Rethinking the role of normalization and residual blocks for spiking neural networks

TL;DR

This paper introduces a novel normalization method called postsynaptic potential normalization for deep spiking neural networks, enabling more effective training of very deep models by controlling spike firing.

Contribution

The paper proposes a simple normalization technique that improves training stability and depth scalability of SNNs, surpassing previous normalization methods.

Findings

Outperforms other normalization techniques in SNNs

Enables training of over 100 layers in deep SNNs

Improves spike firing control and training stability

Abstract

Biologically inspired spiking neural networks (SNNs) are widely used to realize ultralow-power energy consumption. However, deep SNNs are not easy to train due to the excessive firing of spiking neurons in the hidden layers. To tackle this problem, we propose a novel but simple normalization technique called postsynaptic potential normalization. This normalization removes the subtraction term from the standard normalization and uses the second raw moment instead of the variance as the division term. The spike firing can be controlled, enabling the training to proceed appropriating, by conducting this simple normalization to the postsynaptic potential. The experimental results show that SNNs with our normalization outperformed other models using other normalizations. Furthermore, through the pre-activation residual blocks, the proposed model can train with more than 100 layers without other special techniques dedicated to SNNs.