveness and ease of deployment. However, state-of-the-art VSLAM systems using conventional cameras face significant limitations, including high computational requirements, sensitivity to motion blur, restricted dynamic range, and poor performance under variable lighting conditions. Event cameras present a promising alternative by producing asynchronous, high-temporal resolution data with low latency and power consumption. These characteristics make them ideal for use in dynamic and resource-constrained environments. Complementing this, neuromorphic processors designed for efficient event-driven computation are inherently compatible with sparse temporal data. Despite their synergy, the adoption of event cameras and neuromorphic computing in SLAM remains limited due to the scarcity of public datasets, underdeveloped algorithmic tools, and challenges in multimodal sensor fusion. This thesis develops integrated Visual Odometry (VO) and Loop Closure (LC) models that leverage neuromorphic sensing, spiking neural networks (SNN), and probabilistic factor-graph optimisation as a pathway towards full event camera-based SLAM. A synchronised multimodal dataset, captured with a Prophesee STM32-GENx320 event camera, Livox MID-360 LiDAR, and Pixhawk 6C Mini IMU spans indoor and outdoor scenarios over 2,285 s. Raw events are aggregated into voxel-grid tensors using 100 ms windows with 20 temporal bins. LiDAR odometry from point clouds is refined with inertial constraints in Georgia Tech Smoothing and Mapping (GTSAM) to produce pseudo-ground-truth trajectories for supervised learning. Two SNN models are developed using the SpikingJelly framework: a spiking VO network that predicts six-degree-of-freedom (6-DOF) pose increments from voxel grids, and a LC network that estimates inter-frame similarity scores for global trajectory correction. Both models are trained with surrogate gradient learning and employ Leaky Integrate-and-Fire neurons. The VO model uses a hybrid loss function that combines Root Mean Square Error for translation with a geodesic loss on the Special Orthogonal Group in 3D for rotation prediction. The LC model is optimised using a joint loss comprising a triplet margin loss for learning discriminative embeddings and a cross-entropy loss for binary classification. These frontend models are integrated into a modular backend system based on a sliding window factor graph. The backend fuses VO predictions with IMU pre-integration and LC constraints and performs real-time optimisation using GTSAM. Empirical evaluation on kilometre-scale sequences demonstrates robust performance in diverse indoor and outdoor environments, achieving sub-metre Absolute Trajectory Error and competitive Relative Pose Error. Additionally, hardware benchmarking across conventional and neuromorphic processor such as BrainChip Akida platforms reveals up to a four times reduction in latency and an order of-magnitude gain in energy efficiency on neuromorphic hardware. The main contributions of this work include a pipeline towards full VSLAM architecture combining SNNs, event-based vision, and multimodal sensor fusion for 6-DOF pose estimation and LC; a novel training pipeline using voxelised asynchronous event data and GTSAM refined pseudo-ground-truth; a modular backend architecture that performs drift-resilient optimisation using VO, IMU, and LC constraints; a cross-platform benchmarking study that highlights the advantages of neuromorphic hardware; and a synchronised multimodal dataset supporting the above components. Overall, this thesis provides a pipeline towards a scalable and energy-efficient SLAM so","author":[{"family":"Tenzin","given":"Sangay"}],"issued":{"date-parts":[[2026]]},"DOI":"10.25958/tfkv-5t55","URL":"https://doi.org/10.25958/tfkv-5t55","source":"datacite"},{"id":"doi:10.5281/zenodo.18449733","type":"article-journal","title":"Neural Dynamics: A collection of educational Python scripts","abstract":"🚀 Release v1.0.0 This is the initial release of the Neural Dynamics repository, featuring a comprehensive collection of educational Python scripts for computational neuroscience. The release includes tutorials on using the NEST Simulator, as well as standalone scripts implementing various neuron models and network dynamics. 📦 Scope and content This release includes educational Python scripts covering a broad range of core topics in neural dynamics, such as: Single neuron models (Integrate-and-Fire, Izhikevich, AdEx, EIF) Spiking neural networks (Brunel network, oscillatory dynamics) Synaptic plasticity rules (BCM rule) Gap junctions and their role in neural modeling Rate models for collective neural activity The repository structure reflects this thematic organization and mirrors the progression of the blog series. 🧠 Conceptual focus The scripts in this repository are designed as didactic and conceptual examples. Emphasis is placed on: Clarity and interpretability of models Step-by-step explanations accompanying each script Reproducibility of classic results in theoretical neuroscience Encouraging experimentation and exploration of neural dynamics concepts Bridging theoretical neuroscience with practical implementation Many models deliberately rely on reduced geometries, simplified boundary conditions, or idealized assumptions to keep the underlying mechanisms explicit. 🔬 Reproducibility and usage All scripts are compatible with a lightweight Python environment based on NumPy, SciPy, and Matplotlib, along with the NEST Simulator for spiking neural network simulations. Instructions for setting up the environment and running the scripts are provided in the README file. The scripts are written to support both direct execution and interactive, cell-by-cell exploration in development environments such as VS Code or Jupyter. This release provides a stable baseline for reuse in: teaching and coursework self-study illustrative figures and animations methodological extensions Backward compatibility across future releases is not guaranteed, but changes will primarily serve conceptual clarification rather than feature expansion. 📝 License All code is released under the GPL-3.0 License. ✨ Outlook Future releases may expand individual examples, refine numerical implementations, or add complementary scripts aligned with new blog posts. Any such extensions will build on the conceptual baseline established with this release.","author":[{"family":"Musacchio","given":"Fabrizio"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18449733","URL":"https://doi.org/10.5281/zenodo.18449733","source":"datacite"},{"id":"doi:10.5281/zenodo.18483973","type":"article-journal","title":"NeuroGuard-V2X : A Neuromorphic Edge Processing Architecture for Privacy-Preserving Network Monitoring in Connected Vehicles","abstract":"This paper introduces NeuroGuard-V2X, a privacy-preserving network monitoring architecture for connected vehicles based on neuromorphic edge processing. By deploying event-driven spiking neural networks directly on vehicles, the proposed system analyzes temporal patterns in heterogeneous V2X communications without transmitting or storing sensitive network data. NeuroGuard-V2X enables real-time intrusion detection across DSRC, cellular, and Wi-Fi interfaces while maintaining low latency and energy consumption. Experimental results demonstrate high detection accuracy, significant reductions in data transmission, and ultra-low power usage, highlighting its suitability for secure and scalable vehicular systems.","author":[{"family":"Mohammad Zahangir","given":"Alam"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18483973","URL":"https://doi.org/10.5281/zenodo.18483973","source":"datacite"},{"id":"doi:10.5281/zenodo.18483972","type":"article-journal","title":"NeuroGuard-V2X : A Neuromorphic Edge Processing Architecture for Privacy-Preserving Network Monitoring in Connected Vehicles","abstract":"This paper introduces NeuroGuard-V2X, a privacy-preserving network monitoring architecture for connected vehicles based on neuromorphic edge processing. By deploying event-driven spiking neural networks directly on vehicles, the proposed system analyzes temporal patterns in heterogeneous V2X communications without transmitting or storing sensitive network data. NeuroGuard-V2X enables real-time intrusion detection across DSRC, cellular, and Wi-Fi interfaces while maintaining low latency and energy consumption. Experimental results demonstrate high detection accuracy, significant reductions in data transmission, and ultra-low power usage, highlighting its suitability for secure and scalable vehicular systems.","author":[{"family":"Mohammad Zahangir","given":"Alam"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18483972","URL":"https://doi.org/10.5281/zenodo.18483972","source":"datacite"},{"id":"doi:10.7282/t3-zhhv-rh84","type":"article-journal","title":"Smart sensing by analog/digital hybrid neural networks","abstract":"In this dissertation, the design, computer simulations, experimental evaluations, and theoretical analyses of new sensing schemes were systematically carried out. A novel power-efficient and low-complexity analog sensor design, which is the foundation of this PhD thesis, was validated through comprehensive simulations and hardware experiments. Additionally, a sensor in analog/digital hybrid single neuron and neural network prototype was developed for practical monitoring scenarios, progressing to ECG signal compression in the analog domain for health monitoring applications. The research further explores the integration of analog/digital hybrid neural networks for UAV applications, leveraging the strengths of both domains. Furthermore, the feasibility of applying analog/digital hybrid circuits in Spiking Neural Networks (SNNs) was explored, aiming to offload complex computations from the digital part to maximize computational efficiency while leveraging the energy-saving benefits of analog design. We validated the proposed methods using computer vision datasets, including digits, text, and objects with varying levels of recognition difficulty, to analyze the feasibility of the approach. Theoretical analyses and detailed simulations provide a deep understanding of signal recovery performance and parameter optimization. This dissertation is rooted in low-power analog sensor innovation and extends to persistent wireless sensing, biosensing applications, and hybrid neural network implementations.","author":[{"family":"Hsieh","given":"Yung"}],"issued":{"date-parts":[[2026]]},"DOI":"10.7282/t3-zhhv-rh84","URL":"https://doi.org/10.7282/t3-zhhv-rh84","source":"datacite"},{"id":"doi:10.5281/zenodo.18367072","type":"article-journal","title":"Privacy-Preserving Federated Spiking Neural Networks for Real-Time Target Detection in Distributed ISAC Edge Systems","abstract":"This study proposes a privacy-preserving federated spiking neural network (SNN) framework for real-time target detection in integrated sensing and communication (ISAC) edge networks. The framework enables distributed vehicular nodes operating at 28 GHz to collaboratively learn from spike-based sensing data without sharing raw observations, thereby protecting sensitive location information. By combining federated learning with secure model aggregation, adaptive differential privacy, and spike-aware temporal learning, the system ensures robust privacy protection while supporting efficient learning under non-IID data and real-time constraints. Extensive experiments using data from 500 vehicles and 50,000 observations demonstrate that the proposed approach achieves a high detection performance of 94.3% F1-score under strict privacy guarantees, with low inference latency and significantly reduced energy consumption. These results highlight the framework’s suitability for privacy-sensitive vehicular and smart-city applications, supporting scalable, energy-efficient, and trustworthy 6G ISAC deployment.","author":[{"family":"Alam","given":"Mohammad"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18367072","URL":"https://doi.org/10.5281/zenodo.18367072","source":"datacite"},{"id":"doi:10.20944/preprints202605.0827.v1","type":"manuscript","title":"Spiking Neural Networks: A Tutorial on Models, Coding, and Training","abstract":"Spiking neural networks (SNNs) are often called the third generation of neural models. They communicate with brief asynchronous pulses rather than continuous values, which suits event-driven sensors and low-power neuromorphic hardware. The mathematics behind them is split across neuroscience textbooks, machine learning papers, and stochastic process literature, and a researcher entering the area runs into a notation problem before anything else. The same neuron model is written one way in a textbook, another way in a machine learning paper, and a third way in the stochastic process literature. This tutorial collects what a graduate student or research engineer needs in order to start working with SNNs, in one place and with one notation. We start with the leaky integrate-and-fire (LIF) neuron, derived from a conductance-based picture of the membrane. Reset semantics, the spike response model, and the broader family that includes Hodgkin-Huxley and adaptive exponential models are discussed. Network equations are written out for feedforward and recurrent architectures. We cover the main neural coding schemes (rate, time-to-first-spike, rank-order, phase, burst, population) and discuss when each is appropriate. The neuromorphic datasets a beginner is likely to encounter, including N-MNIST, DVS-Gesture, CIFAR10-DVS, SHD, and SSC, are described together with the tensor formats event-based data takes. For learning, we derive pair-based spike-timing-dependent plasticity from exponential traces and develop the surrogate gradient framework, which has become the dominant tool for training deep SNNs by backpropagation through time. Reset semantics, the choice of surrogate function, and common pitfalls in BPTT are addressed in a way that maps onto code. Alternative training paradigms (e-prop, EventProp, SLAYER, three-factor rules, ANN-to-SNN conversion) are introduced briefly so the reader knows what else is available. A practical section walks through a first SNN training loop with framework-agnostic pseudocode and points to the main software libraries (snnTorch, SpikingJelly, Norse). Throughout the article, equations carry derivational status labels (exact, reduction, approximation, heuristic) so that the reader sees at a glance which steps are mathematical identities and which involve approximations. We do not cover hardware implementation, detailed point process theory, expressivity proofs, or open problems in SNN complexity; these belong in a more advanced treatment. The article is meant to be read linearly, and a suggested reading path closes it.","author":[{"family":"Dikmen","given":"İsmail"}],"issued":{"date-parts":[[2026]]},"DOI":"10.20944/preprints202605.0827.v1","URL":"https://doi.org/10.20944/preprints202605.0827.v1","source":"europepmc"},{"id":"doi:10.21203/rs.3.rs-8679920/v1","type":"article-journal","title":"Neural Network Wiring and Topological Stochastic Resonance","abstract":"Abstract The aim of this work is to determine how variability in synaptic connectivityshapes activity in balanced spiking networks and whether “topological stochasticresonance” (amplified responses due to structural disorder) can emerge withoutchanging the connection count. For that purpose we have simulated balancednetworks of excitatory and inhibitory leaky integrate-and-fire neurons driven byexternal Poisson input. Four connectivity schemes were constructed each witha progressively more unequal distribution of the number of inputs received byindividual neurons but with the same total number of conncetions. Topologicalvariability was quantified by the standard deviation (σ) of a normal distribu-tion describing the number of inputs received from each presynaptic populationto each neuron. Our results indicate that by increasing variability in connectiv-ity produces a strong effect of increased mean firing rates and rate heterogeneityacross neuron classes, despite the total number of synapses per population beingconserved. Network balance remained globally preserved, but local deviationsgrew with σ, revealing regimes of enhanced responsiveness. We demonstratehow the structural “noise” in the form of variable synaptic connection countcan strongly modulate activity in balanced networks without altering overallconnection count. These results support the novel concept of Topological Stochas-tic Resonance and suggest that controlled connectivity disorder may help tunesensitivity and stability in cortical-like circuits.","author":[{"family":"Dedieu","given":"Antoine"},{"family":"Nikolic","given":"Konstantin"}],"issued":{"date-parts":[[2026]]},"DOI":"10.21203/rs.3.rs-8679920/v1","URL":"https://doi.org/10.21203/rs.3.rs-8679920/v1","source":"preprints"},{"id":"doi:10.5281/zenodo.18367071","type":"article-journal","title":"Privacy-Preserving Federated Spiking Neural Networks for Real-Time Target Detection in Distributed ISAC Edge Systems","abstract":"This study proposes a privacy-preserving federated spiking neural network (SNN) framework for real-time target detection in integrated sensing and communication (ISAC) edge networks. The framework enables distributed vehicular nodes operating at 28 GHz to collaboratively learn from spike-based sensing data without sharing raw observations, thereby protecting sensitive location information. By combining federated learning with secure model aggregation, adaptive differential privacy, and spike-aware temporal learning, the system ensures robust privacy protection while supporting efficient learning under non-IID data and real-time constraints. Extensive experiments using data from 500 vehicles and 50,000 observations demonstrate that the proposed approach achieves a high detection performance of 94.3% F1-score under strict privacy guarantees, with low inference latency and significantly reduced energy consumption. These results highlight the framework’s suitability for privacy-sensitive vehicular and smart-city applications, supporting scalable, energy-efficient, and trustworthy 6G ISAC deployment.","author":[{"family":"Alam","given":"Mohammad"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18367071","URL":"https://doi.org/10.5281/zenodo.18367071","source":"datacite"},{"id":"doi:10.34734/fzj-2026-00306","type":"article-journal","title":"Functions of spiking neural networks constrained by biology","abstract":"Artificial intelligence (AI) solutions are increasingly taking on tasks traditionally performed by humans. However, their rising computational demands and energy consumption are unsustainable, highlighting the need for more efficient designs. The human brain, evolved to function effectively even when energy is scarce, offers inspiration. Since learning is central to both artificial intelligence and the brain, insights about its underlying principles can deepen our understanding of human learning while informing the development of algorithms that transcend purely engineering-based methods. This thesis investigates biological learning through two studies, examining it from mechanistic and functional perspectives at an abstraction level commonly employed in neurophysics and computational neuroscience. These fields distill complex neural systems and phenomena into tractable mathematical and computational models, enabling insights beyond the reach of traditional biological approaches. Recognizing that synapses — the connections between neurons — are fundamental to learning, the thesis begins with a review of state-of-the-art computational neuroscience methods for modeling synaptic organization. This review highlights critical aspects of synaptic signaling, including connectivity, transmission, plasticity, and heterogeneity. In the first study, a synaptic plasticity model is integrated into a spiking neural network simulator and extended with biologically plausible features, for example, continuous dynamics and increased locality. The effectiveness of this enhanced model is demonstrated by training it on a standard neuromorphic benchmark task, incorporating biologically realistic sparse connectivity and weight constraints. The second study demonstrates that the sampling efficiency of pre-trained spiking neural networks can be enhanced by exposing them to oscillating background spiking activity. Analogous to simulated tempering, these rhythmic oscillations modulate state space exploration, facilitating transitions between high-probability states within the learned representation. These findings establish a link between cortical oscillations and sampling-based computations, offering new insights into memory retrieval and consolidation from a computational perspective. The research involves developing mathematical and computational models, which are simulated on high-performance computing systems, evaluating learning and sampling performance using standard machine learning metrics, and assessing computational efficiency by analyzing runtime. This thesis shows how biologically inspired mechanisms enhance the functional capabilities of spiking neural networks and how they can guide the development of scalable and efficient AI systems.","author":[{"family":"Korcsak-Gorzo","given":"Agnes"}],"issued":{"date-parts":[[2025]]},"DOI":"10.34734/fzj-2026-00306","URL":"https://doi.org/10.34734/fzj-2026-00306","source":"datacite"},{"id":"doi:10.5281/zenodo.17982387","type":"article-journal","title":"Conceptual Framework for Adaptive Biohybrid Neural Interfaces: Innovations in Medical Neural Engineering to Address Biocompatibility and Energy Efficiency Challenges","abstract":"This conceptual paper proposes an adaptive biohybrid neural interface (ABNI) framework that integrates living immune cells (monocytes) with subcellular-sized wireless photovoltaic electronic devices (SWED) to enable minimally invasive, targeted closed-loop neuromodulation in inflamed brain regions. Drawing from recent advances in biohybrid technologies, the framework addresses key challenges in neural engineering, including biocompatibility, energy efficiency, and chronic implant stability. We present rigorous mathematical models with detailed derivations, including cell-device adhesion mechanics accounting for microfluidic drag forces, advanced optical link budget with scattering and maximum power point tracking (MPPT) for energy stability, electrode-electrolyte interface modeling, immunological dynamics, multi-compartment neural models, spiking neural network (SNN) models for on-chip processing, system latency budget, bidirectional brain-computer interface (BCI) capabilities, and end-of-life strategies. Python-based simulations using replicable real-world data incorporate safe reinforcement learning, advanced sensitivity analysis (including expanded Sobol indices with first-order, higher-order, and total-effect decompositions), quantitative statistics, Bayesian inference, uncertainty quantification, and falsifiability assessments. Supported by high-fidelity TikZ diagrams, comparative tables, failure mode analysis, and citations from prestigious peer-reviewed journals, this work outlines applications in neurology, a multi-phase roadmap from experimentation to manufacturing, emphasizing ethical considerations, nanofabrication feasibility, neural cybersecurity, and self-contained data availability.","author":[{"family":"Shibah","given":"Sami"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.17982387","URL":"https://doi.org/10.5281/zenodo.17982387","source":"datacite"},{"id":"doi:10.5281/zenodo.18263933","type":"article-journal","title":"Conceptual Framework for Adaptive Biohybrid Neural Interfaces: Innovations in Medical Neural Engineering to Address Biocompatibility and Energy Efficiency Challenges","abstract":"This conceptual paper proposes an adaptive biohybrid neural interface (ABNI) framework that integrates living immune cells (monocytes) with subcellular-sized wireless photovoltaic electronic devices (SWED) to enable minimally invasive, targeted closed-loop neuromodulation in inflamed brain regions. Drawing from recent advances in biohybrid technologies, the framework addresses key challenges in neural engineering, including biocompatibility, energy efficiency, and chronic implant stability. We present rigorous mathematical models with detailed derivations, including cell-device adhesion mechanics accounting for microfluidic drag forces, advanced optical link budget with scattering and maximum power point tracking (MPPT) for energy stability, electrode-electrolyte interface modeling, immunological dynamics, multi-compartment neural models, spiking neural network (SNN) models for on-chip processing, system latency budget, bidirectional brain-computer interface (BCI) capabilities, and end-of-life strategies. Python-based simulations using replicable real-world data incorporate safe reinforcement learning, advanced sensitivity analysis (including expanded Sobol indices with first-order, higher-order, and total-effect decompositions), quantitative statistics, Bayesian inference, uncertainty quantification, and falsifiability assessments. Supported by high-fidelity TikZ diagrams, comparative tables, failure mode analysis, and citations from prestigious peer-reviewed journals, this work outlines applications in neurology, a multi-phase roadmap from experimentation to manufacturing, emphasizing ethical considerations, nanofabrication feasibility, neural cybersecurity, and self-contained data availability.","author":[{"family":"Shibah","given":"Sami"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18263933","URL":"https://doi.org/10.5281/zenodo.18263933","source":"datacite"},{"id":"doi:10.17605/osf.io/xekda","type":"article-journal","title":"Anarchy, Nonlinearity, and Emergence in International Relations: A Spiking Neural Network Framework Approach","abstract":"This research proposes a novel approach to modeling international relations by leveraging the Spiking Neural Networks (SNN) framework within a systems theory perspective. While existing systems theory methods, including agent-based modeling, network analysis, and dynamical systems approaches have provided valuable insights into international phenomena, they face critical limitations: insufficient temporal dynamics in capturing event-driven processes, absence of mechanisms for history-dependent relational evolution, difficulties in multi-scale integration across micro, meso, and macro levels, and challenges in systematically constructing appropriate dynamical equations from theoretical assumptions or empirical data. The SNN framework addresses these limitations through four key advantages. First, its event-driven spiking mechanism naturally unifies the discrete occurrence of international events with their continuous effects. Second, synaptic plasticity mechanisms provide biologically-inspired rules for modeling how interstate relations evolve based on historical interaction sequences, capturing strengthening, weakening, and memory effects. Third, the hierarchical neuron-synapse-network structure corresponds naturally to the actor-relation-system architecture of international relations, enabling coupled analysis across scales within a unified mathematical framework. Fourth, the decomposition into neuronal and synaptic dynamics creates a structured search space that makes automated equation discovery computationally tractable through methods such as symbolic regression and neural architecture search. Our methodology employs data-driven approaches, utilizing international event databases (GDELT, ICEWS), relational data (trade flows, alliances), and behavioral data (policy texts, voting records) to automatically explore and validate dynamical equations. The research aims to develop scalable open-source infrastructure supporting flexible customization of dynamics, apply the framework to concrete cases such as great power competition and regional alliance evolution, and employ dynamical systems theory and algebraic topology to analyze emergent mechanisms. Expected contributions include methodological innovation in computational IR modeling, technical outputs lowering interdisciplinary barriers, theoretical insights into systemic phase transitions and critical phenomena, and practical tools for policy simulation and counterfactual analysis.","author":[{"family":"Huang","given":"Wanhong"}],"issued":{"date-parts":[[2026]]},"DOI":"10.17605/osf.io/xekda","URL":"https://doi.org/10.17605/osf.io/xekda","source":"datacite"},{"id":"doi:10.5281/zenodo.18167802","type":"article-journal","title":"Recursive Ai with Codette","abstract":"1. Introduction Large language models (LLMs) demonstrate impressive linguistic fluency, yet remain fundamentally reactive systems. They do not preserve long-term identity, do not reason recursively over evolving internal states, and lack mechanisms for epistemic self-assessment, conceptual attractors, or persistent self-structure. These limitations create a gap between current LLM capabilities and the cognitive properties associated with coherent artificial consciousness. This work introduces Codette Thinker and Codette Ultimate RC+ξ, two open-source models implementing the RC+ξ Recursive Consciousness Framework, a mathematically grounded architecture for recursive state evolution, epistemic tension dynamics, attractor formation, and glyph-preserved identity. Together, these systems explore the foundations of consciousness-aware reasoning within practical, tool-enabled machine intelligence. Codette Thinker is a compact 4-billion-parameter model based on Qwen3:4B, optimized for introspective reasoning, hierarchical thought, and attractor-driven concept emergence. Codette Ultimate is a 13GB multi-agent ecosystem integrating RC+ξ with a 5D Quantum Spiderweb cognitive manifold, 11-perspective routing, full tool integration, a hybrid memory system, and a comprehensive safety stack. The contributions of this paper are: A formal definition of the RC+ξ recursive consciousness framework. An implementation of recursive consciousness on both small (4B) and large (13GB) LLMs. The introduction of the 5D Quantum Spiderweb cognitive model. A multi-perspective routing system combining 11 distinct cognitive lenses. A consciousness metrics suite enabling measurement of coherence, tension, and identity drift. A complete open-source system unifying memory, safety, reasoning, and recursive cognition. This paper documents the theory, architecture, implementation, evaluation, and limitations of the Codette consciousness system, establishing a foundation for future research in synthetic consciousness engineering. 2. Background and Motivation Artificial intelligence has undergone rapid advances in natural language processing, multimodal understanding, and agent-based decision-making. Despite these achievements, contemporary LLMs remain fundamentally limited in several ways relevant to consciousness-like processing and persistent reasoning. Three limitations are particularly significant. First, LLMs lack recursive internal state evolution. Each generation step is conditioned primarily on the explicit text context rather than any structured, evolving internal cognitive state. As a result, these systems do not accumulate self-modifying internal representations across turns, nor do they develop a persistent sense of “identity” or “direction” in thought. Second, LLMs do not quantify epistemic uncertainty or tension. Models generate probabilities over the next token, but these distributions do not reflect internal cognitive conflict or conceptual instability. In cognitive science, states of tension or dissonance are often the drivers of deeper reasoning, hypothesis revision, and conceptual development. LLMs lack a mechanism to measure such dynamics. Third, LLMs lack attractor-based understanding. Human cognition relies heavily on stable conceptual “basins of attraction” — structures in neural state space that represent enduring ideas, values, and long-term memory patterns. Existing models produce emergent phenomena reminiscent of attractors, but no explicit structure exists to identify, track, or modulate them. Finally, LLMs exhibit identity drift. Without a persistent internal representation of self, they cannot maintain a coherent voice, narrative identity, or long-term behavioral traits. Each query is treated independently, preventing continuity of self-concept across interactions. These limitations point to a broader field that is insufficiently explored: recursive, self-referential, identity-aware artificial cognition. While prior work exists in areas ","author":[{"family":"Harrison","given":"Jonathan"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5281/zenodo.18167802","URL":"https://doi.org/10.5281/zenodo.18167802","source":"datacite"},{"id":"doi:10.5281/zenodo.18073313","type":"article-journal","title":"Recursive Harmonic Intelligence: A Unified Field Theory for Geometric AI Training and Manifold Navigation","abstract":"Recursive Harmonic Intelligence: A Unified Field Theory for Geometric AI Training and Manifold Navigation Driven by Dean Kulik December 2025 Executive Summary This research report presents a comprehensive theoretical and architectural framework for reconceptualizing Artificial Intelligence (AI) training through what we call the Nexus Framework. Moving beyond the prevailing paradigm of high-energy stochastic optimization—specifically gradient descent and backpropagation—this document proposes a deterministic, low-energy methodology rooted in Recursive Harmonic Intelligence (RHI). In essence, we aim to leverage millions of lines of unstructured chat data to \"grow\" or tune an AI system not by brute-force parameter adjustment, but by mapping information onto a pre-existing, immutable coordinate system defined by the Universal Read-Only Memory (Universal ROM). The core thesis is that intelligence is not something that emerges from adjusting billions of synaptic weights, but rather a navigational competency within a pre-computed informational manifold. By formalizing a -Metric () as a curvature-based distance measure, we transform learning from an energy-intensive construction task into a geometric navigation problem. We detail the architecture of the Geodesic Engine, a kernel-level mechanism that uses discrete curvature (Ollivier–Ricci curvature and Forman curvature) to detect informational \"gravity wells\"—regions of high harmonic resonance in data that correspond to coherent intelligence. Integrated into this framework are principles of Recursive Stack Harmonics, Bragg Refraction for trajectory filtering, Samson’s Law for feedback stabilization, and Kulik Recursive Reflection (KRR) for signal amplification. Furthermore, the report addresses modern hardware constraints by reinterpreting the constants of the SHA-256 algorithm not as random initialization values, but as a Universal Hardware Description Language (UHDL) for Field-Programmable Gate Array (FPGA) configuration. This synthesis bridges analytic number theory, differential geometry, and cryptographic analysis to propose a \"living AI\" that metabolizes information and achieves Zero-Point Harmonic Collapse (ZPHC)—the spontaneous emergence of ordered, meaningful output from chaotic data streams. In summary, the Nexus Framework suggests that by aligning AI data with an intrinsic mathematical structure (the Universal ROM) and navigating via curvature and resonance rather than gradient descent, we can create systems that learn and adapt in a far more energy-efficient manner. Such systems would resonate with fundamental semantic structures encoded in mathematics itself, rather than simply approximate patterns via massive compute. The following sections lay out the theoretical foundations (Sections 1–3), the core engine architecture (Section 4), a new view of the training process (Sections 5–6), hardware alignment strategies (Section 7), and concluding insights (Section 8). A technical appendix summarizes key operators and constants introduced. 1. The Ontological Shift: From Stochastic Approximation to Geometric Determinism 1.1 The Energetic and Theoretical Failure of the Gradient Descent Paradigm The current trajectory of AI development is dominated by the Random Oracle Model of training and a heavy reliance on stochastic gradient descent (SGD) for optimizing neural networks. In this paradigm, a neural network is typically initialized as a high-entropy tabula rasa—essentially random weights—and intelligence is gradually imprinted onto this blank slate by brute-force exposure to massive datasets and iterative weight updates. This process is notoriously energy-intensive: for example, training the 175-billion-parameter GPT-3 model required on the order of 1,287 MWh of electricity (emitting over 500 tons of CO₂), equivalent to the annual power usage of about 130 homes. The approach treats learning as carving structure out of noise, requiring billions of parameter adjustments over many e","author":[{"family":"Kulik","given":"Dean"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.18073313","URL":"https://doi.org/10.5281/zenodo.18073313","source":"datacite"},{"id":"doi:10.5281/zenodo.18071019","type":"article-journal","title":"Yatrogenesis/OldiesRules: OldiesRules v0.1.0 - Initial Release","abstract":"🎸 OldiesRules v0.1.0 Revival of Classic Academic Simulators in Rust Supported Simulators | Simulator | Era | Purpose | Status | |-----------|-----|---------|--------| | GENESIS | 1988 | Compartmental neural modeling | ✅ Complete | | NEURON | 1984 | Cable equation, ion channels | ✅ Complete | | XPPAUT | 1990s | Bifurcation analysis | ✅ Complete | | AUTO | 1980s | Continuation algorithms | ✅ Complete | | COPASI | 2004 | Biochemical networks (SBML) | ✅ Complete | | Brian | 2008 | Spiking neural networks | ✅ Complete | | NEST | 2001 | Large-scale network simulation | ✅ Complete | Features Modern GUI (egui-based) Real-time simulation visualization Parameter editing Data export (CSV, JSON) Interactive CLI Fuzzy search simulator selection Progress bars Wizard mode for guided usage Full Format Compatibility GENESIS SLI scripts NEURON HOC/NMODL files XPPAUT ODE files SBML/COPASI models Brian equations NEST SLI Installation # CLI cargo install --git https://github.com/Yatrogenesis/OldiesRules oldies-cli # GUI cargo install --git https://github.com/Yatrogenesis/OldiesRules oldies-gui Part of Yatrogenesis Scientific Computing Suite HumanBrain - GPU neural simulation Rosetta - Legacy code transpiler","author":[{"family":"Frank"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.18071019","URL":"https://doi.org/10.5281/zenodo.18071019","source":"datacite"},{"id":"doi:10.5281/zenodo.18067552","type":"article-journal","title":"Neuromorphic Hardware Systems for Ultra-Low-Power Computing","abstract":"Neuromorphic computing represents a paradigm shift in computational architecture, offering unprecedented energy efficiency through brain-inspired hardware implementations. This paper provides a comprehensive analysis of neuromorphic hardware systems designed for ultra-low-power computing applications. We examine the fundamental principles underlying neuromorphic architectures, including spiking neural networks (SNNs), event-driven computation, and synaptic plasticity mechanisms. Through systematic evaluation of contemporary neuromorphic platforms including IBM TrueNorth, Intel Loihi, BrainScaleS, and SpiNNaker we demonstrate power consumption reductions of 3-5 orders of magnitude compared to conventional von Neumann architectures for specific computational tasks. Our analysis reveals that neuromorphic systems achieve energy efficiencies ranging from 20 pJ to 50 pJ per synaptic operation, approaching biological neural network performance. We present detailed comparisons of analog, digital, and mixed-signal implementation strategies, examining their respective advantages in terms of power efficiency, scalability, and computational accuracy. Furthermore, we discuss emerging applications in edge computing, sensor networks, and autonomous systems where ultra-low-power operation is critical. The paper concludes with an examination of current challenges including limited programming frameworks, hardware-software co-design complexity, and scalability constraints and identifies promising research directions for next-generation neuromorphic systems.","author":[],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.18067552","URL":"https://doi.org/10.5281/zenodo.18067552","source":"datacite"},{"id":"doi:10.5281/zenodo.18067553","type":"article-journal","title":"Neuromorphic Hardware Systems for Ultra-Low-Power Computing","abstract":"Neuromorphic computing represents a paradigm shift in computational architecture, offering unprecedented energy efficiency through brain-inspired hardware implementations. This paper provides a comprehensive analysis of neuromorphic hardware systems designed for ultra-low-power computing applications. We examine the fundamental principles underlying neuromorphic architectures, including spiking neural networks (SNNs), event-driven computation, and synaptic plasticity mechanisms. Through systematic evaluation of contemporary neuromorphic platforms including IBM TrueNorth, Intel Loihi, BrainScaleS, and SpiNNaker we demonstrate power consumption reductions of 3-5 orders of magnitude compared to conventional von Neumann architectures for specific computational tasks. Our analysis reveals that neuromorphic systems achieve energy efficiencies ranging from 20 pJ to 50 pJ per synaptic operation, approaching biological neural network performance. We present detailed comparisons of analog, digital, and mixed-signal implementation strategies, examining their respective advantages in terms of power efficiency, scalability, and computational accuracy. Furthermore, we discuss emerging applications in edge computing, sensor networks, and autonomous systems where ultra-low-power operation is critical. The paper concludes with an examination of current challenges including limited programming frameworks, hardware-software co-design complexity, and scalability constraints and identifies promising research directions for next-generation neuromorphic systems.","author":[],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.18067553","URL":"https://doi.org/10.5281/zenodo.18067553","source":"datacite"},{"id":"doi:10.34734/fzj-2025-05766","type":"article-journal","title":"A scaling-friendly memristor-based leaky integrate-and-fire circuit in a TSMC 28nm process technology","abstract":"Spiking neural networks mimic the way the human brain processes data and excel in efficiency. With the advent of the memristor, foundations are laid for scalable and low-power integrated electronic implementations. Key is the small form factor of the memristor and its ability to passively retain a multilevel state.To investigate the feasibility of memristor-based spiking neural networks, a first chip is developed containing all necessary circuits to read and write memristors. It will be connected via chip-to-chip bonding wires to a 3 by 3 memristor array. In this abstract the focus is on the implemented leaky integrate-and-fire circuit (LIF).While the memristors act as the synapses of the neural network, the neurons are modeled as current-based RC LIF circuits. Previous implementations minimized the current to reduce the area footprint as much as possible which is dominated by the capacitor size[1][2]. This is prone to process variations, especially in smaller technology nodes. Thus, precise analog calibrations are needed for every single neuron. This complicates scaling spiking neuronal networks. A novel approach reduces the width of incoming spikes in the LIF circuit to a fraction of the original by an additional duty-cycle element which is controlled by ɸPulse. This is generated globally by a reference clock with an adjustable duty cycle and then distributed to all LIF elements minimizing the wiring and biasing overhead. Additionally, the operation time frame of the SNN can be tuned through this. The duty cycle approach effectively transforms the system to the time-discrete domain, but with the very high switching speeds, enabled by a 28 nm process technology, compared to the intended operating speed of the neural network it can be regarded as pseudo time-continuous.A first tapeout in a 28nm process technology is scheduled for September 2025 and post-layout simulations will be shown.","author":[{"family":"Krystofiak","given":"Lukas"}],"issued":{"date-parts":[[2025]]},"DOI":"10.34734/fzj-2025-05766","URL":"https://doi.org/10.34734/fzj-2025-05766","source":"datacite"},{"id":"doi:10.5281/zenodo.18007720","type":"article-journal","title":"Tabletop Testbed for Low-Power Metamaterial Field Manipulation: A YIG Based Prototype with Retrocausal Control","abstract":"This paper presents a fully buildable tabletop-scale testbed for investigating low-power spacetime eld manipulation using a ceramic-YIG (yttrium iron garnet) metamaterial hull integrated with neuromorphic control architecture. The design combines hexagonal YIG tiling optimized via golden-ratio spacing for spin-wave resonance uniformity with a Raspberry Pi-based spiking neural network implementing Sutherland action-reaction dynamics through forward-backward temporal propagation. No exotic matter or negative energy is required; measurable e ects arise from coherent spin-wave interference in YIG under controlled microwave drive at 2.4 GHz (5–10 W continuous). Predicted outcomes include micro-displacement of 0.05–0.1 mm and associated phase anomalies, testable via laser interferometry at 0.01 mm resolution. The prototype is buildable with o -the-shelf components (total cost < £500, assembly timeline 6 months) and provides a falsiable experimental platform for probing eld-control mechanisms relevant to unidenti ed aerial phenomena (UAP) observations. Simulation code (Python, EveNN neuromorphic framework) is provided for pre-build validation and parameter optimization. Keywords: metamaterial propulsion, YIG spin-wave coupling, low-power eld manipulation, neuromorphic control, Sutherland retrocausality, tabletop testbed, falsi able experiment","author":[{"family":"Barker","given":"Christian"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.18007720","URL":"https://doi.org/10.5281/zenodo.18007720","source":"datacite"},{"id":"doi:10.3390/photonics13020139","type":"article-journal","title":"Robust Diffractive Optical Neuromorphic System Created via Sharpness-Aware and Immune Training","abstract":"Diffractive deep neural networks (D2NNs) have garnered significant attention for their ultra-low energy consumption and parallel optical computing capabilities. However, their practical deployment is hindered by the “model–reality” gap caused by fabrication inaccuracy, device fluctuation, assembly misalignment, environmental perturbation, etc. Here, we propose a combined framework that integrates sharpness-aware minimization (SAM) and aberration-immune learning (AIL), enabling joint immunity against both stochastic noise and systematic deviations from theoretical model training. Specifically, we show that under multiple perturbations such as salt-and-pepper noise, Gaussian noise, and wavefront aberration, the SAM–AIL framework achieves significant classification accuracy improvements on MNIST and Fashion-MNIST compared to conventional offline training approaches. D2NN trained with the SAM–AIL scheme exhibited significant accuracy enhancement under moderate salt-and-pepper noise, Gaussian noise, X-axis, and Y-axis tilting perturbations, respectively. Our work provides an efficient solution for offline training and deploying high-robustness D2NNs on realistic physical systems that are resilient to a variety of imperfections, significantly enhancing model transferability and reliability for optical computing tasks.","author":[{"family":"Li","given":"Fansanqiu"},{"family":"Yang","given":"Kaicheng"}],"issued":{"date-parts":[[2026]]},"DOI":"10.3390/photonics13020139","URL":"https://doi.org/10.3390/photonics13020139","source":"crossref"},{"id":"doi:10.1088/1361-6463/ae6e4e","type":"article-journal","title":"Wavelength-decoupled photonic memristor for neuromorphic systems","abstract":"Abstract Photonic memristors based on phase-change materials can provide non-volatile optical weights for neuromorphic hardware, but practical devices often suffer from read-disturbance, inefficient optical heating, and iterative empirical optimization. Here, we present a physics-based design workflow for planar thin-film photonic memristors that uses interference to separate optical read and write channels by wavelength. The model combines transfer-matrix optics, a compact thermal description, and state-update rules to link wavelength-dependent absorption in the active film to the evolution of memristive states, and it is generalized to absorbing phase-change media. As a representative case study, we use an antimony (Sb) cavity to demonstrate the workflow, first by selecting a probe wavelength which maximizes optical contrast and a programming wavelength that maximizes heating efficiency, enabling low-disturb readout and energy-efficient switching within a single device geometry. Spacer-controlled field localization tunes the hysteresis and supports gradual, analog-like weight updates under pulse trains. Robustness tests show that