Towards Understanding the Generative Capability of Adversarially Robust Classifiers
Abstract
Recently, some works found an interesting phenomenon that adversarially robust classifiers can generate good images comparable to generative models. We investigate this phenomenon from an energy perspective and provide a novel explanation. We reformulate adversarial example generation, adversarial training, and image generation in terms of an energy function. We find that adversarial training contributes to obtaining an energy function that is flat and has low energy around the real data, which is the key for generative capability. Based on our new understanding, we further propose a better adversarial training method, Joint Energy Adversarial Training (JEAT), which can generate high-quality images and achieve new state-of-the-art robustness under a wide range of attacks. The Inception Score of the images (CIFAR-10) generated by JEAT is 8.80, much better than original robust classifiers (7.50).
Cite
@article{arxiv.2108.09093,
title = {Towards Understanding the Generative Capability of Adversarially Robust Classifiers},
author = {Yao Zhu and Jiacheng Ma and Jiacheng Sun and Zewei Chen and Rongxin Jiang and Zhenguo Li},
journal= {arXiv preprint arXiv:2108.09093},
year = {2021}
}
Comments
Accepted by ICCV 2021, Oral