On the Robustness of Bayesian Neural Networks to Adversarial Attacks

arXiv:2207.06154v3 Announce Type: replace-cross
Abstract: Vulnerability to adversarial attacks is one of the principal hurdles to the adoption of deep learning in safety-critical applications. Despite significant efforts, both practical and theoretical, training deep learning models robust to adversarial attacks is still an open problem. In this paper, we analyse the geometry of adversarial attacks in the large-data, overparameterized limit for Bayesian Neural Networks (BNNs). We show that, in the limit, vulnerability to gradient-based attacks arises as a result of degeneracy …

adoption adversarial adversarial attacks applications arxiv attacks critical cs.ai cs.cr cs.lg deep learning networks neural networks problem robustness safety safety-critical training vulnerability