A Novel Homeostasis Method to Improve the Learning Efficiency of Spiking Neural Networks

Abstract

Spiking neural networks (SNNs) trained by spike timing dependent plasticity (STDP) is a promising computing paradigm for unsupervised artificial intelligence systems. During the learning procedure of SNNs trained by STDP, homeostasis method is always implemented to mitigate the inhomogeneity induced by inputs and initial synapse weights. If homeostasis is not achieved effectively, the learning efficiency will be affected due to unbalanced training of neurons in the same layer. In this paper, we propose a novel homeostasis method and carry out software simulations to evaluate the learning efficiency and performance of the proposed method compared with two classical SNN learning algorithms. Simulation results on the task of digital recognition of MNIST dataset show that, our proposed method can achieve a ~2 times higher learning efficiency while maintaining comparable performance.