On Adaptive Attacks to Adversarial Example Defenses

Florian Tramèr*, Nicholas Carlini*, Wieland Brendel* and Aleksander Madry   (*joint first authors)

Conference on Neural Information Processing Systems (NeurIPS) 2020


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.

  author   =   {Tram{\`e}r, Florian and Carlini, Nicholas and Brendel, Wieland and Madry, Aleksander},
  title   =   {On Adaptive Attacks to Adversarial Example Defenses},
  booktitle   =   {Conference on Neural Information Processing Systems (NeurIPS)},
  year   =   {2020},
  howpublished   =   {arXiv preprint arXiv:2002.08347},
  url   =   {https://arxiv.org/abs/2002.08347}