Existence and Minimax Theorems for Adversarial Surrogate Risks in Binary Classification
Abstract
Adversarial training is one of the most popular methods for training methods robust to adversarial attacks, however, it is not well-understood from a theoretical perspective. We prove and existence, regularity, and minimax theorems for adversarial surrogate risks. Our results explain some empirical observations on adversarial robustness from prior work and suggest new directions in algorithm development. Furthermore, our results extend previously known existence and minimax theorems for the adversarial classification risk to surrogate risks.
Cite
@article{arxiv.2206.09098,
title = {Existence and Minimax Theorems for Adversarial Surrogate Risks in Binary Classification},
author = {Natalie S. Frank and Jonathan Niles-Weed},
journal= {arXiv preprint arXiv:2206.09098},
year = {2025}
}
Comments
42 pages. version 2: corrects several errors and employs a significantly different proof technique. version 3: modifies the arXiv author list but has no other changes. version 4: improved exposition and fixed typos