Brain-inspired chaotic spiking backpropagation  

作　　者：Zijian Wang Peng Tao Luonan Chen 

机构地区：[1]Key Laboratory of Systems Health Science of Zhejiang Province,School of Life Science,Hangzhou Institute for Advanced Study,University of Chinese Academy of Sciences,Hangzhou 310024,China [2]Key Laboratory of Systems Biology,Shanghai Institute of Biochemistry and Cell Biology,Center for Excellence in Molecular Cell Science,Chinese Academy of Sciences,Shanghai 200031,China [3]Guangdong Institute of Intelligence Science and Technology,Hengqin,Zhuhai 519031,China [4]Pazhou Laboratory(Huangpu),Guangzhou 510555,China

出　　处：《National Science Review》2024年第6期162-172,共11页国家科学评论（英文版）

基　　金：supported by the National Natural Science Foundation of China(T2350003,T2341007,12131020 and 31930022);the Special Fund for Science and Technology Innovation Strategy of Guangdong Province(2021B0909050004 and 2021B0909060002);the National Key R&D Program of China(2022YFA1004800);the Strategic Priority Research Program of the Chinese Academy of Sciences(XDB38040400);JST MoonshotR&D(JPMJMS2021).

摘　　要：Spiking neural networks (SNNs) have superior energy efficiency due to their spiking signal transmission,which mimics biological nervous systems, but they are difficult to train effectively. Although surrogategradient-based methods offer a workable solution, trained SNNs frequently fall into local minima becausethey are still primarily based on gradient dynamics. Inspired by the chaotic dynamics in animal brainlearning, we propose a chaotic spiking backpropagation (CSBP) method that introduces a loss function togenerate brain-like chaotic dynamics and further takes advantage of the ergodic and pseudo-random natureto make SNN learning effective and robust. From a computational viewpoint, we found that CSBPsignificantly outperforms current state-of-the-art methods on both neuromorphic data sets (e.g.DVS-CIFAR10 and DVS-Gesture) and large-scale static data sets (e.g. CIFAR100 and ImageNet) in termsof accuracy and robustness. From a theoretical viewpoint, we show that the learning process of CSBP isinitially chaotic, then subject to various bifurcations and eventually converges to gradient dynamics,consistently with the observation of animal brain activity. Our work provides a superior core tool for directSNN training and offers new insights into understanding the learning process of a biological brain.

关 键 词：spiking neural networks surrogate gradient local minima backpropagation brain-inspired learning CHAOS 

分 类 号：TP212[自动化与计算机技术—检测技术与自动化装置]