SSTDP: Supervised Spike Timing Dependent Plasticity for Efficient Spiking Neural Network Training.

Fangxin Liu,Zongwu Wang,Tao Yang,Li Jiang,Wenbo Zhao,Yongbiao Chen

doi:10.3389/fnins.2021.756876

Abstract

Spiking Neural Networks (SNNs) are a pathway that could potentially empower low-power event-driven neuromorphic hardware due to their spatio-temporal information processing capability and high biological plausibility. Although SNNs are currently more efficient than artificial neural networks (ANNs), they are not as accurate as ANNs. Error backpropagation is the most common method for directly training neural networks, promoting the prosperity of ANNs in various deep learning fields. However, since the signals transmitted in the SNN are non-differentiable discrete binary spike events, the activation function in the form of spikes presents difficulties for the gradient-based optimization algorithms to be directly applied in SNNs, leading to a performance gap (i.e., accuracy and latency) between SNNs and ANNs. This paper introduces a new learning algorithm, called SSTDP, which bridges the gap between backpropagation (BP)-based learning and spike-time-dependent plasticity (STDP)-based learning to train SNNs efficiently. The scheme incorporates the global optimization process from BP and the efficient weight update derived from STDP. It not only avoids the non-differentiable derivation in the BP process but also utilizes the local feature extraction property of STDP. Consequently, our method can lower the possibility of vanishing spikes in BP training and reduce the number of time steps to reduce network latency. In SSTDP, we employ temporal-based coding and use Integrate-and-Fire (IF) neuron as the neuron model to provide considerable computational benefits. Our experiments show the effectiveness of the proposed SSTDP learning algorithm on the SNN by achieving the best classification accuracy 99.3% on the Caltech 101 dataset, 98.1% on the MNIST dataset, and 91.3% on the CIFAR-10 dataset compared to other SNNs trained with other learning methods. It also surpasses the best inference accuracy of the directly trained SNN with 25~32× less inference latency. Moreover, we analyze event-based computations to demonstrate the efficacy of the SNN for inference operation in the spiking domain, and SSTDP methods can achieve 1.3~37.7× fewer addition operations per inference. The code is available at: https://github.com/MXHX7199/SNN-SSTDP.

Highlights

Deep neural networks have made tremendous progress and become a prevalent tool for performing various cognitive tasks such as object recognition (Simonyan and Zisserman, 2015; Sandler et al, 2018), natural language processing (Devlin et al, 2018; Radford et al, 2019), and self-driving (Nedevschi et al, 2012; Liu et al, 2017), etc
To further evaluate the effectiveness of our proposed SSTDP learning algorithm, as shown in Table 2, we compared the top-1 accuracy of SSTDP with recent works that directly train the Spiking Neural Networks (SNNs) from scratch based on BP on the MNIST datasets
We found that our proposed SSTDP achieves better network performance than others in terms of accuracy and latency

Summary

Introduction

Deep neural networks have made tremendous progress and become a prevalent tool for performing various cognitive tasks such as object recognition (Simonyan and Zisserman, 2015; Sandler et al, 2018), natural language processing (Devlin et al, 2018; Radford et al, 2019), and self-driving (Nedevschi et al, 2012; Liu et al, 2017), etc. Spiking Neural Networks (SNNs) are often regarded as third-generation braininspired neural networks, and represent one of the leading candidates for overcoming computational constraints and efficiently exploiting deep learning algorithms in real (or mobile) applications whilst being highly power-efficient (Deng et al, 2020; Rathi et al, 2020; Taherkhani et al, 2020). Training highperformance SNNs with competitive classification accuracy and less latency is a nontrivial problem, limiting their scalability in complex applications (Benjamin et al, 2014; Roy et al, 2019; Sengupta et al, 2019; Comsa et al, 2020; Han et al, 2020; Deng et al, 2021)

Methods

Results

Conclusion