Search Results (540)

Spiking neural networks (SNNs) are well suited for modeling temporally evolving information due to their event-driven and dynamic neuronal mechanisms. Nevertheless, the majority of existing SNN topologies are constructed through algorithmic procedures rather than guided by constraints from biological brain connectivity, which weakens their biological plausibility. In our earlier work, we developed a spiking neural network (SNN) by incorporating topological information from functional brain networks extracted from functional magnetic resonance imaging (fMRI) data of healthy individuals, and named the resulting model fMRISNN. Nevertheless, the fMRI data used in previous work were resting-state fMRI. Compared with resting-state fMRI, task-state fMRI can capture brain-region coordination patterns induced by specific task stimuli, and the resulting functional brain network is therefore more closely related to the corresponding task. Motivated by this advantage, this study replaces the resting-state topology used in previous fMRISNN studies with a task-state, number/digit-related fMRI topology and validates the resulting Task-fMRISNN on handwritten digit recognition. The experimental results demonstrate that the proposed Task-fMRISNN outperforms the Rest-fMRISNN in terms of recognition accuracy, lesion robustness, and noise robustness. In addition, the Task-fMRISNN achieves significantly better performance than several baseline models constructed using algorithmically generated topologies. While deep convolutional neural networks (CNNs) may deliver superior absolute recognition performance, the proposed fMRISNN provides a more compact model structure and shows potential resource-efficiency advantages due to its sparse and event-driven computational characteristics. Full article

Price spikes are rare but economically significant events observed across electricity, financial, commodity, and cryptocurrency markets. Their abrupt magnitude, heavy-tailed distributions, and severe class imbalance make them difficult to forecast using conventional time-series methods. This systematic literature review, conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) framework, synthesises recent deep learning approaches to forward-looking price-spike prediction and classification. Searches of Scopus, Web of Science, and IEEE Xplore identified studies published between 2020 and 2026. Following screening and full-text eligibility assessment of approximately 300 studies, only 20 met the inclusion criteria and were included in the final synthesis, comprising 19 peer-reviewed papers and one doctoral thesis. The review develops a structured taxonomy spanning spike definitions, task formulations, model architectures, input design, and evaluation practices. A central finding is that predictive performance is driven more by problem formulation, label construction, and evaluation design than by model architecture. While architectures have diversified to include recurrent networks, transformers, graph neural networks, and hybrid frameworks, improvements are often attributable to differences in how the prediction problem is defined rather than the models themselves. Key limitations stem from inconsistent spike definitions and insufficient treatment of class imbalance, leading to a misalignment between modelling objectives and evaluation practices, further exacerbated by the absence of standardised benchmarks. These issues hinder comparability and can lead to overstated model performance by masking poor detection of rare but economically critical spike events. The review therefore identifies clear directions for future research, including standardised spike labelling, adoption of rare-event-appropriate evaluation frameworks, and problem formulations that explicitly target extreme-event prediction. Full article