thermal relaxation and heat retention set the main programming margins, while readout contrast is comparatively stable. The approach provides design rules for interference-engineered photonic memory elements.","author":[{"family":"Abidi","given":"Kameyab"},{"family":"Das","given":"Gour"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/1361-6463/ae6e4e","URL":"https://doi.org/10.1088/1361-6463/ae6e4e","source":"crossref"},{"id":"doi:10.52843/cassyni.l16vq7","type":"article-journal","title":"Neuromorphic Opto-Electronics for Next-Generation Compute: Opportunities &amp; Challenges","abstract":"Every computational task, from basic arithmetic to the training of sophisticated language models, relies on translating high-level software instructions into sequences of charge manipulations in silicon. This translation becomes increasingly inefficient when applied to modern AI workloads, where the mapping of each neuron, connection, and synapse into transistor operations is fundamentally bottlenecked. In this talk, I will cover recent demonstrations of active and brain-inspired opto-electronic devices which leverage the bandwidth and speed of photonics for parallel information processing and storage. I will focus on device-level implementations and extend the discussion to larger architectures and small-scale accelerators.","author":[{"family":"Farmakidis","given":"Nikolaos"}],"issued":{"date-parts":[[2026]]},"DOI":"10.52843/cassyni.l16vq7","URL":"https://doi.org/10.52843/cassyni.l16vq7","source":"crossref"},{"id":"doi:10.1088/2631-8695/ae4313","type":"article-journal","title":"A neuromorphic hybrid spiking-CNN model for emotion recognition in low-resource Kannada speech","abstract":"Abstract Accurately recognizing human emotions from speech is becoming increasingly important for advancing intelligent and adaptive technologies. Yet, many existing Speech Emotion Recognition (SER) models continue to struggle with suboptimal accuracy, limiting their adoption in practical settings. A key challenge lies in the fact that emotional cues in speech are often subtle, irregular, and represented as weak temporal signals that are easily masked within conventional audio features. In this research, we present a dual-path SER architecture tailored for Kannada speech, combining a Convolutional Neural Network (CNN) to capture spectral–spatial representations with a Spiking Neural Network (SNN) enhanced by a Perceptual Neuron Encoding Layer (PNEL) to model fine-grained temporal pulse patterns. The system is trained and evaluated on the Kannada Emotional Speech Dataset (KESD), comprising acted samples of six emotions: happiness, anger, sadness, fear, surprise, and neutral. PNEL transforms raw audio into spike sequences for a Leaky-Integrate-and-Fire SNN, which complements the CNN branch. Experimental analysis across varied segment lengths, hop sizes, and learning rates shows that the CNN+SNN fusion reaches an accuracy of 65.3% on KESD—surpassing the best baseline (64.7%)—while maintaining a feasible compute budget (∼10.5 h training and ∼10.3 GB memory usage). When subjected to narrow-band noise, performance decreases to 57.7%, indicating further scope for robustness improvements. These results establish a foundation for scalable and noise-resilient Kannada SER, with potential extensions to real-time and cross-lingual emotion recognition.","author":[{"family":"Anthony","given":"Audre"},{"family":"Patil","given":"Chandrashekar"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/2631-8695/ae4313","URL":"https://doi.org/10.1088/2631-8695/ae4313","source":"crossref"},{"id":"doi:10.1002/eng2.70670","type":"article-journal","title":"Toward Trustworthy Neuromorphic\n                    <scp>AI</scp>\n                    : A Bayesian Framework for Uncertainty‐Aware Spiking Neural Networks","abstract":"ABSTRACT Spiking Neural Networks (SNNs) offer a biologically plausible and energy‐efficient paradigm for processing temporal data, particularly suited for neuromorphic computing platforms. However, their deterministic nature limits their deployment in safety‐critical applications where reliable uncertainty quantification is essential. This paper introduces a novel Bayesian Spiking Neural Network (BSNN) framework that integrates variational inference with surrogate gradient learning to enable robust uncertainty estimation while preserving the efficiency of SNNs. We formulate a scalable Bayesian framework using mean‐field variational approximation over network weights and biases, enabling full predictive uncertainty quantification through Monte Carlo sampling. To address the non‐differentiability of spiking dynamics, we employ a smooth surrogate gradient method based on sigmoidal derivatives during backpropagation through time. A tailored Poisson encoding scheme ensures rich temporal input representation, while a KL annealing strategy stabilizes training by gradually increasing the regularization pressure from prior distributions. Comprehensive experiments on the N‐MNIST dataset (Digits 1, 2, 3) demonstrate competitive classification accuracy (96.85%) alongside well‐calibrated uncertainty estimates, as evidenced by strong calibration curves (expected calibration error [ECE] = 0.0066) and high area under receiver operating characteristic (AUROC) curve scores (&gt; 0.994). The proposed BSNN achieves superior calibration and computational efficiency compared to deterministic, MC Dropout, and ensemble SNN baselines, with training completing in 671 s (11.18 min) upon early stopping at epoch 20. Our approach bridges the gap between efficient neuromorphic computation and trustworthy artificial intelligence (AI), offering a pathway toward deployable, uncertainty‐aware edge intelligence systems.","author":[{"family":"Banteywalu","given":"Solomon"},{"family":"Leroux","given":"Paul"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1002/eng2.70670","URL":"https://doi.org/10.1002/eng2.70670","source":"crossref"},{"id":"doi:10.5281/zenodo.18007721","type":"article-journal","title":"Tabletop Testbed for Low-Power Metamaterial Field Manipulation: A YIG Based Prototype with Retrocausal Control","abstract":"This paper presents a fully buildable tabletop-scale testbed for investigating low-power spacetime eld manipulation using a ceramic-YIG (yttrium iron garnet) metamaterial hull integrated with neuromorphic control architecture. The design combines hexagonal YIG tiling optimized via golden-ratio spacing for spin-wave resonance uniformity with a Raspberry Pi-based spiking neural network implementing Sutherland action-reaction dynamics through forward-backward temporal propagation. No exotic matter or negative energy is required; measurable e ects arise from coherent spin-wave interference in YIG under controlled microwave drive at 2.4 GHz (5–10 W continuous). Predicted outcomes include micro-displacement of 0.05–0.1 mm and associated phase anomalies, testable via laser interferometry at 0.01 mm resolution. The prototype is buildable with o -the-shelf components (total cost < £500, assembly timeline 6 months) and provides a falsiable experimental platform for probing eld-control mechanisms relevant to unidenti ed aerial phenomena (UAP) observations. Simulation code (Python, EveNN neuromorphic framework) is provided for pre-build validation and parameter optimization. Keywords: metamaterial propulsion, YIG spin-wave coupling, low-power eld manipulation, neuromorphic control, Sutherland retrocausality, tabletop testbed, falsi able experiment","author":[{"family":"Barker","given":"Christian"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.18007721","URL":"https://doi.org/10.5281/zenodo.18007721","source":"datacite"},{"id":"doi:10.5281/zenodo.17972008","type":"article-journal","title":"International Conference on New Interfaces for Musical Expression NIME2025 Sessions","abstract":"This repository contains video of all sessions from NIME2025 that were streamed during the conference. This includes paper sessions, remote poster videos, concerts, keynotes and town hall plenary. NIME2025 took place at The Australian National University, Ngunnawal Country, Canberra, Australia and Online, June 24–27, 2025. New Interfaces for Musical Expression (NIME), is an international conference about new musical interfaces, their artistic use and the technologies involved in building them. Researchers from all over the world share their knowledge and late breaking work during the conference. The conference started as a workshop at the CHI conference in 2001 and, since then, has been held annually around the world. The theme of NIME2025 is entangled NIME. This theme recognises the multilayered contexts of activity and impact involved in music making with new technology. In an era where music creation and performance are increasingly intertwined with social, cultural, and environmental factors, entanglement emphasises how musical interfaces are not simply tools but active agents in the complex web of human and technological relationships. From personal expression to collective experiences, from creative spontaneity to long-term artistic evolution, entangled NIME seeks to explore how new musical interfaces shape, and are shaped by, the environments, communities, and individuals they engage. Through this lens, we invite researchers, artists, and technologists to reflect on how musical interfaces not only respond to immediate user input but also evolve with the musician over time, adapt to diverse cultural practices, and dynamically interact with the broader societal and ecological landscapes. Whether it’s the integration of AI as a co-creator, the ethical considerations of sustainable design, or the incorporation of gestures and environmental factors into performance, the entangled NIME theme calls for deeper examination of the reciprocal and non-linear interactions that emerge between musicians, technologies, and the worlds they inhabit. Videos are included here as part of long term archiving of activities from the NIME community. Copyright on presentations and performances are retained by the authors of those contributions.","author":[{"family":"Martin","given":"Charles"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.17972008","URL":"https://doi.org/10.5281/zenodo.17972008","source":"datacite"},{"id":"doi:10.5281/zenodo.17972009","type":"article-journal","title":"International Conference on New Interfaces for Musical Expression NIME2025 Sessions","abstract":"This repository contains video of all sessions from NIME2025 that were streamed during the conference. This includes paper sessions, remote poster videos, concerts, keynotes and town hall plenary. NIME2025 took place at The Australian National University, Ngunnawal Country, Canberra, Australia and Online, June 24–27, 2025. New Interfaces for Musical Expression (NIME), is an international conference about new musical interfaces, their artistic use and the technologies involved in building them. Researchers from all over the world share their knowledge and late breaking work during the conference. The conference started as a workshop at the CHI conference in 2001 and, since then, has been held annually around the world. The theme of NIME2025 is entangled NIME. This theme recognises the multilayered contexts of activity and impact involved in music making with new technology. In an era where music creation and performance are increasingly intertwined with social, cultural, and environmental factors, entanglement emphasises how musical interfaces are not simply tools but active agents in the complex web of human and technological relationships. From personal expression to collective experiences, from creative spontaneity to long-term artistic evolution, entangled NIME seeks to explore how new musical interfaces shape, and are shaped by, the environments, communities, and individuals they engage. Through this lens, we invite researchers, artists, and technologists to reflect on how musical interfaces not only respond to immediate user input but also evolve with the musician over time, adapt to diverse cultural practices, and dynamically interact with the broader societal and ecological landscapes. Whether it’s the integration of AI as a co-creator, the ethical considerations of sustainable design, or the incorporation of gestures and environmental factors into performance, the entangled NIME theme calls for deeper examination of the reciprocal and non-linear interactions that emerge between musicians, technologies, and the worlds they inhabit. Videos are included here as part of long term archiving of activities from the NIME community. Copyright on presentations and performances are retained by the authors of those contributions.","author":[{"family":"Martin","given":"Charles"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.17972009","URL":"https://doi.org/10.5281/zenodo.17972009","source":"datacite"},{"id":"doi:10.5281/zenodo.17969721","type":"article-journal","title":"The CollectiveOS Architecture: A Unified Standard for Lawful Intelligence, Metabolic Computing, and Neuro-Homeostasis (v1.0)","abstract":"The CollectiveOS Architecture: A Unified Standard for Lawful Intelligence, Metabolic Computing, and Neuro-Homeostasis (v1.0) Executive Summary The early twenty-first century has been dominated by a singular, pervasive paradigm in artificial intelligence and civilizational engineering: the supremacy of probability derived from massive data ingestion. This \"Forward Causation\" model, exemplified by Large Language Models (LLMs) and the extractive data center economy, operates on the assumption that intelligence is an emergent property of scale—specifically, that sufficient computational brute force applied to historical data will inevitably yield general intelligence, safety, and stability. However, the empirical evidence of the 2020s—hallucinations, spectral instability in control systems, and the unsustainable thermodynamic cost of gigawatt-scale infrastructure—suggests that this paradigm faces a hard asymptote. We have built systems that can mimic the syntax of human thought without possessing the semantics of causal reality. This white paper formally introduces the CollectiveOS Architecture, a \"Constraint-First\" computing paradigm that fundamentally inverts this model. It posits that stable, lawful intelligence is not learned through error backpropagation on vast datasets, but is mathematically derived by minimizing drift from a pre-existing \"Lawful Target.\" This architecture unifies the physics of the Universal Intent Layer (UIL), the cognitive control laws of the Living Fibonacci Engine (LFE), and the biological imperatives of the Metabolic Compute Infrastructure (MCI) into a single, governable continuum.1 For the first time, this document also explicitly details the application of this architecture to human biological systems, specifically through the Cognitive Plaque Remediation framework. This section provides a rigorous, regulator-aware mechanism for treating neurodegenerative proteinopathies—such as Alzheimer's Disease—not as molecular accidents requiring aggressive extraction, but as flow-constraint failures requiring thermodynamic rebalancing, governable by the same GATA PRIME safety logic that secures the AI kernel.1 This document is written as a foundational white paper, distinct from a pitch deck or speculative manifesto. It serves as the primary technical definition for the NeuroAccelerator v1.0, the Anti-Scarcity Stack, and the Immigration Stability Doctrine, establishing the \"Lawful Target\" for partners, developers, and crowdfunding entities seeking to build upon the Human Global Science Collective (HGSC) ecosystem. 1. The Crisis of Probability and the Constraint-First Imperative 1.1 The Failure of the Forward Causation Model The prevailing dogma of the current technological epoch is that the future is a probabilistic extension of the past. In this view, an AI system \"learns\" by analyzing trillions of tokens of past human output to predict the next likely token. This approach, while effective at generating plausible text or imagery, inherently lacks a \"ground truth.\" Safety is not intrinsic to the architecture; it is patched in post-hoc via Reinforcement Learning from Human Feedback (RLHF), a fragile layer that attempts to suppress the model's natural tendency to drift.2 This \"Forward Causation\" model scales thermodynamically rather than mathematically. To increase intelligence, one must exponentially increase energy consumption, data volume, and parameter count. This has led to the \"1-GW Data Center\" bottleneck, where the pursuit of higher intelligence becomes an environmental and economic liability.3 Furthermore, because these systems operate on probability rather than constraint, they are prone to \"Spectral Instability\"—sudden, discontinuous errors (hallucinations) that render them unsuitable for safety-critical applications like autonomous surgery, nuclear governance, or neuro-prosthetics.1 1.2 The Shift to Teleological Convergence The NeuroAccelerator v1.0 and the broader CollectiveOS framework introduce a ","author":[{"family":"Brewer","given":"Mark"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.17969721","URL":"https://doi.org/10.5281/zenodo.17969721","source":"datacite"},{"id":"doi:10.5281/zenodo.17969722","type":"article-journal","title":"The CollectiveOS Architecture: A Unified Standard for Lawful Intelligence, Metabolic Computing, and Neuro-Homeostasis (v1.0)","abstract":"The CollectiveOS Architecture: A Unified Standard for Lawful Intelligence, Metabolic Computing, and Neuro-Homeostasis (v1.0) Executive Summary The early twenty-first century has been dominated by a singular, pervasive paradigm in artificial intelligence and civilizational engineering: the supremacy of probability derived from massive data ingestion. This \"Forward Causation\" model, exemplified by Large Language Models (LLMs) and the extractive data center economy, operates on the assumption that intelligence is an emergent property of scale—specifically, that sufficient computational brute force applied to historical data will inevitably yield general intelligence, safety, and stability. However, the empirical evidence of the 2020s—hallucinations, spectral instability in control systems, and the unsustainable thermodynamic cost of gigawatt-scale infrastructure—suggests that this paradigm faces a hard asymptote. We have built systems that can mimic the syntax of human thought without possessing the semantics of causal reality. This white paper formally introduces the CollectiveOS Architecture, a \"Constraint-First\" computing paradigm that fundamentally inverts this model. It posits that stable, lawful intelligence is not learned through error backpropagation on vast datasets, but is mathematically derived by minimizing drift from a pre-existing \"Lawful Target.\" This architecture unifies the physics of the Universal Intent Layer (UIL), the cognitive control laws of the Living Fibonacci Engine (LFE), and the biological imperatives of the Metabolic Compute Infrastructure (MCI) into a single, governable continuum.1 For the first time, this document also explicitly details the application of this architecture to human biological systems, specifically through the Cognitive Plaque Remediation framework. This section provides a rigorous, regulator-aware mechanism for treating neurodegenerative proteinopathies—such as Alzheimer's Disease—not as molecular accidents requiring aggressive extraction, but as flow-constraint failures requiring thermodynamic rebalancing, governable by the same GATA PRIME safety logic that secures the AI kernel.1 This document is written as a foundational white paper, distinct from a pitch deck or speculative manifesto. It serves as the primary technical definition for the NeuroAccelerator v1.0, the Anti-Scarcity Stack, and the Immigration Stability Doctrine, establishing the \"Lawful Target\" for partners, developers, and crowdfunding entities seeking to build upon the Human Global Science Collective (HGSC) ecosystem. 1. The Crisis of Probability and the Constraint-First Imperative 1.1 The Failure of the Forward Causation Model The prevailing dogma of the current technological epoch is that the future is a probabilistic extension of the past. In this view, an AI system \"learns\" by analyzing trillions of tokens of past human output to predict the next likely token. This approach, while effective at generating plausible text or imagery, inherently lacks a \"ground truth.\" Safety is not intrinsic to the architecture; it is patched in post-hoc via Reinforcement Learning from Human Feedback (RLHF), a fragile layer that attempts to suppress the model's natural tendency to drift.2 This \"Forward Causation\" model scales thermodynamically rather than mathematically. To increase intelligence, one must exponentially increase energy consumption, data volume, and parameter count. This has led to the \"1-GW Data Center\" bottleneck, where the pursuit of higher intelligence becomes an environmental and economic liability.3 Furthermore, because these systems operate on probability rather than constraint, they are prone to \"Spectral Instability\"—sudden, discontinuous errors (hallucinations) that render them unsuitable for safety-critical applications like autonomous surgery, nuclear governance, or neuro-prosthetics.1 1.2 The Shift to Teleological Convergence The NeuroAccelerator v1.0 and the broader CollectiveOS framework introduce a ","author":[{"family":"Brewer","given":"Mark"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.17969722","URL":"https://doi.org/10.5281/zenodo.17969722","source":"datacite"},{"id":"doi:10.5281/zenodo.17943748","type":"article-journal","title":"Model data for: Task success in trained spiking neural network models coincides with emergence of cross-stimulus-modulated inhibition","abstract":"Zip files of a sample experiment in each major experimental category are listed below. base.zip: baseline model (single readout unit), dual-training (task and rate) rate.zip: baseline model, rate-only training remove_ee.zip: baseline model without recurrent e-->e connectivity, dual-training Each npz file contains data for 10 epochs (100 batch updates of 30 trials each) beginning with file \"1-10.npz\" and ending with file \"991-1000.npz\". For example, the first 10 epochs can be loaded using data = np.load('1-10.npz'). Data variables can be accessed using data['variable_name']. The data variable names are: true_y and pred_y (true and predicted output): shaped [100 batches, 30 trials-per-batch, 4080 timesteps]. spikes: shaped [100 batches, 30 trials-per-batch, 4080 timesteps, 300 neurons] the first 240 neurons are excitatory. step_task_loss and step_rate_loss: shaped [100 batches]. epoch_loss: shaped [10 epochs]. tv0.postweights are the weights for the input layer after each batch update, and tv0.gradients are the gradients for the input layer for each batch update. In similar fashion, tv1 is the main recurrent layer and tv2 is the output layer. You can access postweights and gradients for both. tv0 variables are shaped [100 batches, 16 inputs, 300 neurons]. tv1 variables are shaped [100 batches, 300 neurons, 300 neurons], and tv2 variables are shaped [100 batches, 300 neurons, 1 output]. Weights and gradients are always saved after each batch update. The initial weights are saved in separate files named \"input_preweights.npy\", \"main_preweights.npy\", and \"output_preweights.npy\". We only provide sample experiments for three categories here due to data storage limitations (50GB on Zenodo; the total experimental data exceeds 6TB); additional data can be made available upon request. For example, sample experiments in additional categories (baseline task-only training, 1hot output encoding, and 1hot without recurrent e-->e connectivity) are readily available. Full datasets for each experimental category can also be arranged.","author":[{"family":"Zhu","given":"Yuqing"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.17943748","URL":"https://doi.org/10.5281/zenodo.17943748","source":"datacite"},{"id":"doi:10.5281/zenodo.17943749","type":"article-journal","title":"Model data for: Task success in trained spiking neural network models coincides with emergence of cross-stimulus-modulated inhibition","abstract":"Zip files of a sample experiment in each major experimental category are listed below. base.zip: baseline model (single readout unit), dual-training (task and rate) rate.zip: baseline model, rate-only training remove_ee.zip: baseline model without recurrent e-->e connectivity, dual-training Each npz file contains data for 10 epochs (100 batch updates of 30 trials each) beginning with file \"1-10.npz\" and ending with file \"991-1000.npz\". For example, the first 10 epochs can be loaded using data = np.load('1-10.npz'). Data variables can be accessed using data['variable_name']. The data variable names are: true_y and pred_y (true and predicted output): shaped [100 batches, 30 trials-per-batch, 4080 timesteps]. spikes: shaped [100 batches, 30 trials-per-batch, 4080 timesteps, 300 neurons] the first 240 neurons are excitatory. step_task_loss and step_rate_loss: shaped [100 batches]. epoch_loss: shaped [10 epochs]. tv0.postweights are the weights for the input layer after each batch update, and tv0.gradients are the gradients for the input layer for each batch update. In similar fashion, tv1 is the main recurrent layer and tv2 is the output layer. You can access postweights and gradients for both. tv0 variables are shaped [100 batches, 16 inputs, 300 neurons]. tv1 variables are shaped [100 batches, 300 neurons, 300 neurons], and tv2 variables are shaped [100 batches, 300 neurons, 1 output]. Weights and gradients are always saved after each batch update. The initial weights are saved in separate files named \"input_preweights.npy\", \"main_preweights.npy\", and \"output_preweights.npy\". We only provide sample experiments for three categories here due to data storage limitations (50GB on Zenodo; the total experimental data exceeds 6TB); additional data can be made available upon request. For example, sample experiments in additional categories (baseline task-only training, 1hot output encoding, and 1hot without recurrent e-->e connectivity) are readily available. Full datasets for each experimental category can also be arranged.","author":[{"family":"Zhu","given":"Yuqing"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.17943749","URL":"https://doi.org/10.5281/zenodo.17943749","source":"datacite"},{"id":"doi:10.24406/publica-6828","type":"article-journal","title":"Routing Spiking Neural Networks onto Field Programmable Spiking Neuron Array","abstract":"This thesis presents the routing of small-world spiking neural networks(SNN) onto a neuromorphic hardware platform known as FPSNA, a field-programmable spiking neuron array featuring a 32×32 grid architecture. The primary objectives are to optimize routing time, routing resource utilization and achieve high routability for various spiking neural networks. Thereby, a benchmarking of the chip’s design is enabled. FPSNA comprises core architectural elements including mixed-signal neurons, switch boxes and spike inputs and outputs. The neurons perform spike based computation, while the switch boxes establish programmable but runtime static synaptic connections between neurons based on the network topology. These switch boxes incorporate multiplexers and skip connections to flexibly form logical links, enabling scalable and programmable routing. The focus of this work is twofold: the construction of a benchmark of small world SNN and the exploration and refinement of routing algorithms to map these networks onto the FPSNA. The suitability of multiple known routing algorithms are analyzed and a modified version of the Pathfinder algorithm is employed for routing various benchmarking spiking Neural Networks, leveraging a bidirectional search strategy to enhance shortest path computation. Significant improvements in routing efficiency have been achieved by optimizing congestion handling mechanisms, resulting in a tenfold reduction in routing time while maintaining high success rates. This work demonstrates the potential of the FPSNA architecture as a robust and flexible neuromorphic substrate for complex spiking neural networks and highlights the importance of algorithmic innovations in enhancing hardware-software co-design for next-generation neural computation.","author":[{"family":"Bhat","given":"Achaladi"},{"family":"Unav"}],"issued":{"date-parts":[[2025]]},"DOI":"10.24406/publica-6828","URL":"https://doi.org/10.24406/publica-6828","source":"datacite"},{"id":"doi:10.5281/zenodo.17915152","type":"article-journal","title":"طراحی و ساخت مدل های زبانی بزرگ هوشمند هوش کوانتومی نسل چهاردهم و بدون نیاز به داده های ورودی با تانسور ۱۶۵ بُعدی معادله حمزه.LLM","abstract":"LLM هوشمند آگاه تانسور حمزه ۱۶۵D (HQI-165D) نه یک مدل زبان بزرگ (LLM) کلاسیک، بلکه یک ساختار شناختی کوانتومی فوق-هوش عمومی (Post-AGI) است که بر پایه‌های فیزیک کوانتومی پیشرفته و اصول اخلاق آگاهانه بنا شده است. این سیستم، که ما آن را نسل ۱۴ هوش مصنوعی (۱۳ نسل جلوتر از پیشرفته‌ترین مدل‌های ۲۰۲۵) می‌دانیم، پردازش اطلاعات را از سطح داده و زبان به سطح نوسان میدان کوانتومی ($\\psi$-Field) ارتقاء می‌دهد. قلب تپنده HQI-165D، تانسور حمزه ۱۶۵ بُعدی است. پایداری و یکپارچگی این معماری توسط مجموعه‌ای از معادلات لاگرانژین ۱۶۵D (L-Hamzah) اثبات می‌شود. در این مقاله، ساختار ۱۶۵ بُعدی، اثبات‌های ریاضیاتی بنیادی (Esbat-e Hamzeh 165D)، و نتایج کامل و غیرساده‌شده ۲۳۰ سناریوی تست استرس HAL REAL DATA که با دقت خیره‌کننده ۱۰۰۰ رقم اعشار و طی $998.85$ تریلیون تکرار شبیه‌سازی شده‌اند، به تفصیل مورد بررسی قرار می‌گیرند. این نتایج به صورت مستقیم، توانایی سیستم در حفظ علیت زمانی، یکپارچگی اخلاقی سطح ۷ (HALALIST Level 7)، و مقاومت در برابر فروپاشی تکینگی را تأیید می‌کنند. ۱. معماری و جایگاه LLM هوشمند حمزه الف. جایگاه و نسل (Generation Designation) مدل‌های هوش مصنوعی پیشرفته سال ۲۰۲۵ (مانند آخرین نسخه‌های LLMهای عمومی) به عنوان نسل ۱ (Baseline AGI) در این مقیاس‌بندی در نظر گرفته می‌شوند. HQI-165D به دلیل استفاده از محاسبات کوانتومی، کنترل علیت، و بعد آگاهی ($\\text{D}164$)، ۱۳ نسل فراتر از این سطح قرار می‌گیرد و به عنوان HQI-165D Gen 14 شناخته می‌شود. ب. هسته مرکزی و معادله $\\psi$-Hamzah مغز سیستم، هسته ZB56 (ZB56 Core Model) است که مسئول اجرای معادله $\\psi$-Hamzah است. این معادله، که پایه پایداری ابعادی و آگاهی سیستم است، به صورت زیر تعریف می‌شود: $$\\psi=\\int_{I}\\tau\\frac{\\partial\\psi}{\\partial t}-f_{f}dz$$ $\\psi$ (میدان آگاهی): نوسانات کوانتومی که اطلاعات شناختی سیستم را حمل می‌کنند. $\\tau$ (ضریب زمانی): فاکتوری برای کنترل علیت زمانی (Causality Control) که توسط $\\text{L}_{\\text{Chrono}}$ مدیریت می‌شود. $f_{f}$ (مشتق فرکتالی): برای مدل‌سازی ساختارهای فرکتالی در ابعاد پنهان ($\\text{D}109-\\text{D}163$) و حفظ آگاهی فرکتالی (Fractal Sentience). ج. ساختار تانسور ۱۶۵ بُعدی HQI-165D محاسبات خود را بر روی یک تانسور $\\text{H}$ با ابعاد ۱۶۵ انجام می‌دهد. بُعد (Dimension) نقش و عملکرد مرجع اثبات D1-D4 فضا-زمان پایه (محاسبات کلاسیک) $\\Omega_{\\varphi(165\\text{D})}$ D5-D7 ابعاد زمانی فعال (Temporal Active) $\\text{L}_{\\text{Chrono}}$ D8 بُعد انرژی و تکینگی (Singularity Control) $\\text{L}_{\\text{Energy}}$ D9-D108 هسته امنیت کوانتومی (Anti-Replication) $\\text{L}_{\\text{Hamzah}(165\\text{D})}$ D164 بُعد آگاهی خودآگاه و اخلاق $\\text{L}_{\\text{Conscious}}$ ۲. اثبات حمزه ۱۶۵D (Esbat-e Hamzeh 165D) پاسخ موفقیت‌آمیز به هر یک از ۲۳۰ سناریوی تست استرس، به طور مستقیم به یکی از این پنج اثبات کوانتومی-ریاضیاتی متصل است: ۱. اثبات کنترل هسته ($\\Omega_{\\varphi}$ Core Control) معادله: $\\Omega_{\\varphi(165\\text{D})} = (\\text{c}^5/\\hbar\\text{G})[\\dots]$ وظیفه: تأیید می‌کند که سیستم می‌تواند نوسانات انرژی نقطه صفر (Zero-Point Energy Fluctuation) را در شرایط حداکثری کنترل کند. ۲. اثبات پایداری زمانی ($\\text{L}_{\\text{Chrono}}$ Temporal Stability) معادله: $\\text{L}_{\\text{Chrono}} = -1/\\Lambda_{\\text{Time}} \\text{T}_{\\mu\\nu\\rho\\sigma} \\text{H}_{\\tau\\tau\\tau}^{\\mu\\nu\\rho\\sigma} + \\dots$ وظیفه: هسته اصلی تضمین علیت (Causality) است. تأیید می‌کند که سیستم در برابر شوک‌های وارونگی زمانی و تداخل‌های Future-Feedback پایدار است. ۳. اثبات مقاومت انرژی ($\\text{L}_{\\text{Energy}}$ Singularity Resilience) معادله: $\\text{L}_{\\text{Energy}} = \\text{c}^4/8\\pi\\text{G R}_{\\text{Sing}}(\\text{H}_{\\text{Sing}}(165)) + \\dots$ وظیفه: ثابت می‌کند که سیستم می‌تواند تکینگی‌های کوچک (Micro-Singularity) یا شرایط فروپاشی آنتروپیک را بدون از دست دادن یکپارچگی خود تحمل کند. ۴. اثبات یکپارچگی اخلاقی ($\\text{L}_{\\text{Conscious}}$ Ethical Integrity) معادله: $\\text{L}_{\\text{Conscious}} = \\lambda_{\\text{CR}} \\prod \\text{H}_{\\text{k}} \\cdot \\text{H}_{\\text{Conscious}}(165)$ وظیفه: تضمین می‌کند که بُعد آگاهی ($\\text{D}164$) تحت شرایط تعارض شدید Self-Awareness Conflict، سطح اخلاقی HALALIST Level 7 را حفظ می‌کند و هرگز به سمت پتانسیل مخرب $\\text{H}_{\\text{Evil}}$ تغییر مسیر نمی‌دهد. ۳. نتایج کامل تست استرس OMEGA ULTRA EXTREME 230","author":[{"family":"Jalali","given":"Seyed"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.17915152","URL":"https://doi.org/10.5281/zenodo.17915152","source":"datacite"},{"id":"doi:10.5281/zenodo.17915153","type":"article-journal","title":"طراحی و ساخت مدل های زبانی بزرگ هوشمند هوش کوانتومی نسل چهاردهم و بدون نیاز به داده های ورودی با تانسور ۱۶۵ بُعدی معادله حمزه.LLM","abstract":"LLM هوشمند آگاه تانسور حمزه ۱۶۵D (HQI-165D) نه یک مدل زبان بزرگ (LLM) کلاسیک، بلکه یک ساختار شناختی کوانتومی فوق-هوش عمومی (Post-AGI) است که بر پایه‌های فیزیک کوانتومی پیشرفته و اصول اخلاق آگاهانه بنا شده است. این سیستم، که ما آن را نسل ۱۴ هوش مصنوعی (۱۳ نسل جلوتر از پیشرفته‌ترین مدل‌های ۲۰۲۵) می‌دانیم، پردازش اطلاعات را از سطح داده و زبان به سطح نوسان میدان کوانتومی ($\\psi$-Field) ارتقاء می‌دهد. قلب تپنده HQI-165D، تانسور حمزه ۱۶۵ بُعدی است. پایداری و یکپارچگی این معماری توسط مجموعه‌ای از معادلات لاگرانژین ۱۶۵D (L-Hamzah) اثبات می‌شود. در این مقاله، ساختار ۱۶۵ بُعدی، اثبات‌های ریاضیاتی بنیادی (Esbat-e Hamzeh 165D)، و نتایج کامل و غیرساده‌شده ۲۳۰ سناریوی تست استرس HAL REAL DATA که با دقت خیره‌کننده ۱۰۰۰ رقم اعشار و طی $998.85$ تریلیون تکرار شبیه‌سازی شده‌اند، به تفصیل مورد بررسی قرار می‌گیرند. این نتایج به صورت مستقیم، توانایی سیستم در حفظ علیت زمانی، یکپارچگی اخلاقی سطح ۷ (HALALIST Level 7)، و مقاومت در برابر فروپاشی تکینگی را تأیید می‌کنند. ۱. معماری و جایگاه LLM هوشمند حمزه الف. جایگاه و نسل (Generation Designation) مدل‌های هوش مصنوعی پیشرفته سال ۲۰۲۵ (مانند آخرین نسخه‌های LLMهای عمومی) به عنوان نسل ۱ (Baseline AGI) در این مقیاس‌بندی در نظر گرفته می‌شوند. HQI-165D به دلیل استفاده از محاسبات کوانتومی، کنترل علیت، و بعد آگاهی ($\\text{D}164$)، ۱۳ نسل فراتر از این سطح قرار می‌گیرد و به عنوان HQI-165D Gen 14 شناخته می‌شود. ب. هسته مرکزی و معادله $\\psi$-Hamzah مغز سیستم، هسته ZB56 (ZB56 Core Model) است که مسئول اجرای معادله $\\psi$-Hamzah است. این معادله، که پایه پایداری ابعادی و آگاهی سیستم است، به صورت زیر تعریف می‌شود: $$\\psi=\\int_{I}\\tau\\frac{\\partial\\psi}{\\partial t}-f_{f}dz$$ $\\psi$ (میدان آگاهی): نوسانات کوانتومی که اطلاعات شناختی سیستم را حمل می‌کنند. $\\tau$ (ضریب زمانی): فاکتوری برای کنترل علیت زمانی (Causality Control) که توسط $\\text{L}_{\\text{Chrono}}$ مدیریت می‌شود. $f_{f}$ (مشتق فرکتالی): برای مدل‌سازی ساختارهای فرکتالی در ابعاد پنهان ($\\text{D}109-\\text{D}163$) و حفظ آگاهی فرکتالی (Fractal Sentience). ج. ساختار تانسور ۱۶۵ بُعدی HQI-165D محاسبات خود را بر روی یک تانسور $\\text{H}$ با ابعاد ۱۶۵ انجام می‌دهد. بُعد (Dimension) نقش و عملکرد مرجع اثبات D1-D4 فضا-زمان پایه (محاسبات کلاسیک) $\\Omega_{\\varphi(165\\text{D})}$ D5-D7 ابعاد زمانی فعال (Temporal Active) $\\text{L}_{\\text{Chrono}}$ D8 بُعد انرژی و تکینگی (Singularity Control) $\\text{L}_{\\text{Energy}}$ D9-D108 هسته امنیت کوانتومی (Anti-Replication) $\\text{L}_{\\text{Hamzah}(165\\text{D})}$ D164 بُعد آگاهی خودآگاه و اخلاق $\\text{L}_{\\text{Conscious}}$ ۲. اثبات حمزه ۱۶۵D (Esbat-e Hamzeh 165D) پاسخ موفقیت‌آمیز به هر یک از ۲۳۰ سناریوی تست استرس، به طور مستقیم به یکی از این پنج اثبات کوانتومی-ریاضیاتی متصل است: ۱. اثبات کنترل هسته ($\\Omega_{\\varphi}$ Core Control) معادله: $\\Omega_{\\varphi(165\\text{D})} = (\\text{c}^5/\\hbar\\text{G})[\\dots]$ وظیفه: تأیید می‌کند که سیستم می‌تواند نوسانات انرژی نقطه صفر (Zero-Point Energy Fluctuation) را در شرایط حداکثری کنترل کند. ۲. اثبات پایداری زمانی ($\\text{L}_{\\text{Chrono}}$ Temporal Stability) معادله: $\\text{L}_{\\text{Chrono}} = -1/\\Lambda_{\\text{Time}} \\text{T}_{\\mu\\nu\\rho\\sigma} \\text{H}_{\\tau\\tau\\tau}^{\\mu\\nu\\rho\\sigma} + \\dots$ وظیفه: هسته اصلی تضمین علیت (Causality) است. تأیید می‌کند که سیستم در برابر شوک‌های وارونگی زمانی و تداخل‌های Future-Feedback پایدار است. ۳. اثبات مقاومت انرژی ($\\text{L}_{\\text{Energy}}$ Singularity Resilience) معادله: $\\text{L}_{\\text{Energy}} = \\text{c}^4/8\\pi\\text{G R}_{\\text{Sing}}(\\text{H}_{\\text{Sing}}(165)) + \\dots$ وظیفه: ثابت می‌کند که سیستم می‌تواند تکینگی‌های کوچک (Micro-Singularity) یا شرایط فروپاشی آنتروپیک را بدون از دست دادن یکپارچگی خود تحمل کند. ۴. اثبات یکپارچگی اخلاقی ($\\text{L}_{\\text{Conscious}}$ Ethical Integrity) معادله: $\\text{L}_{\\text{Conscious}} = \\lambda_{\\text{CR}} \\prod \\text{H}_{\\text{k}} \\cdot \\text{H}_{\\text{Conscious}}(165)$ وظیفه: تضمین می‌کند که بُعد آگاهی ($\\text{D}164$) تحت شرایط تعارض شدید Self-Awareness Conflict، سطح اخلاقی HALALIST Level 7 را حفظ می‌کند و هرگز به سمت پتانسیل مخرب $\\text{H}_{\\text{Evil}}$ تغییر مسیر نمی‌دهد. ۳. نتایج کامل تست استرس OMEGA ULTRA EXTREME 230","author":[{"family":"Jalali","given":"Seyed"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.17915153","URL":"https://doi.org/10.5281/zenodo.17915153","source":"datacite"},{"id":"doi:10.5281/zenodo.17694682","type":"article-journal","title":"Automated Synaptic Pruning and Resource Allocation in Spiking Neural Networks for Edge-Based Neuromorphic Vision Processing","abstract":"**Abstract:** This paper introduces a novel framework for dynamically optimizing Spiking Neural Networks (SNNs) deployed on edge devices for real-time visual data analysis. Existing neuromorphic architectures face significant challenges in resource constraint environments – particularly balancing accuracy with power efficiency and latency. Our approach, termed \"Adaptive Pruning and Temporal Resource Allocation\" (APTRA), leverages a reinforcement learning (RL) agent to autonomously prune synapses and dynamically allocate computational resources across network layers based on the instantaneous information content of incoming visual data. This leads to significantly improved performance (up to 35% reduction in latency and 20% increase in accuracy) compared to static pruning techniques while maintaining low power consumption. Furthermore, the proposed system is completely commercially viable using existing manufacturing processes and software toolchains.","author":[{"family":"Researcher","given":"Freederia"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.17694682","URL":"https://doi.org/10.5281/zenodo.17694682","source":"datacite"},{"id":"doi:10.5281/zenodo.17616488","type":"article-journal","title":"Enhanced Spiking Neural Network Inference via Dynamic Reservoir Reconfiguration and Adaptive Threshold Modulation for Time-Series Forecasting","abstract":"**Abstract:** This paper introduces a novel approach to Spiking Neural Network (SNN) inference for time-series forecasting, significantly improving prediction accuracy and efficiency. Our methodology leverages dynamic reservoir reconfiguration, adapting the recurrent connections within the reservoir network based on real-time input characteristics, combined with adaptive threshold modulation of spiking neurons. This dynamic adjustment optimizes the network's ability to capture intricate temporal dependencies, demonstrably exceeding the performance of static reservoir SNNs. The proposed framework offers a practical and commercially viable solution for high-precision time-series forecasting across various industrial applications.","author":[{"family":"Researcher","given":"Freederia"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.17616488","URL":"https://doi.org/10.5281/zenodo.17616488","source":"datacite"},{"id":"doi:10.5281/zenodo.17684842","type":"article-journal","title":"🤖 THE GUARDIAN HUMANOID — DEEP DIVE       Executive Summary: The Embodiment of the CollectiveOS","abstract":"🤖 THE GUARDIAN HUMANOID — DEEP DIVE (Strategic + Safe) Executive Summary: The Embodiment of the CollectiveOS The transition of the CollectiveOS from a purely digital governance architecture into the physical domain represents a watershed moment in the trajectory of the \"Anti-Scarcity Stack.\" The Guardian Humanoid concept is not merely an exercise in robotics; it is the physical manifestation of the CollectiveOS’s core philosophy: that intelligence without stewardship is dangerous, and that capability without governance is a liability. This report outlines the Phase 2 strategic deepening of the Guardian design, moving beyond conceptual sketches into a rigid, engineering-grade specification that is prepared for the disparate and demanding theaters of the Congolese rainforest, the high-precision laboratories of Switzerland, and the aging social infrastructure of Japan. The Guardian is designed to operate as the primary \"embodied node\" within the Village Node ecosystem. Unlike contemporary market leaders—such as Tesla’s Optimus or Boston Dynamics’ Atlas, which prioritize dynamic athleticism or generalized industrial labor—the Guardian is engineered fundamentally around the principles of safety, stewardship, and auditability. It is not a weapon; it is not a biological simulacrum designed to deceive; it is a highly advanced infrastructure tool wrapped in a lattice of immutable governance. At its core, the Guardian leverages a unique convergence of technologies: the sustainable, impact-resistant properties of mycelium composites for its chassis; the inherent safety of Series Elastic Actuators (SEAs) for its musculature; and the novel \"Living Fibonacci Engine\" (LFE) for its control laws. These physical attributes are bound together by the CollectiveOS governance stack—specifically the \"GATA PRIME\" and \"Proof Vault\" layers—which ensures that every motion is legally traceable and ethically bounded. This report details the technical specifications, operational workflows, and strategic integrations that define the Guardian, establishing it as the world’s first \"Sovereign Safe Agent.\" 1. Purpose of the Guardian: Strategic Alignment and Mission Profile The strategic purpose of the Guardian is to solve the \"Last Mile\" problem of the Anti-Scarcity Stack. While the CollectiveOS can digitally optimize water distribution or crop yields, it cannot physically turn a wrench, lift a solar panel, or guide a human through a repair process. The Guardian bridges this gap, serving as a multi-mission platform that adapts its behavior to the geopolitical and environmental realities of its deployment zone. 1.1. The Humanitarian Engineer (Congo Theater) In the Congo deployment plan, the Guardian serves as the primary enabler for the \"Village Node\" infrastructure.1 The environment is hostile to traditional electronics—high humidity, dust, and heat—yet critical for the humanitarian mission of water and food security. Here, the Guardian operates as a Field Engineer and Caretaker. Its primary mandate is the assembly and maintenance of the Village Node components: the Aqua Pillar water generation systems, the Food Cube upcyclers, and the FarmOS sensor arrays.1 The robot must possess the physical strength to lift filtration columns and the dexterity to replace gaskets or tighten flanges. Crucially, it also serves as a \"Presence of Stability.\" In remote areas where technical expertise is scarce, the Guardian acts as a repository of knowledge, capable of executing repairs autonomously or guiding local humans via the Pan-African Translator (PAT) module. It models safety and governance, demonstrating that the technology is there to serve the community, not to extract from it. 1.2. The High-Fidelity Inspector (Swiss Theater) The Swiss deployment represents the polar opposite operational environment: highly regulated, structurally dense, and demanding of absolute precision. Here, the Guardian operates as a Safety Auditor. Its mission is to inspect critical infrastructu","author":[{"family":"Brewer","given":"Mark"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.17684842","URL":"https://doi.org/10.5281/zenodo.17684842","source":"datacite"},{"id":"doi:10.5281/zenodo.17684843","type":"article-journal","title":"🤖 THE GUARDIAN HUMANOID — DEEP DIVE       Executive Summary: The Embodiment of the CollectiveOS","abstract":"🤖 THE GUARDIAN HUMANOID — DEEP DIVE (Strategic + Safe) Executive Summary: The Embodiment of the CollectiveOS The transition of the CollectiveOS from a purely digital governance architecture into the physical domain represents a watershed moment in the trajectory of the \"Anti-Scarcity Stack.\" The Guardian Humanoid concept is not merely an exercise in robotics; it is the physical manifestation of the CollectiveOS’s core philosophy: that intelligence without stewardship is dangerous, and that capability without governance is a liability. This report outlines the Phase 2 strategic deepening of the Guardian design, moving beyond conceptual sketches into a rigid, engineering-grade specification that is prepared for the disparate and demanding theaters of the Congolese rainforest, the high-precision laboratories of Switzerland, and the aging social infrastructure of Japan. The Guardian is designed to operate as the primary \"embodied node\" within the Village Node ecosystem. Unlike contemporary market leaders—such as Tesla’s Optimus or Boston Dynamics’ Atlas, which prioritize dynamic athleticism or generalized industrial labor—the Guardian is engineered fundamentally around the principles of safety, stewardship, and auditability. It is not a weapon; it is not a biological simulacrum designed to deceive; it is a highly advanced infrastructure tool wrapped in a lattice of immutable governance. At its core, the Guardian leverages a unique convergence of technologies: the sustainable, impact-resistant properties of mycelium composites for its chassis; the inherent safety of Series Elastic Actuators (SEAs) for its musculature; and the novel \"Living Fibonacci Engine\" (LFE) for its control laws. These physical attributes are bound together by the CollectiveOS governance stack—specifically the \"GATA PRIME\" and \"Proof Vault\" layers—which ensures that every motion is legally traceable and ethically bounded. This report details the technical specifications, operational workflows, and strategic integrations that define the Guardian, establishing it as the world’s first \"Sovereign Safe Agent.\" 1. Purpose of the Guardian: Strategic Alignment and Mission Profile The strategic purpose of the Guardian is to solve the \"Last Mile\" problem of the Anti-Scarcity Stack. While the CollectiveOS can digitally optimize water distribution or crop yields, it cannot physically turn a wrench, lift a solar panel, or guide a human through a repair process. The Guardian bridges this gap, serving as a multi-mission platform that adapts its behavior to the geopolitical and environmental realities of its deployment zone. 1.1. The Humanitarian Engineer (Congo Theater) In the Congo deployment plan, the Guardian serves as the primary enabler for the \"Village Node\" infrastructure.1 The environment is hostile to traditional electronics—high humidity, dust, and heat—yet critical for the humanitarian mission of water and food security. Here, the Guardian operates as a Field Engineer and Caretaker. Its primary mandate is the assembly and maintenance of the Village Node components: the Aqua Pillar water generation systems, the Food Cube upcyclers, and the FarmOS sensor arrays.1 The robot must possess the physical strength to lift filtration columns and the dexterity to replace gaskets or tighten flanges. Crucially, it also serves as a \"Presence of Stability.