On the tradeoff between robustness and geometric accuracy in adversarial training

Key points

• This rigorously shows that robust classifiers sacrifice at most $O(\sqrt{\varepsilon})$ or $O(\varepsilon)$ in accuracy, depending on the regularity of the underlying Bayes classifier.1
• In contrast to prior work on the incompatibility of robustness and accuracy over low-complexity hypothesis classes, our work shows that high-complexity hypothesis classes allow for robust classifiers with precisely quantified accuracy.1
• Our results apply for adversaries determined by any norm on Euclidean space and in both the infinite and finite data settings.1

Sources (1)

• [1]On the tradeoff between robustness and geometric accuracy in adversarial training
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 16, 05:27 AM

  For non-parametric binary classification in the presence of an adversary, we prove an optimal quantitative estimate for adversarial classifiers controlling geometric accuracy, as measured by the Hausdorff distance to the Bayes classifier, in terms or their robustness, as encoded by the adversarial budget $\varepsilon$.

  This rigorously shows that robust classifiers sacrifice at most $O(\sqrt{\varepsilon})$ or $O(\varepsilon)$ in accuracy, depending on the regularity of the underlying Bayes classifier.

Extractive summary: sentences quoted from the sources.