Towards Robust Spiking Neural Networks:Mitigating Heterogeneous Training Vulnerability via Dominant Eigencomponent Projection

TL;DR

This paper identifies a vulnerability in training Spiking Neural Networks with direct encoding and BPTT, and proposes a novel, hyperparameter-free method called DEP that improves robustness by reducing the Hessian spectral radius.

Contribution

The paper introduces DEP, a new gradient projection technique that enhances SNN robustness against data heterogeneity and training vulnerabilities.

Findings

DEP reduces Hessian spectral radius effectively.

DEP improves robustness against data poisoning.

SNNs trained with DEP outperform baselines in robustness.

Abstract

Spiking Neural Networks (SNNs) process information via discrete spikes, enabling them to operate at remarkably low energy levels. However, our experimental observations reveal a striking vulnerability when SNNs are trained using the mainstream method--direct encoding combined with backpropagation through time (BPTT): even a single backward pass on data drawn from a slightly different distribution can lead to catastrophic network collapse. Our theoretical analysis attributes this vulnerability to the repeated inputs inherent in direct encoding and the gradient accumulation characteristic of BPTT, which together produce an exceptional large Hessian spectral radius. To address this challenge, we develop a hyperparameter-free method called Dominant Eigencomponent Projection (DEP). By orthogonally projecting gradients to precisely remove their dominant components, DEP effectively reduces the Hessian spectral radius, thereby preventing SNNs from settling into sharp minima. Extensive experiments demonstrate that DEP not only mitigates the vulnerability of SNNs to heterogeneous data poisoning, but also significantly enhances overall robustness compared to key baselines, providing strong support for safer and more reliable SNN deployment.