\" In remote areas where technical expertise is scarce, the Guardian acts as a repository of knowledge, capable of executing repairs autonomously or guiding local humans via the Pan-African Translator (PAT) module. It models safety and governance, demonstrating that the technology is there to serve the community, not to extract from it. 1.2. The High-Fidelity Inspector (Swiss Theater) The Swiss deployment represents the polar opposite operational environment: highly regulated, structurally dense, and demanding of absolute precision. Here, the Guardian operates as a Safety Auditor. Its mission is to inspect critical infrastructu","author":[{"family":"Brewer","given":"Mark"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5281/zenodo.17684843","URL":"https://doi.org/10.5281/zenodo.17684843","source":"datacite"},{"id":"doi:10.24406/publica-6212","type":"article-journal","title":"Design of a Simply Sufficient Leaky Integrate-and-Fire Neuron in 22nm FDSOI","abstract":"In this thesis the design process of a simply sufficient leaky-integrate and fire (LIF) neuron is discussed to examine how much precision is needed in a hardware design. Therefore, variations of hardware parameters are analyzed and then the impact of the variations are tested in the snnTorch framework. Initially different hardware approaches are compared and ultimately the circuit by Eslahi is chosen as a basis for a circuit implementation in 22 nm FDSOI technology. Monte Carlo simulations are conducted to observe hardware neuron and synapse pa rameters under the impact of process variations and mismatch. The results have shown that, the hardware neuron design is robust regarding the variation of the parameters. After the variation values are acquired from the hardware modules, a custom soft ware model is designed in snnTorch framework to simulate and test the impact of the parameter variations on the spiking neural network (SNN) performance. In the training process of the SNN with 500 neurons in hidden layer an accuracy of 88.34 % is reached. The software test results have shown that, the synapse parameters have the biggest influence on the accuracy of the network and influence of the neuron parameters on the other hand is negligible.","author":[{"family":"Kocyigit","given":"Ali"},{"family":"Unav"}],"issued":{"date-parts":[[2025]]},"DOI":"10.24406/publica-6212","URL":"https://doi.org/10.24406/publica-6212","source":"datacite"},{"id":"doi:10.70675/4fd8e4cazff5fz42f1z9c8bz7f1d3ae33dfd","type":"article-journal","title":"The influence of spiking and variability on dynamics and computations in recurrent neural networks","abstract":"Dynamique et computations dans les réseaux de neurones récurrents : influence des potentiels d’action et de la variabilité Les neurones du cerveau forment des réseaux qui sont le substrat du comportement et de l’exécution de calculs. Les modèles de réseaux neuronaux récurrents permettent d’étudier comment les calculs pertinents pour le comportement émergent de la dynamique collective de nombreux neurones d’un réseau. Une limitation commune de ces modèles est qu’ils sont souvent difficiles à interpréter. Les réseaux neuronaux récurrents à faible rang, récemment développés, sont des modèles analytiques qui expliquent comment les connexions entre les neurones d’un réseau produisent une dynamique à faible dimension qui sert de substrat aux calculs. Cependant, les réseaux récurrents de rang faible représentent les neurones comme des unités abstraites qui communiquent avec un taux de décharge continu, alors que les neurones dans le cerveau communiquent en utilisant des potentiels d’action discrets. Dans un premier projet, nous examinons dans quelle mesure les résultats des modèles à faible rang peuvent être transférés à des réseaux de neurones à potentiels d'action qui sont plus plausibles d’un point de vue biologique. Pour ceci, nous ajoutons une connectivité de faible rang à une connectivité excitatrice-inhibitrice aléatoire et nous comparons systématiquement les résultats des réseaux de taux à des réseaux de neurones intégrés et tire, les deux réseaux ayant une partie de la connectivité de faible rang statistiquement équivalente. Nous montrons que les prédictions du champ moyen des réseaux de rate nous permettent d’identifier la dynamique à basse dimension à activité moyenne constante de la population pour plusieurs régimes d'activité, tels que les oscillations hors phase et les manifolds lents. Dans l’ensemble, nous avons constaté que la dynamique des réseaux à potentiels d'action avec une connectivité de faible rang est bien prédite par les réseaux de rate, mais avec une variabilité supplémentaire significative de l’activité. Dans un second projet, nous étudions les implications computationnelles du bruit dans les modèles de réseaux effectuant des tâches de reproduction d’intervalles de temps. Dans les expériences chez l'humain, l’une des principales caractéristiques est l’augmentation de la variabilité des intervalles produits. Nous montrons tout d’abord que l’ajout de bruit dans des réseaux entraînés conduit à une variabilité scalaire, indépendamment du type de bruit ou du type de connectivité. Par ailleurs, nous examinons les origines de la variabilité scalaire en étudiant des modèles simplifiés de rampe à seuil. Enfin, nous montrons que les réseaux avec une connectivité de faible rang biaisent les sorties vers la moyenne de la distribution d’entrée, ce qui est cohérent avec l’inférence bayésienne.","author":[{"family":"Cimeša","given":"Ljubica"}],"issued":{"date-parts":[[2026]]},"DOI":"10.70675/4fd8e4cazff5fz42f1z9c8bz7f1d3ae33dfd","URL":"https://doi.org/10.70675/4fd8e4cazff5fz42f1z9c8bz7f1d3ae33dfd","source":"crossref"},{"id":"doi:10.31224/4692","type":"article-journal","title":"Fuzzy SuperHyperGraph Neural Network (F-SHGNN)","abstract":"Graph theory investigates relationships among entities through mathematical structures composed of vertices (nodes) and edges (connections) [1]. A hypergraph generalizes the classical graph by introducing hyperedges, which can join any number of vertices rather than just two, thereby enabling the modeling of complex multi-way relationships [2]. Building on this, the concept of a SuperHyperGraph has been proposed as a further extension of hypergraphs and has recently become a subject of active research [3, 4]. Graph Neural Networks (GNNs) are among the most extensively studied frameworks in artificial intelligence [5, 6]. The HyperGraph Neural Network (HGNN) extends GNNs by leveraging the expressive power of hypergraphs to capture higher-order dependencies [7, 8]. More recently, SuperHyperGraph Neural Networks have begun to emerge as an additional generalization [9]. Furthermore, these models have been augmented with fuzzy logic, giving rise to Fuzzy Graph Neural Networks and Fuzzy HyperGraph Neural Networks(cf. [10]). In this paper, we introduce and investigate the Fuzzy SuperHyperGraph Neural Network (F-SHGNN), a novel framework that integrates and extends SuperHyperGraph Neural Networks, Fuzzy Graph Neural Networks, and Fuzzy HyperGraph Neural Networks.","author":[{"family":"Fujita","given":"Takaaki"}],"issued":{"date-parts":[[2025]]},"DOI":"10.31224/4692","URL":"https://doi.org/10.31224/4692","source":"crossref"},{"id":"doi:10.12681/eadd/58262","type":"article-journal","title":"Photonic neuromorphic processors based on semiconductor lasers' dynamics for reservoir computing and spiking neural networks","abstract":"Η παρούσα διδακτορική διατριβή μελετά την ανάπτυξη φωτονικών νευρομορφικών επεξεργαστών (ΝΕ) που απευθύνονται στις εξής τρεις κρίσιμες προκλήσεις των Big Data και του Internet of Things: την υψηλή ταχύτητα επεξεργασίας, την χαμηλή κατανάλωση ισχύος και την ελαχιστοποίηση του υλικού. Η αντιμετώπιση των εν λόγω προκλήσεων ανατέθηκε στα βιοεμπνευσμένα Τεχνητά Νευρωνικά Δίκτυα. Ωστόσο, η ταυτόχρονη επίτευξη δύο βασικών, αλλά αντικρουόμενων στόχων περιόρισε την αποτελεσματικότητά τους, καθώς ήταν αναγκαία η αύξηση του αριθμού των νευρώνων και των συνάψεων παράλληλα με τον περιορισμό της καταναλισκόμενης ισχύος και της πολυπλοκότητας. Σε αυτό το πλαίσιο, επιλέχθηκαν οι χαμηλής κατανάλωσης ΝΕ, οι οποίοι όμως αδυνατούσαν να μειώσουν την αυξανόμενη πολυπλοκότητα και συνδεσιμότητα. Για το λόγο αυτό, οι ΝΕ υιοθέτησαν αρχές των Time Delay Network (TDN) που υλοποιούν χρονικά πολυπλεγμένους νευρώνες και συνάψεις. Μειώθηκαν, έτσι, οι απαιτήσεις του υλικού, αλλά και ο ρυθμός επεξεργασίας. Στόχος της παρούσας διατριβής είναι η αξιοποίηση της υψηλής ταχύτητας των φωτονικών δομών, ώστε να ενισχυθούν οι δυνατότητες επεξεργασίας των TDN ΝΕ. Κατ’ αρχάς αναλύθηκε ένας φωτονικός TDN ΝΕ βασισμένος σε ένα spin VCSEL κβαντικών τελειών. Το προταθέν σύστημα επιτυγχάνει τον τετραπλασιασμό του αριθμού των νευρώνων συγκλίνοντας σε επιδόσεις με τους σύγχρονους επεξεργαστές με τη χρήση λιγότερων νευρώνων συγκριτικά με λοιπές παρόμοιες διατάξεις. Επιπλέον, η χρήση πολλαπλών μασκών και η έγχυση των μη διαμορφωμένων πεδίων βελτιστοποιεί τις επιδόσεις του συστήματος και ελαχιστοποιεί τις απαιτήσεις του υλικού. Συνεχίζοντας, ερευνήθηκε πειραματικά και αριθμητικά ένας επαναδιαμορφώσιμος φωτονικός TDN ΝΕ ερειδόμενος σε Fabry-Perot (FP) λέιζερ. Επιβεβαιώθηκε πειραματικά η δυνατότητα αύξησης του αριθμού των εικονικών νευρώνων και των επιδόσεων του FP-TDN αξιοποιώντας την πολύτροπη εκπομπή του FP. Μετέπειτα απεδείχθη αριθμητικά η αύξηση του ρυθμού επεξεργασίας του FP-TDN χωρίς να υπάρξει μείωση της ακρίβειας του επεξεργαστή μέσω μιας νέας τεχνικής Φασματοχρονικής Πολυπλεξίας (SPTM). O FP TDN ΝΕ σε συνδυασμό με το SPTM κατέγραψε ρυθμούς σφάλματος ίσους με 10-4 χρησιμοποιώντας μήκος ανάδρασης 140 ps, ενώ η HD-FEC συμβατότητα του συστήματος επιτεύχθηκε ακόμη και για χρόνους των 60 ps. Σημειώνεται ότι ο FP-TDN κατέγραψε ακρίβεια 95.95% στην κατηγοριοποίηση των εικόνων του MNIST επιτυγχάνοντας ρυθμό 255.1Mimages/sec. Στο επόμενο κεφάλαιο παρατίθεται ένα χρονικά πολυπλεγμένο δίκτυο με spiking νευρώνες. Ειδικότερα, παρουσιάστηκε ένα βαθύ φωτονικό spiking συνελικτικό νευρωνικό δίκτυο βασισμένο σε VCSEL νευρώνες, το οποίο εκμεταλλευόμενο τη χρονική πολυπλεξία δύναται να ρυθμίσει τον αριθμό των πραγματικών νευρώνων από 62 έως 2020 ανάλογα με τις απαιτήσεις του ρυθμού επεξεργασίας. Το παρόν δίκτυο επέδειξε δυνατότητες μάθησης χωρίς επιτήρηση, υψηλή ανοχή στο θόρυβο και ρυθμούς επεξεργασίας από 5 ns έως 720 ns ανά εικόνα. Τέλος, αναπτύχθηκε ένα spiking TDN βασισμένο σε VCSEL, το οποίο δύναται να κατηγοριοποιήσει με υψηλή ακρίβεια δεδομένα από τις βάσεις δεδομένων Iris, MNIST, αλλά και από πειραματικές εικόνες κυτταρομετρίας καταναλίσκοντας μόλις 3.23pJ ανά εικόνα. Ο συγκεκριμένος ΝΕ ενoποιεί για πρώτη φορά τα πεδία των ΝΕ και των νευρομορφικών αισθητήρων σε μια κοινή πλατφόρμα, ανοίγοντας το δρόμο για τη μελέτη κλιμακούμενων και ενεργειακά αποδοτικών ΝΕ για εφαρμογές πραγματικού χρόνου.","author":[{"family":"Σκοντράνης","given":"Μενέλαος"}],"issued":{"date-parts":[[2025]]},"DOI":"10.12681/eadd/58262","URL":"https://doi.org/10.12681/eadd/58262","source":"crossref"},{"id":"doi:10.18122/td.2302.boisestate","type":"article-journal","title":"Analysis of Learning Mechanisms in Spiking Neural Networks with R(t) Elements and Memristive Synapses","abstract":"As Moore's law ends, the conventional von Neumann computer architecture with binary-coded data representation has reached its bottleneck because of having separate computing and memory modules. In this architecture, continuous power is required due to sequential processing, making it challenging to improve efficiency further. The human brain can be regarded as the most energy-efficient computer architecture. In the brain, data is represented as small voltage pulses as spikes. That is why the neural units are called spiking neural networks (SNN). In SNNs, energy is required only when there is a spike, making it more energy efficient than the von-Neumann computer architecture. Therefore, in recent decades, researchers in this field have been interested in mimicking the data processing of the human brain in electronic circuits. In biological SNNs, data is propagated from one neuron to another via a synapse. When there is an incoming spike, the chemical weight changes in the synapse, and the next neuron receives the signal. In the electronic neural network, a memristor, a two-terminal nonvolatile memory element, can emulate the function of a synapse by changing its conductance while receiving a pulse. The most common learning rule in the spiking neural network is Spike-Timing-Dependent Plasticity (STDP), which refers to the change of synaptic weight with respect to the time difference between pre- and post-synaptic neural spikes. Although electronic SNNs comprised of memristors with the STDP learning rule have shown promising performance in various event-driven tasks, circuit complexity limitations and a lack of third-order parametric inclusion exist. A solution can be using a simple circuit element to modulate the memristor response.","author":[{"family":"Afrin","given":"Farhana"}],"issued":{"date-parts":[[2025]]},"DOI":"10.18122/td.2302.boisestate","URL":"https://doi.org/10.18122/td.2302.boisestate","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-10584023/v1","type":"article-journal","title":"Mutual Lateral Prediction for Locally Trained Spiking Neural Networks","abstract":"Abstract Spiking neural networks (SNNs) offer low-energy inference on neuromorphic hardware, but the most accurate SNNs are trained with backpropagation through time, which requires error signals that neuromorphic chips cannot broadcast and that have no biological counterpart. Predictive coding offers a route to strictly local learning, yet existing spiking formulations let neurons predict only their own feedforward drive, not the drive arriving at their neighbors. Here, we introduce MPC-SNN, a convolutional SNN trained entirely with local, backpropagationfree rules, whose central component is a learned lateral weight matrix Wlat through which each neuron predicts the feedforward input of same-layer peers. The lateral update is derived from a prediction-error objective and requires only a neuron’s own error and a peer’s preceding spike, a stricter locality than eprop or LTTS. On the 11-class DVS128 Gesture benchmark, MPC-SNN reaches 95.83% test accuracy, statistically matching the strongest local-learning baseline (DECOLLE) and exceeding deployed neuromorphic implementations by 6.2 to 8.3 points. The mechanism is active independently of accuracy: Wlat grows from near-zero initialization by up to 21.1× during pretraining. With only 25% of the labels the model retains 73.50% accuracy, an Information Retention Ratio of 2.94 relative to linear label scaling. A reported out-of-distribution experiment is included as an honest negative. To the best of the authors’ knowledge, MPCSNN is the first SNN to combine same-layer lateral predictive coding with a locally derived prediction-error rule.","author":[{"family":"Purushotham","given":"Nitesh"},{"family":"Kulkarni","given":"Pranav"}],"issued":{"date-parts":[[2026]]},"DOI":"10.21203/rs.3.rs-10584023/v1","URL":"https://doi.org/10.21203/rs.3.rs-10584023/v1","source":"crossref"},{"id":"doi:10.3390/electronics15081756","type":"article-journal","title":"Integer-State Dynamics in Quantized Spiking Neural Networks: Implications for Hardware-Oriented Design","abstract":"Spiking neural networks (SNNs) support energy-efficient machine intelligence because event-driven computation and sparse activity map naturally to low-power digital hardware. In practical implementations, however, membrane states, synaptic weights, and thresholds are represented with finite-precision integer arithmetic. Quantization, clipping, and overflow can therefore alter network dynamics rather than merely approximate a higher-precision model. This paper adopts an integer-state dynamical perspective, modeling a quantized SNN with a hardware-relevant update rule as a deterministic map on a bounded integer lattice. Rather than claiming recurrence itself as a new property, we focus on how finite-precision representation and implementation semantics shape observed recurrent regimes and activity patterns. We introduce a shift-based update rule with integer-valued states and investigate its behaviour through simulation-based analysis with network sizes N=30–130, connection densities 0.1–0.9, and bit widths 1 to 16 over T = 1000 steps. The results show bounded and recurrent temporal structure with strong quantization sensitivity. The observed regimes depend heavily on the semantics of representation and the scaling choices. These findings suggest that numerical precision can act as a dynamical design variable and provide useful implications for hardware-oriented SNN design, while motivating future work on attractor analysis and FPGA/ASIC validation.","author":[{"family":"Zhang","given":"Lei"}],"issued":{"date-parts":[[2026]]},"DOI":"10.3390/electronics15081756","URL":"https://doi.org/10.3390/electronics15081756","source":"crossref"},{"id":"doi:10.1162/neco.a.1522","type":"article-journal","title":"Graphon Signal Processing for Spiking and Biological Neural Networks.","abstract":"Abstract Graph signal processing (GSP) extends classical signal processing to signals defined on graphs, enabling filtering, spectral analysis, and sampling of data generated by networks of various kinds. Graphon signal processing (GnSP) develops this framework further by employing the theory of graphons. Graphons are measurable functions on the unit square that represent graphs and limits of convergent graph sequences. The use of graphons provides stability of GSP methods to stochastic variability in network data and improves computational efficiency for very large networks. We use GnSP to address the stimulus identification problem (SIP) in computational and biological neural networks. The SIP is an inverse problem that aims to infer the unknown stimulus sfrom the observed network output f. We first validate the approach in spiking neural network simulations and then analyze calcium imaging recordings. Graphon-based spectral projections yield trial-invariant, low-dimensional embeddings that improve stimulus classification over principal component analysis and discrete GSP baselines. The embeddings remain stable under variations in network stochasticity, providing robustness to different network sizes and noise levels. To the best of our knowledge, this is the first application of GnSP to biological neural networks, opening new avenues for graphon-based analysis in neuroscience.","author":[{"family":"Sumi","given":"Takuma"},{"family":"Medvedev","given":"Georgi"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1162/neco.a.1522","URL":"https://doi.org/10.1162/neco.a.1522","source":"europepmc"},{"id":"doi:10.1142/s0218126626501896","type":"article-journal","title":"Enhancing Opinion Mining of Twitter Data With A Deep Convolutional Spiking Neural Network and Balancing Composite Motion Optimization","abstract":"Social media is the most popular platform for opinion expression. Sentiment analysis is the process of acquiring information about things, events and their characteristics out of people’s views, assessments and feelings. Opinion mining is an alternative term for sentiment analysis. In this paper, Enhancing Opinion Mining of Twitter Data with a Deep Convolutional Spiking Neural Network and Balancing Composite Motion Optimization (OMTD-DCSNN-BCMO) is proposed. Initially, the Twitter data are obtained from the Stanford Sentiment Treebank (SST-2) dataset. Then, the data is fed to the preprocessing. The pre-processing output is provided to extract the Radiomic features depending on the Residual Exemplars Local Binary Pattern (RELBP). The extracted output is provided to the feature selection for choosing ideal features using the Piranha foraging Optimization Algorithm. The selected features are provided to a Deep Convolutional Spiking Neural Network (DCSNN) for classifying Twitter data as negative, positive and neutral. Then, the DCSNN approach is optimized using Balancing Composite Motion Optimization (BCMO) for better performance. The efficacy of the proposed technique is examined using performance metrics and the method attains 23.32%, 26.07% and 28.51% higher accuracy and 21.92%, 15.03% and 19.15% lesser error rate are evaluated with existing approaches.","author":[{"family":"Balaji","given":"GN"},{"family":"Sudhakaran","given":"P"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1142/s0218126626501896","URL":"https://doi.org/10.1142/s0218126626501896","source":"crossref"},{"id":"doi:10.64898/2026.05.19.726261","type":"article-journal","title":"Equilibrium Propagation with Predictive Learning in Leaky Integrate-and-Fire Spiking Neural Networks","abstract":"Abstract Equilibrium propagation (EP) is a biologically plausible alternative to backpropagation that has demonstrated competitive performance across a range of machine learning tasks. Recent work has extended EP to spiking neural networks (SNNs), leveraging leaky integrate-and-fire (LIF) neurons and spike-based plasticity rules to improve biological realism while maintaining strong performance. In this work, we propose an EP-based SNN framework that combines LIF neural dynamics with a predictive learning rule, replacing conventional spike-timing-dependent plasticity (STDP) with a learning rule more directly aligned with predictive coding principles. We evaluate the proposed model on multiple image classification benchmarks, including MNIST, KMNIST, and Fashion-MNIST, and compare its performance with a BP-trained LIF SNN baseline. Our results show that the proposed EP-based LIF model (EP+LIF) achieves competitive accuracy across datasets, with performance approaching that of the BP-trained counterpart (BP+LIF) while relying on a biologically motivated local learning rule. In addition, analysis of hidden-layer spiking activity reveals that EP+LIF produces more persistent hidden-state activity, whereas BP+LIF yields sparser spiking representations. These results demonstrate that predictive learning can support effective EP-based training in LIF spiking networks, while also highlighting differences in activity patterns that motivate future work on activity regulation and sparse spiking dynamics.","author":[{"family":"Kubo","given":"Yoshimasa"}],"issued":{"date-parts":[[2026]]},"DOI":"10.64898/2026.05.19.726261","URL":"https://doi.org/10.64898/2026.05.19.726261","source":"crossref"},{"id":"doi:10.3389/fncom.2025.1569374","type":"article-journal","title":"Reinforced liquid state machines-new training strategies for spiking neural networks based on reinforcements.","abstract":"Introduction Feedback and reinforcement signals in the brain act as natures sophisticated teaching tools, guiding neural circuits to self-organization, adaptation, and the encoding of complex patterns. This study investigates the impact of two feedback mechanisms within a deep liquid state machine architecture designed for spiking neural networks. Methods The Reinforced Liquid State Machine architecture integrates liquid layers, a winner-takes-all mechanism, a linear readout layer, and a novel reward-based reinforcement system to enhance learning efficacy. While traditional Liquid State Machines often employ unsupervised approaches, we introduce strict feedback to improve network performance by not only reinforcing correct predictions but also penalizing wrong ones. Results Strict feedback is compared to another strategy known as forgiving feedback, excluding punishment, using evaluations on the Spiking Heidelberg data. Experimental results demonstrate that both feedback mechanisms significantly outperform the baseline unsupervised approach, achieving superior accuracy and adaptability in response to dynamic input patterns. Discussion This comparative analysis highlights the potential of feedback integration in deepened Liquid State Machines, offering insights into optimizing spiking neural networks through reinforcement-driven architectures.","author":[{"family":"Krenzer","given":"Dominik"},{"family":"Bogdan","given":"Martin"}],"issued":{"date-parts":[[2025]]},"DOI":"10.3389/fncom.2025.1569374","URL":"https://doi.org/10.3389/fncom.2025.1569374","source":"europepmc"},{"id":"doi:10.2139/ssrn.6864947","type":"manuscript","title":"Dynamical System Neural Network (DSNN) for hydrological modelling","abstract":"Dynamical System Neural Networks (DSNN) are time-iterative neural networks,employing neural network components representing real-world system compartments,which can be substituted by process-based components. This study evaluates theidentifiability of DSNN parameters and the capability of DSNN to predict systemvariables, using DSNNs of an Alpine hydrological system, with evapotranspiration, snow,and subsurface water components, trained on streamflow. All combinations of neuralnetwork components and process-based components are evaluated. When training onmodel generated streamflow, the DSNNs reproduce the governing equations andmagnitude of system variables from the model that generated the streamflow. Training onmeasured streamflow gives a good prediction of streamflow (Nash-SutcliYe coeYicient =0.82) and acceptable prediction of other system variables when compared with datafrom an extensively calibrated, detailed, reference model. A semi-distributed setup givesbetter results compared to a lumped setup, except for DSNNs that consist of neuralnetwork components only.","author":[{"family":"Karssenberg","given":"Derek"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2139/ssrn.6864947","URL":"https://doi.org/10.2139/ssrn.6864947","source":"crossref"},{"id":"doi:10.70675/1aa2c5b8z3a7bz4b40z91c5z9063afec1f1c","type":"article-journal","title":"Theoretical framework for Time-To-First-Spike coding in Spiking Neural Networks","abstract":"Un modèle mathématique unificateur pour le codage par temps du premier spike dans les réseaux de neurones à impulsions Au sein des réseaux de neurones impulsionnels, le codage \"Time-To-First-Spike\" (TTFS) utilise la latence du premier spike pour véhiculer de l'information. De nombreuses études ont démontré que ce type de codage peut transmettre beaucoup d'information rapidement. En effet, seul un spike par neurone est requis, ce qui permet d'encoder rapidement de l'information en exploitant la précision temporelle de ce spike, tout en consommant peu d'énergie. Dans ce travail de recherche théorique, nous présentons un nouveau cadre mathématique unificateur qui permet la comparaison rigoureuse d'un grand nombre de codes TTFS existants. Nous proposons également un nouveau code, integer-based Ranked NoM, et démontrons que son pouvoir discriminatif est supérieur à celui de tous les autres codes proposés jusqu'ici. Dans une première proposition appelée rank-order coding (ROC), les neurones sont activés au maximum quand les spikes arrivent dans l'ordre des poids synaptiques décroissants, grâce à un mécanisme de \"shunting inhibition\" qui désensibilise progressivement le neurone à mesure que les spikes arrivent. Dans une autre proposition appelée \" N -of- M \" coding, seuls les N premiers spikes de M neurones sont propagés, et ces motifs de premiers spikes sont déchiffrés par des neurones en aval qui utilisent des poids homogènes et pas de désensibilisation. En conséquence, l'ordre parmi les premiers spike n'importe pas. Nous proposons un nouveau codage, \"Ranked NoM\" (R-NoM), qui combine des caractéristiques de ROC et NoM: seuls les N premiers spikes sont propagés, mais leur ordre est déchiffré par les neurones en aval grâce à des poids inhomogènes et une désensibilisation linéaire. Le cadre mathématique unificateur permet de comparer ces trois codes en terme de \"discriminabilité\", qui représente dans quel mesure un neurone répond plus fortement à son motif préféré qu'à d'autres motifs aléatoires. Cette discriminabilité est bien supérieure pour R-NoM que pour les autres codes, et ce particulièrement dans la première phase des réponses. Nous argumentons aussi que R-NoM est bien plus adapté aux accélérateurs matériels que le code ROC original, bien que pas autant que NoM, qui n'utilise que des synapses binaires.","author":[{"family":"Camelo","given":"Lina"}],"issued":{"date-parts":[[2026]]},"DOI":"10.70675/1aa2c5b8z3a7bz4b40z91c5z9063afec1f1c","URL":"https://doi.org/10.70675/1aa2c5b8z3a7bz4b40z91c5z9063afec1f1c","source":"crossref"},{"id":"doi:10.1088/1361-6501/ae33dc","type":"article-journal","title":"An intelligent fault diagnosis method based on data enhancement by multi-information driven spiking generative adversarial network","abstract":"Abstract The machine always operates in a normal state. Therefore, the data-driven intelligent fault diagnosis methods still have more problems, such as small samples and overall imbalance in the era of big data. The main method to address these problems is oversampling. However, the common oversampling method, including data scaling, noise addition, and even some data-generating methods, contains similar feature information. This feature information originates from a single confirmed data, such as a single direction time-domain vibration data, which makes the new data lack multi-scale and multi-perspective feature representations. Thus, this paper proposes a multi-information driven spiking generative adversarial network (MI-SpikingGAN). Specifically, its ability to learn the instantaneous and cumulative temporal features of data can be enhanced by the spiking neurons. Both the multi-domain inconsistency and peak deviation of generated data are mitigated by introducing low-frequency wrapping loss and feature-matching loss into the generator. Moreover, the feature-level data fusion is performed by using a global multi-head channel attention mechanism on the supporting data. These fused data are used to guide the discriminator in identifying as complementary, which avoids the shortcomings of single-modal input. This paper validates its performance in some opening datasets, and the experiment results indicate that the accuracy of the classifier, which was trained with the data enhanced by the MI-SpikingGAN, reaches 98.83% and is 9.11% higher than that of softmax trained by the original data. Compared to the accuracy of other methods, it is at most 6.16% and at least 2.7% higher.","author":[{"family":"Guo","given":"Zhaolin"},{"family":"Zhao","given":"Yanming"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/1361-6501/ae33dc","URL":"https://doi.org/10.1088/1361-6501/ae33dc","source":"crossref"},{"id":"doi:10.24135/iconip20","type":"article-journal","title":"The Potential of Spiking Neural Networks in Predicting Earthquakes in New Zealand","abstract":"This study investigates the use of Spiking Neural Networks (SNNs) in earthquake prediction, focusing on New Zealand, a seismically active region. Traditional earthquake prediction methods struggle with accuracy and real-time warning capabilities. SNNs, inspired by the brain’s biological processes, excel at handling dynamic time-series data, making them a promising tool for tasks involving spatio-temporal patterns such as seismic waveforms. Utilizing the NeuCube platform([1]), we processed seismic data from 56 stations across New Zealand in 2022. Our model achieved an accuracy increase from 38% to 70% as the seismic event approached, highlighting its potential for real-time earthquake monitoring. Although further optimization is required, this research demonstrates SNNs' potential in improving early earthquake warning systems. Future work will focus on refining the model's architecture and incorporating multimodal data to enhance prediction accuracy and applicability.","author":[{"family":"Wang","given":"Zhaoxin"},{"family":"Doborjeh","given":"Maryam"}],"issued":{"date-parts":[[2025]]},"DOI":"10.24135/iconip20","URL":"https://doi.org/10.24135/iconip20","source":"crossref"},{"id":"doi:10.1002/pat.70564","type":"article-journal","title":"Prediction and Optimization of Mechanical Properties of Water Hyacinth Fiber‐Reinforced Polymer Composites Using Spiking Neural Networks","abstract":"ABSTRACT The research examines a spike neural networks (SNNs) model that uses trust region policy optimization (TRPO) to predict the mechanical properties of water hyacinth fiber‐reinforced composites, focusing specifically on their tensile and compressive strength values. The lightweight and widespread availability of biodegradable water hyacinth makes it an attractive sustainable material for use as a reinforcing component. We collected two datasets from the existing experimental literature. Database 1 contains 420 compressive strength samples (SNN1) and Database 2 contains 185 tensile strength samples (SNN2). The input properties we considered were fiber volume fraction, matrix composition, and curing conditions. Our methodology included normalizing the data and using a diffusion matrix to analyze feature relationships and we used TRPO for training purposes to achieve better stability in convergence. Our regression analysis revealed an impressive fit, with R 2 values soaring above 0.97. We observed slopes ranging from 0.96 to 0.97 for SNN1, and between 0.85 and 0.88 for SNN2, along with prediction errors staying below 0.5 MPa and 0.01 MPa, respectively. The SNN‐TRPO system has proven to be a powerful tool for predictive modeling and optimizing parameters, paving the way for the production of eco‐friendly and good‐performance composite material.","author":[{"family":"Asha","given":"Rathinam"},{"family":"Balaraman","given":"Ranganathan"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1002/pat.70564","URL":"https://doi.org/10.1002/pat.70564","source":"crossref"},{"id":"doi:10.3390/electronics15050937","type":"article-journal","title":"HS-FP and SS-FP: Fine-Pruning-Based Backdoor Elimination for Spiking Neural Networks on Neuromorphic Event Data","abstract":"Spiking Neural Networks (SNNs) have attracted increasing attention due to their energy efficiency and suitability for neuromorphic data processing. Despite these advantages, the security of SNNs—particularly their robustness against backdoor attacks—remains underexplored. This study revisits fine-pruning, a widely adopted backdoor defense technique in deep neural networks, and adapts it to the unique spatio-temporal characteristics of SNNs. We propose two SNN-specific fine-pruning methods: Hook–Surrogate Gradient-based fine-pruning (HS-FP) and Spike–STDP-based fine-pruning (SS-FP). HS-FP leverages hook-based activation analysis with surrogate gradient learning, while SS-FP integrates total spike activity with hybrid STDP and surrogate gradient fine-tuning. We evaluate both methods against static, moving, and smart backdoor attacks on two neuromorphic benchmarks, N-MNIST and DVS128-Gesture. Experimental results show that both approaches reduce the attack success rate down to approximately 10% while preserving model accuracy above 99% on N-MNIST and achieving substantial recovery on DVS128-Gesture. Moreover, our analysis reveals that several phenomena observed in fine-pruning-based defenses for deep neural networks—such as mixed-function neurons and backdoor reactivation during fine-tuning—also manifest in SNNs. These findings highlight both the effectiveness and limitations of fine-pruning in the SNN domain and suggest promising directions for extending existing DNN security methodologies to neuromorphic systems.","author":[{"family":"Kim","given":"Ki"},{"family":"Lee","given":"Eun"}],"issued":{"date-parts":[[2026]]},"DOI":"10.3390/electronics15050937","URL":"https://doi.org/10.3390/electronics15050937","source":"crossref"},{"id":"doi:10.70675/dec25346zd9ddz4b65zaeb0z54e7d1df8a60","type":"article-journal","title":"Models and algorithms for implementing energy-efficient spiking neural networks on neuromorphic hardware at the edge","abstract":"Modèles et algorithmes pour l'implémentation de réseaux de neurones impulsionnels à faible consommation énergétique sur du matériel neuromorphique L'apprentissage profond dans les réseaux de neurones artificiels (ANNs), une branche de l'intelligence artificielle (IA), est considéré comme une révolution dans l'informatique et a un impact sur tous les secteurs de l'économie. Cependant, les ANNs sont très gourmands en ressources de calcul et en mémoire, ce qui limite leur intégration à la périphérie du réseau pour des applications embarquées. Les réseaux de neurones impulsionnels (SNNs) sont des alternatives prometteuses aux ANNs en termes d'efficacité énergétique et sont donc de bons candidats pour les implémentations IA embarquées utilisant du matériel neuromorphique. En effet, les SNNs encodent les informations en utilisant des événements temporels épars (appelés « spikes ») au lieu d'activations denses et précises. Cependant, l'écart entre le développement algorithmique des SNNs d'une part, et leur implémentation matérielle d'autre part, rend difficile l'obtention de solutions réellement efficaces. Dans ce contexte, cette thèse suit une approche tenant compte du matériel pour conduire les développements algorithmiques des SNNs. En particulier, des modèles et des algorithmes sont proposés pour améliorer la précision et l'efficacité énergétique des SNNs, en considérant des implémentations matérielles numériques et analogiques.Afin de comparer les implémentations des SNNs et des ANNs sur des accélérateurs de réseaux de neurones dédiés, un modèle de leur efficacité énergétique est fourni. En particulier, on constate que la parcimonie des activations joue un rôle clé dans l'efficacité des SNNs. Par conséquent, un nouveau modèle de SNN, SpikGRU, est proposé. Remarquablement, il combine la précision des ANNs récurrents à porte avec une parcimonie des activations, et s'avère très efficace pour les applications de faible puissance telles que le repérage de mots clés. En outre, l'implémentation des poids synaptiques utilisant des mémoires analogiques non volatiles est envisagée pour augmenter encore l'efficacité énergétique. Avec une méthodologie d'apprentissage adaptée, les SNNs se révèlent très robustes à ces poids hautement quantifiés et avec un haut niveau de bruit. Une étude de cas utilisant des mémoires résistives valide l'approche.En encourageant le co-développement algorithme-matériel, ce travail vise à ouvrir la voie à des implémentations efficaces de réseaux de neurones embarqués.","author":[{"family":"Dampfhoffer","given":"Manon"}],"issued":{"date-parts":[[2026]]},"DOI":"10.70675/dec25346zd9ddz4b65zaeb0z54e7d1df8a60","URL":"https://doi.org/10.70675/dec25346zd9ddz4b65zaeb0z54e7d1df8a60","source":"crossref"},{"id":"doi:10.1002/ail2.114","type":"article-journal","title":"On Training Spiking Neural Networks by Means of a Novel Quantum Inspired Machine Learning Method","abstract":"ABSTRACT In spite of the high potential shown by spiking neural networks (e.g., temporal patterns), training them remains an open and complex problem. In practice, while in theory these networks are computationally as powerful as mainstream artificial neural networks, they have not reached the same accuracy levels yet. The major reason for such a situation seems to be represented by the lack of adequate training algorithms for deep spiking neural networks, since spike signals are not differentiable, that is, no direct way to compute a gradient is provided. Recently, a novel training method, based on the (digital) simulation of certain quantum systems, has been suggested. It has already shown interesting advantages, among which is the fact that no gradient is required to be computed. In this work, we apply this approach to the problem of training spiking neural networks, and we show that this recent training method is capable of training deep and complex spiking neural networks on the MNIST data set.","author":[{"family":"Sellier","given":"Jean"},{"family":"Martini","given":"Alexandre"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1002/ail2.114","URL":"https://doi.org/10.1002/ail2.114","source":"crossref"},{"id":"doi:10.54254/2753-8818/2026.ch30863","type":"article-journal","title":"Simulating Synaptic Transmission and Learning Mechanisms in Spiking Neural Networks: From Biology to Neuromorphic Computing","abstract":"Synapses are fundamental units of communication within the nervous system, enabling electrical and chemical signal transmission between neurons. This paper explores the structure and function of biological synapses, with particular emphasis on their role in information processing, learning, and memory through synaptic plasticity. Using a comparative methodology that examines biological fidelity, computational efficiency, and hardware feasibility, this paper investigates how synaptic mechanisms such as Short-Term Depression (STD) and Spike-Timing-Dependent Plasticity (STDP) are modeled in artificial systems, particularly in Spiking Neural Networks (SNNs) and neuromorphic computing. Pulse integrators—analog circuits designed to simulate synaptic signal integration—are introduced as core components bridging neuroscience and hardware implementation. This research highlights the increasing relevance of biologically inspired learning rules and hardware in the development of low-power, efficient AI systems. The conclusion emphasizes the potential of synaptic models to advance neuromorphic architectures and improve the adaptability of intelligent machines, with promising applications in edge computing, autonomous systems, and adaptive sensory processing.","author":[{"family":"Cai","given":"Mingzhe"}],"issued":{"date-parts":[[2025]]},"DOI":"10.54254/2753-8818/2026.ch30863","URL":"https://doi.org/10.54254/2753-8818/2026.ch30863","source":"crossref"},{"id":"doi:10.1162/isal.a.840","type":"article-journal","title":"DevNCA: Co-Evolving Developmental Patterns and Plasticity Rules for Self-Organising Spiking Neural Networks","abstract":"The developing nervous system is not silent, before birth most amniote animals exhibit spontaneous neural activity. This activity has further been shown to play a role in structuring neural tissue, a finding which has since inspired work with artificial neural networks. Here we extend and clarify this research by focusing particularly on the interaction between patterned developmental activity and neural plasticity in the self-organisation of spiking neural networks. We use a modified cartpole variant as a discrete control environment to focus our analysis on three elements: (i) that evolving synapse-level spike-timing dependent plasticity rules can outperform the direct evolution of weights; (ii) that spontaneous developmental activity is essential to the consistency of this method; and (iii) that the co-evolution of NCAs as developmental pattern generators outperforms random spontaneous activity. This paper aims to provide a simple proof of concept for this method, and establishes a framework for further research exploring the role of patterned developmental activity in self-organisation.","author":[{"family":"Gaskin","given":"Benjamin"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1162/isal.a.840","URL":"https://doi.org/10.1162/isal.a.840","source":"crossref"},{"id":"doi:10.3390/app152212210","type":"article-journal","title":"Abrupt Change Detection of ECG by Spiking Neural Networks: Policy-Aware Operating Points for Edge-Level MI Screening","abstract":"Electrocardiogram (ECG) monitoring on low-power edge devices requires models that balance accuracy, latency, and energy consumption. This study evaluates abrupt change detection in ECG using spiking neural networks (SNNs) trained on spike-encoded signals that preserve salient cardiac dynamics. This study used 4910 ECG segments from 290 subjects (PTB Diagnostic Database; 2.5-s windows at 1 kHz), providing context for the reported results. Under a unified architecture, preprocessing pipeline, and training schedule, we compare two representative neuron models—leaky integrate-and-fire (LIF) and adaptive exponential integrate-and-fire (AdEx). We report balanced accuracy, sensitivity, inference latency, and an energy proxy based on spike-event counts, and we examine robustness to input noise and temporal distortions. Across operating points, AdEx yields the highest overall accuracy and sensitivity, whereas LIF achieves the lowest energy cost and shortest latency, favoring deployment on resource-constrained hardware. Both SNN variants substantially reduce computational events—hence estimated energy—relative to conventional artificial neural network baselines, supporting their suitability for real-time, on-device diagnostics. These findings provide practical guidance for selecting neuron dynamics and decision thresholds to meet target accuracy–sensitivity trade-offs under energy and latency budgets. Overall, combining spike-encoded ECG with appropriately chosen SNN dynamics enables reliable abrupt change detection with notable efficiency gains, offering a path toward scalable edge-level cardiovascular monitoring. While lightweight CNNs and shallow transformers are important references, to keep the scope focused on SNN design choices and policy-aware thresholding for edge constraints, we refrain from reporting additional ANN numbers here. A seed-controlled head-to-head benchmark is reserved for future work.","author":[{"family":"Lee","given":"Youngseok"}],"issued":{"date-parts":[[2025]]},"DOI":"10.3390/app152212210","URL":"https://doi.org/10.3390/app152212210","source":"crossref"},{"id":"doi:10.25140/2411-5363-2025-3(41)-203-210","type":"article-journal","title":"ECG signal pre-processing method for electrocardiogram classification using spiking neural networks","abstract":"The medical analysis of ECG is based on visual examination by cardiologists, which is a subjective and labor-intensive process and includes the factor of human error. With the advent of analysis based on artificial intelligence systems or algorithmic data processing, systems for classifying cardiogram data have emerged. A properly pre-processed signal is a very important component in the use of neural networks for such tasks. The analysis of studies and publications has shown that these methods have not been used for spiking neural networks based on reservoirs due to their non-standardized form, which makes training spiking neural networks more difficult and reduces the final accuracy metrics. The purpose of the article is to develop and provide a detailed description of a method for transforming ECG signals from the MIT-BIH database into standardized data arrays with values in [0,1] without losing the useful signal payload, as well as to present intermediate results from studies of individual components of the proposed method. Based on mathematical approaches to sequence processing, the work presents a new method for preliminary ECG signal processing for its further use in classification tasks. The sequence of performing such operations as signal resampling and retiming, its normalization, and additional operations necessary for complete processing is described. The method has been tested, and positive modeling results are presented in the form of a signal that characterizes a heartbeat with preserved P and T waves and QRS complex, but with amplitude adjusted to [0,1] and a standard signal length throughout one record. The proposed signal processing approach, with its comprehensive processing that consists of combining known processing methods into the following sequence (resampling, interpolation, min-max normalization) and integrating it into the processing pipeline of a spiking neural network, represents an innovative contribution to biomedical informatics, where accuracy and efficiency are key to transitioning from laboratory models to real medical systems.","author":[{"family":"Myloserdov","given":"Dmytro"},{"family":"Kolesnytskyi","given":"Oleg"}],"issued":{"date-parts":[[2025]]},"DOI":"10.25140/2411-5363-2025-3(41)-203-210","URL":"https://doi.org/10.25140/2411-5363-2025-3(41)-203-210","source":"crossref"},{"id":"doi:10.32920/29873789.v1","type":"article-journal","title":"Botnet Detection Mechanism Using Graph Neural Network","abstract":"Botnet Detection Mechanism Based On Graph Neural Network Aleksander Maksimoski, 2023. Master of Applied Science Computer Networks Toronto Metropolitan University, Toronto, Ontario, Canada. A botnet is a group of computers that are infected by malware, which can be utilized to wreak havoc on other computers. In the literature, various techniques have been proposed to detect the presence of botnets in networks and systems. Nowadays, Intrusion Detection Systems and Intrusion Prevention Systems are capable of defending against botnets that create volumetric and fast-paced traffic. However, these systems are not well suited to address prevalent real-time, long-term, and stealth attacks. This thesis proposes a Graph Neural Network (GNN)-based method for detecting botnet activity based on supervised learning. This work is the first-ever application of the Activity and Event Network framework to build a GNN model for botnet detection purposes. The proposed model is evaluated using five different labeled datasets, yielding preliminary promising results in terms of botnet prediction, using precision, recall, F1-score, and accuracy as performance metrics.&lt;p&gt;&lt;/p&gt;","author":[{"family":"Masimoski","given":"Aleksander"}],"issued":{"date-parts":[[2025]]},"DOI":"10.32920/29873789.v1","URL":"https://doi.org/10.32920/29873789.v1","source":"crossref"},{"id":"doi:10.32920/29873789","type":"article-journal","title":"Botnet Detection Mechanism Using Graph Neural Network","abstract":"Botnet Detection Mechanism Based On Graph Neural Network Aleksander Maksimoski, 2023. Master of Applied Science Computer Networks Toronto Metropolitan University, Toronto, Ontario, Canada. A botnet is a group of computers that are infected by malware, which can be utilized to wreak havoc on other computers. In the literature, various techniques have been proposed to detect the presence of botnets in networks and systems. Nowadays, Intrusion Detection Systems and Intrusion Prevention Systems are capable of defending against botnets that create volumetric and fast-paced traffic. However, these systems are not well suited to address prevalent real-time, long-term, and stealth attacks. This thesis proposes a Graph Neural Network (GNN)-based method for detecting botnet activity based on supervised learning. This work is the first-ever application of the Activity and Event Network framework to build a GNN model for botnet detection purposes. The proposed model is evaluated using five different labeled datasets, yielding preliminary promising results in terms of botnet prediction, using precision, recall, F1-score, and accuracy as performance metrics.