Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Fast Adversarial Training with Adaptive Step Size

About

While adversarial training and its variants have shown to be the most effective algorithms to defend against adversarial attacks, their extremely slow training process makes it hard to scale to large datasets like ImageNet. The key idea of recent works to accelerate adversarial training is to substitute multi-step attacks (e.g., PGD) with single-step attacks (e.g., FGSM). However, these single-step methods suffer from catastrophic overfitting, where the accuracy against PGD attack suddenly drops to nearly 0% during training, destroying the robustness of the networks. In this work, we study the phenomenon from the perspective of training instances. We show that catastrophic overfitting is instance-dependent and fitting instances with larger gradient norm is more likely to cause catastrophic overfitting. Based on our findings, we propose a simple but effective method, Adversarial Training with Adaptive Step size (ATAS). ATAS learns an instancewise adaptive step size that is inversely proportional to its gradient norm. The theoretical analysis shows that ATAS converges faster than the commonly adopted non-adaptive counterparts. Empirically, ATAS consistently mitigates catastrophic overfitting and achieves higher robust accuracy on CIFAR10, CIFAR100 and ImageNet when evaluated on various adversarial budgets.

Zhichao Huang, Yanbo Fan, Chen Liu, Weizhong Zhang, Yong Zhang, Mathieu Salzmann, Sabine S\"usstrunk, Jue Wang• 2022

Related benchmarks

TaskDatasetResultRank
Image ClassificationCIFAR-100
Clean Accuracy68.18
90
Image ClassificationCIFAR-10
Clean Accuracy87.38
89
Image ClassificationPathMNIST
Clean Accuracy62.5
60
Medical Image ClassificationPathMNIST
Clean Accuracy66.49
48
Image ClassificationTissueMNIST MedMNIST v2 (test)
Clean Accuracy48.39
29
Image ClassificationTissueMNIST
Clean Accuracy31.95
20
Image ClassificationCIFAR-100 (test)
Clean Accuracy62.48
16
Image ClassificationCIFAR-10 (test)
Accuracy (Clean)89.07
16
Showing 8 of 8 rows

Other info

Follow for update