Efficiency Attacks on Spiking Neural Networks
TimeTuesday, July 12th5:06pm - 5:30pm PDT
Location3006, Level 3
Event Type
Research Manuscript
AI/ML Security/Privacy
DescriptionSpiking Neural Networks are a class of artificial neural networks that process information as discrete spikes. The time and energy consumed in SNN implementations is strongly dependent on the number of spikes processed. We explore this sensitivity from an adversarial perspective and propose a completely new class of attacks on SNNs. These attacks impact the efficiency of SNNs via imperceptible perturbations that increase the overall spike activity of the network, leading to increased time and energy consumption. Across several SNN benchmarks, our attack achieves up to 2.5X increase in spike activity, leading to 2.2x increase in latency.