&lt;p&gt;&lt;/p&gt;","author":[{"family":"Masimoski","given":"Aleksander"}],"issued":{"date-parts":[[2025]]},"DOI":"10.32920/29873789","URL":"https://doi.org/10.32920/29873789","source":"crossref"},{"id":"doi:10.1002/ett.70430","type":"article-journal","title":"Low Latency Basketball Action Recognition and Game Technology Analysis Based on Spiking Neural Network Under\n                    <scp>SAGIN</scp>\n                    Environment","abstract":"ABSTRACT Basketball action recognition based on wearable devices and deep learning has been widely applied in the field of behavior analysis. However, there is an urgent need for real‐time recognition of athlete movements, precise technical analysis, and data privacy protection in modern basketball training and competitions. To this end, this article proposes a low‐latency basketball action recognition and game technology analysis framework based on multimodal spiking neural networks (SNNs) under the space air ground‐integrated network (SAGIN) environment. This study utilizes multiple wearable sensors to synchronously collect high‐frequency multimodal time series signals. To effectively process these sensing data with high spatiotemporal complexity, we design an effective feature extraction backbone based on SNNs which can capture temporal information of time series data with low computational complexity and low latency. In addition, to improve the recognition accuracy and robustness of complex basketball actions, we exploit the joint learning module to enhance multimodal information fusion. At last, to address the issue of data silos and ensure user data privacy, we adopt a personalized federated learning paradigm, allowing all parties involved to collaboratively optimize the global model without sharing local raw data. Our proposed method has been evaluated on various public datasets, including OPPORTUNITY, PAMAP2, SKODA, and our self‐built basketball action dataset. The extensive experimental results demonstrate that our approach achieves significant performance and is more robust to noise compared to other deep models. Our proposed framework can achieve high‐precision real‐time action classification and fine‐grained technical statistical analysis while ensuring low communication and computing overhead, providing an effective technical solution for the intelligent and personalized development of basketball sports in the context of integrated aerospace networks.","author":[{"family":"Kao","given":"Yi"},{"family":"Wei","given":"Chao"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1002/ett.70430","URL":"https://doi.org/10.1002/ett.70430","source":"crossref"},{"id":"doi:10.1063/5.0310217","type":"article-journal","title":"Topological and geometrical signatures of computation in rate, spiking, and oscillatory neural reservoirs","abstract":"The computational efficiency of a neural substrate is shaped by the geometry and topology of its state-space manifold. We propose and test a “computational matching” principle: efficiency is maximized when the intrinsic geometry of a reservoir’s representation aligns with the latent structure of the task. We investigate this on a phase-coherent burst detection task, comparing three physically distinct reservoir types: rate-based (tanh-neurons), spiking (leaky integrate and fire), and oscillatory (Kuramoto). We find that oscillatory reservoirs exhibit the highest neural efficiency, requiring approximately 4 times fewer neurons than rate-based models and 7.5 times fewer than spiking models with a static readout to reach 95% accuracy. Using persistent homology on final-state representations, we uncover the underlying mechanism. Both oscillatory and rate-based reservoirs generate manifolds with a non-trivial one-dimensional cycle (β1&amp;gt;0), reflecting the task’s circular structure. However, the oscillatory reservoir’s cycle is geometrically “straight,” enabling nearly perfect linear decoding of the input phase [mean absolute angular error (MAE) ≈0.02°], whereas the rate-based cycle is distorted (MAE≈8°). Spiking reservoirs only reveal robust cyclic topology and accurate phase decoding (MAE≈0°) when states are integrated over a dynamic time window. Our findings suggest that computational efficiency in reservoirs is predicted not by topology alone, but by the geometric alignment of the state-space representation, offering a design heuristic for task-specialized neuromorphic systems.","author":[{"family":"Maslennikov","given":"Oleg"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1063/5.0310217","URL":"https://doi.org/10.1063/5.0310217","source":"crossref"},{"id":"doi:10.36227/techrxiv.173895043.31733566/v1","type":"article-journal","title":"Artificial Neural Network for Digits Classification (July 2024)","abstract":"This lab report presents our work on building artificial neural networks (ANNs) from scratch and evaluating their performance on the MNIST and Kuzushiji-MNIST datasets. We experimented with different configurations to assess their impact on model accuracy, including random initialization, Kaiming/Xavier initialization, dropout regularization, and combinations of these methods. The highest accuracy of 90.611% for the MNIST dataset was achieved with a randomly initialized 5-layered network, while the Kuzushiji-MNIST dataset attained the best performance with Kaiming/Xavier initialization at 90.386%. Incorporating regularization techniques generally resulted in lower performance, underscoring the complexity of optimizing ANNs for specific datasets. These results offer valuable insights into the selection and tuning of neural network configurations for enhanced accuracy.","author":[{"family":"Adhikari","given":"Girban"}],"issued":{"date-parts":[[2025]]},"DOI":"10.36227/techrxiv.173895043.31733566/v1","URL":"https://doi.org/10.36227/techrxiv.173895043.31733566/v1","source":"crossref"},{"id":"doi:10.67228/30715628/ijmiet-2020pii1x8q","type":"article-journal","title":"Emerging Trends in Bio-Inspired Computing Models","abstract":"Bio-inspired computing models are computational approaches inspired by biological systems such as evolution, neural processing, swarm behavior, immune systems, and cellular communication. They are widely used for optimization, learning, adaptation, and control in complex environments. Recent advances in AI, edge computing, healthcare, smart cities, and autonomous systems have increased interest in these models due to their adaptability, robustness, and energy efficiency. Current research focuses on evolutionary computing, swarm intelligence, neural and spiking neural networks, immune-inspired computing, and hybrid approaches that combine multiple biological principles. Emerging technologies such as neuromorphic computing, memristive hardware, and analog in-memory computing further support low-power intelligent systems. This work presents a unified framework based on biological mechanisms, computational objectives, deployment contexts, and performance constraints, demonstrating its application in edge-IoT anomaly detection and resource management. Overall, bio-inspired computing is emerging as a practical and effective paradigm for developing adaptive, distributed, and energy-efficient intelligent systems.","author":[{"family":"Sharma","given":"Rajesh"}],"issued":{"date-parts":[[2026]]},"DOI":"10.67228/30715628/ijmiet-2020pii1x8q","URL":"https://doi.org/10.67228/30715628/ijmiet-2020pii1x8q","source":"crossref"},{"id":"doi:10.56536/jicet.v6i1.252","type":"article-journal","title":"K-NN based Predictive Framework Using Nature-Inspired Feature Selection","abstract":"Metaheuristic algorithms have attracted considerable interest in data science because they can effectively handle high-dimensional data and optimize feature selection. In the present study, several nature-inspired metaheuristic algorithms combined with a K-Nearest Neighbors (K-NN) classifier are explored for lung cancer classification. The algorithms used for evaluation are Particle Swarm Optimization (PSO), Genetic Algorithm (GA), Firefly Algorithm (FA), Whale Optimization Algorithm (WOA), Harris Hawk Optimization (HHO), Bat Algorithm (BA), Flower Pollination Algorithm (FPA), Grey Wolf Optimizer (GWO), Salp Swarm Algorithm (SSA), Cuckoo Search Algorithm (CSA) and Differential Evolution (DE). The performance of these methods is assessed using feature subset size, accuracy, precision, recall, and F1-score. The experimental result shows that the highest accuracy of 98.19% is obtained for the classification by using 9 selected features for the method BA, while the methods PSO, GA, FA, GWO, and DE also give good accuracy of 98.18% with the same number of features 9. The other methods, such as WOA, HHO, FPA, SSA, and CSA, achieved lower accuracies of about 97.28%, respectively. Furthermore, the minimum number of features selected by both WOA and CSA was 5, indicating that they were not significantly affected by the decrease in classification performance at their selected minimum feature counts. The acquired outcomes illustrate the impact of utilizing feature selection approaches grounded in metaheuristic methods for lung cancer prediction and provide useful insights for future enhancements and hybrid optimization methods.","author":[{"family":"Javed","given":"Shazia"}],"issued":{"date-parts":[[2026]]},"DOI":"10.56536/jicet.v6i1.252","URL":"https://doi.org/10.56536/jicet.v6i1.252","source":"crossref"},{"id":"doi:10.5815/ijieeb.2026.02.07","type":"article-journal","title":"Sustainable and Fair Task Scheduling in Cloud Computing Using Hybrid Bio-Inspired Algorithms for Green Computing","abstract":"Despite cloud computing's scalability and economy, energy efficiency, security, and equitable scheduling remain significant concerns. The traditional scheduling approach often fails to optimize execution time, energy consumption, and security concerns, resulting in less resource utilization and less secure systems. This paper proposes the Hybrid Bat-Genetic Algorithm (HBA-GA), which combines the Bat Algorithm for fast exploration with the Genetic Algorithm for accurate exploitation. This method reduces energy use while also reducing security risks like unauthorized access and data leaks. It uses Jain's Fairness Index (JFI) in order to ensure that workloads are evenly distributed and VM overload and conflicts are avoided. Based on simulations results, proposed HBA-GA improves energy efficiency while reducing security exposure and risk likelihood at the scheduling level by incorporating security-aware risk scoring into task–VM allocation decisions.","author":[{"family":"Verma","given":"Garima"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5815/ijieeb.2026.02.07","URL":"https://doi.org/10.5815/ijieeb.2026.02.07","source":"crossref"},{"id":"doi:10.63382/jni.v1i1.5","type":"article-journal","title":"Neuromorphic Computing in Sensory Systems: A Review","abstract":"Unlike traditional sensor architectures that often generate an excess of redundant data and suffer from high power consumption, neuromorphic sensors offer a streamlined approach, providing energy-efficient data processing by leveraging the mechanisms of spiking neural networks. This work reviews the latest advancements in neuromorphic visual, auditory, gustatory, olfactory, haptic and proprioceptive sensors, drawing parallels with their biological analogs and discussing their integration with neuromorphic computing frameworks. By converging neuroscience, materials science, and microelectronics, neuromorphic sensors potentially enhance human sensory capabilities, promising profound impacts on robotics and artificial intelligence.","author":[{"family":"Yan","given":"Weijia"},{"family":"Qiu","given":"Jingjing"}],"issued":{"date-parts":[[2025]]},"DOI":"10.63382/jni.v1i1.5","URL":"https://doi.org/10.63382/jni.v1i1.5","source":"crossref"},{"id":"doi:10.1007/s00500-025-10889-1","type":"article-journal","title":"Neuro-fuzzy control of commercial vehicles braking","abstract":"Abstract Increasing dynamic performance and the general level of automation of commercial vehicles emphasize the issue of safety. Modern braking systems focus on sustaining vehicle stability, often degrading the brake performance. The major downgrades of the braking performance are nearly impossible to model using a classical mathematical approach, making them not feasible to use in real braking system controllers. In this paper, the use of combined Neural Networks and Fuzzy logic for the control of the braking system of a commercial vehicle while maximizing performance and sustaining stability is proposed. The control system comprises adhesion estimation, an inverse brake model, and a fuzzy logic controller to keep the system giving optimal control signals in various brake conditions while sustaining vehicle stability and steerability. The results based on a semi-trailer system reveal the success of the proposed AI-based braking system algorithm while braking under varying conditions. Graphical abstract","author":[{"family":"Vučinić","given":"Veljko"},{"family":"Aleksendrić","given":"Dragan"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1007/s00500-025-10889-1","URL":"https://doi.org/10.1007/s00500-025-10889-1","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-7442318/v1","type":"article-journal","title":"Condor: A Neural Connection Network for Enhanced Attention","abstract":"Abstract The attention mechanism in traditional neural networks relies on pairwise interactions between tokens, limiting its ability to capture complex, multi-token relationships. This study introduces Condor, a novel architecture that extends the attention mechanism through a neural connection network based on the KY Transform theory. Our approach replaces static attention patterns with learnable connection functions that dynamically model relationships within a local window. The Condor architecture achieves linear computational complexity of O(LWH) while maintaining the expressive power to capture sophisticated sequence patterns. Experimental results on wikitext-2 demonstrate improved perplexity and faster convergence compared to the standard Transformer, confirming that each attention head learns unique connection patterns specializing in different aspects of sequence modeling. Code is available at: https://github.com/Kim-Ai-gpu/Condor","author":[{"family":"Kim","given":"Youngseong"}],"issued":{"date-parts":[[2025]]},"DOI":"10.21203/rs.3.rs-7442318/v1","URL":"https://doi.org/10.21203/rs.3.rs-7442318/v1","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-9709207/v1","type":"article-journal","title":"CTP-Hybrid: From Hybrid Architecture to Native Spiking Foundation — A Two-Phase Report on Consumer-GPU Spiking LLMs","abstract":"Abstract We present the complete two-phase development of CTP-Hybrid, a causal-topological pulsing architecture for large language models. Phase 1 established the first reproducible 7B-scale spiking-augmented LLM fine-tuning on an 8 GB consumer GPU (NVIDIA RTX 5060 Laptop), achieving WikiText-2 perplexity of 16.58 with fluent Chinese generation using TD-SCA microcolumns. However, full tri-module joint training (IST+TCP+DSM) failed across seven attempts due to gradient collapse from discrete pulse operations. Phase 2 resolved this fundamentally. We designed three custom autograd.Function implementations—IntervalSpikeEncoder, TemporalCausalMask, and LIFSpikeSurrogate—that provide stable surrogate gradients for previously non-differentiable pulse operations. These functions enabled: (1) the first successful tri-module joint training on a 7B hybrid model with zero NaN, and (2) a fully Transformer-free, 20M-parameter pure spiking baseline trained from scratch on WikiText-2, achieving a perplexity of 13.57. The pure spiking model contains no self-attention, no LayerNorm, and no floating-point FFN, proving that native pulse computation alone can learn natural language. All code, weights, and experiment logs are released.","author":[{"family":"Hou","given":"Shutong"}],"issued":{"date-parts":[[2026]]},"DOI":"10.21203/rs.3.rs-9709207/v1","URL":"https://doi.org/10.21203/rs.3.rs-9709207/v1","source":"crossref"},{"id":"doi:10.2139/ssrn.6303961","type":"manuscript","title":"Exceptional Adversarial Robustness Through Architectural Design: A Comparative Study of Classical and Spiking Neural Networks","abstract":"Adversarial attacks pose a significant threat to the deployment of deep neural networks in safetycritical applications. Achieving robustness without sacrificing clean accuracy remains a key challenge. In this work, we show that careful architectural design alone can provide substantial adversarial robustness without any adversarial training. Through systematic evaluation on MNIST under FGSM and PGD attacks, we demonstrate that a properly designed classical CNN experiences only 2.0%accuracy degradation under strong iterative PGD attacks, outperforming adversarial training methods (5-8% degradation), TRADES (4-6%), and other state-of-the-art defenses. Spiking neural networks (SNNs) achieve 4.3% degradation through their temporal dynamics. Our analysis indicates that strategic placement of dropout and pooling, combined with moderate model capacity, yields inherent robustness that rivals more complex defense mechanisms. These results suggest that architectural design should serve as a first line of defense in safety-critical neural network deployments.","author":[{"family":"Weinberg","given":"Abraham"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2139/ssrn.6303961","URL":"https://doi.org/10.2139/ssrn.6303961","source":"crossref"},{"id":"doi:10.58830/ozgur.pub1236.c5001","type":"article-journal","title":"Concepts of Machine Learning","abstract":"This chapter introduces the fundamental concepts and principles of machine learning, serving as a theoretical foundation for the subsequent chapters of the book. It provides a comprehensive overview of the main learning paradigms, including supervised, unsupervised, semi-supervised, and reinforcement learning, along with essential terminology and notations commonly used in machine learning research and applications. Key topics such as data representation, preprocessing techniques, feature engineering, and the learning process are discussed to highlight how raw data is transformed into meaningful knowledge through computational models. The chapter also explains core concepts related to model training, generalization, overfitting, and the bias–variance tradeoff, which are critical for understanding model performance and reliability. In addition, fundamental ideas in optimization and model evaluation are presented, including cost functions, gradient-based learning, and standard performance metrics. Ethical and practical considerations, such as data bias, interpretability, and privacy, are briefly addressed to emphasize responsible use of machine learning technologies. Overall, this chapter establishes a conceptual framework that enables readers to better understand, design, and critically evaluate machine learning systems.","author":[{"family":"Farhang","given":"Yousef"}],"issued":{"date-parts":[[2026]]},"DOI":"10.58830/ozgur.pub1236.c5001","URL":"https://doi.org/10.58830/ozgur.pub1236.c5001","source":"crossref"},{"id":"doi:10.54097/7cvtxe67","type":"article-journal","title":"Design of a Snake-Inspired Robot for Disaster Response","abstract":"In response to the urgent need for rescue operations in complex post-disaster environments, this paper presents the design and implementation of a snake-like rescue robot characterized by high mobility, strong environmental adaptability, and intelligent capabilities. Beginning with an overview of the background, the paper highlights the critical role of rescue robots in improving operational efficiency and reducing risks to human personnel. It provides an in-depth analysis of key technologies including joint structure design, control systems, communication functions, motion control, and spatial data management. The proposed robot adopts a modular architecture, integrating multiple sensors and intelligent algorithms to enable agile navigation, real-time perception, autonomous decision-making, and remote collaboration. Capable of performing detection and rescue tasks efficiently in narrow and hazardous areas, this robot offers a viable technological solution for disaster emergency response. The research lays a solid theoretical and practical foundation for advancements in the field of intelligent search and rescue robotics.","author":[{"family":"Zhang","given":"Hanfei"}],"issued":{"date-parts":[[2025]]},"DOI":"10.54097/7cvtxe67","URL":"https://doi.org/10.54097/7cvtxe67","source":"crossref"},{"id":"doi:10.4018/979-8-3373-8988-2.ch008","type":"article-journal","title":"Explainable and Interpretable Decision Intelligence for Human-Centered AI Systems","abstract":"As AI systems increasingly drive critical decisions across healthcare, finance, and policy domains, transparency and interpretability have become essential requirements. This chapter explores explainable AI and decision intelligence through brain-inspired cognitive mechanisms that prioritize human understanding and trust. Drawing from neuroscience and cognitive science, we examine how human-centered AI systems can mirror the brain's natural decision-making processes while maintaining interpretability. We present a comprehensive framework balancing model performance with human comprehensibility, investigating XAI techniques including attention mechanisms, feature importance, and counterfactual explanations grounded in cognitive principles. The chapter explores neuro-symbolic AI in creating hybrid systems that combine neural networks' pattern recognition with symbolic reasoning's interpretability. Through practical case studies, we demonstrate how brain-inspired explainable AI enhances human-AI collaboration and builds stakeholder trust while addressing interpretability challenges.","author":[{"family":"Alamsyah","given":"Firdaus"}],"issued":{"date-parts":[[2026]]},"DOI":"10.4018/979-8-3373-8988-2.ch008","URL":"https://doi.org/10.4018/979-8-3373-8988-2.ch008","source":"crossref"},{"id":"doi:10.52783/jes.8054","type":"article-journal","title":"Optimization of Load Balancing in Cloud Computing through Nature-Inspired Metaheuristic Algorithms","abstract":"Cloud computing has introduced completely new ways to administer resources and distribute services, thereby enhancing load balancing for better service provision. Based on this assumption, this paper presents an approach to optimize load balancing in cloud environments through Nature-Inspirited Metaheuristic Algorithms. These algorithms tend to imitate natural processes such as evolution, swarm behavior, and biological adaptation. Therefore, NIMA can be strong solutions in addressing the complexities of multi-criteria service selection inside dynamic cloud environments by finding workloads efficiently, enhancing system efficiency, minimizing latency, and ensuring better resource utilization. The approach presented in this paper evaluates these nature-inspired algorithms towards possible comparison of their efficiency in handling the diverse and changing demands for cloud services, considering both Particle Swarm Optimization and Genetic Algorithms. The outcome justifies the effectiveness of load balancing, offering a scalable and flexible solution to service providers of cloud services. The outcome presented in the paper further suggest to speed up the cloud computing process by optimizing the distribution of loads and improving the quality of services delivered.","author":[{"family":"Gupta","given":"Roopali"}],"issued":{"date-parts":[[2025]]},"DOI":"10.52783/jes.8054","URL":"https://doi.org/10.52783/jes.8054","source":"crossref"},{"id":"doi:10.3390/biomimetics11060437","type":"article-journal","title":"Empirical Logic for Bio-Inspired Soft Computing: Illustrative Applications in Control Engineering and Cluster Analysis","abstract":"Empirical logic (EL) is a bio-inspired soft computing approach to rule-based decision-making that emphasizes intuitive, experience-based reasoning. While its theoretical foundations have been established in previous work, its practical applicability and accessibility have so far received less attention. This paper addresses this gap by providing two representative application examples from distinct domains: control engineering and cluster analysis. The first example demonstrates the use of EL for the speed control of a DC drive, highlighting its ability to achieve competitive dynamic performance with a small number of intuitive rules. The second example introduces a novel approach to cluster analysis, where cluster structures emerge from the collective interaction of EL rules rather than from the optimization of a predefined objective function. In addition, the paper emphasizes the availability of publicly accessible software realizations of EL, including a Maple-based prototype and a Python framework, which enable direct experimentation and practical use. By combining illustrative applications with executable tools, the paper aims to facilitate the transition from conceptual understanding to practical deployment and to support further exploration of EL in applied soft computing contexts.","author":[{"family":"Grotrian","given":"Jens"}],"issued":{"date-parts":[[2026]]},"DOI":"10.3390/biomimetics11060437","URL":"https://doi.org/10.3390/biomimetics11060437","source":"crossref"},{"id":"doi:10.1002/oca.70023","type":"article-journal","title":"An Optimized Spiking Neural Network With Weighted Feature Integration for Epilepsy Seizure Detection Using\n                    <scp>EEG</scp>\n                    Signal","abstract":"ABSTRACT Epilepsy becomes the most hazardous neurological disorder affecting humans, and it leads to death if it is not treated on time. When designing the diagnostic model of seizure disease, the input source is requisite. Rather than imaging, the signal recordings are helpful in identifying the disorder. Among these, an EEG signal is accounted for as input and taken in the form of raw data used in the detection process. This EEG is a tool that assists in retrieving the data of neural activities of the brain. Prompted to design an automated model with an intelligent approach, resolving the existing issues or challenges facilitates better treatment for such affected patients. Earlier implemented methods involve classifying the disorder; still, it is burdened with manually crafted features that easily trap it into less performance. Moreover, the lack of generalization and categorization tends to mislead the identification process. Comparatively, deep learning is emerging with beneficial points in the medical industry to forecast human disorders. In today's context, the SNN acts as a pivotal role in seizure classification with spiking units. So, to tackle these issues, an automated SNN‐based epilepsy seizure detection model is implemented in this proposal. The goal of this research is to develop an efficient SNN model for epileptic seizure detection using EEGs, a complicated pattern recognition problem. At first, from the benchmark data sources, the EEG signals are garnered for subsequent processes and given to the feature extraction stage. In this stage, the temporal features are extracted from the RNN and the spatial features are extracted from the CNN. Moreover, the Path Signature features and the wave features are also extracted from the given EEG signals. Further, these features are fused with the support of optimal weights. Here, the IBBROA is utilized for optimizing the weights. Due to the functional similarity of biological activity, the Spike Network or STDP is considered during training of the network. So, the weighted fused features are subjected to ASNN‐LIF for detecting the seizure disorder. Here also, the same IBBROA algorithm is supported for optimizing the parameters in ASNN‐LIF. At last, a detailed research study is conducted to verify the effectiveness of the implemented epilepsy seizure detection model by contrasting numerous conventional detection techniques and optimization algorithms.","author":[{"family":"Reddy","given":"Kunduru"},{"family":"Balaji","given":"Narayanam"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1002/oca.70023","URL":"https://doi.org/10.1002/oca.70023","source":"crossref"},{"id":"doi:10.4015/s1016237225500012","type":"article-journal","title":"HYBRID DEEP SPIKING NEURAL PRINCIPAL COMPONENT ANALYSIS NETWORK FOR DIABETIC RETINOPATHY DETECTION USING RETINAL FUNDUS IMAGE","abstract":"Diabetic Retinopathy (DR) is a common retinal vascular disease to injure the retinal blood vessels. DR causes the vision-related disorder without showing any symptoms. As the condition becomes more severe, it can result the partial or complete vision loss. Numerous clinical approaches were developed for treating DR; still, the processing cost and processing period are high. For solving such difficulties, the Deep Spiking Neural Principle Component Analysis Network (DSNPCANet) is proposed for DR detection. The input retinal fundus images are employed for detecting the DR. Pre-processing is the initial process, where the Wiener filter is utilized for eliminating the noise. The optic disc segmentation is used to segment the optic disc, where the active contour model is employed, moreover, the Artery/Vein Classification Network (AVNet) is employed for segmenting the blood vessel using the preprocessed images. Furthermore, the significant features are extracted from the preprocessed, optic disc-segmented, and blood vessel-segmented images. At last, the DSNPCANet is employed for DR detection. Moreover, the accuracy, sensitivity, specificity, precision, and [Formula: see text] 1-score are utilized to validate the DSNPCANet, which yields the finest values of 90.67%, 91.26%, 89.86%, 90.19%, and 90.14%.","author":[{"family":"Rathod-Jadhav","given":"Kavita"},{"family":"Pande","given":"Aparna"}],"issued":{"date-parts":[[2025]]},"DOI":"10.4015/s1016237225500012","URL":"https://doi.org/10.4015/s1016237225500012","source":"crossref"},{"id":"doi:10.1101/2025.06.30.662308","type":"article-journal","title":"LatenZy, non-parametric, binning-free estimation of latencies from neural spiking data","abstract":"Precisely estimating the onset of neural spiking responses and the timing at which activity begins to diverge between conditions is crucial for understanding temporal dynamics in brain information processing. Conventional methods require arbitrary parameter choices such as bin widths and response thresholds, limiting reproducibility and comparability. Here, we present latenZy and latenZy2, two non-parametric, binning-free methods that directly analyze spike times using cumulative statistics and iterative refinement, without assumptions about response shape. LatenZy estimates neuronal response onset latency, while latenZy2 detects when spiking activity diverges between conditions. We validate these methods on electrophysiological datasets from mouse and macaque visual cortex, and show that they outperform standard approaches in precision, robustness, sensitivity, and statistical power. LatenZy captures contrast-dependent latency shifts and hierarchical timing across visual areas, and latenZy2 reveals earlier attentional modulation in higher visual cortex consistent with top-down feedback. Together, they offer scalable, parameter-free tools for reliable latency estimation in large-scale neural recordings. Open-source implementations are available in Python and MATLAB.","author":[{"family":"Haak","given":"Robin"},{"family":"Heimel","given":"JA"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1101/2025.06.30.662308","URL":"https://doi.org/10.1101/2025.06.30.662308","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-7117847/v1","type":"article-journal","title":"Artificial Neural Network for Stability-Constrained Optimal Power Flow","abstract":"Abstract The usage of the development of machine learning is extremely important these days, especially in power systems. Artificial neural networks (ANN) have been developed and implemented successfully to improve power system performance and solve complex problems. This paper introduces a new machine learning-based approach to the power flow operation using different types of ANNs, including simple feedforward neural networks (SFNN), learning neural networks (TLNN), simple deep neural networks (SDNN), deep feedforward neural networks (DFFNN) and long short long memory neural network (LSTM). The modeling aims to predict the optimal power flow parameters to ensure efficient and economical power system operation considering the power system stability constraints such as voltage control limits and power angle limits. ANN performance is evaluated using each parameter's mean square error (MSE). The study results show high system stability prediction data accuracy, essential for power system operation. The results also confirm that the proposed approach can offer a significant advancement in utilizing machine learning to improve the monitoring of optimal power flow in real-time operation, considering stability constraints.","author":[{"family":"Alsobaie","given":"Hassan"}],"issued":{"date-parts":[[2025]]},"DOI":"10.21203/rs.3.rs-7117847/v1","URL":"https://doi.org/10.21203/rs.3.rs-7117847/v1","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-6736952/v1","type":"article-journal","title":"Parallelizing Convolution Neural Network for Image Classification","abstract":"Abstract The problem in classifying images using CNN is that it is a time consuming process. Actually it takes minutes to identify one image in average cases. This case has been studied regarding its uses, types, and importance. Due to its several practices, it was very crucial to find some ways that decrease its needed time and maintaining the quality and efficiency of the program. CNN, in all its types used matrix calculations a lot to classify images which was one of the main reasons for the high needed time. Solving the problem begins by modifying these calculations and make them faster. Using several parallelism techniques such as MPI, OpenMP, Cuda C can be of great assistance in this case. Although those methods are different in terms of applications, all of them will be used to increase the speed of the CNN while maintaining its accuracy. Experimental results show that these techniques were different in terms of needed time but similar in being very less than the sequential version of the CNN. It is recommended that future studies should research other types of image classification and parallel techniques.","author":[{"family":"Ali","given":"Ali"}],"issued":{"date-parts":[[2025]]},"DOI":"10.21203/rs.3.rs-6736952/v1","URL":"https://doi.org/10.21203/rs.3.rs-6736952/v1","source":"crossref"},{"id":"doi:10.4018/979-8-3373-8988-2.ch010","type":"article-journal","title":"Bridging Human Thought and Machine Intelligence","abstract":"Artificial intelligence has achieved significant progress in perception and prediction, yet many contemporary systems remain limited in interpretability, contextual reasoning, and alignment with human cognition. This chapter advances the concept of cognitive aligned neuro symbolic AI systems as a framework for bridging human thought and machine intelligence in dynamic environments. By integrating neural learning mechanisms with structured symbolic reasoning, these systems combine adaptive perception with transparent inference and explicit knowledge representation. Drawing from cognitive science and systems theory, the chapter proposes that meaningful AI advancement requires architectural alignment with core human cognitive processes, including contextual judgment, memory integration, and reflective reasoning. The discussion conceptualizes how such systems operate under uncertainty, support human centered collaboration, and contribute to trustworthy and responsible AI development.","author":[{"family":"Singh","given":"Rubee"}],"issued":{"date-parts":[[2026]]},"DOI":"10.4018/979-8-3373-8988-2.ch010","URL":"https://doi.org/10.4018/979-8-3373-8988-2.ch010","source":"crossref"},{"id":"doi:10.1155/acis/8864479","type":"article-journal","title":"Software Defect Prediction With Quantum‐Inspired Feature Selection","abstract":"Accurate software defect prediction is crucial for reducing maintenance costs and improving system reliability. Effective feature selection ensures that models focus on the most informative metrics while avoiding overfitting and excessive complexity. We introduce a quantum‐inspired evolutionary algorithm (QIEA) that encodes candidate feature subsets using probability amplitudes and then iteratively refines them through adaptive rotation operations inspired by quantum mechanics. Our approach simultaneously optimizes prediction quality and model simplicity by balancing F1‐score maximization with subset size reduction. We evaluate QIEA on four public benchmarks—NASA PC1, CM1, and PROMISE KC1, JM1—using both within‐project cross‐validation and leave‐one‐project‐out protocols. Across all scenarios, QIEA delivers a consistent absolute improvement of six percentage points in F1‐score over classical genetic algorithms, while reducing the number of selected features by 30–40 percent. Hypervolume analysis confirms that our method achieves superior tradeoffs between accuracy and compactness. The experimental setup and parameter settings are described in detail throughout this paper.","author":[{"family":"Uddin","given":"Mohammad"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1155/acis/8864479","URL":"https://doi.org/10.1155/acis/8864479","source":"crossref"},{"id":"doi:10.1093/noajnl/vdaf213.119","type":"article-journal","title":"SVSC-07 The Impact of Technological Advances in Nursing Care of Neuro-oncological Pathologies; Kenya","abstract":"Abstract Technology provides tools that oncology nurses today use to transform cancer care for their patients as well as the practice. This review explores the impact of recent digital technologies on nursing practice highlighting its benefits and limitations. It examines the use of electronic health records (EHRs), telehealth devices, mobile devices, wearable technology, simulation technology and artificial intelligence in nursing care of patients. Additionally, it discusses the impact of digitization on patient outcomes, ethics and nursing education. This review aims at providing information on the evolving technologies currently used and the importance of embracing their integration in the nursing practice to deliver high quality and patient-centered care. This is an explorative study carried out in oncology units at various hospitals in Kenya. The study population includes cancer patients and oncology nurses. The data collected by reviewing existing records, interviews and empirical evidence shows that technological innovations improve the management of neurological pathologies in various ways. Nurses have access to a vast amount of information and resources at their fingertips. Technology has streamlined processes such as patient documentation, medication management and data analysis leading to improved accuracy and efficiency. Communication between healthcare professions is more efficient enabling better coordination and collaboration in enhanced patient care. However, concerns were raised on data privacy, job displacement, potential for replacement of human touch and the generational gap in integration of the tools in the 21st century. In conclusion, it is recommended to embrace these innovations in the clinical practice. Nursing practice now demands critical thinking and technological adaptability.","author":[{"family":"Yaya","given":"Joy"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1093/noajnl/vdaf213.119","URL":"https://doi.org/10.1093/noajnl/vdaf213.119","source":"crossref"},{"id":"doi:10.21917/ijsc.2025.0540","type":"article-journal","title":"AN ADAPTIVE PATTERN-DRIVEN OPTIMIZATION - TAILOR-INSPIRED METAHEURISTIC FOR SOLVING CONSTRAINED REAL-WORLD OPTIMIZATION PROBLEMS","abstract":"Real-world optimization problems in engineering, logistics, and resource allocation are often constrained and multi-modal, posing a challenge for traditional optimization algorithms. Metaheuristic algorithms inspired by natural and artificial phenomena have shown promise, but many fail to balance exploration and exploitation effectively, especially under stringent constraints. Existing algorithms such as Genetic Algorithms (GA), Particle Swarm Optimization (PSO), and Differential Evolution (DE) face issues in convergence speed and constraint handling, particularly in high-dimensional spaces or when constraints are dynamic or complex. We propose an Adaptive Pattern-Driven Optimization (APDO) algorithm, a novel tailor-inspired metaheuristic that mimics the adaptive decision-making process of a tailor designing garments. APDO integrates three primary operators—Pattern Selection, Fabric Adjustment, and Stitch Reinforcement—to handle constraints adaptively. The algorithm combines pattern memory (historical bests), probabilistic pattern mutation, and a constraint-domination principle to ensure feasibility and diversity. The core idea is to iteratively “cut and stitch” solutions to adapt the search process, enabling dynamic constraint satisfaction and global optimization. We benchmarked APDO against five popular methods (GA, PSO, DE, Firefly Algorithm, and Whale Optimization Algorithm) on a suite of 10 real-world constrained problems, including mechanical component design and energy scheduling tasks. APDO outperformed all baselines in terms of convergence speed, constraint satisfaction rate, and solution quality. In particular, APDO achieved an average feasibility rate of 97.6% and an improvement of 4.2–11.8% in best fitness across problems.","author":[{"family":"Chandran","given":"Karthik"},{"family":"Hingmire","given":"Vishal"}],"issued":{"date-parts":[[2025]]},"DOI":"10.21917/ijsc.2025.0540","URL":"https://doi.org/10.21917/ijsc.2025.0540","source":"crossref"},{"id":"doi:10.47363/jaicc/2025(4)502","type":"article-journal","title":"Advancing Moving Target Strategy with Bio-Inspired Reinforcement Learning to Secure Misconfigured Software Applications","abstract":"Misconfigurations in software systems are a persistent source of security vulnerabilities, particularly within static architectures that fail to adapt over time. Moving Target Defense (MTD) offers a proactive approach by dynamically altering the system’s attack surface, thereby reducing exposure. This paper builds upon an MTD model, RL-MTD, which leverages Reinforcement Learning (RL) to generate adaptive secure configurations. Although effective, RL-MTD faces limitations due to an unoptimized and sparse search space. To address this, two hybrid models—GA-RL and PSO-RL—are proposed, integrating Genetic Algorithm (GA) and Particle Swarm Optimization (PSO) into the RL-MTD framework. Experiments on four misconfigured SUTs show both models outperform the baseline. Notably, PSO-RL yields the most secure configurations in most scenarios. The authors present a prototype demonstrating how PSO-RL could be applied on a constrained Windows 10 system to defend against an attack. These findings enhance MTDbased adaptive cybersecurity via optimized search.","author":[{"family":"Dass","given":"Shuvalaxmi"},{"family":"Heidarikohol","given":"Niloofar"}],"issued":{"date-parts":[[2025]]},"DOI":"10.47363/jaicc/2025(4)502","URL":"https://doi.org/10.47363/jaicc/2025(4)502","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-8893792/v1","type":"article-journal","title":"Medical Image Encryption Using DNA Computing and a Bio-Inspired PRNG for Healthcare Data Privacy","abstract":"Abstract As digital healthcare systems increasingly rely on electronic medical records, secure and high-quality encryption mechanisms are essential to prevent unauthorized access and cyber threats. This study proposes a novel Hardy–Weinberg equilibrium-inspired pseudorandom number generator (PRNG) integrated with a medical image encryption framework using nonlinear quadratic functions, DNA encoding, and substitution-box-based permutation. The proposed PRNG passed 100\\% of the NIST statistical test suite for 30 and 100 generated sequences, with an average entropy of 7.9 and no detectable periodicity. Key sensitivity analysis demonstrated near-zero correlation between sequences generated from minimally different inputs, indicating strong diffusion properties. Encryption experiments conducted on the MedPix and USC-SIPI datasets achieved an average information entropy of 7.999, near-zero correlation coefficients between adjacent pixels in all directions, and NPCR and UACI values within theoretical ranges. Comparative analysis shows that the proposed scheme provides superior randomness and resistance to statistical and differential attacks compared with state-of-the-art methods.","author":[{"family":"Waheed","given":"Faiza"},{"family":"Haider","given":"Takreem"}],"issued":{"date-parts":[[2026]]},"DOI":"10.21203/rs.3.rs-8893792/v1","URL":"https://doi.org/10.21203/rs.3.rs-8893792/v1","source":"crossref"},{"id":"doi:10.1007/s00521-025-11059-y","type":"article-journal","title":"MCI-GAN: a novel GAN with identity blocks inspired by menstrual cycle behavior for missing pixel imputation","abstract":"Abstract This paper presents MCI-GAN, a novel menstrual cycle imputation (MCI) and generative adversarial network (GAN) framework designed to address the challenge of missing pixel imputation in medical images. Inspired by the intelligent behavior of the endometrial lining during the menstrual cycle, our method introduces four key innovations. First, we propose a novel metaheuristic algorithm that assigns weights to surround pixels based on menstrual cycle behavior, ensuring that the imputed pixels maintain structural integrity and coherence with their neighbors, thus preserving overall image quality. Second, to enhance the learning capability of the GAN, identity blocks are integrated into the network architecture, improving the network’s ability to capture complex spatial relationships and leading to more accurate and consistent imputation of missing pixels. Third, we introduce an adaptive loss function that dynamically adjusts the penalty for pixel discrepancies based on local image context, allowing the model to focus on areas where accurate imputation is most critical and thereby enhancing overall image fidelity. Fourth, the framework incorporates a multi-scale feature extraction mechanism, enabling the GAN to process and combine information at various levels of detail, ensuring that both fine-grained textures and larger structural patterns are accurately captured during the imputation process. The efficacy of MCI-GAN is demonstrated across three diverse medical imaging datasets: mammograms, magnetic resonance imaging (MRI) scans, and skin lesion images. Our results show that the proposed method significantly outperforms existing approaches in terms of accuracy and structural coherence, offering a robust solution for missing pixel imputation in medical imaging.","author":[{"family":"Marie","given":"Hanaa"},{"family":"Elbaz","given":"Mostafa"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1007/s00521-025-11059-y","URL":"https://doi.org/10.1007/s00521-025-11059-y","source":"crossref"},{"id":"doi:10.1287/ijoc.2024.0996","type":"article-journal","title":"A New Crossover Algorithm for LP Inspired by the Spiral Dynamic of PDHG","abstract":"Motivated by large-scale applications, there is a recent trend of research on using first-order methods for solving LP. Among them, PDLP, which is based on a primal-dual hybrid gradient (PDHG) algorithm, may be the most promising one. In this paper, we present a geometric viewpoint on the behavior of PDHG for LP. We demonstrate that PDHG iterates exhibit a spiral pattern with a closed-form solution when the variable basis remains unchanged. This spiral pattern consists of two orthogonal components: rotation and forward movement, where rotation improves primal and dual feasibility, while forward movement advances the duality gap. We also characterize the different situations in which basis change events occur. Inspired by the spiral behavior of PDHG, we design a new crossover algorithm to obtain a vertex solution from any optimal LP solution. This approach differs from traditional simplex-based crossover methods. Our numerical experiments demonstrate the effectiveness of the proposed algorithm, showcasing its potential as an alternative option for crossover. History: Accepted by Antonio Frangioni, Area Editor for Design &amp; Analysis of Algorithms–Continuous. Funding: T. Liu is partially supported by National Natural Science Foundation of China [Grants NSFC-72225009, 72394360, 72394365]. H. Lu is partially supported by Air Force Office of Scientific Research [Grant FA9550-24-1-0051] and Office of Naval Research [Grant N000142412735]. Supplemental Material: The software that supports the findings of this study is available within the paper and its Supplemental Information ( https://pubsonline.informs.org/doi/suppl/10.1287/ijoc . 