Towards Understanding the Generative Capability of Adversarially Robust Classifiers
적의 온건 분류 기의 생 성 능력 을 이해 하 다
Comprensión de la capacidad de generación de clasificadores robustos enemigos
Vers une compréhension de la capacité générative des classificateurs robustes à l'adversaire
понимание способности противника к созданию стабилизатора
Yao Zhu ¹, Jiacheng Ma ², Jiacheng Sun ², Zewei Chen ², Rongxin Jiang 蒋荣欣 ¹, Zhenguo Li ²
¹ Zhejiang University
² Huawei Noah’s Ark Lab
arXiv, 20 August 2021

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). In particular, we achieve new state-of-the-art robustness on CIFAR-10 (from 57.20% to 62.04%) and CIFAR-100 (from 30.03% to 30.18%) without extra training data.
Reviews and Discussions
An overview on the principle of inkjet printing technique and its application in micro-display for augmented/virtual realities
Nonlinear optics with structured light
Femtosecond laser-induced periodic structures: mechanisms, techniques, and applications
Giant and light modifiable third-order optical nonlinearity in a free-standing h-BN film
New approach for the digital reconstruction of complex mine faults and its application in mining
p62/SQSTM1 Participates in the Innate Immune Response of Macrophages Against Candida albicans Infection
Configurable topological beam splitting via antichiral gyromagnetic photonic crystal
Single-molecule optoelectronic devices: physical mechanism and beyond
Functional nonlinear optical nanoparticles synthesized by laser ablation
The m6A methylation regulates gonadal sex differentiation in chicken embryo
A new species in the genus Synanthedon (Lepidoptera: Sesiidae) from China
A review on the forward osmosis applications and fouling control strategies for wastewater treatment

Previous Article                                Next Article
Copyright © Hot Paper