Wearable ECG monitors enable continuous cardiac surveillance, but most still rely on cloud-based analysis with limited on-device support for multi-class arrhythmia detection. Spiking neural networks (SNNs) are promising for low-power edge inference, yet it remains unclear how class-imbalance loss design interacts with RR-interval features in directly trained quantized SNNs, and FPGA validation in this setting is largely unexplored. We propose a quantized convolutional spiking neural network (QCSNN) for real-time arrhythmia detection on resource-constrained hardware. The model uses a dual-head architecture that jointly trains binary and four-class classifiers, subsequently reorganized into a cascaded pipeline that routes only abnormal beats to the second stage. At inference, beats classified as Normal exit at Stage 1; only beats classified as Abnormal are routed to the four-class head, so the bulk of the inference cost is absorbed by Stage 1. We evaluate two loss functions, Cross-Entropy and Focal Loss, under four RR-feature routing strategies. Without RR features, Focal Loss improves macro F1 by 2.3–2.5% over Cross-Entropy (mean Δ = +0.013 in Stage-2 macro F1; Wilcoxon two-sided p = 0.031). With RR features, this advantage largely disappears (Wilcoxon two-sided p ≥ 0.219 at all RR routings); meanwhile, RR features at the strongest routing improve Stage-2 macro F1 by +0.028 to +0.034 depending on loss function—a gain that exceeds the entire Focal-Loss-over-Cross-Entropy advantage, suggesting that RR features provide discriminative information that compensates for class imbalance at the input level. Based on clinically prioritized sensitivity, the CE:RR→Both configuration was deployed on a PYNQ-Z2 FPGA, achieving 99.02% cascaded accuracy, 11.54 ms per-beat latency, and 0.33 W accelerator power—a 31.66× power reduction and 4.01× energy reduction versus GPU inference, within 1% macro F1. These results demonstrate quantized SNNs as a practical solution for real-time edge arrhythmia monitoring that operates independently of cloud connectivity—removing the network-dependent latency, connectivity-dropout failure modes, and continuous-transmission energy burden that constrain current wearable monitors and, to our knowledge, represent one of the first systematic studies of loss-function/RR-feature interactions in directly trained SNN arrhythmia classification and one of the first FPGA deployments of a fully quantized, directly trained SNN for multi-class ECG arrhythmia detection. All code generated and used in this study has been made publicly available. Full article

Microsatellites operate in highly dynamic thermal environments due to severe physical constraints, making temperature telemetry a critical onboard health indicator. Conventional threshold-based monitoring fails to distinguish normal operational mode transitions from genuine faults, causing excessive false alarms. To address this, we propose a two-stage framework integrating unsupervised thermal mode discovery with mode-specific deep learning prediction. Raw temperature telemetry is downsampled and segmented into orbital cycles. Unsupervised clustering identifies two nominal thermal regimes and four canonical fault-type libraries (step, spike, drift, and noise), each corresponding to distinct in-orbit failure mechanisms. For each nominal mode, a Convolutional Neural Network–Long Short-Term Memory (CNN-LSTM) is trained on 7-day historical windows to forecast 3-day temperature evolution. Post-downlink, incoming cycle mode is inferred via nearest-neighbor DTW classification; anomalies are flagged when prediction residuals exceed mode-adaptive thresholds. Validation on Macau Science Satellite-1B (MSS-1B, COSPAR 2023-069-B, NORAD 56732) in-orbit telemetry from a 41° inclination low-Earth orbit—where solar illumination dominates external thermal loading and internal heat from the data-communication module and scientific payload constitutes the primary internal thermal source—shows the method reduces anomaly flags by 96.6% and improves prediction mean absolute error by 51.3% compared to a non-classified global baseline under nominal operating conditions, correctly detecting a known operational transient while suppressing spurious alarms. A synthetic fault injection experiment with four anomaly types and five baseline methods further confirms the framework’s detection capability, achieving an overall F1 score of 0.725 vs. 0.258 for the global baseline—a 2.8× improvement driven primarily by a 4× precision gain. Sensitivity analysis reveals that the two-stage advantage is most pronounced for low-magnitude and short-duration faults, where mode-specific context is essential. This work advances microsatellite autonomous health management by providing reliable anomaly detection with quantified fault detection performance. Full article