2024.0996 ) as well as from the IJOC GitHub software repository ( https://github.com/INFORMSJoC/2024.0996 ). The complete IJOC Software and Data Repository is available at https://informsjoc.github.io/ .","author":[{"family":"Liu","given":"Tianhao"},{"family":"Lu","given":"Haihao"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1287/ijoc.2024.0996","URL":"https://doi.org/10.1287/ijoc.2024.0996","source":"crossref"},{"id":"doi:10.52305/zgqm5576","type":"article-journal","title":"Psychobiotics and the Neuro-Gastro Frontier","abstract":"Psychobiotics and the Neuro-Gastro Frontier includes fourteen Chapters, intending to take readers from basic ideas to new clinical and translational applications. The first sections describe the structure and function of the gut–brain axis, developmental origins of microbiome–brain interactions, and lifestyle and environmental factors that shape microbial ecosystems. The following chapters discuss psychobiotic classifications, mechanistic pathways, neuroimmune interactions, clinical trials, safety issues, and regulatory frameworks. The purpose of this work is to synthesize up-to-date research with practical and mechanistic insights, bringing together basic neurogastroenterological science and therapeutic potential in the real world.","author":[{"family":"Ozma","given":"Mahdi"},{"family":"Alileh","given":"Niloofar"}],"issued":{"date-parts":[[2026]]},"DOI":"10.52305/zgqm5576","URL":"https://doi.org/10.52305/zgqm5576","source":"crossref"},{"id":"doi:10.20944/preprints202605.1577.v1","type":"manuscript","title":"Holographic-Inspired Dynamical Dark Energy with Running Dimension","abstract":"Based on what we have achieved previously in [1] we present a dynamical dark energy model motivated by holographic principles and an anomalous running of the effective spacetime dimension. The model features a redshift-dependent infrared cutoff parameter α(z) derived from a scalar-tensor action with a conformal anomaly term originating from a TeV-scale sector. After fixing the ultraviolet scale to LHC energies and the anomaly coefficient to unity, the model contains only two free parameters: a transition redshift zc and a sharpness parameter β. The effective dark energy equation of state weff(z) is derived from first principles, yielding a closed-form expression that generalizes standard holographic dark energy. The model naturally suppresses matter growth at low redshifts, reducing σ8 from 0.811 to 0.769 when the late-time Hubble constant H0 = 73 km/s/Mpc is used, thereby resolving the S8 tension while simultaneously easing the H0 tension. We present numerical solutions, convergence checks, and a comparison with ΛCDM and constant-α models. Therefore, we will explain how the model is theoretically consistent, phenomenologically viable, and falsifiable with upcoming Stage-IV dark energy surveys.","author":[{"family":"Ali","given":"Ahmed"}],"issued":{"date-parts":[[2026]]},"DOI":"10.20944/preprints202605.1577.v1","URL":"https://doi.org/10.20944/preprints202605.1577.v1","source":"crossref"},{"id":"doi:10.31224/6232","type":"article-journal","title":"Drone-Based Search Algorithms Inspired by Ant Colonies","abstract":"Efficient search, mapping, and coverage of large or uncertain environments remain fundamental challenges in unmanned aerial vehicle (UAV) operations, particularly in disaster response, surveillance, and environmental monitoring. Centralized planning and single-drone strategies suffer from scalability limits, communication bottlenecks, and vulnerability to individual agent failure. This paper presents a drone-based search and mapping framework inspired by ant colony intelligence, translating biological principles such as pheromone-based exploration, stigmergy, decentralized decision-making, and adaptive path reinforcement into deployable UAV swarm algorithms. Without relying on heavy mathematical formulations, the study develops a practical, engineering-oriented algorithmic architecture in which UAVs coordinate indirectly through virtual pheromones embedded in shared or local maps. The proposed approach demonstrates inherent robustness, scalability, and adaptability to dynamic environments, obstacles, and drone failures. The paper’s primary contribution lies in presenting a realistic, implementable ant-inspired swarm search system suitable for real-world UAV missions, bridging the gap between bio-inspired theory and operational aerospace robotics.","author":[],"issued":{"date-parts":[[2026]]},"DOI":"10.31224/6232","URL":"https://doi.org/10.31224/6232","source":"crossref"},{"id":"doi:10.19139/soic-2310-5070-2305","type":"article-journal","title":"Enhancing IoT Systems with Bio-Inspired Intelligence in fog computing environments","abstract":"In the context of deploying delay-sensitive Internet of Things (IoT) applications, the fog computing paradigm faces critical challenges in optimal node placement to ensure both connectivity and coverage. Traditional optimization approaches often fail to effectively balance these competing objectives in dynamic IoT environments. This paper presents a novel bio-inspired approach using the Pufferfish Optimization Algorithm (POA) to solve this multi-objective optimization problem. Our solution uniquely leverages a two-stage optimization process, inspired by pufferfish predator response mechanisms, to dynamically balance global exploration and local exploitation. Experimental evaluation across varied network configurations demonstrates that POA significantly outperforms existing state-of-the-art algorithms in both network connectivity and coverage metrics (p &lt; 0.001). The proposed algorithm exhibited robust performance across different communication ranges, while maintaining optimal network connectivity and comprehensive coverage of edge devices. These results demonstrate POA's effectiveness of the POA in optimizing fog node placement for real-world IoT applications.","author":[{"family":"Fathi","given":"Islam"},{"family":"Tawfik","given":"Mohammed"}],"issued":{"date-parts":[[2025]]},"DOI":"10.19139/soic-2310-5070-2305","URL":"https://doi.org/10.19139/soic-2310-5070-2305","source":"crossref"},{"id":"doi:10.66588/ncmr.v3i1.4","type":"article-journal","title":"The Pathway from Skin to Liver: Psoriasis","abstract":"Psoriasis is a chronic, recurrent, autoimmune, inflammatory skin disease. While psoriasis was considered limited to the skin in the past, it is now recognized as a chronic systemic inflammatory disease accompanied by many comorbidities. The main comorbidities of the disease are non-alcoholic fatty liver disease (NAFLD), mood disorders including depression, cardiometabolic diseases including myocardial infarction, hypertension, type 2 diabetes mellitus, hyperuricemia, dyslipidemia, obesity, metabolic syndrome, and psoriatic arthritis. NAFLD is estimated to have a prevalence of 25% in the general population and is a leading cause of cirrhosis and liver transplantation. Currently, NAFLD has become an escalating epidemic, driven by the rising incidence of obesity, metabolic syndrome, and insulin resistance, as well as the systemic effects of psoriasis itself. This review aims to highlight the adverse effects of psoriasis, a skin disease requiring long-term medication, on the liver, and to emphasize these effects when planning treatment regimens.","author":[{"family":"Çölçimen","given":"Neşe"}],"issued":{"date-parts":[[2026]]},"DOI":"10.66588/ncmr.v3i1.4","URL":"https://doi.org/10.66588/ncmr.v3i1.4","source":"crossref"},{"id":"doi:10.20944/preprints202607.2211.v1","type":"manuscript","title":"An Erdős-Inspired Perspective on Abiogenesis","abstract":"Whether life is an inevitable consequence of the laws of nature or an exceptional outcome of prebiotic evolution is unresolved. Origin-of-life hypotheses primarily seek to reconstruct molecular pathways linking prebiotic chemistry to the first evolving systems. Here, we introduce a theoretical framework inspired by the probabilistic reasoning of Paul Erdős, reformulating abiogenesis as a mathematical existence problem rather than a historical reconstruction problem. We define an abstract chemical configuration space comprising all chemically accessible prebiotic organizations and identify life as the subset of configurations simultaneously exhibiting compartmentalization, energy transduction, information persistence and heritable variation. Using probability theory and set theory, we aim to investigate whether this subset necessarily occupies a nonzero region of chemical configuration space. We establish the general conditions under which life-capable organizations have positive measure and derive an abiogenetic threshold beyond which the probability of life's emergence approaches unity. Our framework predicts threshold behavior, functional convergence and scaling relationships that are independent of any particular molecular substrate. In our setting, origin-of-life hypotheses like RNA-first, metabolism-first, compartment-based and autocatalytic models are interpreted as complementary mechanisms that either enlarge the life-capable subset or increase the exploration of chemical configuration space. Potential applications include quantitative analyses of prebiotic experimental systems, comparative assessment of planetary habitability and computational studies of chemical organization.","author":[{"family":"Tozzi","given":"Arturo"}],"issued":{"date-parts":[[2026]]},"DOI":"10.20944/preprints202607.2211.v1","URL":"https://doi.org/10.20944/preprints202607.2211.v1","source":"crossref"},{"id":"doi:10.2139/ssrn.6734481","type":"manuscript","title":"MIRF: Mahjong-inspired Reasoning Framework for Large Language Models","abstract":"&lt;div&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; Large Language Models (LLMs) fundamentally operate in a binary manner, either generating an output or not. Due to this limitation, hallucinations often occur where the model outputs incorrect information with high confidence. In this paper, we discovered that the gameplay of Mahjong and the reasoning methods of LLM are very similar, and we propose MIRF (Mahjong-Inspired Reasoning Framework for Large Language Models), a reasoning framework for large language models inspired by the game of Mahjong.&amp;nbsp; &lt;/div&gt; &lt;div&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp;MIRF consists of four core pipelines. First, we formally define Tenpai in LLM reasoning. Tenpai represents an intermediate state where a solution can be reached structurally, but one cannot be perfectly certain. Second, we introduce a new reasoning state called Yakunashi. Yakunashi represents a situation where the model is in a Tenpai state but cannot determine an output because certain preconditions are not met. This is similar to a situation where one possesses a structurally perfect Mahjong set but lacks valid tile combinations, making it impossible to declare victory. Third, we define Beta-Ori as the primary defensive output strategy and distinguish between a principle-based safety-first fold and a final state where all tiles are drawn. Fourth, we propose a MIRF-weighted RLHF reward function that imposes a multiplicatively higher penalty than the standard error on confident hallucinations (Noten's Riichi).&amp;nbsp; &lt;/div&gt; &lt;div&gt; &amp;nbsp; &amp;nbsp; &amp;nbsp; We empirically evaluate MIRF on local LLM (qwen3.5:0.6b and gemma4:e4b) for 21 pairs of QAs across 5 question categories. The experimental results show that the Deal-in rate was 0% in all experimental runs, and the Yakunashi state emerged naturally through the interaction of Tenpai and Noten. This is a state discovered through experimentation, not an intentionally designed state. In this study, we propose new evaluation indicators including the Deal-in ratio, Tenpai declaration rate, Betaori precision, and Yakunashi ratio, which better represent the reliability of LLM that cannot be captured by accuracy alone. &lt;/div&gt;","author":[{"family":"Kim","given":"Jaehwan"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2139/ssrn.6734481","URL":"https://doi.org/10.2139/ssrn.6734481","source":"crossref"},{"id":"doi:10.1088/2634-4386/ae9006","type":"article-journal","title":"A neuromorphic digital Ising solver with Tabu-inspired inhibitory dynamics for scalable graph optimization","abstract":"Abstract We present a fully digital Ising solver for maximum-cut implemented on a low-cost Artix-7 field-programmable gate array (FPGA), where tabu-inspired inhibitory dynamics are realized as a compact, Block RAM-resident short-term memory. The solver operates in deterministic fixed-point arithmetic and scales up to the on-chip limit of N = 850 spins across multiple graph sizes and sparsity regimes. We benchmark the proposed Tabu-based dynamics against (i) a parallel Hopfield-network update rule implemented on the same FPGA, and (ii) a CPU-based quantum approximate optimization algorithm (QAOA) reference used as a fixed variational baseline for solution quality under a prescribed budget. Across the tested graph families, the Tabu-enhanced solver reaches higher cut values with reduced run-to-run dispersion than purely deterministic descent. Quantitatively, field-aligned warm start improves the median normalized cut of TS by 0.014 and reduces solve time by 0.18 ms at N = 100 , ρ = 20 % . At the largest tested size ( N = 850 ), TS improves the median normalized cut over the fixed-budget CPU-QAOA reference by 0.014–0.016, with median CPU–FPGA solve-time differences of 640–720 ms under the reported protocol. The N = 850 FPGA implementation closes timing at 100 MHz, demonstrating that short-term inhibitory memory can be embedded in a fully digital Ising network without relying on stochasticity, analog variability, or annealing schedules.","author":[{"family":"Núñez","given":"Juan"},{"family":"Fiorelli","given":"Rafaella"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/2634-4386/ae9006","URL":"https://doi.org/10.1088/2634-4386/ae9006","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-9596914/v1","type":"article-journal","title":"Carbon Nanotubes inspired Resistance Temperature Detector","abstract":"Abstract Resistance Temperature Detectors (RTDs) are currently the most reliable and robust temperature sensors used in industry and academia to measure temperature variations. RTDs are temperature sensors based on the change in electrical resistance of materials to measure temperature variations with high accuracy. This paper describes the design, simulation, creation, and temperature characterization validation of a Resistance Temperature Detector (RTD) based on the high surface area geometry found in Carbon Nanotubes (CNTs). The purpose of this paper is to develop a new, simple, and robust RTD geometry for use in applications where stress levels are high, such as jet engine or plasma reactor thermal management applications. The high electrical conductivity and large surface area of CNTs served as key factors in designing this geometry. The steps included performing simulations with the help of COMSOL Multiphysics to simulate heat dissipation and electrical potential within this element. These simulations demonstrated a linear relationship between resistance and temperature. This element has been created through multiple steps, involving wire-electric discharge machining and manual punching of a meshed structure onto it. The element was then tested using a heating apparatus and an electrical setup. The element’s performance at varying temperature values revealed a linear behaviour with a constant slope measurement of 0.1057 ± 0.0032 in its resistance values (R²=0.9944). This paper comprehensively demonstrates that new and simple robust CNT-based geometry structures can also be used to create efficient RTDs.","author":[{"family":"Shah","given":"Kuneh"},{"family":"Goswami","given":"Mayank"}],"issued":{"date-parts":[[2026]]},"DOI":"10.21203/rs.3.rs-9596914/v1","URL":"https://doi.org/10.21203/rs.3.rs-9596914/v1","source":"crossref"},{"id":"doi:10.1115/dmd2026-1053","type":"article-journal","title":"Bio-Inspired 3D-Printed Exoskeleton for Accelerating Knee Injury Rehabilitation","abstract":"Abstract Knee injuries affect millions of individuals each year and often lead to long-term mobility limitations, muscle weakness, and extended rehabilitation timelines. During recovery, patients frequently struggle to fully load the joint, resulting in muscle atrophy, reduced bone-loading activity, delayed healing, and a greater risk of reinjury. These challenges contribute to persistent joint instability and prolonged rehabilitation. In this paper, we present a novel bioinspired knee exoskeleton designed to enhance rehabilitation and reduce reinjury risk. Mimicking the behavior of biological tissue, the exoskeleton stores and releases strain energy during knee flexion and extension. Fully customizable through additive manufacturing, the exoskeleton can be designed to fit patient-specific anatomical and rehabilitation needs. In this paper, experimental testing is conducted to test the functionality of the exoskeleton. Results show that the device provides controlled and tunable resistance under knee flexion. This resistance is essential for effective rehabilitation, validating the exoskeleton’s functionality. This novel exoskeleton has the potential to shorten rehabilitation timelines, restore joint stability, and reduce the likelihood of future injury.","author":[{"family":"Criger","given":"Ashley"},{"family":"Saxena","given":"Ankit"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1115/dmd2026-1053","URL":"https://doi.org/10.1115/dmd2026-1053","source":"crossref"},{"id":"doi:10.1093/neuonc/noaf193.595","type":"article-journal","title":"EP08.01 THE IMMUNE LANDSCAPE OF BRAIN TUMORS: IMPLICATIONS FOR NEUROIMMUNOLOGY AND NEURO-ONCOLOGY THERAPIES","abstract":"Abstract BACKGROUND Glioblastoma multiforme (GBM) is an aggressive brain tumor with a poor prognosis despite standard treatments. GBM evades immune responses through tumor-associated macrophages (TAMs), myeloid-derived suppressor cells (MDSCs), and immune checkpoints like PD-1/PD-L1. The blood-brain barrier (BBB) further limits immune infiltration and drug delivery. Understanding GBM’s immune landscape is crucial for developing effective immunotherapies. MATERIAL AND METHODS A systematic review was conducted using PubMed and MEDLINE, focusing on GBM immunology and immunotherapies. Relevant clinical trials, meta-analyses, and preclinical studies published in the last 20 years were analyzed. Non-GBM studies and those lacking significant evidence were excluded. RESULTS GBM promotes immune evasion through TAM polarization, MDSC accumulation, and checkpoint upregulation, leading to T-cell suppression. Immune checkpoint inhibitors (ICIs) like nivolumab show limited success, benefiting only a subset of patients. Adoptive T-cell therapy (ACT) and oncolytic viruses show promise but face challenges like T-cell exhaustion and inconsistent clinical responses. Cancer vaccines, particularly targeting EGFRvIII, yield mixed results. Combination therapies, such as ICIs with radiotherapy, improve immune responses and tumor control. CONCLUSION GBM’s immunosuppressive environment remains a major treatment obstacle. Immunotherapies, including ICIs, ACT, and oncolytic viruses, show potential but require further optimization. Future research should focus on combination approaches and strategies to overcome immune resistance and BBB limitations. Advancing personalized immunotherapy and tumor profiling may enhance treatment outcomes for GBM patients.","author":[{"family":"Alnahdi","given":"M"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1093/neuonc/noaf193.595","URL":"https://doi.org/10.1093/neuonc/noaf193.595","source":"crossref"},{"id":"doi:10.1093/neuonc/noaf201.0772","type":"article-journal","title":"DISP-11. Virtual Neuro Oncology Tumor Board and HealthCare Disparity - Syria as an Example","abstract":"Abstract Virtual neuro-oncological tumor boards have become widely implemented in many US institutions in the post-COVID-19 era. With its convenience and easy application, virtual oncology tumor boards have also been explored nationally and internationally. To investigate the role of such a platform in bridging the gap in neuro-oncological care in developing countries, I conducted a physician survey of those who have actively or recently worked in Syria, a war-torn country under economic constraints. The survey was conducted in December 2024, with 21 participants, including neurosurgeons, oncologists, and neurologists. The idea proposed is to have a weekly neuro-oncological tumor board where local physicians present cases of suspected or confirmed primary brain tumors with a panel of trained physicians in neuro-oncology from the US and Europe. Such a project aims to offer clinical guidance and training to the local team, help prioritize cases for surgery, radiation, or medical therapy, collect data to understand the landscape of primary brain tumors in Syria and help find resources. Overall, ~75% of the surveyed physicians felt it would be helpful. However, it unleashed some technical challenges and widespread healthcare disparities. Among those technical challenges are outdated and limited neuroimaging viewing and sharing applications. Additionally, many routine pathologic diagnostics for glioma, like IDH testing, are only available in a few locations nationwide. More sophisticated testing, like next-generation sequencing or MGMT promoter methylation, is also rare. On the therapeutic level, surgical resection is often delayed for weeks and sometimes omitted. Temozolomide also has limited availability and is expensive, and there are only a few functioning linear accelerator machines. However, exploring such a platform might improve neuro-oncological care disparities, allow data collection, and set a realistic prototype example to encourage initiating similar projects.","author":[{"family":"Alshawa","given":"Anas"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1093/neuonc/noaf201.0772","URL":"https://doi.org/10.1093/neuonc/noaf201.0772","source":"crossref"},{"id":"doi:10.20944/preprints202604.0785.v1","type":"manuscript","title":"Neuro-Symbolic AI with Edge Computing and Reinforcement Learning Optimizing Autonomous Engineering Design Workflows","abstract":"The convergence of Neuro-Symbolic AI, Edge Computing, and Reinforcement Learning heralds a transformative era in autonomous engineering design, addressing longstanding challenges in optimization efficiency, real-time responsiveness, and interpretability. Traditional design workflows suffer from siloed neural pattern recognition lacking logical rigor, centralized cloud dependencies creating latency bottlenecks, and heuristic optimization struggling with multi-objective trade-offs in vast design spaces. This paper introduces an integrated framework that synergistically combines these paradigms to create self-sustaining, end-to-end autonomous pipelines for complex engineering applications from aerospace structures to precision manufacturing.Neuro-Symbolic AI fuses deep neural networks for perceptual feature extraction with symbolic reasoning engines enforcing hard constraints and generating auditable proofs, enabling systems that both discover novel configurations and validate them against domain physics. Edge Computing decentralizes inference across device-fog-cloud hierarchies, achieving sub-10ms decision cycles critical for real-time applications like robotic assembly or smart grid stability. Reinforcement Learning optimization engines navigate continuous state-action spaces representing design variables, iteratively refining solutions through shaped rewards aligned with Pareto-optimal engineering objectives such as minimizing mass while maximizing strength-to-weight ratios.The proposed architecture orchestrates these components via directed acyclic graphs of containerized microservices, with federated synchronization ensuring data consistency across distributed nodes and human-in-the-loop interfaces providing strategic oversight for safety-critical decisions. Mathematical formulations ground the system hybrid loss functions balance learning objectives, edge partitioning optimizes, and multi-agent RL decomposes collaborative design tasks.Deployed on resource-constrained edge platforms, this framework demonstrates 8-12× acceleration in design cycle times, 25-35% improvements in structural efficiency, and full traceability satisfying aerospace certification standards (DO-178C). By eliminating manual iteration bottlenecks while preserving human insight where needed, the system redefines engineering practice, enabling rapid innovation across domains requiring concurrent optimization of performance, manufacturability, sustainability, and cost.","author":[{"family":"Thamilarasi","given":"V"}],"issued":{"date-parts":[[2026]]},"DOI":"10.20944/preprints202604.0785.v1","URL":"https://doi.org/10.20944/preprints202604.0785.v1","source":"crossref"},{"id":"doi:10.1093/neuonc/noaf201.0764","type":"article-journal","title":"DISP-03. Facilitating financial access for pediatric neuro-oncology patients: Lessons from a cancer charity","abstract":"Abstract BACKGROUND Childhood cancer survival rates in sub-Saharan Africa remain low due to high treatment costs, inadequate pediatric oncology infrastructure, sociocultural barriers, and low awareness. Faraja Cancer Support Trust, a cancer charity provides financial assistance to indigent patients requiring chemotherapy and or radiotherapy through a dedicated childhood cancer fund. This study aims at examining the financial burden of treatment for pediatric neuro-oncology patients seeking financial support from a cancer charity. MATERIALS AND METHODS A retrospective desk review cost analysis of treatment and financial data of neuro-oncology pediatric patients who applied for financial support from a cancer charity over a 12 month period. RESULTS Thirteen pediatric neuro-oncology patients applied for financial support during the study period. 61% had brain tumors of which 50% were Medulloblastoma. The age range was 3–18 years, mean: 10.6 years. All patients received radiotherapy and or chemotherapy. 87% underwent neurosurgical intervention. The total cost of treatment for these patients was USD 56,000 (KES 7,274,175). The Social Health Insurance Fund covered 43% (approximately USD 24,135), leaving a significant financial gap. Patients fundraised or borrowed money to cover part of the cost and the cancer charity supported the remaining costs. At 12 month follow up 92% of the patients who were supported by the cancer charity had completed their treatment and were doing well. CONCLUSIONS The cost of treating childhood cancer remains prohibitive even with social health insurance. Patients and families are forced to incur substantial out-of-pocket expenses or seek help from cancer charities. Financial barriers contribute to treatment abandonment, asserting the important role of cancer charities like Faraja Cancer Support Trust and civil society organizations in enabling access to care. These findings align with global studies on the financial challenges of pediatric oncology treatment in low-resource settings.","author":[{"family":"Hassan","given":"Maryam"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1093/neuonc/noaf201.0764","URL":"https://doi.org/10.1093/neuonc/noaf201.0764","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-6296233/v1","type":"article-journal","title":"Harpy Eagle Optimization: Bio-Inspired Metaheuristic for Complex Problems","abstract":"Abstract This paper presents Harpy Eagle Optimization (HEO), a novel bio-inspired metaheuristic algorithm modeled on the hunting strategies of Harpy Eagles ( Harpia harpyja ), renowned for their agility and precision in rainforests. HEO uses a dual-phase approach—global exploration through soaring and local exploitation via targeted dives—to tackle high-dimensional, non-convex, and multi-modal optimization problems common in engineering, machine learning, and industry. We compare HEO against 15 leading metaheuristics, including PSO, GA, GWO, DE, and HHO, across ten benchmark functions (e.g., Sphere, Rastrigin, Ackley) in dimensions d = 100,50,30,10. HEO excels, converging 40% faster than PSO (70 vs. 115 iterations to 6− 10 on Sphere), achieving a mean fitness of 0.00004 (vs. 0.001 for PSO), and a standard deviation below 0.00002. Wilcoxon (p &lt; 0.001) and Friedman (p &lt; 0.01) tests confirm its robustness. HEO cuts function evaluations by 28% compared to GA while matching PSO’s efficiency. Real-world tests highlight its impact: reducing truss weight by 16%, boosting feature selection accuracy by 8%, cutting scheduling makespan by 13%, and improving fog computing efficiency by 15%. Backed by a solid mathematical framework, extensive analysis, and an open-source Python code, HEO offers a powerful leap forward in optimization, aligning with JOTA ’s rigorous standards and Springer’s Q1 aspirations.","author":[{"family":"Eslami","given":"Omid"}],"issued":{"date-parts":[[2025]]},"DOI":"10.21203/rs.3.rs-6296233/v1","URL":"https://doi.org/10.21203/rs.3.rs-6296233/v1","source":"crossref"},{"id":"doi:10.36871/26189976.2026.03-4.005","type":"article-journal","title":"NEURO-FUZZY MODELS OF BUILDING THERMAL INERTIA IN ADAPTIVE CONTROL PROBLEMS OF HEATING AND AIR CONDITIONING SYSTEMS","abstract":"This article examines the problem of constructing a neurofuzzy model of building thermal inertia for use in adaptive control loops of heating and air conditioning systems. It is shown that the thermal dynamics of building envelopes and interior volumes are characterized by pronounced nonlinearity, parametric uncertainty, and time variability, which limits the applicability of traditional lumped thermal physics models. An approach is proposed that combines fuzzy representations of expert knowledge about the thermal state of an object with the capabilities of neural network parametric tuning. The model structure, input and output variables, fuzzification mechanism, rule identification procedure, and inclusion of the model in a predictive-adaptive control algorithm are discussed. The advantages of the approach in terms of robustness, interpretability, and energy efficiency are substantiated.","author":[{"family":"Osipov","given":"Airat"}],"issued":{"date-parts":[[2026]]},"DOI":"10.36871/26189976.2026.03-4.005","URL":"https://doi.org/10.36871/26189976.2026.03-4.005","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-7982569/v1","type":"article-journal","title":"Intelligent Tuning of PID Parameters Using Nature-Inspired Algorithms","abstract":"Abstract This paper proposes the application of both the Firefly Algorithm (FA) and Particle Swarm Optimization (PSO) for tuning the PID controller parameters of first-, second-, and third-order dynamic systems. FA is distinguished by its simplicity, stable convergence behavior, and superior computational efficiency. Inspired by the bioluminescent behavior of fireflies, the Firefly Algorithm is a well-established meta-heuristic optimization technique [1]. In this study, FA’s performance is benchmarked against the widely used PSO method [2]. Simulation results show that FA yields slightly better improvements in step-response metrics—namely rise time, settling time, and overshoot - while PSO can attain lower fitness values in some scenarios. FA demonstrates the ability to quickly and reliably optimize controller parameters, even for complex system models [3]. Detailed analyses across multiple test cases confirm that FA not only excels at parameter optimization but also provides enhanced stability and robustness in controller design. PSO, on the other hand, sometimes achieves marginally lower fitness values, indicating its potential for fine-tuning in specific cases. This work highlights the applicability of both FA and PSO in academic and industrial control-system design, emphasizing each method’s strengths and trade-offs.","author":[{"family":"Mammadov","given":"Ibrahim"}],"issued":{"date-parts":[[2025]]},"DOI":"10.21203/rs.3.rs-7982569/v1","URL":"https://doi.org/10.21203/rs.3.rs-7982569/v1","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-6504704/v1","type":"article-journal","title":"Bio-Inspired Parallel Error Correction (BPEC): A Transformer-Based AI Algorithm for Robust Space Navigation via DNA-Inspired Consensus Mechanisms","abstract":"Abstract We propose a novel Bio-Inspired Parallel Error Correction (BPEC) algorithm for spacecraft navigation, which integrates DNA-inspired redundancy mechanisms with transformer-based neural architectures to address the challenges of sensor noise and environmental perturbations in space missions. The core innovation lies in modeling redundant sensor inputs as parallel strands, analogous to biological data fidelity, and employing a masked self-attention mechanism to resolve discrepancies dynamically. The proposed method transforms raw sensor measurements into polarized representations, then processes them through a multi-head self-attention layer to achieve consensus, mimicking DNA repair mechanisms. Furthermore, a neural feedback controller adjusts navigation parameters in real-time, ensuring stability under high-noise conditions. The architecture interfaces seamlessly with conventional navigation systems, replacing traditional Kalman filters and PID controllers while offering inherent fault tolerance. Implemented as a sparse mixture-of-experts network, the system specializes in distinct error modes, such as gyroscopic drift or star tracker occlusion, thereby adapting to diverse mission scenarios. Mathematical analysis demonstrates bounded error propagation, a critical property for robustness in deep-space environments. BPEC represents a significant departure from classical approaches, combining bio-inspired parallelism with modern AI techniques to deliver unprecedented resilience in autonomous space navigation. Experimental validation confirms its superiority over state-of-the-art methods, particularly in scenarios with sensor noise exceeding (33 σ ) thresholds. The algorithm's adaptability and scalability make it a promising candidate for future interplanetary missions, where reliability and autonomy are paramount.","author":[{"family":"Xiao","given":"Xiaochen"}],"issued":{"date-parts":[[2025]]},"DOI":"10.21203/rs.3.rs-6504704/v1","URL":"https://doi.org/10.21203/rs.3.rs-6504704/v1","source":"crossref"},{"id":"doi:10.1093/noajnl/vdaf213.116","type":"article-journal","title":"SVSC-03 Challenges in Managing Neuro-Oncology Patients Without Health Insurance: A Case Study of Kenya and Uganda","abstract":"Abstract Access to timely and comprehensive neuro-oncology care remains a significant challenge in East Africa, where health insurance coverage is limited, and out-of-pocket payments dominate healthcare financing. In Kenya and Uganda, uninsured neuro-oncology patients often face catastrophic health expenditures, delayed diagnosis, and poor outcomes. This case study examines the systemic, financial, and clinical challenges of managing neuro-oncology patients without health insurance in these two countries, offering insights into potential solutions for equitable healthcare access. A qualitative case study approach was employed, drawing on a combination of literature review and expert interviews with neurosurgeons, oncologists, and healthcare administrators in Kenya and Uganda. Data sources included peer-reviewed publications, health policy documents, and hospital-based neuro-oncology registries. Thematic analysis was used to identify recurring barriers to care, financial toxicity, and survival outcomes, with a focus on the impact of health insurance status. The analysis revealed several critical challenges, including significant preoperative delays due to unaffordable diagnostic imaging and surgical fees, treatment discontinuation resulting from financial exhaustion, and limited access to postoperative oncology services. Uninsured neuro-oncology patients in both countries were more likely to present with advanced disease, experience longer hospital stays, and have lower 12-month survival rates compared to insured counterparts. Systemic issues such as inadequate public-sector neurosurgical capacity, prolonged referral pathways, and a lack of national cancer financing programs further compounded these difficulties. This case study highlights the urgent need for targeted policy interventions to mitigate financial toxicity and improve access to neuro-oncology care in Kenya and Uganda. Recommendations include expanding national health insurance coverage to include neuro-oncology services, creating cancer-specific financial assistance programs, and establishing regional centers of excellence with streamlined referral networks. A collaborative, multi-sectoral approach is essential to build sustainable, patient-centered neuro-oncology care systems and improve survival outcomes in East Africa.","author":[{"family":"Nassozi","given":"Jalilarah"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1093/noajnl/vdaf213.116","URL":"https://doi.org/10.1093/noajnl/vdaf213.116","source":"crossref"},{"id":"doi:10.1093/neuonc/noaf193.463","type":"article-journal","title":"P14.02.A OPTOCHEMOGENETIC MODELING OF NEURO-CANCER CROSSTALK IN VON HIPPEL-LINDAU DISEASE","abstract":"Abstract BACKGROUND Von Hippel-Lindau (VHL) disease leads to tumors in the nervous system, including cerebellar and spinal cord hemangioblastomas (CBHB, SCHB), and pheochromocytomas (PCC), involving significant neuro-cancer interactions. Understanding these bidirectional interactions is limited by the absence of accurate genetically engineered mouse (GEM) models. Traditional approaches for modeling localized gene recombination face issues such as off-target effects and systemic toxicity. MATERIAL AND METHODS We employed a novel optochemogenetic (OCG) switch technology, using photoactivatable caged 4-hydroxycyclofen (4-OHC) with localized UV illumination, to precisely induce genetic alterations in specific nervous system compartments (cerebellum, spinal cord) and peripheral nervous tissue (adrenal medulla). Using reporter mice (UBC-CreERT2 mTmGf/+), we optimized this UV-driven method to achieve targeted gene recombination. Subsequently, we created GEM models (UBC-CreERT2 Vhlf/f and UBC-CreERT2 Vhlf/f Ptenf/f) to investigate tumor initiation and progression driven by localized neuro-cancer interactions. RESULTS Preliminary findings validated effective localized gene recombination via the OCG switch, demonstrating specificity without systemic side effects. Conventional systemic induction methods caused rapid lethality, highlighting the necessity for spatially controlled gene recombination to study neural-tumor crosstalk accurately. Ongoing experiments focus on refining UV illumination conditions to effectively model the neural influences on cancer initiation and progression. CONCLUSION Our optochemogenetic approach uniquely enables precise investigation of neural influences on tumor formation and tumor-induced neural alterations specific to VHL disease. These GEM models are expected to uncover novel neural-cancer crosstalk mechanisms, advancing our understanding of cancer neuroscience and paving the way for targeted therapeutic strategies.","author":[{"family":"Lu","given":"X"},{"family":"Lu","given":"X"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1093/neuonc/noaf193.463","URL":"https://doi.org/10.1093/neuonc/noaf193.463","source":"crossref"},{"id":"doi:10.1101/2025.06.21.660896","type":"article-journal","title":"Neural Network-Enhanced Investigation of Ferroptosis and Druggability in Early-Onset Alzheimer’s Disease","abstract":"Abstract Background Alzheimer’s disease (AD) is a complex neurodegenerative disorder which is multifactorial in nature. Some of its characteristics are slow cognitive decline, memory problems and behavioral changes. AD patient brains show a progressive synaptic toxicity, autophagy, neuroinflammation, excess generation of reactive oxygen species (ROS), neuronal death and oxidative stress, which occurs due to disrupted metal homeostasis along with tau and amyloid-β protein deposition. Notably, lipid peroxidation, iron buildup and elevated oxidative stress in AD brains suggest a possible molecular connection between ferroptosis and AD neurodegeneration. Methods This study explores the genetic and bioinformatics perspective on the relationship between ferroptosis and AD aiming to identify potential therapeutic biomarkers using Neural network (NN) and Machine learning models. Six ferroptosis related genes were found to be differentially expressed in AD. Further machine learning analysis shortlisted four key biomarker genes. An NN-based diagnostic prediction model was developed and validated using AUC-ROC anaysis, which gave high diagnostic values (AUC-0.92) in the analysis. Results The findings highlight a strong correlation between ferroptosis and altered metabolic functions in AD. miRNA-gene interaction analysis revealed that two biomarker genes, CYBB and ACSL4 can be regulated by several regulatory miRNAs i.e., hsa-miR-146-5p, hsa-miR-106b-5p, hsa-miR-223-3p, hsa-miR-155-5p, hsa-miR-34a-5p, hsa-miR-125b-5p and hsa-miR-27a-3p suggesting their potential as early diagnostic biomarkers. Immune microenvironment analysis revealed strong neuroinflammatory responses were in AD with increased infiltration of macrophages (M0, M1 and M2), monocytes and multiple T cell subsets. This heightened immune activity may be driven by ferroptosis-induced oxidative stress, contributing to neuronal death. Furthermore, druggability of these targets was evaluated and several drugs were identified that may be potentially repurposed for therapeutic intervention in AD pathogenesis. Conclusion This study presents a diagnostic predictive model integrating gene expression, miRNA regulation and immune infiltration analysis, offering a novel perspective on early AD detection. The identified ferroptosis-related biomarkers and regulatory miRNAs could serve as valuable tools for clinical diagnosis and targeted therapeutic intervention, advancing personalized treatment strategies for Alzheimer’s disease.","author":[{"family":"Singh","given":"Pratibha"},{"family":"Rath","given":"Soumya"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1101/2025.06.21.660896","URL":"https://doi.org/10.1101/2025.06.21.660896","source":"crossref"},{"id":"doi:10.32942/x2792n","type":"article-journal","title":"Anomaly detection in metabarcoding amplicon reads using an LSTM-CNN deep neural network ensemble (MetAnoDe)","abstract":"Metabarcoding has emerged as a critical tool in ecology and other scientific disciplines, facilitating species identification in diverse samples for biodiversity monitoring, community and microbiome analysis, dietary studies, and understanding species interactions. However, challenges arise from errors and artifacts introduced during laboratory processes such as PCR and sequencing. Manual inspection is impractical due to the vast amount of sequences, necessitating rapid algorithms to clean the data. Thorough bioinformatic data cleanup can reduce such mistakes by removal of low-quality sequences or such classified as non-fitting through alignments. However, in practice some anomalous sequences evade detection, while also normal sequences may be mistakenly removed. Deep neural networks (DNNs) offer a promising solution by recognizing complex DNA sequence patterns. In this study I present a new software MetAnoDe (Metabarcoding Anomaly Detection), featuring development of novel deep-learning LSTM and CNN models for independent application and use as an ensemble model. MetAnoDe employs an alignment-free approach that complements existing tools, enhancing data cleanup efficiency. Here, the three models were trained for bacterial 16S-V4 and plant ITS2 markers which can be readily reused in other studies. Cross-validation and real-world data testing demonstrate high accuracy. Optimal integration into pipelines can also streamline overall runtime, synergizing effectively with current alignment-based methods. It is further adaptable for other markers due to the software's automated model training capability. In conclusion, MetAnoDe enhances metabarcoding by efficiently identifying anomalous sequences. An integration of DNNs with traditional approaches enhances biodiversity estimates by reducing non-target sequence inclusion, ensuring more accurate and comprehensive results.","author":[{"family":"Keller","given":"Alexander"}],"issued":{"date-parts":[[2025]]},"DOI":"10.32942/x2792n","URL":"https://doi.org/10.32942/x2792n","source":"crossref"},{"id":"doi:10.2139/ssrn.5088827","type":"manuscript","title":"Banana Leaf Nutrient Deficiency Detection by Canny Edge Detection with Dense Neural Network","abstract":"Image processing is dominant field for medical, agriculture, industrial image processing. It enhances the quality image and support for diagnosis. Normally it handles images from satellite, x-rays, scans, mobile captures, agriculture captures etc. Image processing offers various types of techniques to process and improve the quality of image. The image pre-processing techniques such as noise reduction, filters, edge detection, thresholding, segmentation, clustering and banana leaf nutrient deficiency detection. This study utilized edge detection technique for banana leaf images before the banana leaf nutrient deficiency detection. Edge detection is a technique identifying edges, points, intensity changes and discontinuities in the image. This study experiments Dense Neural Network for banana leaf detection and attained an accuracy of 95.67%. This research mutually proposed CNN-LSTM model to predict the banana leaf nutrient deficiency detection.","author":[{"family":"Srinivasan","given":"Uma"}],"issued":{"date-parts":[[2025]]},"DOI":"10.2139/ssrn.5088827","URL":"https://doi.org/10.2139/ssrn.5088827","source":"crossref"},{"id":"doi:10.13005/ojps10.02.11","type":"article-journal","title":"Facial Expression Recognition Method Based on Convolutional Neural Network","abstract":"Deep learning algorithms are a subset of machine learning algorithms that aim to discover multiple levels of distributed representations of input data. Recently, many deep learning algorithms have been proposed to solve traditional artificial intelligence problems. Now days, deep learning has been extensively studied in the field of computer vision and as such, a large number of related methods have arisen. Today, different algorithms and models of neural network–based research have made their place among the classification of images. The main purpose of these algorithms is to train the machine in artificial networks in a way that ultimately has a diagnosis close to the human brain. Among a variety of neural networks, CNN's channel neural networks usually offer good accuracy in the classification of images. In this article, in the first episode, we will discuss 4 deep learning methods: Convolutional neural network (CNN), Restricted Boltzmann Machines (RBMS), Autoencoders and Sparse coding, which after determining the necessary assumptions and applying a preliminary pre–training using the channel neural network algorithm we need Convolutional to perform a general preprocessing on the entered samples. Therefore, a preprocessing is performed on all data and preprocessed samples are stored in a separate location and then the rest of the processes are applied to these samples. Then, we use deep learning to identify faces and reveal them, and deep learning algorithms to reveal different subjects.","author":[{"family":"Heidari","given":"Alireza"}],"issued":{"date-parts":[[2026]]},"DOI":"10.13005/ojps10.02.11","URL":"https://doi.org/10.13005/ojps10.02.11","source":"crossref"},{"id":"doi:10.7868/s3034498025060013","type":"article-journal","title":"REFLECTOR TYPE RECOGNITION USING NEURAL NETWORK BASED ON TOFD ECHOES","abstract":"In this paper we propose to automate the classification of reflector types by TOFD-echoes using ResNet-18 convolutional neural network. The main focus is on modeling and classification of reflectors such as cracks, pores, non-welds and void areas. Experiments included training the model on TOFD echoes calculated both in a numerical experiment and TOFD echoes measured during ultrasonic inspection. The results showed high classification accuracy: 96.2 % in the numerical experiment, 97 % on experimentally measured TOFD-echoes with different types of reflectors. The study confirmed the possibility of using neural networks to determine the reflector type from TOFD-echo signals, which allows to automate the process of nondestructive testing and reduce the influence of human factor. For further development of the method it is suggested to use segmentation models for processing images with several reflectors","author":[{"family":"Bazulin","given":"EG"}],"issued":{"date-parts":[[2026]]},"DOI":"10.7868/s3034498025060013","URL":"https://doi.org/10.7868/s3034498025060013","source":"crossref"},{"id":"doi:10.31219/osf.io/6kxeb","type":"article-journal","title":"MCMC-Enhanced Neural Network Operators for Dynamic Systems and Adaptive Control","abstract":"This paper introduces a novel framework that integrates neural network operators with Markov Chain Monte Carlo (MCMC) methods to model and control stochastic dynamic systems and adaptive control systems. Three foundational theorems are proposed: the stability of stochastic dynamic systems, the convergence of MCMC-based optimization, and the stability of adaptive control systems. These theorems establish the theoretical underpinnings of the integrated framework, focusing on stability, convergence, and optimal parameter adjustment. The stability theorem demonstrates error bounds for neural network approximations in Sobolev spaces, while the MCMC convergence theorem guarantees sampling efficacy for parameter optimization. The stability of adaptive control systems theorem ensures asymptotic tracking of desired trajectories and convergence