Recent work has extensively shown that randomized perturbations of neural networks can improve robustness to adversarial attacks. The literature is, however, lacking a detailed compare-and-contrast of the latest proposals to understand what classes of perturbations work, when they work, and why they work. We contribute a detailed evaluation that elucidates these questions and benchmarks perturbation based defenses consistently. In particular, we show five main results: (1) all input perturbation defenses, whether random or deterministic, are equivalent in their efficacy, (2) attacks transfer between perturbation defenses so the attackers need not know the specific type of defense -- only that it involves perturbations, (3) a tuned sequence of noise layers across a network provides the best empirical robustness, (4) perturbation based defenses offer almost no robustness to adaptive attacks unless these perturbations are observed during training, and (5) adversarial examples in a close neighborhood of original inputs show an elevated sensitivity to perturbations in first and second-order analyses.
تحميل البحث