Machine learning’s computational demands necessitate optimal performance and utilization across diverse hardware architectures. This research compares computing as spiking neural networks (CSNNs, or simulated neuromorphic computing) and regular CNNs on Apple Silicon M3 Pro with Metal Performance Shaders (MPS), and NVIDIA RTX 3070 GPU with CUDA. We run Convolutional Spiking Neural Networks (CSNNs) and traditional CNNs on two datasets (frame-based CIFAR-10; and sequential event-based DVS) to evaluate the suitability of neural net architectures and platforms for different data problems. For both CSNNs and traditional CNNs, Apple Silicon with MPS delivers better energy efficiency but longer processing times for training and inference. NVIDIA with CUDA offers faster computation in training and inference at higher energy costs for CNNs. For CSNNs, frame-based data (CIFAR-10) significantly degraded performance when proper temporal encoding was absent, while event-based data (DVS) proved more naturally suited to the CSNN architecture than frame-based inputs. Though CNNs still achieved higher empirical accuracy in the reported experiments. CSNNs also performed better on Apple Silicon (with MPS) for the sequential event-based data. RAM utilization patterns favored Apple Silicon (with MPS) across both data experiments. The CSNN architecture demanded higher memory resources than CNN, regardless of platform and dataset. NVIDIA (with CUDA) was less energy efficient for spiking neural networks (CSNNs) as compared to Apple Silicon (with MPS). We also compared how the number of time steps affects accuracy and energy consumption across hardware platforms, finding that higher accuracy correlates with energy costs as time steps increase; the accuracy-energy relation seems linear for frame-based data, while for event-based data the energy consumption remains stable increasing at higher time steps. Our cross-platform performance analysis of spiking and regular neural network architectures highlight the importance of matching platform-architecture combinations to a dataset and application requirements. Full article

Recently, artificial neural networks (ANNs) have shown impressive performance in the compressive hyperspectral imaging (CHI) reconstruction task, but the high energy consumption limits their deployment on energy-constrained devices. This paper develops a novel spiking neural network (SNN), termed spiking spectral-weighting reconstruction network (SSWR-Net), to significantly improve the energy–efficiency ratio in CHI reconstruction. Firstly, a spiking spectral-weighting convolution block is proposed to adaptively modulate the spiking signals, enabling the SNN to fit continuous spectral correlation curves. Secondly, a residual feature reuse module with more direct connections is designed to achieve efficient and lightweight spatial–spectral feature extraction. Thirdly, customized feature scaling architectures are introduced to resolve the dimensional mismatch issue and enhance information flow. Finally, we propose a novel temporal-wise progressive training method to optimize the multi-timestep SSWR-Net, which can significantly improve both training efficiency and reconstruction quality. Both simulation and real experiments demonstrate the superiority of the proposed method in both CHI reconstruction performance and energy efficiency. Specifically, SSWR-Net outperforms its ANN-based counterpart by 0.87 dB at a 19.74% energy cost. Full article

Safe and intelligent collision avoidance technology is essential for the autonomous underwater vehicle (AUV) to navigate in underwater environments. Most existing spike methods are constrained by a fixed static threshold and are unable to dynamically adjust to threshold changes reasonably, leading to difficulties in robustly adapting to external dynamic interference and thus resulting in insufficient homeostasis and generalization. To address these limitations, inspired by the dynamic threshold changes in biological neural systems, a bioinspired dynamic adaptive threshold (BDAT) is proposed. Combining the spiking neural network with deep reinforcement learning, a novel bioinspired dynamic adaptive threshold planner (BDAT-Planner) framework is constructed for underwater dynamic collision avoidance tasks performed by AUVs in complex, unknown environments. The proposed BDAT-Planner consists of the spiking dynamic adaptive actor network (SDAAN) and the deep critic normal network (DCNN). The BDAT is deployed to each spiking neuron in the SDAAN, dynamically adjusting the spike firing rate through threshold changes and avoiding excessive excitation or inhibition, thus maintaining homeostasis. The spiking encoder and spiking decoder are designed to convert continuous information and spiking sequences. Experimental results from both the training process and evaluation process (ablation studies, comparison experiments, and homeostasis experiments) demonstrate that the proposed BDAT-Planner has achieved superior performance in dynamic collision avoidance and model homeostasis compared to static threshold methods and existing comparison methods. The novel idea of bioinspired dynamic adaptive threshold can maintain model homeostasis and effectively enhance its adaptability to external dynamic interference, which offers significant development potential for promoting the efficient and stable operation of AUVs in marine environments. Full article