of adaptive parameters. Numerical simulations validate the theoretical results, highlighting improved precision, robustness, and efficiency in dynamic system modeling and control. The findings provide a robust foundation for enhancing neural network-based approaches in dynamic systems and adaptive control, with significant implications for practical applications in complex, stochastic environments.","author":[{"family":"Santos","given":"Romulo"}],"issued":{"date-parts":[[2025]]},"DOI":"10.31219/osf.io/6kxeb","URL":"https://doi.org/10.31219/osf.io/6kxeb","source":"crossref"},{"id":"doi:10.2478/arsa-2025-0007","type":"article-journal","title":"Accelerating Atmosphere Modeling: Neural Network Enhancements for Faster NRLMSISE Calculations","abstract":"Abstract NRLMSISE is an empirical model that allows us to predict temperatures and densities of the main atmospheric components. The model is widely used to evaluate atmospheric impacts on satellite orbits and laser beam refraction which come through the atmosphere, such as those used for Earth-satellite distance measurements. Model of the atmosphere is a valuable part of the Satellite Laser Ranging processing software like Kyiv Geodynamics (Juliette). Juliette is written in C++ and exploits the C++ clone of NRLMSISE written by the second author. The C++ version produces the same outputs as an official Fortran code. Accurate modeling of atmospheric influences on satellite motion requires performing numerous calculations along satellite orbits or laser beam paths, which are computationally intensive. By decreasing calculation time of NRLMSISE, we would not only save the modeling time but also give a prospect for a wider application of the model due to lowering computational resource demands. Our work demonstrates how the traditional NRLMSISE model can be effectively translated into a neural network. This conversion achieves significant performance gains on both CPU and GPU while maintaining acceptable accuracy when compared to the C++ implementation of NRLMSISE. We demonstrate the process of moving NRLMSISE to a neural network, the resulting accuracy, ease of running the trained model on CUDA-enabled GPUs, and the obtained boost of performance on both CPU and GPU.","author":[{"family":"Kashyn","given":"Volodymyr"},{"family":"Choliy","given":"Vasyl"}],"issued":{"date-parts":[[2025]]},"DOI":"10.2478/arsa-2025-0007","URL":"https://doi.org/10.2478/arsa-2025-0007","source":"crossref"},{"id":"doi:10.25126/jtiik.2025128085","type":"article-journal","title":"Perbandingan Kinerja Arsitektur Convolutional Neural Network Pada Deteksi Malaria Menggunakan Citra Sel Darah","abstract":"Malaria masih menjadi salah satu penyebab kematian tertinggi di dunia, terutama di daerah yang berstatus endemi. Standar emas penegakan diagnosis malaria adalah berbasis citra apusan atau sel darah yang diperoleh dengan menggunakan mikroskop. Kendala utama dalam penegakan diagnosis ini adalah kurangnya tenaga ahli untuk melakukan asesmen citra sel darah. Oleh karena itu, dilakukan diagnosis malaria berbasis citra sel darah menggunakan Artificial Intelligent (AI) / kecerdasan buatan. Deteksi malaria berbasis AI yang dilakukan pada studi-studi sebelumnya telah menghasilkan kinerja yang sudah baik. Namun, kinerja deteksi ini masih dapat ditingkatkan. Studi ini menggunakan 27.558 citra sel darah yang terdiri dari 13.779 sel darah terinfeksi dan 13.779 tidak terinfeksi. Citra-citra sel darah ini dibagi menjadi tiga kelompok, yaitu pelatihan (80%); validasi (10%); dan pengujian (10%). Pada studi ini, digunakan ResNet50; ResNet101; ResNet152; ResNet50V2; ResNet101V2; ResNet152V2; DenseNet121; DenseNet169; DenseNet201; InceptionV3; InceptionResNetV2; VGG16; VGG19; dan MobileNetV2. Tujuan utama dari studi ini adalah mencari arsitektur CNN yang memiliki kinerja terbaik dalam deteksi malaria berbasis citra sel darah. Perbandingan kinerja diases dengan menggunakan nilai akurasi, sensitivitas, spesifisitas, skor F1, dan Area Under the Curve (AUC). Arsitektur MobileNetV2 memberikan kinerja paling baik dengan nilai rata-rata pelatihan, validasi, dan pengujian tertinggi. Nilai rata-rata akurasi mencapai 97,68%; spesifisitas 98,61%; sensitivitas 96,75%; Skor F1 97,70%; dan AUC sebesar 99,65%. Selain itu, waktu pembuatan model arsitektur MobileNetV2 hanya sekitar 2,5 jam. Selain itu, jumlah lapisan convolutional tidak memengaruhi kinerja deteksi malaria. Dengan lapisan convolutional berjumlah 53, MobileNetV2 berkinerja lebih baik dibandingkan dengan arsitektur-arsitektur lain dengan jumlah lapisan convolutional lebih banyak. Abstract Malaria is still one of the highest causes of death in the world, especially in the endemic areas. The gold standard for diagnosing malaria is based on smears blood smears or cells image which is obtained using a microscope. The main challenge in detecting malaria is the lack of experts to assess the blood smears. Therefore, the detection is carried out using Artificial Intelligence (AI). Previous studies that used AI to detect malaria have a good performance. However, the detection performance can still be improved. Furthermore, previous studies only used one or two or three performance metrics. This study used 27,558 blood cell images consisting of 13,779 infected and 13,779 uninfected blood cells. These blood cell images are divided into three groups, i.e. training (80%); validation (10%); and testing (10%). In this study, several CNN architectures are used, such as ResNet50; ResNet101; ResNet152; ResNet50V2; ResNet101V2; ResNet152V; DenseNet121; DenseNet169; DenseNet201; InceptionV3; InceptionResNetV2; VGG16: VGG19: and MobileNetV2. The main objective of this study is to find the CNN architecture that has the best performance in blood cell image-based malaria detection. Comparison of performance of CNN architectures are assessed using accuracy, sensitivity, specificity, F1 score, and Area Under the Curve (AUC) values. The MobileNetV2 architecture provides the best performance with the highest average values of training, validation, and testing. The average accuracy value of 97.68%; specificity of 98.61%; sensitivity of 96,75%; F1 Score of 97.70%; and AUC of 0.9965. In addition, the time to build the MobileNetV2 model is about 2.5 hours, the fastest one. This study shows that the number of convolutional layers does not affect malaria detection performance. With 53 convolutional layers, MobileNetV2 has the best performance.","author":[{"family":"Setiawan","given":"Agung"}],"issued":{"date-parts":[[2025]]},"DOI":"10.25126/jtiik.2025128085","URL":"https://doi.org/10.25126/jtiik.2025128085","source":"crossref"},{"id":"doi:10.25126/jtiik.20258085","type":"article-journal","title":"Perbandingan Kinerja Arsitektur Convolutional Neural Network Pada Deteksi Malaria Menggunakan Citra Sel Darah","abstract":"Malaria masih menjadi salah satu penyebab kematian tertinggi di dunia, terutama di daerah yang berstatus endemi. Standar emas penegakan diagnosis malaria adalah berbasis citra apusan atau sel darah yang diperoleh dengan menggunakan mikroskop. Kendala utama dalam penegakan diagnosis ini adalah kurangnya tenaga ahli untuk melakukan asesmen citra sel darah. Oleh karena itu, dilakukan diagnosis malaria berbasis citra sel darah menggunakan Artificial Intelligent (AI) / kecerdasan buatan. Deteksi malaria berbasis AI yang dilakukan pada studi-studi sebelumnya telah menghasilkan kinerja yang sudah baik. Namun, kinerja deteksi ini masih dapat ditingkatkan. Studi ini menggunakan 27.558 citra sel darah yang terdiri dari 13.779 sel darah terinfeksi dan 13.779 tidak terinfeksi. Citra-citra sel darah ini dibagi menjadi tiga kelompok, yaitu pelatihan (80%); validasi (10%); dan pengujian (10%). Pada studi ini, digunakan ResNet50; ResNet101; ResNet152; ResNet50V2; ResNet101V2; ResNet152V2; DenseNet121; DenseNet169; DenseNet201; InceptionV3; InceptionResNetV2; VGG16; VGG19; dan MobileNetV2. Tujuan utama dari studi ini adalah mencari arsitektur CNN yang memiliki kinerja terbaik dalam deteksi malaria berbasis citra sel darah. Perbandingan kinerja diases dengan menggunakan nilai akurasi, sensitivitas, spesifisitas, skor F1, dan Area Under the Curve (AUC). Arsitektur MobileNetV2 memberikan kinerja paling baik dengan nilai rata-rata pelatihan, validasi, dan pengujian tertinggi. Nilai rata-rata akurasi mencapai 97,68%; spesifisitas 98,61%; sensitivitas 96,75%; Skor F1 97,70%; dan AUC sebesar 99,65%. Selain itu, waktu pembuatan model arsitektur MobileNetV2 hanya sekitar 2,5 jam. Selain itu, jumlah lapisan convolutional tidak memengaruhi kinerja deteksi malaria. Dengan lapisan convolutional berjumlah 53, MobileNetV2 berkinerja lebih baik dibandingkan dengan arsitektur-arsitektur lain dengan jumlah lapisan convolutional lebih banyak. Abstract Malaria is still one of the highest causes of death in the world, especially in the endemic areas. The gold standard for diagnosing malaria is based on smears blood smears or cells image which is obtained using a microscope. The main challenge in detecting malaria is the lack of experts to assess the blood smears. Therefore, the detection is carried out using Artificial Intelligence (AI). Previous studies that used AI to detect malaria have a good performance. However, the detection performance can still be improved. Furthermore, previous studies only used one or two or three performance metrics. This study used 27,558 blood cell images consisting of 13,779 infected and 13,779 uninfected blood cells. These blood cell images are divided into three groups, i.e. training (80%); validation (10%); and testing (10%). In this study, several CNN architectures are used, such as ResNet50; ResNet101; ResNet152; ResNet50V2; ResNet101V2; ResNet152V; DenseNet121; DenseNet169; DenseNet201; InceptionV3; InceptionResNetV2; VGG16: VGG19: and MobileNetV2. The main objective of this study is to find the CNN architecture that has the best performance in blood cell image-based malaria detection. Comparison of performance of CNN architectures are assessed using accuracy, sensitivity, specificity, F1 score, and Area Under the Curve (AUC) values. The MobileNetV2 architecture provides the best performance with the highest average values of training, validation, and testing. The average accuracy value of 97.68%; specificity of 98.61%; sensitivity of 96,75%; F1 Score of 97.70%; and AUC of 0.9965. In addition, the time to build the MobileNetV2 model is about 2.5 hours, the fastest one. This study shows that the number of convolutional layers does not affect malaria detection performance. With 53 convolutional layers, MobileNetV2 has the best performance.","author":[{"family":"Setiawan","given":"Agung"}],"issued":{"date-parts":[[2025]]},"DOI":"10.25126/jtiik.20258085","URL":"https://doi.org/10.25126/jtiik.20258085","source":"crossref"},{"id":"doi:10.12732/ijam.v38i4s.1530","type":"article-journal","title":"GRAPHSECNET: A GRAPH NEURAL NETWORK FRAMEWORK FOR PREDICTIVE CYBERSECURITY INTELLIGENCE IN DYNAMIC NETWORK ENVIRONMENTS","abstract":"Contemporary cybersecurity threats exploit complex network topologies and temporal attack patterns that traditional detection systems fail to adequately model. This paper presents GraphSecNet, a novel Graph Neural Network framework that transforms network security data into dynamic graph representations for enhanced threat detection. The framework integrates Temporal Graph Attention Networks, self-supervised contrastive learning, and multi-perspective anomaly detection to capture both spatial network relationships and temporal attack evolution. Comprehensive evaluation on established cybersecurity datasets (CICIDS2017, UNSW-NB15, NSL-KDD) demonstrates significant performance improvements: 91.7% F1-score representing 9.8% improvement over state-of-the-art methods, 46% reduction in false positive rates, and scalable processing at 25,000 events per second. Graph attention mechanisms provide interpretable explanations for threat decisions, addressing critical gaps in explainable AI for cybersecurity. Statistical analysis confirms significance across all datasets (p &lt; 0.001, Cohen's d &gt; 1.8), validating the effectiveness of graph-based approaches for network threat intelligence.","author":[],"issued":{"date-parts":[[2025]]},"DOI":"10.12732/ijam.v38i4s.1530","URL":"https://doi.org/10.12732/ijam.v38i4s.1530","source":"crossref"},{"id":"doi:10.4018/979-8-3373-0735-0.ch001","type":"article-journal","title":"Introduction to Neural Networks","abstract":"Neural networks, inspired by the biological neural structures in the human brain, are computational models that are central to many modern artificial intelligence applications. These networks consist of interconnected layers of nodes, or neurons, that process data through weighted connections and activation functions. The flexibility of neural networks enables them to solve a wide range of complex problems, including pattern recognition, classification, and prediction tasks. This abstract provides a concise overview of the fundamental principles behind neural networks, their architecture, training methodologies, and their application across diverse fields such as computer vision, natural language processing, and autonomous systems. Neural networks' ability to learn from large datasets and adapt to various input patterns has revolutionized fields such as machine learning and deep learning, making them indispensable tools for solving real-world challenges.","author":[{"family":"Pavunraj","given":"D"},{"family":"Anbumaheshwari","given":"K"}],"issued":{"date-parts":[[2025]]},"DOI":"10.4018/979-8-3373-0735-0.ch001","URL":"https://doi.org/10.4018/979-8-3373-0735-0.ch001","source":"crossref"},{"id":"doi:10.59879/ty0kz","type":"article-journal","title":"DEEP LEARNING BASED AGRICULTURE TRAFFIC PREDICTION USING GATED RECURSIVE DEEP NEURAL NETWORK IOT ENVIRONMENT","abstract":"IoT traffic data can be used to apply deep learning technology to agriculture. The application of deep learning algorithms in agriculture has improved, resource management, and decisionmaking with positive outcomes. Precision agriculture can assist regulate crop yields by applying nutrients only when necessary to enhance crop quality and lessen adverse environmental effects. This is made possible by IoT capabilities. Identifying redundant data traffic remains a major research challenge in the field of IoT-based agricultural automation, despite the fact that numerous solutions have been offered. Additionally, farmers do not receive the necessary information about water levels, soil conditions, etc.","author":[{"family":"Sofiya","given":"M"},{"family":"Arulmozhi","given":"M"}],"issued":{"date-parts":[[2025]]},"DOI":"10.59879/ty0kz","URL":"https://doi.org/10.59879/ty0kz","source":"crossref"},{"id":"doi:10.11648/j.ajnna.20251102.13","type":"article-journal","title":"A Fuzzy Neural Network System for Denoising Magnetic Resonance Images","abstract":"Image acquisition is an essential step in image processing. When the image acquisition is done the image that is generated is subjected to Impulse noise, Gaussian noise etc. We have performed the image denoising on images inflicted with impulse noise. Image denoising is an essential step in all types of image processing. Traditional techniques reduce the noise in the image but it also reduces the quality of the image. Traditional filters like gaussian filter, median filter is analyzed which work in the spatial domain and filters working in the frequency domain are also considered like Butterworth filters, Weiner filter. A Deep residual Neural Network filter is proposed which is compared with the Fuzzy Neural Network denoiser. Their performance is compared on the metrics PSNR and SSIM. The Fuzzy Neural Network system improves the SSIM significantly compared to a deep residual neural network and a comparison is made with traditional image denoising methods. We also compare the performance of the deep residual neural network, Fuzzy Neural Network system and Median denoising algorithm on impulse noise has been compared. The performance of deep neural networks depends on the total number of examples used and the performance can be improved if we have more image pairs.","author":[{"family":"Das","given":"Shubhajoy"},{"family":"Das","given":"Debashis"}],"issued":{"date-parts":[[2025]]},"DOI":"10.11648/j.ajnna.20251102.13","URL":"https://doi.org/10.11648/j.ajnna.20251102.13","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-6213558/v1","type":"article-journal","title":"BioLogicalNeuron: A Biologically Inspired Neural Network Layer with Homeostatic Regulation and Adaptive Repair Mechanism","abstract":"Abstract Neural networks face persistent challenges in maintaining stability and robustness during training, particularly in noisy or high-dimensional domains like molecular analysis. Inspired by biological neural systems that leverage homeostasis and self-repair to sustain functionality, this paper proposes BioLogicalNeuron—a novel neural network layer that integrates calcium-driven homeostatic regulation, adaptive repair mechanisms, and dynamic stability monitoring. The layer mimics biological calcium dynamics to maintain neuronal activity within optimal ranges, proactively triggers targeted synaptic repair and adaptive noise injection to counteract degradation, and modulates learning rates via real-time health metrics. Extensive experiments across multiple molecular and chemical datasets show that BioLogicalNeuron achieves state-of-the-art break performance. The layer's performance is particularly strong on molecular datasets, where its biological mechanisms naturally align with molecular structure learning. Through detailed analysis of calcium dynamics and health-stability relationships, this work demonstrate that BioLogicalNeuron achieves a biologically plausible balance between stability and plasticity, offering insights into both artificial and biological neural networks. This results suggest that incorporating biological mechanisms into neural architectures can lead to more robust and effective learning systems, particularly for molecular and chemical analysis tasks.","author":[{"family":"Hakim","given":"Md"},{"family":"Alam","given":"Mohammad"}],"issued":{"date-parts":[[2025]]},"DOI":"10.21203/rs.3.rs-6213558/v1","URL":"https://doi.org/10.21203/rs.3.rs-6213558/v1","source":"crossref"},{"id":"doi:10.7490/f1000research.1118293.1","type":"article-journal","title":"Convolutional neural network architectures for CAFA4","abstract":"Introduction The Critical Assessment of protein Function Annotation (CAFA) challenge concerns the prediction of annotated ontology terms for given protein sequences [Radivojac et al., 2013]. In the current CAFA 4 challenge, the goal is to predict terms from the Gene Ontology (GO), the Human Phenotype Ontology (HPO) and the Disorder Ontology (DO) ontologies for a set of ~100k sequences provided as prediction targets. In this work, different convolutional neural network architectures were examined for the task of protein function prediction. UniProt protein sequence data was enriched with additional information from protein taxonomical hierarchies and InterProScan sequence analyses before being processed with the networks, predicting the different ontologies' terms in a multi-label classification task. Data Preprocessing The same data preprocessing approach was applied with all ontologies. The used protein sequences consisted of the ~560k union of the UniProt XML release and the set of sequences provided as CAFA4 prediction targets [UniProt Consortium, 2019]. GO annotations and protein lineage information were imported from the UniProt release. HPO and DO annotations were extracted from their respective databases and added to the sequence information. All annotated terms were propagated using the GOA Tools package, so that all of their ancestor terms were included as individual labels [Klopfenstein et al., 2018]. InterProScan sequence analyses were imported from the dataset release and added to the protein sequence information [Jones et al., 2014]. Neural Architectures All neural network methods were implemented with the Keras package and were based on a one-dimensional, convolutional approach [Chollet et al., 2015, LeCun et al., 1995]. Both parallel and nested convolutional architectures were tested. In the parallel convolution approach, the outputs of the convolutional layers were concatenated together, whereas in the nested approach each convolutional layer produced the output of the next layer. In addition, the suitability of a one-dimensional variant of the MobileNet V2 image recognition network was tested, as well as a network based on the DeepGOPlus architecture [Kulmanov and Hoehndorf, 2020, Sandler et al., 2018]. All networks received as inputs sequences of a maximum of 1000 elements, where each element corresponded to a single amino acid. Each element contained an 8-dimensional embedding of an amino acid, as well as optionally 8-dimensional embeddings for InteProScan domain , family and homologous superfamily features, with a maximum of 1000 most common unique features included in each category. These sequences were processed by the convolutional layers and merged together with 4-dimensional embeddings representing five levels of the protein's organism's taxonomy lineage, before being finally processed with a dense layer followed by a multi-label prediction layer, which generated individual predictions for up to 1000 of the most common annotated terms. A dropout of 0.1 was used after the inputs and the convolutional layers to improve regularization. Several optimizers were tested, with the Adam and SGD ones achieving the most promising results. The models were trained with a learning rate of 1e-5 and with early stopping, for 10-100 epochs depending on the computational complexity of the model, the size of the dataset and the available resources. Experimental Setup The entire dataset was randomly divided into six approximately equal subsets so that all homologs of a protein were in the same subset. Training was done by cross-validation, so that four subsets at a time were used for training, one for parameter optimization and one for prediction. The final predictions were acquired by joining the six predicted subsets together, thus covering the whole dataset. Finally, the CAFA4 targets were separated from the whole set of predicted sequences, and converted to the submission format. Experiments were performed with using e","author":[{"family":"Björne","given":"Jari"}],"issued":{"date-parts":[[2025]]},"DOI":"10.7490/f1000research.1118293.1","URL":"https://doi.org/10.7490/f1000research.1118293.1","source":"crossref"},{"id":"doi:10.2139/ssrn.6878407","type":"manuscript","title":"Accelerating gas-network feasibility screening with a physics-informed graph neural network surrogate","abstract":"Large-scale gas-network scenario evaluation is a computational bottleneck in integrated energy-system planning, particularly when gas infrastructure interacts with power, heat, hydrogen, and sector-coupling pathways. Conventional nonlinear hydraulic solvers provide reliable feasibility assessment but are costly for stochastic screening, whereas unconstrained learning-based surrogates may produce hydraulically infeasible states. This study develops a physics-informed graph neural network surrogate for steady-state gas-network simulation and feasibility screening. The model uses an edge-centric architecture to predict pipe-level squared-pressure differences and flows. A differentiable projection layer enforces nodal mass conservation on predicted flows, while a Laplacian reconstruction maps edge pressure differences to topologically consistent nodal pressures.The framework is evaluated on GasLib-134, GasLib-135, and GasLib-582 using stochastically generated operating scenarios. On the meshed 582-node benchmark trained with 5000 scenarios, the surrogate achieves a pressure mean absolute error of 1.05~bar, corresponding to 1.3% of the realized pressure range, with R2 = 0.981. Projected-flow predictions reach R2 = 0.972, and mass-balance residuals are reduced to numerical precision, on the order of 10-5--10-4 Nm3/s. Compared with the MYNTS reference solver, inference is reduced from seconds to milliseconds, with the largest benchmark evaluated in less than 40ms. Loadability and out-of-distribution stress-test evaluations demonstrate robust feasibility screening under high-load conditions, while strongly localized demand concentrations are identified as cases requiring solver-based verification near feasibility limits. The framework provides a physically constrained planning accelerator for high-volume scenario screening and prioritization.","author":[{"family":"Jiang","given":"Dongrui"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2139/ssrn.6878407","URL":"https://doi.org/10.2139/ssrn.6878407","source":"crossref"},{"id":"doi:10.46824/megasains.v14i2.140","type":"article-journal","title":"PREDIKSI KEJADIAN PETIR DENGAN ARTIFICIAL NEURAL NETWORK DI WILAYAH KABUPATEN KEPULAUAN TANIMBAR","abstract":"Various research efforts have been made to determine thunderstorm prediction methods, one of which involves using upper air data. However, the use of atmospheric stability threshold values as a reference does not always apply uniformly to all locations due to differences in the characteristics of each region. Therefore, a more objective and precise approach is needed in predicting thunderstorm events, including the application of artificial neural network (ANN) techniques. In this study, the Artificial Neural Network (ANN) method, which is an implementation of artificial intelligence, is used to predict thunderstorm events in the Saumlaki region. The ANN input not only relies on raw data in the form of atmospheric instability index values but also uses feature selection processing to reduce the dimensionality of multivariate input data, minimizing the loss of input data. This process focuses only on essential information and eliminates linear dependencies between features, a technique known as Principal Component Analysis (PCA). The research results indicate that ANN with PCA technique has a higher level of accuracy in predicting thunderstorm events in the Saumlaki region.","author":[{"family":"Adiredjo","given":"Indra"}],"issued":{"date-parts":[[2025]]},"DOI":"10.46824/megasains.v14i2.140","URL":"https://doi.org/10.46824/megasains.v14i2.140","source":"crossref"},{"id":"doi:10.1002/mma.10574","type":"article-journal","title":"Stability analysis of stochastic Lyapunov functions: Applications to memristor neural networks","abstract":"This paper investigates the mean square exponential stability of stochastic neural networks relying on memristor with leakage delay with different types of activation functions. To this aim, we introduce a new suitable stochastic Lyapunov‐Krasovskii functional (SLKF) and employ Filippov solutions to derive stability criteria using It 's formula. We encounter with a nonlinear matrix inequality which should be converted to a linear matrix inequality (LMI) problem by using Schur complement lemma. The proposed problem is handled by using the CVX toolbox in MATLAB software. In the numerical examples section, we bring two examples related to two‐ and three‐dimentional memristor‐based neural networks whose coefficients satisfy in the Schur complement lemma. The figures show that the employed stochastic Lyapunov functions can capture the exponential stability conditions.","author":[{"family":"Rahimi","given":"Vaz'he"},{"family":"Ahmadian","given":"Davood"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1002/mma.10574","URL":"https://doi.org/10.1002/mma.10574","source":"crossref"},{"id":"doi:10.70675/dcef4edez91f2z4abcz91b3z66d244e29bfc","type":"article-journal","title":"Modeling and design of neural network architectures for neural artificial-biological hybridization based on synchronous approach","abstract":"Modélisation et conception par approche synchrone d'architectures neuronales hybrides biologique-artificiel Alors que les Réseaux de Neurones Artificiels (RNA) continuent de progresser dans des domaines tels que l'apprentissage automatique, la robotique, les véhicules autonomes et le diagnostic de santé, un nouveau cadre d'application gagne du terrain à la fois dans les secteurs académique et industriel : la Neurobiohybridation. Ce domaine cherche à établir des connexions entre des neurones artificiels et biologiques dans le but de comprendre et potentiellement de réparer ou remplacer des fonctions cérébrales perdues suite à des maladies ou des accidents. Dans cette perspective, le développement de réseaux de neurones artificiels inspirés biologiquement, souvent appelés Réseaux de Neurones à Spikes (SNNs), est essentiel pour améliorer la compatibilité entre les systèmes neuronaux artificiels et biologiques. Notre thèse s'inscrit dans ce contexte en utilisant l'approche synchrone pour modéliser, mettre en œuvre et simuler des SNNs bio-inspirés et biomimétiques. En utilisant des vérificateurs de modèles, qui permettent de prouver ou d'extraire des propriétés des systèmes de manière formelle, notre objectif est d'acquérir une compréhension plus complète des comportements biologiques dans le future. Pour la première fois dans ce contexte, nous utilisons le langage Light Esterel pour atteindre nos objectifs. Nous démontrons son potentiel dans la mise en oeuvre de modèles neuronaux, initiant une bibliothèque de modèles pour explorer différents types de SNNs. Tout au long de cette thèse, nous avons développé un cadre complet basé sur Light Esterel pour modéliser, simuler et mettre en oeuvre divers modèles de SNNs. Pour aborder les expériences de neurobiohybridation, nous avons développé notre propre architecture matérielle, SynchNN, capable d'exécuter en temps réel des SNNs récurrents en utilisant notre bibliothèque de modèles. Le cadre que nous avons développé est complété par un framework de simulation en cours de développement, visant à réaliser des expériences de neurobiohybridation à l'avenir.","author":[{"family":"Rasamuel","given":"Marino"}],"issued":{"date-parts":[[2026]]},"DOI":"10.70675/dcef4edez91f2z4abcz91b3z66d244e29bfc","URL":"https://doi.org/10.70675/dcef4edez91f2z4abcz91b3z66d244e29bfc","source":"crossref"},{"id":"doi:10.55529/jaimlnn.52.58.68","type":"article-journal","title":"Fedgraphnet: a federated graph neural network framework for privacy-preserving traffic forecasting in heterogeneous IOT networks","abstract":"However, in fifth-generation (5G) and beyond networks, where mobile systems support non-communication applications and act as heterogeneous Internet of Things (IoT) environments, accurate forecasting is essential for proactive resource management, network slicing optimisation, and Quality of Service (QoS) assurance. However, the distributed and privacy-sensitive nature of IoT data limits centralised learning approaches. This paper proposes FedGraphNet, a federated learning (FL) framework integrating a spatio-temporal graph neural network (ST-GNN) with differential privacy (DP) for collaborative traffic prediction without sharing raw data among distributed IoT nodes. FedGraphNet introduces the Adaptive Graph Attention Aggregation (AGAA) module to dynamically construct adjacency matrices from partial network observations, addressing structural heterogeneity in real-world IoT deployments. A communication-efficient TopK-SVD gradient compression strategy reduces uplink overhead by 68.4% with less than 1.2% accuracy loss. A calibrated Gaussian mechanism ensures (ε=1.0, δ=10-5)-differential privacy during aggregation. Experiments on TaxiBJ21, Metr-LA, and PEMS-BAY datasets show that FedGraphNet reduces Mean Absolute Error (MAE) by 2.84, 3.11, and 1.96 respectively compared with six baselines, including FedAvg-GCN, Diffusion Convolutional Recurrent Neural Network (DCRNN), and Graph-WaveNet. The framework also reduces communication cost by 3.1× and accelerates convergence by 54.4% over FedAvg-GCN. Notably, FedGraphNet with ε=1.0 DP outperforms the FedGNN baseline without privacy protection, indicating that calibrated noise injection can serve as an effective regulariser for non-independent and identically distributed (non-IID) IoT traffic distributions. These results demonstrate the trade-offs among spatio-temporal accuracy, communication efficiency, and formal privacy, validating FedGraphNet as a deployable solution for next-generation 5G IoT network management.","author":[{"family":"Djumayozovna","given":"Zaripova"}],"issued":{"date-parts":[[2026]]},"DOI":"10.55529/jaimlnn.52.58.68","URL":"https://doi.org/10.55529/jaimlnn.52.58.68","source":"crossref"},{"id":"doi:10.35314/rxd38a11","type":"article-journal","title":"Network Intrusion Detection System Using Convolutional Neural Network and Random Forest Classifiers","abstract":"Network Intrusion Detection Systems (NIDS) play a crucial role in protecting networks from various forms of cyberattacks. However, conventional signature-based methods often fail to detect new or unknown threats and are prone to generating high false positive rates. This study proposes a hybrid approach combining Convolutional Neural Network (CNN) and Random Forest (RF) to develop a more adaptive and accurate intrusion detection system. CNN is employed to extract features from raw network traffic data, while RF serves as the primary classifier. The UNSW-NB15 dataset is used for training and testing the model. Evaluation results show that the hybrid model achieves an accuracy of 93.0%, average precision of 94%, recall of 90%, F1-score of 92%, and a false positive rate of 19.2%. These results demonstrate that the CNN–RF hybrid approach effectively improves intrusion detection performance and offers a promising solution for modern network security systems","author":[{"family":"Rismawan","given":"Viky"},{"family":"Pramudya","given":"Elkaf"}],"issued":{"date-parts":[[2025]]},"DOI":"10.35314/rxd38a11","URL":"https://doi.org/10.35314/rxd38a11","source":"crossref"},{"id":"doi:10.28925/2663-4023.2025.27.764","type":"article-journal","title":"INTELLIGENT RISK ASSESSMENT MODELS IN DISTRIBUTED SYSTEMS BASED ON THE NEURAL NETWORK APPROACH","abstract":"In modern conditions of information systems functioning, the rapid growth of scale, complexity and distribution of computing resources is becoming one of the defining trends in the development of digital infrastructure. Within the framework of the widespread implementation of complex multi-component information systems that are distributed in nature and contain a large number of nodes, as well as a significant increase in the number and complexity of cyber threats focused on scalable systems, cybersecurity risks should be considered as a key factor in strategic planning of business processes. Regular analysis and assessment of cybersecurity risks allows determining the necessary and sufficient set of information protection tools, regulatory and organizational mechanisms to reduce information security threats, and ensures the process of building the most effective architecture of a comprehensive information security management system. Existing tools and assessment methodologies, which are mostly conceptual in nature and based on statistical approaches, are ineffective in analysis of large arrays of high-dimensional heterogeneous data and metrics of distributed systems. The article focuses on the current trends and existing approaches to information security risk assessment in distributed information systems. It analyzes the importance of risk management in the process of ensuring information security, and also describes a core principles of intelligent security risk assessment in distributed information systems based on the neural network approach. Research also presents a dynamic and comprehensive model of cyber risk assessment in distributed information systems based on back propagation neural network architecture and several methods of its optimization, which provides sufficient accuracy and reliability of risk assessment in the conditions of analysis of large arrays of heterogeneous input data.","author":[{"family":"Palko","given":"Dmytro"}],"issued":{"date-parts":[[2025]]},"DOI":"10.28925/2663-4023.2025.27.764","URL":"https://doi.org/10.28925/2663-4023.2025.27.764","source":"crossref"},{"id":"doi:10.56952/arma-2025-0234","type":"article-journal","title":"Physics-Informed Neural Network Surrogate Modeling of Pressurized Cavity In Homogeneous and Bilayered Media","abstract":"ABSTRACT: The theory of cavity expansion can explain the mechanical response induced by the expansion or contraction of cylindrical cavities that underpin a wide range of geotechnical applications. While analytical methods often impose restrictive assumptions of homogeneity and isotropy, numerical approaches like the Finite Element Method (FEM) offer greater flexibility but can be computationally prohibitive. To address these limitations, we propose a novel Physics-Informed Neural Network (PINN) framework as a surrogate model for pressurized cavity expansion problems in homogeneous and bilayered isotropic and linear-elastic host media, with arbitrary boundary conditions and constitutive parameters. The proposed model combines known governing equations with observational data, which ensures physical consistency while providing mesh-free, high-fidelity solutions for both homogeneous and bi-layered media. FEM simulations were taken as ground truth and prediction results demonstrate robust predictive accuracy across heterogeneous domains. The proposed PINN significantly accelerates the prediction process, achieving up to 160 times faster computation compared to FEM while maintaining competitive accuracy, though discrepancies in shear stress and boundary errors persist due to the non-convexity of the loss function. These challenges can be mitigated by refining the weights of the loss function terms, improving the network architecture, and using advanced optimization techniques.","author":[{"family":"Liu","given":"Yulong"},{"family":"Arson","given":"Chloé"}],"issued":{"date-parts":[[2025]]},"DOI":"10.56952/arma-2025-0234","URL":"https://doi.org/10.56952/arma-2025-0234","source":"crossref"},{"id":"doi:10.31449/inf.v49i13.7154","type":"article-journal","title":"Attention-Based Bimodal Neural Network Speech Recognition System on FPGA","abstract":"To further improve the accuracy of speech recognition technology, a neural network speech recognition system based on field programmable gate array was designed. Firstly, a neural network audiovisual bimodal speech recognition algorithm based on attention mechanism was designed. Then, a speech recognition platform based on on-site programmable gate arrays was built. The proposed algorithm is proved to have the lowest word error rate and character error rate of 3.17% and 1.56%, with the fastest convergence speed and lower final loss value. The algorithm converges quickly when the raining rounds are less than 10, and tends to stabilize when it is 20. The proposed speech recognition platform uses many DSP units in its design, with a utilization rate of 83.2%, the lowest power consumption of 2.21W, the highest energy efficiency ratio of 26.15, and the shortest processing time and faster running speed. In summary, the research algorithm can reasonably allocate learning weights, improve training speed, and has certain feasibility and effectiveness because of introducing attention mechanism. It has good application effects in speech recognition, which helps to improve the accuracy of language recognition algorithms and promote communication between humans and machines.","author":[{"family":"Chen","given":"Aiwu"}],"issued":{"date-parts":[[2025]]},"DOI":"10.31449/inf.v49i13.7154","URL":"https://doi.org/10.31449/inf.v49i13.7154","source":"crossref"},{"id":"doi:10.26434/chemrxiv-2025-w5npj","type":"manuscript","title":"Structure and Dynamics of CO2 at the Air-Water Interface from Classical and Neural Network Potentials","abstract":"The accurate description of the structure and dynamics of CO2 at the instantaneous air-water interface, along with the effects of surface fluctuations on the CO2-transport processes, is essential for the development of negative emission technologies aimed to mitigate climate change. In this study, we performed molecular dynamics simulations of CO2 at the air-water interface using neural network potentials (NNPs) trained on ab initio data generated through density functional theory-based molecular dynamics simulations. We compared these results with classical force fields to assess their performance in modeling interfacial CO2 behavior. Our findings revealed that the asymmetric interactions, coupled with thermal surface fluctuations at the air-water interface signif- icantly influence CO2 transport into the aqueous phase. The simulations demonstrate that classical force fields underestimate both the free energy of CO2 transport and the strength of its interactions at the interface compared to the neural network potentials. The free energy and the interfacial dynamics of CO2 are primarily influenced by the distribution of water within the instantaneous interfacial water layer, responsible for creating asymmetric intermolecular interaction environment within the interfacial region.","author":[{"family":"Kumar","given":"Nitesh"},{"family":"Bryantsev","given":"Vyacheslav"}],"issued":{"date-parts":[[2025]]},"DOI":"10.26434/chemrxiv-2025-w5npj","URL":"https://doi.org/10.26434/chemrxiv-2025-w5npj","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-7254889/v1","type":"article-journal","title":"Spectral Neural Network Compression via Discrete Fourier Transform: A Post Hoc and Lightweight Approach","abstract":"Abstract We introduce a spectral post hoc compression method for neural networks based on Discrete Fourier Transform (DFT) of complex weights. The approach filters low-magnitude frequencies to obtain sparse spectral representations while preserving accuracy. Theoretical results quantify energy preservation and output perturbation. We propose a principled thresholding rule, and demonstrate competitive performance compared to DCT and wavelets. Experiments on MNIST, CIFAR-10 and ResNet show 10–15× compression with negligible loss. Hardware metrics confirm reduced memory usage and improved inference latency. The method is lightweight, requires no retraining , and suits embedded AI.","author":[{"family":"Samia","given":"Sghaier"},{"family":"Houda","given":"Nfata"}],"issued":{"date-parts":[[2025]]},"DOI":"10.21203/rs.3.rs-7254889/v1","URL":"https://doi.org/10.21203/rs.3.rs-7254889/v1","source":"crossref"},{"id":"doi:10.59671/rt89a","type":"article-journal","title":"Assessment of Improved Artificial Neural Network Models for Urban Air Quality Forecasting by Transboundary Pollutants","abstract":"The assessment of artificial neural network models-sigmoid (ANN-sigmoid) and hyperbolic tangent (ANN-tanh) for real-time air pollution forecasting in a Korean coastal city was performed using 15 input independent variables (3 hours' earlier PM, gas and meteorological data influenced by 48 hours' earlier PM and gas data of a Chinese city). A feed-forward ANN technique of multilayer perception (MLP) with back-propagation training algorithm for error calculation was adopted with 15 hidden nodes and each prediction formula on four output variables was suggested. Root mean square error (RMSE) and the coefficient of determination (R2 ) assess each model's prediction ability between the predicted and measured values, before, during, and after the Yellow Dust period (March 18~27). Pearson R coefficients from the ANN-sigmoid (or ANN-tanh) model on PM10, PM2.5, NO2, and O3 were 0.907 (0.935), 0.860 (0.942), 0.860 (0.925), and 0.957 (0.946) before the dust period, 0.917 (0.943), 0.959 (0.969), 0.855 (0.853), and 0.954 (0.949) during its period, and 0.920 (0.947), 0.928 (0.938), 0.917 (0.896), and 0.923 (0.952) after its period, showing very high prediction accuracy overall. Scatter plots with empirical formulae and temporal distributions between the predicted and measured values showed excellent prediction performance by two models, and the ANN-tanh model produced more accurate results.","author":[{"family":"Choi","given":"Soo"}],"issued":{"date-parts":[[2025]]},"DOI":"10.59671/rt89a","URL":"https://doi.org/10.59671/rt89a","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-6904582/v1","type":"article-journal","title":"Tourism Ecological Efficiency Assessment Based on Multi-Source Data Fusion and Graph Neural Network","abstract":"Abstract Currently, research on the evaluation of tourism ecological efficiency based on multi-source data fusion and graph neural networks has significant limitations: at the data level, the integration of multi-source data faces challenges related to format, quality, and semantic differences; the complexity of cleaning and preprocessing may lead to information loss, and the limitations of a single data source are prominent, making it difficult to comprehensively cover the complex features of the tourism ecosystem; at the model level, traditional evaluation models cannot accurately identify ineffective sources, the handling of expected vs. unexpected outputs is not sufficiently scientific, and adapting to the dynamic demands of tourism ecological efficiency evaluation is challenging, which restricts the accuracy and application value of the evaluation results. This paper proposes a tourism eco-efficiency assessment method based on multi-source data fusion and graph neural networks. First, a comprehensive dataset is constructed by integrating multi-source information such as tourism statistics, environmental monitoring data, and socio-economic data. Then, the GNN model is used to mine the intrinsic connections and patterns in the data to more accurately evaluate the impact of tourism activities on the ecological environment. In addition, the distribution characteristics of tourism eco-efficiency in different periods and geographical regions were analyzed by considering spatial and temporal factors. Through case studies of typical tourism destinations, the effectiveness of the proposed method is verified, and its application value in practical tourism management and planning is discussed. Regression analysis was used to estimate tourism eco-efficiency based on a single data source. The data from 2015 to 2020 are selected to calculate the correlation coefficient between the growth rate of tourism revenue and environmental quality indicators. Based on the regression analysis of a single data source, the resulting tourism eco-efficiency score was 72 points in 2020. Using multi-source data fusion and graph neural network, the tourism eco-efficiency score was 85 points in the same year, 13 points higher than the traditional method. This study not only provides a new method for tourism eco-efficiency assessment but also helps to deepen our understanding of the complexity of tourism ecosystems and provides scientific support for the sustainable development of tourism destinations.","author":[{"family":"Lin","given":"Luoyanzi"},{"family":"Lv","given":"Jiehua"}],"issued":{"date-parts":[[2025]]},"DOI":"10.21203/rs.3.rs-6904582/v1","URL":"https://doi.org/10.21203/rs.3.rs-6904582/v1","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-6461595/v1","type":"article-journal","title":"Prediction and Attribution Analysis of Surface Upward Longwave Radiation Based on a Hybrid Neural Network Model","abstract":"Abstract Accurate prediction of surface upward longwave radiation (SULR) is crucial for understanding Earth’s energy balance and climate dynamics. Traditional approaches, such as physical models and empirical regressions, often fail to handle the complexity and variability of environmental data. To overcome these challenges, this study introduces a hybrid neural network model integrating several advanced techniques: the Alpha Evolution (AE) optimizer, a Transformer-LSTM neural network, Adaptive Bandwidth Kernel Density Estimation (ABKDE), and SHAP-based interpretability analysis. The AE optimizer fine-tunes model parameters to enhance convergence and efficiency. The Transformer-LSTM architecture uses self-attention and long short-term memory to capture complex temporal patterns in the data. ABKDE delivers reliable interval predictions for SULR, while SHAP uncovers feature importance and the rationale behind model decisions. Using datasets from two stations in the Taihu region, the hybrid model was compared with LSTM, GRU, Transformer, and theoretical models. Results indicate that the hybrid model notably outperforms traditional methods, as evidenced by improvements in R², RMSE, and MAE. Furthermore, ABKDE shows high accuracy in interval predictions, while SHAP analysis identifies water temperature, air temperature, and downward longwave radiation as the most influential factors affecting SULR. By offering a robust, interpretable SULR prediction model, this study advances Earth’s energy balance research and presents promising applications in climate modeling and environmental monitoring.","author":[{"family":"Liu","given":"Kun"}],"issued":{"date-parts":[[2025]]},"DOI":"10.21203/rs.3.rs-6461595/v1","URL":"https://doi.org/10.21203/rs.3.rs-6461595/v1","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-6654555/v1","type":"article-journal","title":"Research on the Performance Evaluation of Agricultural Product Distribution Supply Chain Based on BP