Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

STCSNN: High energy efficiency spike-train level spiking neural networks with spatio-temporal conversion

Changqing Xu, Yi Liu, Yintang Yang

TL;DR

The experiment results show that the proposed STCSNN outperforms the state-of-the-art accuracy on nearly all datasets, using fewer time steps and being highly energy-efficient.

Abstract

Brain-inspired spiking neuron networks (SNNs) have attracted widespread research interest due to their low power features, high biological plausibility, and strong spatiotemporal information processing capability. Although adopting a surrogate gradient (SG) makes the non-differentiability SNN trainable, achieving comparable accuracy for ANNs and keeping low-power features simultaneously is still tricky. In this paper, we proposed an energy-efficient spike-train level spiking neural network with spatio-temporal conversion, which has low computational cost and high accuracy. In the STCSNN, spatio-temporal conversion blocks (STCBs) are proposed to keep the low power features of SNNs and improve accuracy. However, STCSNN cannot adopt backpropagation algorithms directly due to the non-differentiability nature of spike trains. We proposed a suitable learning rule for STCSNNs by deducing the equivalent gradient of STCB. We evaluate the proposed STCSNN on static and neuromorphic datasets, including Fashion-Mnist, Cifar10, Cifar100, TinyImageNet, and DVS-Cifar10. The experiment results show that our proposed STCSNN outperforms the state-of-the-art accuracy on nearly all datasets, using fewer time steps and being highly energy-efficient.

STCSNN: High energy efficiency spike-train level spiking neural networks with spatio-temporal conversion

TL;DR

The experiment results show that the proposed STCSNN outperforms the state-of-the-art accuracy on nearly all datasets, using fewer time steps and being highly energy-efficient.

Abstract

Brain-inspired spiking neuron networks (SNNs) have attracted widespread research interest due to their low power features, high biological plausibility, and strong spatiotemporal information processing capability. Although adopting a surrogate gradient (SG) makes the non-differentiability SNN trainable, achieving comparable accuracy for ANNs and keeping low-power features simultaneously is still tricky. In this paper, we proposed an energy-efficient spike-train level spiking neural network with spatio-temporal conversion, which has low computational cost and high accuracy. In the STCSNN, spatio-temporal conversion blocks (STCBs) are proposed to keep the low power features of SNNs and improve accuracy. However, STCSNN cannot adopt backpropagation algorithms directly due to the non-differentiability nature of spike trains. We proposed a suitable learning rule for STCSNNs by deducing the equivalent gradient of STCB. We evaluate the proposed STCSNN on static and neuromorphic datasets, including Fashion-Mnist, Cifar10, Cifar100, TinyImageNet, and DVS-Cifar10. The experiment results show that our proposed STCSNN outperforms the state-of-the-art accuracy on nearly all datasets, using fewer time steps and being highly energy-efficient.
Paper Structure (26 sections, 11 equations, 9 figures, 4 tables)

This paper contains 26 sections, 11 equations, 9 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. ANN-to-SNN conversion
  4. Direct training
  5. Methodology
  6. Spatio-temporal conversion block (STCB)
  7. Spike train encoding block(STEB) and Current accumulator block (CAB)
  8. Spike train encoding block(STEB)
  9. Current accumulator Block (CAB)
  10. Error Backpropagation of STCB
  11. Network Architecture
  12. Experiments and results
  13. Experiment settings
  14. Datasets and training details
  15. FashionMnist
  16. ...and 11 more sections

Figures (9)

  • Figure 1: Spatio-temporal conversion block (STCB)
  • Figure 2: The structure of STEB and CAB. (a) STEB (b) CAB
  • Figure 3: Forward and backward pass of STCB
  • Figure 4: Diagram of VGG and ResNet based on STCB. (a) VGG, (b) ResNet. Cov refers to the convolutional layer, $N\times N$ Cov/STCB means the Convolutional layer or STCB with $N \times N$ filters, BN means Batch Norm, AP is the average pooling layer, AAP is the adaptive average pool, and FC refers to the fully connected layers.
  • Figure 5: Performance analysis. (a) Initial threshold voltages (b) length of spike trains
  • ...and 4 more figures