Subspace Defense: Discarding Adversarial Perturbations by Learning a Subspace for Clean Signals

arXiv:2403.16176v1 Announce Type: new
Abstract: Deep neural networks (DNNs) are notoriously vulnerable to adversarial attacks that place carefully crafted perturbations on normal examples to fool DNNs. To better understand such attacks, a characterization of the features carried by adversarial examples is needed. In this paper, we tackle this challenge by inspecting the subspaces of sample features through spectral analysis. We first empirically show that the features of either clean signals or adversarial perturbations are redundant and span in low-dimensional linear …

abstract adversarial adversarial attacks adversarial examples arxiv attacks cs.cl cs.cr cs.lg defense examples features networks neural networks normal paper type vulnerable