Imbalanced Adversarial Training with Reweighting

TL;DR

This paper investigates the challenges of adversarial training on imbalanced datasets, revealing issues with class performance and proposing a novel reweighting method to improve class separability and robustness.

Contribution

The paper introduces Separable Reweighted Adversarial Training (SRAT), a new approach that enhances adversarial training on imbalanced data by learning more separable features.

Findings

SRAT improves performance on under-represented classes.

Traditional reweighting strategies are less effective for adversarial training.

Theoretical analysis links data separability to class imbalance issues.

Abstract

Adversarial training has been empirically proven to be one of the most effective and reliable defense methods against adversarial attacks. However, almost all existing studies about adversarial training are focused on balanced datasets, where each class has an equal amount of training examples. Research on adversarial training with imbalanced training datasets is rather limited. As the initial effort to investigate this problem, we reveal the facts that adversarially trained models present two distinguished behaviors from naturally trained models in imbalanced datasets: (1) Compared to natural training, adversarially trained models can suffer much worse performance on under-represented classes, when the training dataset is extremely imbalanced. (2) Traditional reweighting strategies may lose efficacy to deal with the imbalance issue for adversarial training. For example, upweighting the under-represented classes will drastically hurt the model's performance on well-represented classes, and as a result, finding an optimal reweighting value can be tremendously challenging. In this paper, to further understand our observations, we theoretically show that the poor data separability is one key reason causing this strong tension between under-represented and well-represented classes. Motivated by this finding, we propose Separable Reweighted Adversarial Training (SRAT) to facilitate adversarial training under imbalanced scenarios, by learning more separable features for different classes. Extensive experiments on various datasets verify the effectiveness of the proposed framework.