Timezone: »

GAT: Guided Adversarial Training with Pareto-optimal Auxiliary Tasks
Salah GHAMIZI · Jingfeng ZHANG · Maxime Cordy · Mike Papadakis · Masashi Sugiyama · YVES LE TRAON

Thu Jul 27 04:30 PM -- 06:00 PM (PDT) @ Exhibit Hall 1 #114

While leveraging additional training data is well established to improve adversarial robustness, it incurs the unavoidable cost of data collection and the heavy computation to train models. To mitigate the costs, we propose *Guided Adversarial Training * (GAT), a novel adversarial training technique that exploits auxiliary tasks under a limited set of training data. Our approach extends single-task models into multi-task models during the min-max optimization of adversarial training, and drives the loss optimization with a regularization of the gradient curvature across multiple tasks. GAT leverages two types of auxiliary tasks: self-supervised tasks, where the labels are generated automatically, and domain-knowledge tasks, where human experts provide additional labels. Experimentally, under limited data, GAT increases the robust accuracy on CIFAR-10 up to four times (from 11% to 42% robust accuracy) and the robust AUC of CheXpert medical imaging dataset from 50% to 83%. On the full CIFAR-10 dataset, GAT outperforms eight state-of-the-art adversarial training strategies. Our large study across five datasets and six tasks demonstrates that task augmentation is an efficient alternative to data augmentation, and can be key to achieving both clean and robust performances.

Author Information

Salah GHAMIZI (University of Luxembourg)
Jingfeng ZHANG (University of Auckland/RIKEN AIP)
Maxime Cordy (University of Luxemburg)
Mike Papadakis (University of Luxemburg)
Masashi Sugiyama (RIKEN / The University of Tokyo)
YVES LE TRAON (University of Luxemburg)

More from the Same Authors