Background: Spiking neural networks (SNNs) have attracted significant attention in the field of brain–computer interfaces owing to their distinctive biological plausibility and energy efficiency advantages. However, the discrete nature of spikes renders gradient-based differentiation infeasible, making it difficult to directly obtain well-trained SNNs. A common approach is to transfer the weights from artificial neural networks (ANNs) to SNNs. However, this process introduces conversion errors that pose significant challenges. Methods: To address these challenges, we propose the self-rectifying integrate-and-fire (SRIF) neuron, which employs negative spikes to reduce asynchronism error and rectification spikes to diminish clipping error. Concomitantly, we propose a collaborative trim (CT) training framework that introduces a quantized network to perceive the weights and results of SNNs, which can further improve performance. Result: The proposed training methodology enables SNNs to achieve performance metrics comparable to those of ANNs in EEG-based motor imagery (MI) classification. Conclusions: Experimental results demonstrate that our method not only preserves the superior classification performance of ANNs but also leverages the superior energy efficiency and lower computational complexity of SNNs. Full article

Collision detection in dynamic environments demands perception systems that are both computationally efficient and robust to diverse motion patterns. Biological vision systems, particularly those of insects, offer efficient neural architectures capable of rapid motion interpretation under strict resource constraints. This work presents a systematic comparative evaluation of three biologically inspired models: the Lobula Giant Movement Detector (LGMD), the Elementary Motion Detector (EMD), and a hybrid Spiking Neural Network (SNN) incorporating LGMD and EMD-derived motion processing pathways, evaluated on programmatically generated synthetic stimuli with frame-level ground truth. The hybrid SNN achieved an accuracy of 73–87% across stimulus types, consistently exceeding the 75.0% held-out test set baseline, with a precision of 1.0 throughout and a substantially lower runtime than the LGMD implementation. LGMD demonstrated rate-based sensitivity consistent with biological spike-frequency adaptation, while the EMD correctly produced near-zero responses to looming stimuli, confirming its role as a directional rather than collision detector. These results demonstrate that hybridising biologically inspired motion detectors within a trainable spiking framework produces a promising and reproducible approach to collision prediction, while identifying the sim-to-real generalisation gap as a key challenge for future deployment. Full article

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. Full article

Spiking neural networks (SNNs) have emerged as a promising paradigm for energy-efficient neuromorphic computing, particularly when processing asynchronous event streams from dynamic vision sensors (DVSs). However, SNNs often suffer from limited representational capacity and suboptimal feature recalibration compared to their artificial counterparts. To address these challenges, we propose SE-SNN, a novel architecture that integrates Squeeze-and-Excitation (SE) blocks into deep residual SNNs, enabling channel-wise attention without spike generation. Furthermore, we introduce a Robust Parametric Leaky Integrate-and-Fire (RobustPLIF) neuron model with learnable membrane time constant ($\tau$) and firing threshold ($v_{\mathrm{th}}$), allowing adaptive temporal dynamics in each layer. Our model is trained on the CIFAR10-DVS dataset.The experimental results demonstrate that SE-SNN achieves an accuracy of $78.8$% on CIFAR10-DVS with 16 time steps, outperforming baseline SNNs while maintaining biological plausibility and hardware efficiency. Ablation studies confirm the individual contributions of the SE blocks and learnable neuron parameters to the performance gains. Full article

Integrating on-device learning into autonomous systems requires neural network frameworks that achieve both high energy efficiency and low latency. While spiking neural networks (SNNs) provide a promising event-driven paradigm, implementing hardware-efficient learning remains a challenge due to the computational overhead of error signaling and global gradients. This paper introduces a hardware-oriented hierarchical spiking predictive coding (SPC) framework designed for end-to-end event-driven systems. The proposed architecture implements an implicit prediction error encoding mechanism through local lateral and supervisory feedback connections, eliminating the need for dedicated error-storage memory or complex inter-layer error communication. The entire framework is structured and parameterized for physical implementation, utilizing digital-aligned simulations and arithmetic operations. We evaluate the system on neuromorphic datasets using a fixed 1 ms temporal resolution to mirror real-time hardware constraints. Experimental results demonstrate that the SPC framework can effectively identify stimuli from transient event streams, achieving stable on-device learning. Our work provides a practical path toward deploying low-power, scalable hierarchical spiking networks in resource-constrained environments. Full article