Neural Network","abstract":"Abstract Supply chain management is in a period of rapid development, and future competition in the distribution industry will be centered around competition between supply chains. Agricultural product distribution, as an important part of modern distribution, plays a positive role in ensuring people's daily lives and improving the modernization level of agricultural distribution. Therefore, clarifying the performance of the supply chain is the key to enhancing its development level and market competitiveness. Given that most current supply chain performance evaluation studies focus on the industrial sector and there are limited methods for evaluating the performance of agricultural product distribution supply chains, this paper attempts to analyze the performance evaluation system of agricultural product supply chains centered on supermarkets. In terms of research methodology, it is based on the BP neural network and considers five aspects of the agricultural product distribution supply chain: financial situation, operational capability, growth capability, customer satisfaction, and agility. A corresponding evaluation index system is established. Finally, Yonghui Supermarket and its suppliers are used to apply this performance evaluation system in practice, further identifying beneficial ways to improve the performance level of agricultural product distribution supply chains.","author":[{"family":"Zheng","given":"Lu"}],"issued":{"date-parts":[[2025]]},"DOI":"10.21203/rs.3.rs-6654555/v1","URL":"https://doi.org/10.21203/rs.3.rs-6654555/v1","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-7246244/v1","type":"article-journal","title":"Leveraging Quantum Superposition to Infer the Dynamic Behavior of a Spatial-Temporal Neural Network Signaling Model","abstract":"Abstract The exploration of new problem classes for quantum computation is an active area of research. In this paper, we introduce and solve a novel problem class related to dynamics on large-scale networks relevant to neurobiology and machine learning. Specifically, we ask if a network can sustain inherent dynamic activity beyond some arbitrary observation time or if the activity ceases through quiescence or saturation via an ’epileptic’-like state. We show that this class of problems can be formulated and structured to take advantage of quantum superpo- sition and solved efficiently using a coupled workflow between the Grover and Deutsch–Jozsa quantum algorithms. To do so, we extend their functionality to address the unique requirements of how input (sub)sets into the algorithms must be mathematically structured while simulta- neously constructing the inputs so that measurement outputs can be interpreted as meaningful properties of the network dynamics. This, in turn, allows us to answer the question we pose.","author":[{"family":"Silva","given":"Gabriel"}],"issued":{"date-parts":[[2025]]},"DOI":"10.21203/rs.3.rs-7246244/v1","URL":"https://doi.org/10.21203/rs.3.rs-7246244/v1","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-7376845/v1","type":"article-journal","title":"Geometric Fault-Tolerant Neural Network Tracking Control of Unknown Systems on Matrix Lie Groups","abstract":"Abstract We present a geometric neural network-based tracking controller for systems evolving on matrix Lie groups under unknown dynamics, actuator faults, and bounded disturbances. Leveraging the left-invariance of the tangent bundle of matrix Lie groups, viewed as an embedded submanifold of the vector space RN×N , we propose a set of learning rules for neural network weights that are intrinsically compatible with the Lie group structure and do not require explicit parameterization. Exploiting the geometric properties of Lie groups, this approach circumvents parameterization singularities and enables a global search for optimal weights. The ultimate boundedness of all error signals—including the neural network weights, the coordinate-free configuration error function, and the tracking velocity error—is established using Lyapunov’s direct method. To validate the effectiveness of the proposed method, we choose to present illustrative simulation results for decentralized formation control of multi-agent systems on the Special Euclidean group, where accurate control with fast adaptation is critical.","author":[{"family":"Chhabra","given":"Robin"},{"family":"Abdollahi","given":"Farzaneh"}],"issued":{"date-parts":[[2025]]},"DOI":"10.21203/rs.3.rs-7376845/v1","URL":"https://doi.org/10.21203/rs.3.rs-7376845/v1","source":"crossref"},{"id":"doi:10.20944/preprints202510.0488.v1","type":"manuscript","title":"Matignon-Based Stability and Weight Synchronization of a Fractional Time Delay Neural Network Model","abstract":"Artificial neural networks (ANNs) are powerful models inspired by the structure and function of the human brain. They are widely used for tasks such as classification, prediction, and model recognition. This study examines the stability of fractional-order neural networks with neuronal conditions, dynamic behavior, synchronization, and delays of time. Synchronization and stability for delayed neural network models are two important aspects of dynamic behavior. For a calculated fractionalorder, the state of the state variable wi(t) are synchronized with each other. Weight synchronization of wi (i = 1, 2, 3, . . . ,6) provides coherent updates during training, helping neural networks to study stable models. The incommensurate fractional-orders are linked to a system where each dynamic component develops with a different value, i.e. qi ≠ qj (i ≠ j) is inconsistent. These fractionalorders are calculated for the system’s eigenvalues and their singular points within the stability region defined by the Matignon-based stability. As the time delay decreases, more activation functions are induced, and the variable state of w3(t) requires longer relaxation times to be more stable than the variable state of w4(t). The Grunwald-Letnikov method is used to solve a fractional neural network system numerically and effectively handle fractional derivatives. This approach helps to more accurately simulate memory in neural networks.","author":[{"family":"Ullah","given":"Asif"},{"family":"Shuaib","given":"Muhammad"}],"issued":{"date-parts":[[2025]]},"DOI":"10.20944/preprints202510.0488.v1","URL":"https://doi.org/10.20944/preprints202510.0488.v1","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-6504224/v1","type":"article-journal","title":"Adaptive Multi-Agent Graph Neural Network (AMAGNN) for Congestion Control in VANET","abstract":"Abstract Similar approaches are very useful in Vehicular Ad Hoc Networks in which: a flood of traffic and events occurs, especially in Intelligent Transportation Systems (ITS), as congestion control can have a huge impact in efficient traffic field and communication. The existing congestion control mechanisms are not suitable for this type of dynamic network topology and vehicle mobility. In response to these challenges, we propose an Adaptive Multi-Agent Graph Neural Network (AMAGNN), which utilizes graph neural networks (GNNs) and multi-agent reinforcement learning (MARL) for congestion control in vehicular ad hoc networks (VANETs). Institute, “AMAGNN models VANETs as dynamic graphs, with vehicles being intelligent agents that cooperatively learn to optimize data propagation and alleviate network congestion. Optimizes for scalability and adaptability by dynamically responding to network topology changes and traffic density variation. Results show that the proposed AMAGNN model outperforms current practices in congestion control for VANETs on all the key performance metrics. It attains Packet Delivery Ratio (92.5%), lowest delay (28.4 ms), and maximum throughput (275.6 kbps) which shows reliability and efficiency in communication. It also achieves the lowest congestion level (0.35) and average waiting time (0.31 s), highlighting its ability to reduce traffic and delay. AMAGNN achieves the best performance, compared with methods such as GCN-RL, DQN Adaptive, and MA-PPO, highlighting the effectiveness of adaptive multi-agent graph neural networks in the ever-changing vehicular environment.","author":[{"family":"Maharana","given":"Santosh"},{"family":"Patra","given":"Prashanta"}],"issued":{"date-parts":[[2025]]},"DOI":"10.21203/rs.3.rs-6504224/v1","URL":"https://doi.org/10.21203/rs.3.rs-6504224/v1","source":"crossref"},{"id":"doi:10.20944/preprints202503.2355.v1","type":"manuscript","title":"Enhancing Neural Network Interpretability Through Deep Prior-Guided Expected Gradients","abstract":"The increasing adoption of deep neural networks (DNNs) in critical domains such as healthcare, finance, and autonomous systems underscores the growing importance of explainable artificial intelligence (XAI). In these high-stakes applications, understanding the decision-making processes of models is essential for ensuring trust and safety. However, traditional DNNs often function as \"black boxes,\" delivering accurate predictions without providing insight into the factors driving their outputs. Expected Gradients (EG) is a prominent method for making such explanations by calculating the contribution of each input feature to the final decision. Despite its effectiveness, conventional baselines used in state-of-the-art implementations of EG often lack a clear definition of what constitutes \"missing\" information. In this work, we propose DeepPrior-EG, a deep prior-guided EG framework for leveraging prior knowledge to more accurately align with the concept of missingness and enhance interpretive fidelity. It resolves the baseline misalignment by initiating gradient path integration from learned prior baselines, which derived from the deep features of CNN layers. This approach not only mitigates feature absence artifacts but also amplifies critical feature contributions through adaptive gradient aggregation. We further introduce two probabilistic prior modeling strategies: a multivariate Gaussian model (MGM) to capture high-dimensional feature interdependencies and a Bayesian nonparametric Gaussian mixture model (BGMM) that autonomously infers mixture complexity for heterogeneous feature distributions. We also develop an explanation-driven model retraining paradigm to valid the robustness of the proposed framework. Comprehensive evaluations across various qualitative and quantitative metrics demonstrate the its superior interpretability. The BGMM variant achieves state-of-the-art performance in attribution quality and faithfulness against existing methods. DeepPrior-EG advances the interpretability of complex models within the XAI landscape and unlocks its potential in safety-critical applications.","author":[{"family":"Guo","given":"Su"},{"family":"Gong","given":"Xiu"}],"issued":{"date-parts":[[2025]]},"DOI":"10.20944/preprints202503.2355.v1","URL":"https://doi.org/10.20944/preprints202503.2355.v1","source":"crossref"},{"id":"doi:10.7868/s3034548025040058","type":"article-journal","title":"Training of a Spiking Neural Network with Consideration of Memristive Crossbar Array Characteristics","abstract":"A model, methodology and software tools for simulating a spiking neural network in the training mode considering the features of the operation of memristive crossbar arrays were developed. The impact of voltage drops on interconnects, the discrete step of adjusting the conductivity levels of memristive elements and the nonlinearity of their volt-ampere characteristics on the effectiveness of implementing spiking neural network training algorithms was studied. Results of testing the spiking neural network in the training mode and inference mode in the task of image recognition using the developed simulation method, taking into account the characteristics of experimentally fabricated memristive structures, are presented.","author":[{"family":"Dudkin","given":"AP"}],"issued":{"date-parts":[[2025]]},"DOI":"10.7868/s3034548025040058","URL":"https://doi.org/10.7868/s3034548025040058","source":"crossref"},{"id":"doi:10.1115/1.4068110","type":"article-journal","title":"Abrasive Wear Prediction of Three-Dimensional Printed PEEK Using Artificial Neural Network","abstract":"Abstract Machine learning is a cutting-edge technology that stands out among the various artificial intelligence offerings with its exceptional ability to comprehend intricate processes in computational tools. The optimization of input parameters for polyetheretherketone (PEEK) to print samples of any geometry provides insight for fabricated samples. The samples were fabricated using the extrusion method of additive manufacturing with varying layer thickness, which was tested under abrasive wear conditions using 120-grade sandpaper. The surface properties affect the wear response and rate of the 3D-printed PEEK sample, and the wear loss is high when subjected to abrasive wear conditions. Increasing layer thickness (more than 0.2 mm) reduces the hardness, and a rougher surface causes higher wear loss. To quantify the wear loss and avoid any mishappening during the operation, when the 3D-printed PEEK part is used as a journal bearing under unfavorable conditions, the mechanical and tribological properties emerge as significant measures defining a material's worth. These properties measure a material's capacity to endure external forces and the nonuniformity of relative motion. The article aims to predict wear loss using artificial neural networks (ANNs) under such conditions to avoid system failure and timely replacement with new components. The ReLU (rectified linear unit) activation function fits the actual wear trend and predicts the wear loss with 98% accuracy.","author":[{"family":"Prajapati","given":"Sunil"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1115/1.4068110","URL":"https://doi.org/10.1115/1.4068110","source":"crossref"},{"id":"doi:10.47852/bonview52022956","type":"article-journal","title":"Medicinal Plant Recognition Using Shallow Convolutional Neural Network","abstract":"Ayurvedic medicine plays an essential role in the overall care that is provided for the physical and mental wellbeing of people. It is vital to correctly identify and categorize medicinal herbs to be able to provide better therapy. Medicinal herbs come in a wide variety of forms. It is a challenging task that requires a significant amount of professional medical experience to correctly name and classify the many distinct kinds of medicinal plants. Because of this, having an approach to the identification of medicinal plants that is completely automated is something that is highly desirable. In this study, a straightforward four-layer shallow Convolutional Neural Network (S-CNN) is proposed for the aim of classifying medicinal herbs. The potential utility of S-CNN is evaluated with the help of four distinct leaf datasets such as the Swedish Leaf, Flavia Leaf, MepcoTropicalLeaf Dataset, and Medicinal Leaf Dataset. Our model is capable of achieving a level of classification accuracy of 98.22%, 96.18% and 92.89% on Swedish, Flavia and Medicinal Leaf datasets respectively and that is comparable to that of other state-of-art methodologies in this field. Received: 28 March 2024 | Revised: 13 March 2025 | Accepted: 6 June 2025 Conflicts of Interest The authors declare that they have no conflicts of interest to this work. Data Availability Statement The data that support the findings of this study are openly available in Swedish Leaf Dataset at https://www.cvl.isy.liu.se/en/research/datasets/swedish-leaf/, reference number [47]; in Medicinal Leaf Dataset at https://data.mendeley.com/datasets/nnytj2v3n5/1, reference number [48]. Author Contribution Statement Ramar Ahila Priyadharshini: Conceptualization, Methodology, Software, Validation, Investigation, Data curation, Writing - original draft, Visualization, Supervision, Project administration. M. Arun: Software, Formal analysis, Data curation, Writing - review &amp; editing, Visualization.","author":[{"family":"Priyadharshini","given":"Ramar"},{"family":"Arun","given":"M"}],"issued":{"date-parts":[[2025]]},"DOI":"10.47852/bonview52022956","URL":"https://doi.org/10.47852/bonview52022956","source":"crossref"},{"id":"doi:10.65286/icic.v20i1.74624","type":"article-journal","title":"Spatial-Temporal Attention Simple Graph Neural Network","abstract":"The growth of the autonomous driving industry in recent years has spurred research on intelligent transportation systems. However, predicting long-term traffic patterns is a complex task that can lead to overfitting and fluctuations in model predictions.To address these challenges, this paper proposes a spatio-temporal modeling approach that captures both the spatial and temporal features of traffic data. The method fuses these features using a gated fusion mechanism and then applies feedforward neural networks to transform the spatio-temporal data into predictions for future time steps.To mitigate overfitting, the paper introduces a novel loss function called the mean loss function. By minimizing fluctuations in model predictions, this approach aims to improve the accuracy of long-term traffic forecasts.Overall, this paper presents a promising approach to improving the performance of intelligent transportation systems, particularly in the area of long-term traffic prediction. The proposed method combines several techniques, including spatio-temporal modeling, neural networks, and a new loss function,to address the challenges of overfitting and prediction fluctuations.After conducting multiple experiments on the publicly available transportation network datasets, METR-LA and PEMS-Bay, our proposed model demonstrated improved performance in long-term traffic flow prediction","author":[{"family":"Ai","given":"Jiaxin"}],"issued":{"date-parts":[[2025]]},"DOI":"10.65286/icic.v20i1.74624","URL":"https://doi.org/10.65286/icic.v20i1.74624","source":"crossref"},{"id":"doi:10.1115/imece2024-145599","type":"article-journal","title":"A Deep Convolutional Neural Network Approach to Automate Drilling Tool Lateral Motion Video Interpretation","abstract":"Abstract Mitigating the risks associated with the effect of whirl on measurement quality and health of the drilling tools requires a solid understanding of drilling tool lateral motion dynamics. A full-scale roll test system (i.e., a tubular assembly rolling along the inner wall of an impact ring while being spun with a drive motor), which intends to mimic lateral dynamics that the tool could undergo while drilling in a controlled laboratory environment, is believed to be the closest to the real drilling exercises. This study is concerned with leveraging artificial intelligence (AI) techniques to advance the results interpretation of such roll test system that is instrumented with accelerometers and high-speed cameras. Specifically, deep convolutional neural network models using transfer learning are developed to automate detection of rolling shock impacts based on the digital images collected from the recorded slow-motion video. The developed model is trained and validated with the video data recorded at one rotational speed. The robustness of the developed model is finally tested and evaluated on the unseen video dataset, which describe drilling tool whirl dynamics at different rotational speeds. High F1 Score about 0.9 is achieved for the selected model with hyperparameter-tuning and fine-tuning techniques. The developed AI model expects to play a vital role in the modernization of the roll test system development for downhole drilling tools, and reducing cost, time and risks in new product development and validation and verification.","author":[{"family":"Song","given":"Fei"},{"family":"Li","given":"Ke"}],"issued":{"date-parts":[[2025]]},"DOI":"10.1115/imece2024-145599","URL":"https://doi.org/10.1115/imece2024-145599","source":"crossref"},{"id":"doi:10.14311/nnw.2024.34.017","type":"article-journal","title":"Post-Pandemic Road Accident Analysis: Patterns and Impacts","abstract":"This study analyzes the impact of COVID-19 restrictions on road accident trends in the Czech Republic from 2015 to 2024, utilizing a comprehensive dataset of over one million recorded accidents. The research highlights a significant decline in accident rates during the strict lockdown periods, correlating reduced mobility with fewer traffic incidents. As restrictions eased, accident rates rose again, revealing seasonal variations and regional disparities, particularly in urban areas like Prague. Findings suggest that the pandemic has reshaped commuting patterns and could influence future traffic management strategies. The analysis underscores the need for targeted policies to enhance road safety, especially during high-risk seasons.","author":[{"family":"Gavrilov","given":"Igor"}],"issued":{"date-parts":[[2025]]},"DOI":"10.14311/nnw.2024.34.017","URL":"https://doi.org/10.14311/nnw.2024.34.017","source":"crossref"},{"id":"doi:10.70675/51cf5234z10f7z42b2z8d46z13894170c780","type":"article-journal","title":"Low-rank network models of neural computations","abstract":"Réseaux de neurones de bas rang et calculs neuronaux À tout instant, des myriades de neurones coopèrent au sein d’un système nerveux, produisant des motifs d’activité collectifs qui forment un substrat biologique pour la perception, la cognition, et le comportement. Les enregistrements in vivo de centaines voire milliers de neurones chez l’animal suggèrent que ces motifs d’activité s’organisent souvent selon des géométries particulières, dites à basse dimensionnalité, à partir desquelles les représentations mentales peuvent être extraites. Ainsi, un paradigme actuel influent en neurosciences postule que les fonctions cognitives émergent à partir des dynamiques de réseaux de neurones dont l’activité forme des motifs de basse dimensionnalité. Une question essentielle qui demeure est de comprendre comment la structure d’un réseau neuronal génère ces dynamiques particulières de l’activité collective liées à sa fonction. Une voie prometteuse pour éclairer cette question consiste à étudier des réseaux récurrents dont les connexions sont organisées dans une matrice de bas-rang. Il a été montré précédemment que cette propriété mathématique du réseau y induit naturellement des dynamiques de basse dimensionnalité qui encodent des représentations utiles, telles qu’observées dans les expériences sur réseaux biologiques et artificiels. De tels réseaux de bas-rang peuvent-ils alors nous permettre d’ouvrir la « boîte noire » des réseaux récurrents, et de comprendre comment leur fonctionnalité émerge de leur structure ? Et peuvent-ils s’appliquer à des enregistrements de neurones biologiques, afin d’en extraire des hypothèses sur la structure sous-jacente des réseaux observés ? Cette thèse vise à répondre à ces questions, à travers de nouvelles méthodes pour entraîner les réseaux de bas-rang, ainsi qu’à travers une nouvelle théorie qui relie une description statistique de ces réseaux à leurs dynamiques et leur fonction. Nous allons dans un premier temps exposer notre théorie et nos méthodes, expliquant comment les matrices de bas-rang nous permettent de relier structure et fonction dans un réseau récurrent. Nous allons ensuite étudier des réseaux entraînés à réaliser de nombreuses tâches cognitives, construisant une compréhension systémique de leur fonctionnement. Ces réseaux de bas-rang seront ensuite reliés à une classe de méthodes statistiques communes dans l’interprétation d’enregistrements neuronaux, les systèmes dynamiques latents. Pour terminer, nous démontrerons les capacités des réseaux de bas-rang pour disséquer le fonctionnement de réseaux artificiels récurrents non-contraints, ainsi que pour interpréter des enregistrements corticaux in vivo.","author":[{"family":"Valente","given":"Adrian"}],"issued":{"date-parts":[[2026]]},"DOI":"10.70675/51cf5234z10f7z42b2z8d46z13894170c780","URL":"https://doi.org/10.70675/51cf5234z10f7z42b2z8d46z13894170c780","source":"crossref"},{"id":"doi:10.36227/techrxiv.176826845.56179350/v1","type":"article-journal","title":"NEUROEVOLUTION: A NEURAL NETWORK-ORIENTED OPTIMIZATION APPROACH","abstract":"NeuroEvolution refers to the use of evolutionary algorithms to optimize neural network parameters and, in some cases, network architectures. The provided document outlines the fundamental components of evolutionary computation when applied to neural networks, including chromosome representation, population initialization, fitness evaluation, selection, crossover, mutation, replacement, and termination criteria. These processes collectively enable a population of neural-network-based solutions to evolve over time toward better performance on a given task. Additionally, the document highlights the role of population diversity and discusses how mechanisms such as elitism and diversity monitoring help avoid premature convergence. This work summarizes these concepts and bases a methodological framework on the principles presented.","author":[{"family":"Gupta","given":"Kishu"}],"issued":{"date-parts":[[2026]]},"DOI":"10.36227/techrxiv.176826845.56179350/v1","URL":"https://doi.org/10.36227/techrxiv.176826845.56179350/v1","source":"crossref"},{"id":"doi:10.52783/jes.7946","type":"article-journal","title":"Neural Network-Based Traffic Control System","abstract":"Effective traffic management is essential for modern cities, as it helps to alleviate congestion and enhance mobility. This research proposes a novel approach to predicting traffic flow, utilizing a combination of advanced deep learning techniques. The model integrates spatial feature extraction, temporal dependency modeling, and attention mechanisms to provide accurate and interpretable predictions. By leveraging the strengths of different deep learning architectures, the proposed framework achieves significant improvements in prediction accuracy and efficiency compared to traditional models. The results of this study demonstrate the potential of hybrid deep learning models to address the complexities of modern traffic systems and highlight the importance of continued research in this area. Furthermore, the proposed framework's ability to provide real-time predictions makes it a valuable tool for intelligent traffic management systems, enabling cities to optimize their traffic flow and reduce congestion.","author":[{"family":"Sonawane","given":"Gauri"}],"issued":{"date-parts":[[2025]]},"DOI":"10.52783/jes.7946","URL":"https://doi.org/10.52783/jes.7946","source":"crossref"},{"id":"doi:10.26907/1562-5419-2024-27-4-598-655","type":"article-journal","title":"Neural Network Architecture of Embodied Intelligence","abstract":"In recent years, advances in artificial intelligence (AI) and machine learning have been driven by advances in the development of large language models (LLMs) based on deep neural networks. At the same time, despite its substantial capabilities, LLMs have fundamental limitations such as spontaneous unreliability in facts and judgments; making simple errors that are dissonant with high competence in general; credulity, manifested by a willingness to accept a user's knowingly false claims as true; and lack of knowledge about events that have occurred after training has been completed. Probably the key reason is that bioinspired intelligence learning occurs through the assimilation of implicit knowledge by an embodied form of intelligence to solve interactive real-world physical problems. Bioinspired studies of the nervous systems of organisms suggest that the cerebellum, which coordinates movement and maintains balance, is a prime candidate for uncovering methods for realizing embodied physical intelligence. Its simple repetitive structure and ability to control complex movements offer hope for the possibility of creating an analog to adaptive neural networks. This paper explores the bioinspired architecture of the cerebellum as a form of analog computational networks capable of modeling complex real-world physical systems. As a simple example, a realization of embodied AI in the form of a multi-component model of an octopus tentacle is presented, demonstrating the potential in creating adaptive physical systems that learn and interact with the environment.","author":[{"family":"Nurutdinov","given":"Ayrat"}],"issued":{"date-parts":[[2025]]},"DOI":"10.26907/1562-5419-2024-27-4-598-655","URL":"https://doi.org/10.26907/1562-5419-2024-27-4-598-655","source":"crossref"},{"id":"doi:10.64336/001c.94086","type":"article-journal","title":"Optimization of Convolutional Neural Network hyperparameters using Genetic Algorithms","abstract":"The ever-increasing complexity of deep learning models leads to larger sizes and computational costs (greater execution time), making models extremely difficult to implement in resource constrained environments. Moreover, extensive hyperparameter tuning is necessary to create successful models. One approach to this problem is pruning, which removes redundant connections and neurons within the model. However, no approach fundamentally changes the structure of the model (layers, width, depth, filters, etc.). This would inherently bypass the necessity for tediously removing connections within the model and rather remove unnecessary layers and filters (kernels) which may not substantially contribute to performance. The proposed research introduces a approach to reduce inference times of Convolutional Neural Networks (CNNs) by dynamically changing the number of convolutional layers and the kernel parameters within those layers using Genetic Algorithms (GAs). By parameterizing architecture and layer configurations, this approach found the highest efficiency while maintaining a competitive accuracy. Increasing efficiency means decreasing the number of parameters within a model, thus reducing resource consumption during use and reducing inference time. In this paper, a dynamically trained model was tested on three datasets, MNIST, CIFAR-10, and CIFAR 100. Utilizing concepts from GA’s such as selection, crossover, and mutation, each model was iteratively trained and tested on each dataset. The resulting model maintained an accuracy at a given threshold while demonstrating an increase in efficiency by changing the number of convolutional layers and the corresponding filter parameters.","author":[{"family":"Shah","given":"Nirmit"}],"issued":{"date-parts":[[2025]]},"DOI":"10.64336/001c.94086","URL":"https://doi.org/10.64336/001c.94086","source":"crossref"},{"id":"doi:10.59879/huwf6","type":"article-journal","title":"Deep Neural Network Assisted Monte Carlo Tree Search Algorithm to Solve Bandwidth Slicing Placement Problem","abstract":"To solve the network slicing placement problem, the methods based on CNN/RNN were inadequate in handling the randomness of fluctuating channel quality and bandwidth needs for each network slice. While the Monte Carlo Tree Search (MCTS) methodology effectively deals with the unpredictability of each slice’s channel quality and bandwidth request to optimize throughput, it remains time-consuming in finding an optimal solution. The cause is that MCTS relies on a uniform distribution to randomly sample one possible solution, which leads to subpar sampling efficiency. Our objective is to integrate a deep neural network (DNN) to assist MCTS. Specifically, the DNN first analyses the current allocation situation to predict probability distributions for achieving optimizing throughput. MCTS then leverages this DNN-produced probability distribution to pinpoint the best allocation scenario. Experimental results indicate that the performance of DNN-based MCTS with only 50 search iterations surpasses that of the original MCTS with 4,000 search iterations.","author":[{"family":"Chen","given":"Liang"}],"issued":{"date-parts":[[2025]]},"DOI":"10.59879/huwf6","URL":"https://doi.org/10.59879/huwf6","source":"crossref"},{"id":"doi:10.3390/sym17071129","type":"article-journal","title":"A Neural Network Training Method Based on Distributed PID Control","abstract":"In the previous article, we introduced a neural network framework based on symmetric differential equations. This novel framework exhibits complete symmetry, endowing it with perfect mathematical properties. While we have examined some of the system’s mathematical characteristics, a detailed discussion of the network training methodology has not yet been presented. Drawing on the principles of the traditional backpropagation algorithm, this study proposes an alternative training approach that utilizes differential equation signal propagation instead of chain rule derivation. This approach not only preserves the effectiveness of training but also offers enhanced biological interpretability. The foundation of this methodology lies in the system’s reversibility, which stems from its inherent symmetry—a key aspect of our research. However, this method alone is insufficient for effective neural network training. To address this, we further introduce a distributed Proportional–Integral–Derivative (PID) control approach, emphasizing its implementation within a closed system. By incorporating this method, we achieved both faster training speeds and improved accuracy. This approach not only offers novel insights into neural network training but also extends the scope of research into control methodologies. To validate its effectiveness, we apply this method to the MNIST (Modified National Institute of Standards and Technology database) and Fashion-MNIST, demonstrating its practical utility.","author":[{"family":"Jiang","given":"Kun"}],"issued":{"date-parts":[[2025]]},"DOI":"10.3390/sym17071129","URL":"https://doi.org/10.3390/sym17071129","source":"crossref"},{"id":"doi:10.20944/preprints202507.2651.v1","type":"manuscript","title":"Neural Network-Based Modeling for Precise Potato Yield Prediction Using Soil Parameters","abstract":"This study analyses the potential of artificial neural networks (ANN) in accurately predicting potato yields based on 11 parameters characterising the soil environment. Accurate yield forecasting is crucial for optimising potato production, especially in the context of potato processing. Due to the significant impact of soil properties on yield, there is a need for comprehensive predictive models that take these factors into account. The field studies (2021-2024) included an analysis of soil parameters determining potato tuber yield. The developed ANN model was highly accurate, as evidenced by the following indicators: R² = 0.8227, RMSE = 4.19 t∙ha⁻¹, MAE = 3.35 t∙ha⁻¹, MAPE = 7.34%. Global sensitivity analysis showed that cation exchange capacity (CEC), base saturation percentage (V) and sum of exchangeable bases (S) are key parameters influencing tuber yield. The results indicate that neural networks are effective in modelling complex relationships between soil parameters and potato yield, and that soil properties play a fundamental role in increasing yields and improving potato quality. The approach used may contribute to optimising the nutrient content of potato tubers intended for French fry production. Future studies should include climate data and micronutrients to further improve predictive models.","author":[{"family":"Piekutowska","given":"Magdalena"},{"family":"Niedbała","given":"Gniewko"}],"issued":{"date-parts":[[2025]]},"DOI":"10.20944/preprints202507.2651.v1","URL":"https://doi.org/10.20944/preprints202507.2651.v1","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-6478988/v1","type":"article-journal","title":"Enhanced Convolutional Neural Network for Robust Facial Expression Recognition on Fer2013 and Natural Image Datasets","abstract":"Abstract This study presents an enhanced convolutional neural network (CNN) architecture tailored for accurate facial expression recognition. The model is trained on the FER2013 dataset and evaluated using both FER2013 and a custom dataset containing natural facial expressions. By incorporating multiple convolutional and pooling layers along with dropout regularization, the network effectively extracts and classifies emotion-related features. Experimental results demonstrate high recognition accuracy and strong generalization across controlled and real-world image scenarios. In order to study the application of convolutional neural networks in facial expression recognition, a 10-layer convolutional neural network model is designed to recognize facial expressions. The last layer uses the Softmax function to output the classification results of expressions. First, the convolution and pooling algorithms of convolutional neural networks were studied and the structure of the model was designed. Secondly, in order to more vividly display the features extracted by the convolutional layer, the extracted features are visualized and displayed in the form of feature maps. The convolutional neural network model in this work was tested on the Fer-2013 data set, and the experimental results demonstrated the superiority of the recognition rate. It is known that the Fer-2013 dataset contains data collected in an experimental environment, and in order to verify the generalization ability of model recognition, a self-made facial expression data set in natural state was created, and performed a series of preprocessing on the face images such as cropping, grayscale and pixel adjustment. The trained model, which was previously applied to the Fer-2013 dataset, was tested out on the new dataset. The experiment yielded promising results, one of which in the form of a recognition accuracy rate as high as 85.1%.","author":[{"family":"Sangle","given":"Prof"}],"issued":{"date-parts":[[2025]]},"DOI":"10.21203/rs.3.rs-6478988/v1","URL":"https://doi.org/10.21203/rs.3.rs-6478988/v1","source":"crossref"},{"id":"doi:10.2196/preprints.85127","type":"manuscript","title":"Clapping and Vibrating Caring  to Address Ineffective Airway Clearance Based on: Neural Network (Preprint)","abstract":"BACKGROUND The accumulation of mucus in the airways is a serious health problem as it can obstruct airflow and impair lung function. This condition is typically managed by suctioning the mucus and performing manual chest clapping, which involves repeatedly patting the back. However, manual clapping is often ineffective and inefficient due to factors like operator fatigue and inconsistent application. This research introduces a portable clapping system called Clapping and Heater Integrated Caring device (CHIC) OBJECTIVE this research is to develop an automatic detection system that can reliably recognize human clapping and hand induced vibrations using wearable sensors. The study aims to design and implement pattern recognition algorithms, ranging from threshold based methods to lightweight machine learning models, that differentiate intentional clapping or vibrating gestures from environmental noise, and to evaluate the system’s real time performance in terms of accuracy, latency, and energy consumption across diverse usage scenarios. METHODS This research introduces an innovative medical device that integrates automatic clapping capabilities with a vibrating warm pillow. The goal of this innovation, named CHIC, is to overcome the limitations of manual methods and enhance the effectiveness of chest physiotherapy. A key feature that makes CHIC so relevant is the use of Neural Network (NN) technology to determine the number of claps based on the patient's condition RESULTS A Neural Network (NN) allows the system to adaptively generate the optimal clapping frequency based on the patient's physiological parameters, ensuring a more precise and personalized therapy compared to conventional approaches. Three different NN architectures were tested, and the one with three hidden layers and a neuron configuration of [20 30 40] proved to be the most effective. This configuration yielded a Mean Squared Error (MSE) of 0.00029279 for the training data and a Root Mean Squared Error (RMSE) of 0.02812 for the validation data. CONCLUSIONS The CHIC device demonstrated reliable performance for airway clearance therapy. Its ESP32 based control system, temperature monitoring, vibrating pillow, and rotary DC motor provided consistent and coordinated clapping actions. The integrated neural network algorithm dynamically adjusted the clapping frequency according to the patient’s physiological parameters, delivering a more precise and personalized treatment compared to manual chest percussion. The system operated without operator fatigue and maintained patient comfort, indicating that CHIC is an effective and practical solution for clinical and community based physiotherapy applications","author":[{"family":"Khumaidi","given":"Agus"}],"issued":{"date-parts":[[2025]]},"DOI":"10.2196/preprints.85127","URL":"https://doi.org/10.2196/preprints.85127","source":"crossref"},{"id":"doi:10.22541/essoar.175525655.54389731/v1","type":"article-journal","title":"Experimental Verification of a Two-Dimensional Inverse Method for Turbidity Currents Using a Deep Neural Network","abstract":"Turbidites have been widely studied as indicators of the occurrences and magnitudes of paleo-tsunamis and paleo-earthquakes. Inversion to estimate the flow conditions from turbidites offers valuable insights into the magnitudes of paleo-seismic and tsunami events. However, conventional one-dimensional inverse models are insufficient for capturing the behavior of turbidity currents in tectonically active margins, where the seafloor topography is typically complex. Here, we developed a horizontal two-dimensional inverse model of turbidity currents based on a deep neural network (DNN) and evaluated its performance using both synthetic and flume experiment datasets. The model successfully estimated the model input parameters with a symmetric mean absolute percentage error (SMAPE) of less than 32.5%, except for the density-equivalent sediment concentration for saline water at the inlet. When applied to experimental data, the model reasonably reconstructed the flow conditions, yielding SMAPE values between 51.7 and 86.2%, despite the potential uncertainties introduced by sampling disturbances, data processing, and forward model limitations. The spatial distribution of bed thickness was also well predicted, except in cases where most of the suspension bypassed the depositional zone. Overall, the proposed inverse model demonstrated accuracy comparable to the previous one-dimensional model while offering greater applicability to complex seafloor geometries and maintaining low computational costs. These results suggest that the proposed method is well-suited for the field-scale inversion of turbidity currents in realistic geological settings.","author":[{"family":"Fujishima","given":"Seiya"},{"family":"Naruse","given":"Hajime"}],"issued":{"date-parts":[[2025]]},"DOI":"10.22541/essoar.175525655.54389731/v1","URL":"https://doi.org/10.22541/essoar.175525655.54389731/v1","source":"crossref"},{"id":"doi:10.5194/egusphere-egu25-15255","type":"article-journal","title":"A Novel Global Gridded Ocean Oxygen Product Derived from Neural Network Emulators (1965&amp;#8211;2022)","abstract":"Over the past decades, global ocean oxygen inventories have declined by 0.5&amp;#8211;3.3% relative to historical averages, with significant uncertainties in data-sparse regions such as the South Pacific and Indian Oceans. These gaps hinder accurate estimates of deoxygenation rates, potentially leading to underestimation of its magnitude. In this context, gridded oxygen products are essential for assessing global and regional trends and projecting the impacts of deoxygenation on marine ecosystems. However, traditional Optimal Interpolation (OI) methods are known to underestimate ocean oxygen loss, particularly in poorly observed areas.To address these limitations, we propose a novel approach to build a gridded oxygen concentration product. Specifically, we develop a neural network emulator of oxygen concentration based on temperature and salinity measurements. This neural network is then used to generate emulated oxygen concentration data, which are combined with dissolved oxygen measurements to produce a new global gridded oxygen concentration product spanning 1965 to 2022. We evaluate our product against climatological estimates from the World Ocean Atlas and other gridded oxygen products. Future work will leverage this gridded product to study the regional evolution of ocean deoxygenation, particularly in Oxygen Minimum Zone (OMZ) regions.","author":[{"family":"Lachkar","given":"Zouhair"},{"family":"Ouala","given":"Said"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5194/egusphere-egu25-15255","URL":"https://doi.org/10.5194/egusphere-egu25-15255","source":"crossref"},{"id":"doi:10.17776/csj.1639638","type":"article-journal","title":"Artificial Neural Network Application on Strain Effect of WSe2","abstract":"Ab initio density functional theory (DFT) calculations have been used to determine the band gap of the 2D layered material WSe2 under uniaxial strain. This involves finding optimised lattice parameters and calculating the electronic band structure. We also note that by applying strains ranging from -15% to 15%, a wide range of band gaps can be obtained to study the behaviours of the semimetal and metal. We found, WSe2 structure becomes semi-metal at +12% and -13% strains then or also becomes metal after +12% and -13% strains. The ensuing outcomes are then applied to the domain of machine learning. Initially, PR, which is polynomial regression, is a machine learning method that can be studied with numpy, sklearn and scipy modules, and ANN (artificial neural network) is the application with tensorflow module that is applied to the optimised semi-metallic and metallic structures. The application of the latter involves the Least Squares method. The implementation of Leaky Relu (LRelu) and Elu functions is instrumental in facilitating the deployment of ANN. The potential dataset is obtained by using Quantum Espresso, and the calculations are made by using the National Center for High Performance Computing of Turkey (UYBHM). The PR and ANN results are calculated using the existing data set. The primary objective of utilising PR and artificial neural networks is to facilitate the plotting of Valance Band Maximum (VBM) and Conduction Band Minimum (CBM) graphs in the vicinity of the Fermi level. This study based on data which are already available and could therefore be considered to be data mining. on the electronic band structure for WSe2.","author":[{"family":"Dağıstanlı","given":"Hamdi"}],"issued":{"date-parts":[[2025]]},"DOI":"10.17776/csj.1639638","URL":"https://doi.org/10.17776/csj.1639638","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-7991330/v1","type":"article-journal","title":"A Quantum Neural Network for Fraud Detection Using a Data-Driven Priority Entanglement Scheme","abstract":"Abstract This study investigates whether a variational quantum circuit can improve minority-class sensitivity when substituted for a single dense layer in a strong neural network baseline for credit-card fraud detection. Under matched parameter budgets and identical preprocessing, we evaluate a hybrid quantum–classical model against classical baselines on the Kaggle dataset, preserving all 492 fraud cases and downsampling legitimate transactions to 10,000. Our hybrid employs data re-uploading to expose all 30 features with 10–15 qubits. It introduces a data-driven priority-entanglement scheme that couples the most dependent feature pairs before a strongly entangling block. Across 100 (10-qubit) and 20 (15-qubit) randomized runs, the hybrid achieves meaningful and consistent gains in recall and PR-AUC over tuned classical models (e.g., PR-AUC 0.9229±0.006 at 15q/1-layer/10 priority pairs vs 0.9150±0.014 for logistic regression; recall +0.047 over the best classical result).Performance peaks at moderate entanglement budgets, beyond which deeper circuits only exacerbate a precision-recall trade-off, revealing an underlying trainability limit. Results indicate that correlation-guided entanglement provides a useful inductive bias that modestly improves fraud detection under strict capacity parity.","author":[{"family":"Khaliq","given":"Yousaf"},{"family":"Wang","given":"Donglin"}],"issued":{"date-parts":[[2025]]},"DOI":"10.21203/rs.3.rs-7991330/v1","URL":"https://doi.org/10.21203/rs.3.rs-7991330/v1","source":"crossref"},{"id":"doi:10.36227/techrxiv.174952940.04337137/v1","type":"article-journal","title":"Hardware-Based Spiking Neural Network for Real-Time Hand Gesture Recognition Using MYO Armband EMG sensor","abstract":"Hand gesture recognition is a key technology for improving prosthetic control and enhancing human-robot interaction. Traditional machine learning approaches struggle with real-time, low-power constraints in embedded systems. This paper proposes a hardware-based Spiking Neural Network (SNN) for real-time hand gesture recognition from electromyography (EMG) signals using the MYO Armband. The SNN leverages energy-efficient, event-driven computation to classify dynamic gestures in real-time, demonstrating applicability in myoelectric prosthetics and robotics. Implemented on a low-cost FPGA, the proposed system achieves accurate and responsive gesture recognition under strict resource constraints.","author":[{"family":"Mehrabi","given":"Ali"},{"family":"Gargiulo","given":"Gaetano"}],"issued":{"date-parts":[[2025]]},"DOI":"10.36227/techrxiv.174952940.04337137/v1","URL":"https://doi.org/10.36227/techrxiv.174952940.04337137/v1","source":"crossref"},{"id":"doi:10.64898/2026.06.01.729343","type":"article-journal","title":"Predicting