Adversarial Attacks on Spiking Convolutional Neural Networks for Event-based Vision

Abstract

Event-based sensing using dynamic vision sensors is gaining traction in lowpower vision applications. Spiking neural networks work well with the sparse nature of event-based data and suit deployment on low-power neuromorphic hardware. Being a nascent field, the sensitivity of spiking neural networks to potentially malicious adversarial attacks has received very little attention so far. In this work, we show how white-box adversarial attack algorithms can be adapted to the discrete and sparse nature of event-based visual data, and to the continuous-time setting of spiking neural networks. We test our methods on the N-MNIST and IBM Gestures neuromorphic vision datasets and show adversarial perturbations achieve a high success rate, by injecting a relatively small number of appropriately placed events. We also verify, for the first time, the effectiveness of these perturbations directly on neuromorphic hardware. Finally, we discuss the properties of the resulting perturbations and possible future directions.

Sniper Backdoor: Single Client Targeted Backdoor Attack in Federated Learning

“…As stated, we are the first to test FL with SNNs and neuromorphic data. Specific to the security of SNNs, [33,9,28] evaluated adversarial examples using both regular and neuromorphic data. On backdoor attacks, recent works have evaluated SNNs with neuromorphic data [1,2], even with neu-romorphic triggers that change with time, which are invisible to the human eye.…”

“…In the last few years, new threats have been discovered [17], such as adversarial examples, model inversion, and backdoor attacks, to name a few. Regarding SNNs, recent investigations have concluded that SNNs are also vulnerable to some of these attacks, i.e., adversarial examples [28,33,9] and backdoor attacks [2,1]. In the context of FL, security and privacy evaluations have also been in the scope of security experts [3,25], concluding that FL is vulnerable to privacy attacks, such as membership inference, and security attacks, such as backdoor attacks.…”

Toward Robust Spiking Neural Network Against Adversarial Perturbation

“…In adversarial attacks, the attacker generates adversarial examples to fool a model predicting wrong. Currently, SNNs are demonstrated can be attacked through adversarial examples (Sharmin et al, 2019;Liang et al, 2021a;Büchel et al, 2021;Marchisio et al, 2021). It is urgent to explore an efficient way to improve the robustness of SNN models.…”