Improving Low-Latency Learning Performance in Spiking Neural Networks via a Change-Perceptive Dendrite-Soma-Axon Neuron

TL;DR

This work tackles the challenge of low-latency, energy-efficient learning in spiking neural networks by introducing the Change-Perceptive Dendrite-Soma-Axon (CP-DSA) neuron. It combines a biologically inspired dendrite-soma-axon structure with a soft reset and a change-perceptive mechanism that leverages adjacent-time-step changes to regulate firing and learning, all trained via STBP with gradient-auxiliary techniques. The CP-DSA framework, including two parallel DSA streams for enhanced and weakened information, yields state-of-the-art or competitive accuracies on neuromorphic datasets at short time steps, with ablations confirming the CP mechanism as the key contributor. The results demonstrate strong low-latency performance and point toward broader potential for energy-efficient neuromorphic hardware implementations.

Abstract

Spiking neurons, the fundamental information processing units of Spiking Neural Networks (SNNs), have the all-or-zero information output form that allows SNNs to be more energy-efficient compared to Artificial Neural Networks (ANNs). However, the hard reset mechanism employed in spiking neurons leads to information degradation due to its uniform handling of diverse membrane potentials. Furthermore, the utilization of overly simplified neuron models that disregard the intricate biological structures inherently impedes the network's capacity to accurately simulate the actual potential transmission process. To address these issues, we propose a dendrite-soma-axon (DSA) neuron employing the soft reset strategy, in conjunction with a potential change-based perception mechanism, culminating in the change-perceptive dendrite-soma-axon (CP-DSA) neuron. Our model contains multiple learnable parameters that expand the representation space of neurons. The change-perceptive (CP) mechanism enables our model to achieve competitive performance in short time steps utilizing the difference information of adjacent time steps. Rigorous theoretical analysis is provided to demonstrate the efficacy of the CP-DSA model and the functional characteristics of its internal parameters. Furthermore, extensive experiments conducted on various datasets substantiate the significant advantages of the CP-DSA model over state-of-the-art approaches.

Figures (5)

• Figure 1: Illustration of the CP-DSA model. (a) Change-Perception mechanism. $\varepsilon$ denotes the step function. The subtraction operation determines the decrement and subtraction according to the CP mechanism. (b) Dendrite-Soma-Axon model. $S$ denotes output spike function.
• Figure 2: Evolution of the STBP training structure from soft-reset LIF neuron to DSA neuron.
• Figure 3: The impact of the change perceptive mechanism on the DSA training process.
• Figure 4: The performance of the CP-DSA model at different time steps.The circles correspond to experimentally measured data, which demonstrate that, even at very low time steps, CP-DSA still outperforms other methods listed in Table \ref{['tab4']}.
• Figure 5: The hyperparameter variation tracking curve of the CP-DSA neurons. Labels are used to distinguish neurons in different layers, which are represented by curves of different colors in the graph. DSA0 corresponds to the neurons in the direct encoding layer, while DSA1-1 represents the first neuron in the sequence of the first convolutional structure. These hyperparameters gradually converge after 100 generations.