Reliable evaluation of adversarial robustness with an ensemble of diverse parameter-free attacks — mentioned by orange3xchicken

Wed, 22 Dec 2021 02:54:53 GMT

I'm less familiar with poisoning, but at least for test-time robustness, the current benchmark for image classifiers is AutoAttack [0,1]. It's an ensemble of adaptive & parameter-free gradient-based white-box and gradient-free black-box attacks. Submitted academic work is typically considered incomplete without an evaluation on AA (and sometimes deepfool [2]). It's good to see that both are included in ART. [0] https://arxiv.org/abs/2003.01690 [1] https://github.com/fra31/auto-attack [2] https://arxiv.org/abs/1511.04599

Paper and discussion history. A mention is not an endorsement.

Reliable evaluation of adversarial robustness with an ensemble of diverse parameter-free attacks · HackerPapers discussions

Reliable evaluation of adversarial robustness with an ensemble of diverse parameter-free attacks — mentioned by orange3xchicken