Neuromorphic Data Augmentation for Training Spiking Neural Networks

TL;DR

This paper introduces Neuromorphic Data Augmentation (NDA), a geometric augmentation technique designed for event-based datasets, which stabilizes SNN training, reduces overfitting, and enables unsupervised contrastive learning, leading to significant accuracy improvements.

Contribution

The paper proposes NDA, a novel data augmentation method tailored for event-based datasets, enhancing SNN training stability and performance, and demonstrating the first use of unsupervised contrastive learning with SNNs.

Findings

NDA improves SNN accuracy on CIFAR10-DVS by 10.1%.

NDA improves SNN accuracy on N-Caltech 101 by 13.7%.

NDA enables unsupervised contrastive learning for SNNs.

Abstract

Developing neuromorphic intelligence on event-based datasets with Spiking Neural Networks (SNNs) has recently attracted much research attention. However, the limited size of event-based datasets makes SNNs prone to overfitting and unstable convergence. This issue remains unexplored by previous academic works. In an effort to minimize this generalization gap, we propose Neuromorphic Data Augmentation (NDA), a family of geometric augmentations specifically designed for event-based datasets with the goal of significantly stabilizing the SNN training and reducing the generalization gap between training and test performance. The proposed method is simple and compatible with existing SNN training pipelines. Using the proposed augmentation, for the first time, we demonstrate the feasibility of unsupervised contrastive learning for SNNs. We conduct comprehensive experiments on prevailing neuromorphic vision benchmarks and show that NDA yields substantial improvements over previous state-of-the-art results. For example, the NDA-based SNN achieves accuracy gain on CIFAR10-DVS and N-Caltech 101 by 10.1% and 13.7%, respectively. Code is available on GitHub https://github.com/Intelligent-Computing-Lab-Yale/NDA_SNN