On Adaptive Attacks to Adversarial Example Defenses

TL;DR

This paper highlights the importance of comprehensive adaptive attack evaluations for adversarial defenses and demonstrates that many recent defenses can be circumvented despite adaptive testing.

Contribution

It provides a detailed methodology for conducting adaptive attacks and shows that previous defenses often fail under thorough adaptive evaluation.

Findings

Thirteen recent defenses can be bypassed with proper adaptive attacks

Current evaluation practices are often incomplete and insufficient

Guidelines are provided for effective adaptive attack implementation

Abstract

Adaptive attacks have (rightfully) become the de facto standard for evaluating defenses to adversarial examples. We find, however, that typical adaptive evaluations are incomplete. We demonstrate that thirteen defenses recently published at ICLR, ICML and NeurIPS---and chosen for illustrative and pedagogical purposes---can be circumvented despite attempting to perform evaluations using adaptive attacks. While prior evaluation papers focused mainly on the end result---showing that a defense was ineffective---this paper focuses on laying out the methodology and the approach necessary to perform an adaptive attack. We hope that these analyses will serve as guidance on how to properly perform adaptive attacks against defenses to adversarial examples, and thus will allow the community to make further progress in building more robust models.