Spiking neural networks (SNNs) operating with asynchronous discrete events show higher energy efficiency sparse computation. A popular approach for implementing deep SNNs is artificial network (ANN)-SNN conversion combining both efficient training of ANNs and inference SNNs. However, the accuracy loss usually nonnegligible, especially under few time steps, which restricts applications SNN on latency-sensitive edge devices greatly. In this article, we first identify that such performance...

Spiking neural networks (SNNs) are widely applied in various fields due to their energy-efficient and fast-inference capabilities. Applying SNNs reinforcement learning (RL) can significantly reduce the computational resource requirements for agents improve algorithm's performance under resource-constrained conditions. However, current spiking (SRL) algorithms, simulation results of multiple time steps only correspond a single-step decision RL. This is quite different from real temporal...

Spiking neural networks (SNNs) are increasingly applied to deep architectures. Recent works developed apply spatio-temporal backpropagation directly train SNNs. But the binary and non-differentiable properties of spike activities force trained SNNs suffer from serious gradient vanishing. In this paper, we first analyze cause vanishing problem identify that gradients mostly backpropagate along synaptic currents. Based on that, modify current equation leaky-integrate-fire neuron model propose...

The temporal credit assignment (TCA) problem, which aims to detect predictive features hidden in distracting background streams, remains a core challenge biological and machine learning. Aggregate-label (AL) learning is proposed by researchers resolve this problem matching spikes with delayed feedback. However, the existing AL algorithms only consider information of single timestep, inconsistent real situation. Meanwhile, there no quantitative evaluation method for TCA problems. To address...

Spiking neural networks (SNNs) operating with asynchronous discrete events show higher energy efficiency sparse computation. A popular approach for implementing deep SNNs is ANN-SNN conversion combining both efficient training of ANNs and inference SNNs. However, the accuracy loss usually non-negligible, especially under a few time steps, which restricts applications SNN on latency-sensitive edge devices greatly. In this paper, we first identify that such performance degradation stems from...

Event cameras, with their high dynamic range and temporal resolution, are ideally suited for object detection, especially under scenarios motion blur challenging lighting conditions. However, while most existing approaches prioritize optimizing spatiotemporal representations advanced detection backbones early aggregation functions, the crucial issue of adaptive event sampling remains largely unaddressed. Spiking Neural Networks (SNNs), which operate on an event-driven paradigm through sparse...

Spiking neural networks (SNNs) are widely applied in various fields due to their energy-efficient and fast-inference capabilities. Applying SNNs reinforcement learning (RL) can significantly reduce the computational resource requirements for agents improve algorithm's performance under resource-constrained conditions. However, current spiking (SRL) algorithms, simulation results of multiple time steps only correspond a single-step decision RL. This is quite different from real temporal...

Given a pair of point clouds, the goal assembly is to recover rigid transformation that aligns one cloud other. This task challenging because clouds may be non-overlapped, and they have arbitrary initial positions. To address these difficulties, we propose method, called SE(3)-bi-equivariant transformer (BITR), based on SE(3)-bi-equivariance prior task: it guarantees when inputs are rigidly perturbed, output will transform accordingly. Due its equivariance property, BITR can not only handle...

It is difficult to automatically recognize defects using digital image processing method in X-ray radiograph that tested from lap joint of unequal thickness plates. The continuous change the wall for workpiece brings about very different gray levels background image. Besides, due shape and fixturing workpiece, distribution weld seam not vertical results an existence a certain angle with direction. This makes it automatic defect detection localization. In this paper, correction based on...

It is difficult to automatically recognize defects using digital image processing methods in X-ray radiographs of lap joints made from plates unequal thickness. The continuous change the wall thickness joint workpiece causes very different gray levels an background image. Furthermore, due shape and fixturing workpiece, distribution weld seam radiograph not vertical which results angle between direction. This makes automatic defect detection localization difficult. In this paper, a method...

Neuromorphic computing holds the promise to achieve energy efficiency and robust learning performance of biological neural systems. To realize promised brain-like intelligence, it needs solve challenges neuromorphic hardware architecture design substrate amicable algorithms with spike-based encoding learning. Here we introduce a spike coding model termed spiketrum, characterize transform time-varying analog signals, typically auditory into computationally efficient spatiotemporal patterns....