Current browse context:
cs
Change to browse by:
References & Citations
Computer Science > Machine Learning
Title: Increasing Confidence in Adversarial Robustness Evaluations
(Submitted on 28 Jun 2022)
Abstract: Hundreds of defenses have been proposed to make deep neural networks robust against minimal (adversarial) input perturbations. However, only a handful of these defenses held up their claims because correctly evaluating robustness is extremely challenging: Weak attacks often fail to find adversarial examples even if they unknowingly exist, thereby making a vulnerable network look robust. In this paper, we propose a test to identify weak attacks, and thus weak defense evaluations. Our test slightly modifies a neural network to guarantee the existence of an adversarial example for every sample. Consequentially, any correct attack must succeed in breaking this modified network. For eleven out of thirteen previously-published defenses, the original evaluation of the defense fails our test, while stronger attacks that break these defenses pass it. We hope that attack unit tests - such as ours - will be a major component in future robustness evaluations and increase confidence in an empirical field that is currently riddled with skepticism.
Submission history
From: Roland Zimmermann [view email][v1] Tue, 28 Jun 2022 13:28:13 GMT (669kb,D)
Link back to: arXiv, form interface, contact.