Towards Understanding the Generative Capability of Adversarially Robust Classifiers

TL;DR

This paper explains why adversarially robust classifiers can generate high-quality images by analyzing their energy landscape, and introduces a new training method that improves both image quality and robustness.

Contribution

The paper provides a novel energy-based explanation for the generative ability of robust classifiers and proposes JEAT, a new training method that enhances image quality and robustness.

Findings

JEAT achieves an Inception Score of 8.80 on CIFAR-10.

JEAT outperforms previous robust classifiers in image quality.

The energy landscape of adversarial training is key to generative capability.

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).