Revisiting adversarial training for the worst-performing class

Thomas Pethick, Grigorios Chrysos, Volkan Cevher

Abstract: Despite progress in adversarial training (AT), there is a substantial gap between the top-performing and worst-performing classes in many datasets. For example, on CIFAR10, the accuracies for the best and worst classes are 74% and 23%, respectively. We argue that this gap can be reduced by explicitly optimizing for the worst-performing class, resulting in a min-max-max optimization formulation. Our method, called class focused online learning (CFOL), includes high probability convergence guarantees for the worst class loss and can be easily integrated into existing training setups with minimal computational overhead. We demonstrate an improvement to 32% in the worst class accuracy on CIFAR10, and we observe consistent behavior across CIFAR100 and STL10. Our study highlights the importance of moving beyond average accuracy, which is particularly important in safety-critical applications.

Submission Length: Regular submission (no more than 12 pages of main content)

Code: https://github.com/LIONS-EPFL/class-focused-online-learning-code

Assigned Action Editor: ~Yair_Carmon1

License: Creative Commons Attribution 4.0 International (CC BY 4.0)

Submission Number: 439