Raw event-based adversarial attacks for Spiking Neural Networks with configurable latencies

Xiao Du, Wanli Shi, Xiaohan Zhao, Yang Cao, Bin Gu, Tieru Wu

Abstract: Highlights•Proposed an adversarial attack framework for SNNs that handles configurable latencies.•Innovatively mitigated latency effects on attack success, enhancing robustness.•Introduced an adaptive random latency strategy, overcoming fixed-latency limitations.•Experiments show superior performance of our method over fixed-latency attacks.•Explored initial steps towards black-box attacks on raw event streams for SNNs.

External IDs:doi:10.1016/j.neunet.2025.108026