Cross Knowledge Distillation between Artificial and Spiking Neural Networks

TL;DR

This paper introduces a novel cross knowledge distillation method that enhances Spiking Neural Networks' performance on event-based data by leveraging well-performing Artificial Neural Networks and addressing cross-modality and cross-architecture challenges.

Contribution

It proposes a new cross knowledge distillation approach that effectively bridges the gap between ANNs and SNNs across different data formats and architectures.

Findings

Outperforms current State-of-the-Art methods on neuromorphic datasets

Effectively mitigates cross-modality challenges in SNN training

Improves SNN performance using knowledge transfer from ANNs

Abstract

Recently, Spiking Neural Networks (SNNs) have demonstrated rich potential in computer vision domain due to their high biological plausibility, event-driven characteristic and energy-saving efficiency. Still, limited annotated event-based datasets and immature SNN architectures result in their performance inferior to that of Artificial Neural Networks (ANNs). To enhance the performance of SNNs on their optimal data format, DVS data, we explore using RGB data and well-performing ANNs to implement knowledge distillation. In this case, solving cross-modality and cross-architecture challenges is necessary. In this paper, we propose cross knowledge distillation (CKD), which not only leverages semantic similarity and sliding replacement to mitigate the cross-modality challenge, but also uses an indirect phased knowledge distillation to mitigate the cross-architecture challenge. We validated our method on main-stream neuromorphic datasets, including N-Caltech101 and CEP-DVS. The experimental results show that our method outperforms current State-of-the-Art methods. The code will be available at https://github.com/ShawnYE618/CKD