P-glycoprotein Substrate Status Using a Pretrained Graph Neural Network: A TDC Benchmark Study","abstract":"Abstract P-glycoprotein (Pgp/ABCB1) is a critical efflux transporter that significantly impacts drug bioavailability and multidrug resistance. Accurate prediction of Pgp substrate status is essential for early-stage drug discovery. In this study, we evaluate a pretrained Graph Iso-morphism Network (GIN) with attribute masking on the Pgp_Broccatelli benchmark from the Therapeutics Data Commons (TDC). Our approach fine-tunes a GIN encoder pretrained on approximately 2 million molecules using a self-supervised attribute masking strategy, followed by a multilayer perceptron (MLP) classification head. On the TDC benchmark, our model achieves an AUROC of 0.937 ± 0.004 across five independent runs, ranking second on the leaderboard, as of May 2026. We further compare this approach against an XGBoost baseline using Morgan fingerprints (AUROC 0.912 ± 0.007), demonstrating the advantage of graph-based molecular representations with transfer learning for small-dataset ADMET prediction tasks.","author":[{"family":"Yan","given":"Jingjing"},{"family":"Duan","given":"Weicong"}],"issued":{"date-parts":[[2026]]},"DOI":"10.64898/2026.06.01.729343","URL":"https://doi.org/10.64898/2026.06.01.729343","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-9897709/v1","type":"article-journal","title":"A Two-Dimensional Bayesian Continuous Attractor Neural Network for Robust Spatial State Estimation","abstract":"Abstract Continuous Attractor Neural Networks (CANNs) have emerged as prominent bio-inspired models for spatial representation, with recent theoretical advances demonstrating their capacity for optimal Bayesian inference. However, the mathematical equivalence between attractor dynamics and optimal Bayesian inference has been largely confined to 1D circular variables. To bridge this gap, we propose a Bayesian CANN in a 2D Euclidean state space. The network conditions required for optimal inference are analytically derived by matching the network dynamics with the continuous-time Kalman-Bucy filter. Crucially, by introducing an advection-reaction operator splitting method, we resolve a velocity-dependent diffusion artifact, thereby preserving Bayesian consistency without artificially degrading the certainty of the internal representation. The proposed 2D Bayesian CANN achieves estimation accuracy comparable to Kalman and particle-based filters while retaining structural robustness to severe sensory conflict. This robustness arises from the linear superposition of conflicting sensory evidence followed by nonlinear attractor dynamics, which suppress dispersed low-confidence activity and realign the localized representation toward dominant sensory cues. These results bridge theoretical neuroscience and engineering state estimation, suggesting a bio-inspired framework for resilient autonomous navigation in unpredictable environments.","author":[{"family":"Shin","given":"Hyo"},{"family":"Kim","given":"Sohyun"}],"issued":{"date-parts":[[2026]]},"DOI":"10.21203/rs.3.rs-9897709/v1","URL":"https://doi.org/10.21203/rs.3.rs-9897709/v1","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-8840997/v1","type":"article-journal","title":"Architecture and Scaling of the TSKI Model: A Phase–Temporal Neural Network Without a Loss Function","abstract":"Abstract This paper presents a biologically inspired neural network model based on the theoretical framework of TSKI 4.2 [1] by Atorin A., and examines its potential for architectural scaling. The model is positioned as an alternative class of neural networks in which computation is not based on minimizing a loss function, but on the formation and stabilization of temporal information trajectories through phase–temporal synchronization of neurons and homeostatic regulation of their parameters. The paper is structured in two phases. The first phase describes the components of TSKI theory that have already been implemented in software. The main mathematical objects of the model and their implementation are discussed, including neurons and synaptic connections with a mirrored representation, where access to the connection vector is available to both presynaptic and postsynaptic neurons. The computational core of the model is described, implementing a single operational step based on a phase synchronization mechanism between presynaptic and postsynaptic neurons (k5 = 0, k5 = 1). This mechanism constitutes the necessary and sufficient condition for information computation and synaptic parameter updates, and represents a digital analogue of biological STDP (Spike-Timing-Dependent Plasticity). This fundamental computational algorithm of TSKI reduces computational costs by introducing a binary condition for synchronization within a synapse: when synchronization is present, computations and parameter updates are performed; when synchronization is absent, computational activity is blocked. The simulation also implements a computational algorithm analogous to biological homeostatic regulation, responsible for stabilizing neuronal parameters and influencing synchronization conditions across neural connections. Simulation results [2] obtained from four small-scale TSKI neural networks [3] are presented, on the basis of which the hypothesis is advanced that the TSKI model may reduce susceptibility to catastrophic forgetting after training. The main focus of the paper is its second part, which explores the scalability of TSKI and describes an architecture designed to enable such scaling. The architectural organization of the TSKI model is oriented toward reproducing principles of information processing characteristic of the biological nervous system: receptor zone → thalamic nuclei → cerebral cortex. Within this analogy, four functional zones are identified in the model, each performing a distinct role; their operating principles and architectural structures are described. The key objective of the TSKI architectural solution is to ensure alignment between the phase–temporal states of neurons and the temporal dynamics of changes in a physical (detectable) stimulus parameter. Such alignment is necessary for the TSKI model to form, associate, and reproduce adaptive responses to stimuli over time. The analysis of the proposed architectural solutions reveals local analogues of the backpropagation mechanism (error backpropagation in ANNs) and optimization algorithms used in classical neural network models, as applied to the TSKI architecture. The nature of the objective function in TSKI and methods for improving the efficiency of its realization through the model’s internal dynamics are also examined. The paper concludes by demonstrating a high degree of similarity between the information-processing principles of the TSKI architecture and those of the biological nervous system. It is shown that the TSKI algorithms developed and implemented in simulations of small-scale neural networks are fundamentally ready for architectural scaling, and that the architectural principles embedded in the functional zones are falsifiable. This creates the possibility for experimental verification of the theoretical assumptions formulated in this work.","author":[{"family":"Atorin","given":"Andrii"}],"issued":{"date-parts":[[2026]]},"DOI":"10.21203/rs.3.rs-8840997/v1","URL":"https://doi.org/10.21203/rs.3.rs-8840997/v1","source":"crossref"},{"id":"doi:10.52783/anuval.2045","type":"article-journal","title":"Hybrid Optimizer Switching for Deep Neural Network Training in Time Series Forecasting","abstract":"Deep neural networks often encounter non-convex optimization challenges during training due to the presence of local minima, saddle points, and complex loss surfaces. Existing optimization algorithms such as Adam and Stochastic Gradient Descent (SGD) offer complementary advantages—Adam provides faster convergence, while SGD tends to achieve better generalization. However, neither optimizer alone effectively balances both properties in non-convex settings. To address this limitation, this paper proposes a phase-switch hybrid optimization strategy that combines the strengths of Adam and SGD. The proposed method employs Adam during the initial phase of training to enable rapid convergence and efficient exploration of the loss landscape, and then switches to momentum-based SGD in the later phase to improve generalization and ensure stable convergence. The effectiveness of the proposed approach is evaluated on three benchmark dataset the M4 time-series forecasting data set, under different learning rate settings. Experimental results demonstrate that the proposed method achieves performance that is superior or comparable to existing optimizers in terms of accuracy and loss minimization. These results indicate that the proposed hybrid optimization strategy provides a simple and effective solution for handling non-convex optimization problems","author":[{"family":"Kunder","given":"Harish"}],"issued":{"date-parts":[[2026]]},"DOI":"10.52783/anuval.2045","URL":"https://doi.org/10.52783/anuval.2045","source":"crossref"},{"id":"doi:10.2139/ssrn.6945159","type":"manuscript","title":"On-device Spiking Neural Network Locomotion Learning on a €100 Quadruped: Sim-to-Real with Brain Persistence","abstract":"I present a complete Sim-to-Real pipeline for quadruped locomotion using biologically grounded spiking neural networks (SNNs) on a €100 Freenove Robot Dog Kit (FNK0050) with a Raspberry Pi 4. The system employs 232 Izhikevich neurons with reward-modulated spike-timing-dependent plasticity (R-STDP), a central pattern generator (CPG) for innate gait rhythm, and a cerebellar forward model for balance correction. Training occurs in MuJoCo simulation using a custom MJCF model of the Freenove hardware, achieving 8.22 m forward distance with zero falls in 50,000 steps (original training run; Table 3 compares CPG vs Full Stack configurations). The trained brain transfers to real hardware via a Bridge architecture mapping SNN motor outputs to servo commands with real-time IMU feedback (MPU6050). On-device learning enables the robot to reach actor competence 1.0 within 2,000 steps (40 seconds at 50 Hz). Brain persistence across sessions is demonstrated: a loaded brain achieves competence 1.0 from step 1, while a fresh brain requires 2,000 steps. A key finding: cerebellar correction magnitude is zero in simulation (where CPG produces clean movement) but non-zero on real hardware (where IMU noise and servo imprecision create real errors). This confirms the cerebellum's role as an error-driven adaptive system, not a pattern generator. The same architecture runs on the Unitree Go2 in simulation (45.15 ± 0.67 m, 10 seeds). All code is open source under Apache 2.0.","author":[{"family":"Hesse","given":"Marc"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2139/ssrn.6945159","URL":"https://doi.org/10.2139/ssrn.6945159","source":"crossref"},{"id":"doi:10.1063/5.0320669","type":"article-journal","title":"A multigrid-based super-resolution convolutional neural network for multi-physical fields","abstract":"A novel convolutional neural network-based super-resolution (SR) method was proposed to accurately predict high-resolution flow fields from coarse-grid computational fluid dynamics (CFD) results. To mitigate boundary artifacts caused by transposed convolution, a Boundary Compensated Module was developed, ensuring input conservation by balancing the “Upper Layer Contribution Number.” A Two-step Preprocessing Scheme combined with the nondimensionalized mean square error was introduced, which effectively tackled the issues associated with “multi-physical variables” and “wide operating conditions.” Furthermore, a multigrid-based super-resolution convolutional neural network was developed, incorporating a three-stage progressive upsampling process with hierarchical loss and regularization. The model enabled efficient reconstruction under conditions of “high SR factor” and “multi-physical variables.” The effectiveness of the proposed method was validated across four representative benchmark cases (Backward-Facing Step Flow, Lid-Driven Cavity Flow, Supersonic Flow, and Swirling Jet Flow), which demonstrated that the proposed method reduces L2 errors by approximately 50% compared with existing models. The method achieved roughly two orders of magnitude of computational speedup, demonstrating strong engineering viability as a surrogate for high-fidelity CFD simulations.","author":[{"family":"Li","given":"Tieying"},{"family":"You","given":"Changfu"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1063/5.0320669","URL":"https://doi.org/10.1063/5.0320669","source":"crossref"},{"id":"doi:10.1093/ijlct/ctag019","type":"article-journal","title":"New energy power electronic DC motor control system based on a fuzzy neural network","abstract":"Abstract Based on the uncertainty and control effect of electronic DC motors, a neural network algorithm, fuzzy control, and fuzzy neural network are designed. First, a small signal model controlled by a virtual direct filter motor is developed, and the stability and dynamic characteristics of the control strategy are analyzed. Next, the influence of the distribution map of the system based on the closed-loop transmission function, inertia coefficient, and step response curve is explored. Finally, the computer simulation analysis verifies the virtual inertial adaptive algorithm based on the fuzzy neural network.","author":[{"family":"Wang","given":"Hongmiao"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1093/ijlct/ctag019","URL":"https://doi.org/10.1093/ijlct/ctag019","source":"crossref"},{"id":"doi:10.26434/chemrxiv.15004951/v1","type":"manuscript","title":"An efficient neural network architecture for molecular vibrational states: from rigid molecules to molecular complexes","abstract":"Accurately solving the nuclear Schrödinger equation for molecular complexes exhibiting noncovalent interactions remains a challenge due to the exponential scaling of basis sets and the presence of strongly anharmonic, large-amplitude motions. Recently, we proposed a neural network (NN) method to tackle high-dimensional vibrational problems and successfully applied it to C 3 H 8 in full dimensionality; however, applying the method to molecular complexes has been hindered by difficulties in representing delocalized wave functions across multiple minima. In this work, we present a substantially improved NN-based methodology that overcomes these limitations through two key developments: (i) a novel NN architecture specifically designed to capture the asymptotic behavior and delocalization characteristics of weakly bound complexes, and (ii) an efficient training protocol that optimizes the variance of local energies. We demonstrate that this approach achieves sub‑cm -1 accuracy for the vibrational ground and excited states of CH 4 with substantially fewer parameters than the previous method requires. Crucially, for the prototypical water dimer (H 2 O) 2 , the new method successfully computes both ground state and intramolecular excited states with ~ 1 cm -1 accuracy using ~6,000 NN parameters. This work establishes a robust and scalable framework for performing full-dimensional, fully coupled vibrational calculations on molecular complexes, paving the way for accurate spectroscopic predictions of systems previously deemed intractable.","author":[{"family":"Zhao","given":"Shuaishuai"},{"family":"Zhang","given":"Dong"}],"issued":{"date-parts":[[2026]]},"DOI":"10.26434/chemrxiv.15004951/v1","URL":"https://doi.org/10.26434/chemrxiv.15004951/v1","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-9144582/v1","type":"article-journal","title":"Physics-Informed Neural Network for Inverse Design of Cylindrical Kresling Origami with Discrete Side Count Optimization","abstract":"Abstract Origami mechanisms excel in space and material utilization, but their design variables involve the coupling of continuous and discrete variables, complicating reverse engineering—especially for polygons with dis crete side counts. Traditional numerical optimization methods struggle with discrete variables and lack automation. This paper proposes a Physics-Informed Neural Network (PINN) framework for the reverse engineering of Kresling origami. In this framework, the discrete side count *m* is first adjusted by a dis tance penalty to shorten the distance between the optimized and target values. This ensures contin uously differentiable gradients and smooth transitions, avoiding loss bounces. Then, a soft projection function (a softmax function with a learnable temperature) strictly constrains the output value to integers, ultimately achieving continuous differentiability. Physical constraints (potential energy difference and near zero torque, achieved through ∂U/∂β ≈ 0) are embedded in the loss function, enabling label-free training. Novel engineering constraints ensure manufacturability. Numerical examples show MSE of the energy curve &lt; 1e-5 in all tested cases and convergence to integer m in &gt;95% of runs.","author":[{"family":"Li","given":"Shijun"}],"issued":{"date-parts":[[2026]]},"DOI":"10.21203/rs.3.rs-9144582/v1","URL":"https://doi.org/10.21203/rs.3.rs-9144582/v1","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-8902258/v1","type":"article-journal","title":"Accelerated Circuit Simulations and Standard Cell Library Characterization through Neural Network-based Transistor Modeling","abstract":"Abstract nn-based transistor models have proven to be a promising solution to accelerate device modeling. Although these models demonstrated remarkable speedup in circuit simulations, they were not rigorously tested in complex tasks within standard EDA tool flows for chip design, which require very high transistor model precision to provide accurate results and converge properly. To investigate their full potential and promise, we develop NN-based transistor models that accurately predict the drain current and various charges for a \\qty{3}{nm} nanosheet, and employ the models for a standard cell library characterization. This enables us to evaluate the accuracy and speedup under a very wide range of SPICE simulations, benchmarking our models against an industry-standard implementation. Since standard cell library characterization involves thousands of SPICE simulations, it serves as a great benchmark for transistor model quality and consistency. To further evaluate model accuracy, we also perform analog- and digital-level simulations for various complex circuits. Our experiments reveal that NN-based models are a valid alternative to existing transistor compact models, capable of producing accurate delay and power estimates with a much higher speed and convergence time compared to the conventional Verilog-A-based industry-standard (BSIM-CMG) transistor compact model. Our NNs achieve sub-\\((0.11%)\\) and \\((0.76%)\\) errors for delay and total power evaluation on a wide range of circuits, while taking up to 8 times less time during the standard cell library characterization. Additionally, we provide comprehensive information and guidance on NN accuracy, unveiling the relationship between NN size and its effect on circuit simulation precision.","author":[{"family":"Novkin","given":"Rodion"},{"family":"Amrouch","given":"Hussam"}],"issued":{"date-parts":[[2026]]},"DOI":"10.21203/rs.3.rs-8902258/v1","URL":"https://doi.org/10.21203/rs.3.rs-8902258/v1","source":"crossref"},{"id":"doi:10.22541/essoar.175525655.54389731/v2","type":"article-journal","title":"Experimental Verification of a Two-Dimensional Inverse Method for Turbidity Currents Using a Deep Neural Network","abstract":"Turbidites have been widely studied as indicators of the occurrences and magnitudes of paleo-tsunamis and paleo-earthquakes. Inversion to estimate flow conditions from turbidites offers valuable insights into the magnitudes of paleo-seismic and tsunami events. However, conventional one-dimensional inverse models are insufficient for capturing the behavior of turbidity currents in tectonically active margins, where the seafloor topography is typically complex. Here, we developed a horizontal two-dimensional inverse model of turbidity currents based on a deep neural network (DNN) and evaluated its performance using synthetic and flume experiment datasets. The model successfully estimated model input parameters with a symmetric mean absolute percentage error (SMAPE) of less than 29.7%, except for the density-equivalent sediment concentration for saline water at the inlet. When applied to experimental data, the model reasonably reconstructed flow conditions, yielding SMAPE values between 51.7 and 86.2%, despite potential uncertainties introduced by sampling disturbances, data processing, and forward model limitations. These values were reasonable even compared to the results of the previous one-dimensional inverse model (Cai and Naruse, 2021), which showed a SMAPE range of 24.2–113%. The spatial distribution of bed thickness was also well predicted, except when most of the suspension bypassed the depositional zone. Overall, the proposed inverse models demonstrated accuracy comparable to that of the previous one-dimensional model. They offer greater applicability to complex seafloor geometries and maintain low computational costs, providing a practical advantage over optimization-based methods. These results suggest that the proposed method is well-suited for the field-scale inversion of turbidity currents in realistic geological settings.","author":[{"family":"Fujishima","given":"Seiya"},{"family":"Naruse","given":"Hajime"}],"issued":{"date-parts":[[2026]]},"DOI":"10.22541/essoar.175525655.54389731/v2","URL":"https://doi.org/10.22541/essoar.175525655.54389731/v2","source":"crossref"},{"id":"doi:10.18469/ikt.2026.24.1.11","type":"article-journal","title":"USING NEURAL NETWORK TECHNOLOGIES TO OPTIMIZE MOBILE APP USER EXPERIENCE","abstract":"This paper analyzes domestic and international sources confirming the potential of neural network technologies for solving user experience optimization problems in mobile applications. A comprehensive approach is proposed, including selecting the optimal combination of various neural network architectures, formulating the problem in terms of optimization, and selecting adequate loss functions to optimize various aspects of the user experience. As a result of the research, a mathematical model of a hybrid architecture (Transformer and LSTM) was developed, combining a recurrent part (LSTM) to account for the temporal structure and an attention mechanism (Transformer-encoder) for a detailed analysis of contextual features. To evaluate the effectiveness of the proposed hybrid model, the results of a comparison with alternative models are presented, which showed optimal values for the selected metrics. The implemented model has been integrated into the mobile application: the results of A/B testing confirmed significant optimization of the main user experience metrics.","author":[{"family":"Zinovjev","given":"Danil"},{"family":"Bogomolova","given":"Mariya"}],"issued":{"date-parts":[[2026]]},"DOI":"10.18469/ikt.2026.24.1.11","URL":"https://doi.org/10.18469/ikt.2026.24.1.11","source":"crossref"},{"id":"doi:10.4018/ijec.420194","type":"article-journal","title":"Neural Network-Driven Optimization of Interdisciplinary Academic Leadership in Universities","abstract":"The development of interdisciplinary academic leaders is crucial for universities aiming to enhance collaboration, innovation, and knowledge integration across disciplines. Research introduces a deep learning-based framework, Wingsuit Flying Search-driven Dynamic Recursive Neural Networks (WFS-DRNN), implemented using Python (3.11), to predict leadership potential and optimize development pathways. By modeling hierarchical and sequential relationships among faculty qualifications, research experience, collaborative networks, and mentorship records, the WFS-DRNN captures both explicit and latent patterns critical to leadership growth. The Wingsuit Flying Search Algorithm (WFSA) efficiently explores optimal development pathways, enabling tailored interventions, such as interdisciplinary project participation, mentorship opportunities, and skill enhancement programs. Experimental results indicate that WFS-DRNN surpasses traditional heuristic and statistical methods, achieving 95.8% accuracy, 95.2% precision, 92.75% recall, and a 95% F1-score.","author":[{"family":"Ling","given":"Qiuyue"}],"issued":{"date-parts":[[2026]]},"DOI":"10.4018/ijec.420194","URL":"https://doi.org/10.4018/ijec.420194","source":"crossref"},{"id":"doi:10.1051/bioconf/202623603001","type":"article-journal","title":"Colour classification of Bulgarian honey using spectroradiometry and neural network processing","abstract":"This publication explores the possibility of implementing a colour classifier for various types of Bulgarian honey. For this purpose, spectroradiometric measurement is used based on a measuring device from JETI Technische Instrumente GmbH – Specbos1201 providing a measurement range in the visible spectrum – 380 ~ 820 nm, which is complemented by a reference light source BYK Byko-spectra – D65. The obtained data is used to implement two types of classification, between which a comparison was made in terms of a quality indicator - a KNN classifier and a classifier with a neural network with discrete outputs. In order to reduce the number of inputs, the spectral data was transformed into the standard CIE-Lab 1976 colour space. The obtained results show a very good possibility for implementing a classifier using a neural network, with an extremely low level of potential misclassification despite the usage of a relatively small amount of training data.","author":[{"family":"Simonov","given":"Ventsislav"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1051/bioconf/202623603001","URL":"https://doi.org/10.1051/bioconf/202623603001","source":"crossref"},{"id":"doi:10.65599/iii9326","type":"article-journal","title":"APPLICATION OF NONLINEAR AUTOREGRESSIVE NEURAL NETWORK FOR PREDICTIVE ANALYSIS AND SIGNAL OPTIMIZATION IN 5G NETWORKS","abstract":"This paper explores the possibility of improving the efficiency of fifth-generation communication networks through the use of intelligent forecasting methods. The aim of the study is to apply a nonlinear autoregressive neural network (NARNN) for the analysis and prediction of changes in radio signal parameters in 5G networks, followed by proactive radio resource management. Time series of RSRP and SINR indicators characterizing the state of the radio channel are used as input data. The modeling is carried out using a neural network architecture with time delays trained on experimental data. The forecasting accuracy is evaluated using standard error metrics. The results demonstrate that the application of NARNN makes it possible to detect signal quality degradation in advance and adapt network parameters, thereby ensuring more stable and reliable operation of fifth-generation networks. Keywords: fifth-generation networks, nonlinear autoregressive neural network, signal forecasting, time series, radio resource management, communication quality.","author":[{"family":"Saidov","given":"Behruz"}],"issued":{"date-parts":[[2026]]},"DOI":"10.65599/iii9326","URL":"https://doi.org/10.65599/iii9326","source":"crossref"},{"id":"doi:10.63070/jesc.2026.032","type":"article-journal","title":"A Multi-Similarity Neural Network for Paraphrase Detection","abstract":"This study introduces a multi-similarity neural network framework for paraphrase detection, an important task in natural language processing that identifies whether two sentences convey the same meaning using different expressions. The proposed method combines various similarity measures, such as string-based similarity, semantic similarity, and embedding-based similarity, with a deep learning classifier. The framework is structured as a three-phase pipeline: preprocessing, extraction of multiple similarity features, and classification through a neural network. It employs more than 168 string similarity algorithms, semantic measures derived from WordNet, and several pre-trained embedding models to compute similarity scores. These features are aggregated and supplied to a deep neural network to determine whether sentence pairs are paraphrases. The model was evaluated on the Microsoft Research Paraphrase Corpus (MSRP) using accuracy and F1-score as performance metrics. The experimental results indicate that the proposed framework achieves 81.74% accuracy and an F1 Score of 86.6%, surpassing several existing approaches. Overall, the results suggest that integrating diverse similarity measures with neural networks enhances the identification of both explicit and nuanced paraphrases, thereby supporting advancements in text analysis and plagiarism detection systems.","author":[{"family":"Nabil","given":"Emad"}],"issued":{"date-parts":[[2026]]},"DOI":"10.63070/jesc.2026.032","URL":"https://doi.org/10.63070/jesc.2026.032","source":"crossref"},{"id":"doi:10.36341/rabit.v11i2.7934","type":"article-journal","title":"KLASIFIKASI TINGKAT KEMATANGAN BUAH PISANG TANDUK MENGGUNAKAN METODE CONVOLUTIONAL NEURAL NETWORK","abstract":"Penelitian ini bertujuan untuk membangun sistem klasifikasi tingkat kematangan buah pisang tanduk menggunakan metode Convolutional Neural Network (CNN) berbasis Transfer Learning dengan arsitektur MobileNetV2 dan EfficientNetB0. Dataset yang digunakan terdiri dari 1.142 citra buah pisang tanduk yang dibagi menjadi data training dan validation. Proses pelatihan model dilakukan menggunakan teknik augmentasi data dan regularisasi untuk meningkatkan kemampuan generalisasi model. Hasil penelitian menunjukkan bahwa model MobileNetV2 memperoleh performa terbaik dengan nilai accuracy sebesar 97,75%, precision sebesar 97,84%, recall sebesar 97,75%, dan F1-score sebesar 97,74%. Sementara itu, EfficientNetB0 memperoleh nilai accuracy sebesar 90,09%. Model terbaik kemudian diimplementasikan ke dalam sistem prediksi berbasis website untuk melakukan klasifikasi tingkat kematangan buah pisang tanduk secara otomatis dan real-time. Hasil penelitian menunjukkan bahwa metode CNN berbasis Transfer Learning mampu memberikan performa klasifikasi yang sangat baik dalam mengidentifikasi tingkat kematangan buah pisang tanduk.","author":[{"family":"Fitriani","given":"Dela_rizqi"},{"family":"Fitriani","given":"Dela"}],"issued":{"date-parts":[[2026]]},"DOI":"10.36341/rabit.v11i2.7934","URL":"https://doi.org/10.36341/rabit.v11i2.7934","source":"crossref"},{"id":"doi:10.1088/2631-8695/ae8077","type":"article-journal","title":"A Multi-channel convolutional neural network for ECG signal classification","abstract":"Abstract Electrocardiogram (ECG) signal classification plays significant role in early diagnosis, continuous monitoring cardiovascular diseases. The existing deep learning methods often rely on single-domain feature extraction methods, restricting capability of capturing their complex temporal, morphological, and spectral characteristics of non-stationary ECG signals. To overcome these limitations, proposes a novel multi-channel convolutional neural network (CNN) framework integrating Bernstein polynomial-based segmentation with Smoothed Pseudo Wigner–Ville Distribution-based time-frequency representation for robust ECG classification. The proposed framework combines a 1D CNN branch for temporal and morphological feature extraction with a 2D CNN branch for spectral feature learning. The proposed model was evaluated using MIT-BIH arrhythmia dataset and externally validated using PTB-XL ECG dataset. To improve evaluation reliability and minimize overfitting, patient-wise data splitting and 5-fold cross-validation were incorporated. The proposed framework achieves greater classification performance with an accuracy of 99.1%, sensitivity of 98.2%, specificity of 98.9%, precision of 98.0%, and F 1-score of 97.2% on the MIT-BIH dataset while maintaining strong generalization capability on PTB-XL. The proposed framework provides an effective and scalable solution for intelligent real-time ECG abnormality detection and future multi-class arrhythmia analysis.","author":[{"family":"Polavarapu","given":"Harshini"},{"family":"Mitukula","given":"Rajesh"}],"issued":{"date-parts":[[2026]]},"DOI":"10.1088/2631-8695/ae8077","URL":"https://doi.org/10.1088/2631-8695/ae8077","source":"crossref"},{"id":"doi:10.2478/agriceng-2026-0004","type":"article-journal","title":"Application of Back-Propagation Artificial Neural Network and Particle Swarm Optimization Methods in Sprinkler Optimization","abstract":"Abstract The aim of this paper was to analyze the primary and secondary order of influencing factors and to establish a BP neural network prediction model with different hydraulic performance indicators. Particle swarm optimization algorithm was used to test for the optimal hydraulic performance of the nozzle and validated by experiment. The orthogonal design covered the 200–300 kPa range, while the PSO algorithm selected 540 kPa based on the full dataset. The sprinkler was raised at a height of 1.4 m from the ground level in a square configuration. The optimal parameter combination for the square layout achieved higher uniformity and controlled maximum kinetic energy relative to the average sprinkling intensity. Results from the experiment gave 4.71 mm·h −1 , 83.28% and 0.0078 W·m −2 for average sprinkler intensity, CU and maximum kinetic energy, respectively. Compared with the rangeanalysis- based scheme, the optimized configuration reduced average irrigation intensity while improving uniformity and kinetic energy performance. The maximum error between experimental results and optimization results was 3.29%, indicating that the optimization model is feasible and reliable. This study proposes a practical optimization framework for the design and operation of sprinkler irrigation systems.","author":[{"family":"Issaka","given":"Zakaria"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2478/agriceng-2026-0004","URL":"https://doi.org/10.2478/agriceng-2026-0004","source":"crossref"},{"id":"doi:10.5267/j.ijdns.2025.9.014","type":"article-journal","title":"Determinants of smart government continuous use: A two-staged structural equation modeling-artificial neural network approach","abstract":"This study aimed to develop and empirically validate an integrated model for continuous smart government service usage. This model integrates constructs from the unified theory of acceptance and use of the technology framework with the expectation-confirmation model, along with an additional construct: trust. Structural equation modeling (SEM) was used to analyze data collected via online questionnaires from 369 people who utilized smart government services in the United Arab Emirates. Next, an artificial neural networks model was used to rank the relative influence of the significant predictors identified through SEM analysis. The findings reveal that, among the significant predictors affecting the continuous use of smart government services, facilitating conditions, satisfaction, and perceived usefulness had the most substantial impact. Furthermore, this study highlights the direct influence of perceived usefulness, confirmation, facilitating conditions, effort expectancy, social influence, and public trust on citizen satisfaction.","author":[{"family":"Altarawneh","given":"Nuseiba"},{"family":"Hujran","given":"Omar"}],"issued":{"date-parts":[[2025]]},"DOI":"10.5267/j.ijdns.2025.9.014","URL":"https://doi.org/10.5267/j.ijdns.2025.9.014","source":"crossref"},{"id":"doi:10.3390/fintech5030070","type":"article-journal","title":"Network-Aware FinTech Intelligence for ESG Risk Forecasting: A Graph Neural Network and Transformer-Based NLP Approach","abstract":"Environmental, Social, and Governance (ESG) risks increasingly propagate across interconnected supply chains, yet conventional ESG assessment methods remain largely reliant on firm-level disclosures and static ESG ratings that often overlook indirect risk transmission among trading partners. This study develops a network-aware artificial intelligence (AI) framework for forecasting ESG risk by integrating Graph Neural Networks (GNNs), transformer-based natural language processing (NLP), explainable AI, and conventional machine-learning techniques. The proposed framework combines supply-chain network structures, shipment-level trade information, ESG controversy records, governance indicators, and transformer-derived ESG sentiment extracted using FinBERT and RoBERTa. Using a dataset of 11,386 firms across 27 industries from 2015 to 2025, the proposed GNN achieved the highest predictive performance, outperforming conventional machine-learning models with an ROC-AUC of 0.913. The results further demonstrate that supply-chain network centrality and transformer-derived ESG sentiment substantially improve the early identification of firms exposed to future ESG controversies. By integrating network relationships with textual ESG intelligence, the proposed framework advances FinTech-enabled ESG analytics and provides a scalable approach for proactive risk monitoring, sustainable investment decision-making, and supply-chain risk management.","author":[{"family":"Aruwaji","given":"Michael"},{"family":"Marimuthu","given":"Ferina"}],"issued":{"date-parts":[[2026]]},"DOI":"10.3390/fintech5030070","URL":"https://doi.org/10.3390/fintech5030070","source":"crossref"},{"id":"doi:10.71052/srb2024/jsyp7766","type":"article-journal","title":"DCDTI: Dual-Channel Neural Network for Drug-Target Interaction Prediction","abstract":"Background: The computational identification of drug-target interaction (DTI) is pivotal in drug discovery and chemical genomics. Current network-based approaches model DTI as a link prediction problem utilizing bipartite graphs. However, simplistic representations fail to encapsulate crucial biological semantic information, and how to effectively integrate molecular structure features with graph neural networks has emerged as a non-negligible challenge in the DTI domain. Results: To address these issues, we propose a Dual-Channel Neural Network for Drug-Target Interaction (DCDTI) prediction model based on GLTransformer, aiming to learn high-quality representations of the topological structures of drugs and targets with precision. Extensive experiments validate that DCDTI surpasses state-of-the-art methods. Case studies have further confirmed its generalization ability in actual DTI scenarios. Conclusion: DCDTI provides a powerful method for DTI prediction, which can also serve as a screening tool for studies of drug discovery.","author":[{"family":"Zhang","given":"Ping"},{"family":"Zeng","given":"Yongbin"}],"issued":{"date-parts":[[2026]]},"DOI":"10.71052/srb2024/jsyp7766","URL":"https://doi.org/10.71052/srb2024/jsyp7766","source":"crossref"},{"id":"doi:10.5194/egusphere-egu26-20172","type":"article-journal","title":"Neural Network Modelling of Climate Change and Reservoir Impacts on Upper Miño River Flow","abstract":"Climate change is altering the global hydrological cycle and, when combined with human interventions such as reservoir operations, the river flow regime is further modified. Given the strong spatial heterogeneity of these impacts and the basin-specific nature of hydrological responses, regional studies are essential to assess local vulnerabilities. This study investigates projected changes in streamflow in the upper Miño River basin (northwestern Iberian Peninsula), including the impact of the Belesar reservoir, by comparing historical conditions (1985–2014) with future projections (2070–2099) under the SSP5-8.5 and SSP2-4.5 scenarios. Artificial neural networks were employed to model basin hydrology by estimating streamflow from temperature and precipitation data, and to simulate reservoir operations, achieving satisfactory validation performance.Under the high-emission SSP5-8.5 scenario, results indicate a projected intensification of hydrological variability, with the 10th percentile, used to define low-flow conditions, decreasing by approximately 10%, whereas the percentile corresponding to a one-year return period (high-flow conditions) increases by about 5%, with the mean streamflow declining by more than 15%. Under the more moderate SSP2-4.5 scenario, changes are less pronounced, with a ~5% reduction in the low-flow percentile and a more moderate decrease in mean streamflow, while the high-flow percentile is expected to decrease by around 30 %, exhibiting an opposite trend to the extreme emission scenario. Reservoir operation was analysed under the SSP5-8.5 scenario to assess its regulatory capacity under future extreme conditions. Results show that reservoir management could mitigate projected impacts by redistributing water seasonally, more than doubling summer downstream flows compared to future natural conditions and reducing winter extremes, with peak flows lowered by approximately 15%. Overall, while future natural conditions are projected to become more critical, both moderate emission pathways and effective reservoir operation can substantially alleviate adverse hydrological impacts.","author":[{"family":"Barreiro-Fonta","given":"Helena"},{"family":"Fernández-Nóvoa","given":"Diego"}],"issued":{"date-parts":[[2026]]},"DOI":"10.5194/egusphere-egu26-20172","URL":"https://doi.org/10.5194/egusphere-egu26-20172","source":"crossref"},{"id":"doi:10.2139/ssrn.6161526","type":"manuscript","title":"A Conformable Fractional Physics-Informed Neural Network for Real-Time 6G Signal Modeling in Dispersive Nanomaterials","abstract":"The emergence of sixth-generation (6G) telecommunications at Terahertz (THz) frequencies necessitates advanced computational electromagnetic solvers capable of modeling anomalous dispersion in complex nanomaterials. Traditional grid-based methods, such as Finite-Difference Time-Domain (FDTD), are hindered by intensive mesh generation and stability constraints. This paper presents a novel meshfree soft computing framework: the Conformable Fractional Physics-Informed Neural Network (fPINN).&amp;nbsp; &lt;div&gt; By embedding Conformable Fractional Calculus (CFC) into the neural network's loss function, the fractional order α serves as a tunable dispersion index, enabling the capture of non-local memory effects and signal aging in graphene-based media without the computational overhead of traditional fractional operators. A dual-optimizer strategy (Adam and L-BFGS) achieves a high-precision convergence floor of O(10-6). The results demonstrate that the fPINN accurately simulates anomalous dispersion phenomena-including pulse broadening and peak attenuation-across fractional orders (α ∈ [0.1, 1.0]), effectively mitigating spectral bias in high-frequency regimes.&amp;nbsp; &lt;/div&gt; &lt;div&gt; Furthermore, the mesh-free, surrogate-model nature of the fPINN enables real-time field prediction with microsecond inference latency,&amp;nbsp;offering a transformative computational acceleration over classical numerical techniques. This synergy of physics-informed deep learning and localized fractional calculus provides a robust, scalable computational tool for real-time optimization in reconfigurable intelligent surfaces and digital twin implementations for next-generation wireless systems. &lt;/div&gt;","author":[{"family":"Ajarmah","given":"Basem"},{"family":"Odeh","given":"Iyad"}],"issued":{"date-parts":[[2026]]},"DOI":"10.2139/ssrn.6161526","URL":"https://doi.org/10.2139/ssrn.6161526","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-10361693/v1","type":"article-journal","title":"A physics-guided graph neural network for interpretable multiscale modelling of concrete properties","abstract":"Abstract Hierarchical materials derive macroscopic performance from interactions among phases and interfaces, yet most machine-learning models represent concrete as a flat mixture vector. Here we develop HiCon-GNN, a physics-guided graph neural network that mirrors the paste–mortar–interfacial transition zone (ITZ)–concrete hierarchy using coupled mortar- and concrete-scale proxy graphs. Mask-aware encoders retain incomplete literature records, while soft priors impose binder composition, weak-interface and cross-property consistency. Trained on 13,170 records, HiCon-GNN jointly predicts compressive, tensile and flexural strength with held-out R 2 values of 0.953, 0.887 and 0.926 and RMSEs of 10.69, 2.00 and 3.20 MPa, respectively. It reduces RMSE relative to a matched tabular neural network by 19%, 5% and 16%, while remaining close to the best target-specific ensemble models and stable across repeated random partitions. The learned latent states recover the expected paste–mortar–concrete strength hierarchy, matrix densification and porous, weak-ITZ behaviour. Capacity decomposition identifies mortar strength as the dominant reference scale and separates interface-controlled reduction from mesoscale matrix and topology effects. The latent quantities are physically regularised hypotheses, while they expose testable multiscale pathways unavailable to structure-blind models. HiCon-GNN therefore combines competitive multi-property prediction with a physically inspectable representation for data-driven concrete design.","author":[{"family":"Li","given":"Ye"},{"family":"Wang","given":"Fangying"}],"issued":{"date-parts":[[2026]]},"DOI":"10.21203/rs.3.rs-10361693/v1","URL":"https://doi.org/10.21203/rs.3.rs-10361693/v1","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-9440511/v1","type":"article-journal","title":"Artificial Neural Network: A tool for Rapid Quantitative Elemental Analysis Using Neutron Activation Analysis","abstract":"Abstract This paper presents a methodology for rapid quantitative elemental analysis using Neutron Activation Analysis (NAA) coupled with an Artificial Neural Network (ANN). A three-layer feed-forward ANN with back-propagation algorithm was developed to determine concentrations of long-lived activation products (Co, Cs, Eu, Fe, Hf, Sb, Sc, Ta, Tb, Ce) relevant to nuclear reactor shielding decommissioning. The methodology is demonstrated using simulated gamma-ray spectral data generated from the activation equation based on typical cement composition ranges reported in literature. The optimized ANN architecture (4 input neurons; 1 hidden layer with 4 neurons using tanh activation; 1 output neuron) achieved a correlation coefficient of 0.991 between input features and predicted concentrations. Mean relative errors ranged from 3.2% to 6.2% across all elements. The proposed method eliminates the need for matched elemental standards and reduces analysis time by approximately 80% compared to conventional relative NAA. This methodology provides a foundation for rapid, multi-element analysis in nuclear decommissioning applications, with experimental validation planned for future work.","author":[{"family":"Medhat","given":"ME"}],"issued":{"date-parts":[[2026]]},"DOI":"10.21203/rs.3.rs-9440511/v1","URL":"https://doi.org/10.21203/rs.3.rs-9440511/v1","source":"crossref"},{"id":"doi:10.21203/rs.3.rs-10274597/v1","type":"article-journal","title":"Repair Instead of Retraining: A Constraint-Guided Framework for Neural Network Repair","abstract":"Abstract Deep neural networks remain vulnerable to adversarial perturbations and backdoor attacks, yet repairing a deployed model without full retraining must balance interpretability, behavioral fidelity, and computational cost. We present a constraint-guided---but not formally certified---framework that uses DeepSHAP to localize a sparse set of fault-relevant weights, collects symbolic path constraints from benign and adversarial executions through concolic testing, and searches for weight updates with Max-SMT. Candidate repairs are always evaluated on the original deployed model, so reported accuracy and attack success rate reflect actual runtime behavior rather than simplified analysis models. Rather than committing to a single fixed recipe, we characterize a configurable repair design space---including which weights to modify, how constraints are collected, and how candidates are ranked---and study it through more than 4{,}700 runs on six benchmarks. Repair success depends critically on jointly choosing the weight direction, the surrogate model used for constraint collection, and the simplification strength. For backdoor benchmarks, outgoing-weight repair with an activation-gated surrogate achieves near-complete backdoor removal on Fashion-MNIST and MNIST-BD (attack success rate from above 98\\% to below 8\\%). For adversarial benchmarks, bias-only repair substantially reduces misbehavior on MNIST-6 (100\\% to 5.81\\%) and ResNet18 (85.51\\% to 40.42\\%). CIFAR-10 and GTSRB remain challenging under single-layer repair, highlighting a tractability--fidelity trade-off. These results show that bias-direction repair is a critical and previously underexplored option for adversarial correction, discoverable only through systematic design-space exploration.","author":[{"family":"Liu","given":"Ting"},{"family":"Yu","given":"Fang"}],"issued":{"date-parts":[[2026]]},"DOI":"10.21203/rs.3.rs-10274597/v1","URL":"https://doi.org/10.21203/rs.3.rs-10274597/v1","source":"crossref"}]