Neuromorphic technologies are attracting increasing interest in neuroengineering, as they provide an event-driven, spike-based computational framework that is well suited to temporally structured, sparse, and resource-constrained biological systems. Compared with conventional computing pipelines, neuromorphic approaches enable tighter integration of sensing, encoding, inference, feedback, and actuation under low-power and low-latency conditions. These features make them particularly relevant for wearable, implantable, and other edge-native neuroengineering applications. This review examines neuromorphic neuroengineering from four closely related perspectives: neuromorphic neurostimulation and adaptive actuation; tactile and sensory biointerfaces; spiking neural network (SNN)-based biosignal processing and state decoding; and wearable or implantable neuromorphic platforms. Across these domains, we highlight how neuromorphic systems may facilitate edge-native, closed-loop architectures that operate closer to the body and respond selectively to meaningful state changes. Neurorehabilitation is further discussed as an important translational context, as it involves long-term use, multimodal sensing, adaptive intervention, and substantial real-world deployment constraints. At present, however, the evidence base remains fragmented and is still largely dominated by device demonstrations and proof-of-concept studies rather than robust translational validation. Overall, neuromorphic approaches offer a promising systems-level pathway toward neuroengineering platforms that are not only computationally efficient but also adaptive, deployable, and responsive in real-world settings. Full article

Memristors, as programmable resistive switching devices, offer two fundamental computational modalities for artificial intelligence: static conductance for parallel data processing and dynamic switching for temporal, logical, and stochastic operations. This Review systematically distinguishes these two modalities and evaluates their respective hardware implementations. In terms of our review scope, we first examine how static conductance modality is exploited in analog matrix computing, which encompasses matrix–vector multiplication and matrix equation solving, and discuss how these primitives enable efficient neural network inference and training. Second, we survey dynamic switching modality and its algorithmic applications, including stateful logic for digital in-memory acceleration, attractor networks for associative memory, reservoir computing and spatiotemporal signal processing using transient device dynamics, biologically inspired spike-timing-dependent plasticity, and stochastic computation. In addition, we discuss key challenges such as device variability, stochastic switching, interconnect parasitics, peripheral circuit overhead, and endurance limitations. We also highlight opportunities for future development, emphasizing algorithm–hardware co-design to leverage application-specific error tolerance and mitigate device non-idealities. Finally, we outline promising research directions aimed at realizing robust, scalable, and energy-efficient memristor-based AI systems. Full article

Radio Frequency Fingerprinting (RFF) enables passive physical-layer device authentication by exploiting unintentional hardware variations in wireless transmitters. Neuromorphic implementations are attractive, given their potential for low-latency, energy-efficient inference capability under Size, Weight, and Power (SWaP) constraints at the edge. A new RFF capability is demonstrated here using recently introduced Radio Frequency Resonate-and-Fire (RF-RAF) neurons and eight WirelessHART devices. Performance is evaluated for RF-RAF-generated fingerprints against the established Gabor Transform (GTX) baseline using three classifier architectures: Random Forest (RndF), Convolutional Neural Network (CNN), and a Time-Incremented Spiking Neural Network (TI-SNN). The results show that RF-RAF fingerprints achieve an average classification accuracy of 96.7% across all three classifier types and consistently outperform GTX fingerprints at all evaluated fingerprint sizes. This performance persists under time-span-matched conditions, and the RF-RAF versus GTX benefit is not solely attributable to input data utilization. The TI-SNN surpasses 94% classification accuracy using M = 4 time step RF-RAF fingerprints with approximately 100 spikes per inference—a 4× larger GTX fingerprint requires approximately 1000 spikes to achieve the same classification accuracy. RF-RAF fingerprints offer two additional benefits: they are natively non-negative, which supports efficient neuromorphic hardware implementation, and they provide greater flexibility in fingerprint size selection. It is concluded that RF-RAF neurons provide an efficient neuromorphic-native encoding pathway for device RFF discrimination and offer improved accuracy–efficiency tradeoffs in training and inference for various classifier architectures. Full article