Title: Deep spiking neural networks for image classification

Authors: Zhuo Li; Lin Meng

Addresses: Graduate School of Science and Engineering, Ritsumeikan University, Kusatsu, Shiga, Japan ' College of Science and Engineering, Ritsumeikan University, Kusatsu, Shiga, Japan

Abstract: Artificial intelligence (AI) has made great progress with the foundations of computer science. Furthermore, brain-like computing, which simulates the human brain through computers, has been a hot research area. As an important part of brain-like computing, spiking neural networks (SNNs) have received great attention due to their high biological plausibility and low energy consumption. The technique is very suitable for the research and implementation of brain-like computing and provides promising solutions for image classification tasks on resource-constrained mobile devices. However, it typically suffers from severe performance degradation and network architecture complexity. In this paper, we first design a deep spiking neural network with a simple architecture. Then, we propose an improved back-propagation algorithm for the approximate derivation of the spiking neural network. The experimental results show that the proposal achieves recognition accuracy of 98.43%, 96.70%, 98.81% and 93.61% respectively on the image classification datasets MNIST, Fashion-MNIST, Kuzushiji-MNIST and CIFAR-10. The number of parameters and the recognition accuracy show that the proposals have great advantages for image classification. This research has important implications for the implementation of intelligent brain-like computing.

Keywords: brain-like computing; spiking neural networks; SNNs; image classification; approximate derivation.

DOI: 10.1504/IJHFMS.2023.130146

International Journal of Human Factors Modelling and Simulation, 2023 Vol.8 No.1, pp.21 - 35

Received: 21 Jun 2022
Accepted: 04 Sep 2022
Published online: 05 Apr 2023 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article