Bilateral Adversarial Training: Towards Fast Training of More Robust Models Against Adversarial Attacks
Authors:
- 1 October 2019
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 6628-6637
Abstract
In this paper, we study fast training of adversarially robust models. From the analyses of the state-of-the-art defense method, i.e., the multi-step adversarial training~\cite{madry2017towards}, we hypothesize that the gradient magnitude links to the model robustness. Motivated by this, we propose to perturb both the image and the label during training, which we call Bilateral Adversarial Training (BAT). To generate the adversarial label, we derive an closed-form heuristic solution. To generate the adversarial image, we use one-step targeted attack with the target label being the most confusing class. In the experiment, we first show that random start and the most confusing target attack effectively prevent the label leaking and gradient masking problem. Then coupled with the adversarial label part, our model significantly improves the state-of-the-art results. For example, against PGD100 white-box attack with cross-entropy loss, on CIFAR10, we achieve 63.7\% versus 47.2\%; on SVHN, we achieve 59.1\% versus 42.1\%. At last, the experiment on the very (computationally) challenging ImageNet dataset further demonstrates the effectiveness of our fast method.Keywords
All Related Versions
This publication has 23 references indexed in Scilit:
- MagNet2017
- Towards Evaluating the Robustness of Neural Networks2017
- Practical Black-Box Attacks against Machine Learning2017
- DeepFool: A Simple and Accurate Method to Fool Deep Neural Networks2016
- Rethinking the Inception Architecture for Computer Vision2016
- Distillation as a Defense to Adversarial Perturbations Against Deep Neural Networks2016
- Adversarial machine learning2011
- Adversarial classification2004
- Gradient-based learning applied to document recognitionProceedings of the IEEE, 1998
- Flat MinimaNeural Computation, 1997