Fooling Adversarial Training with Inducing Noise

Zhirui Wang, Yifei Wang, Yisen Wang

Adversarial training is widely believed to be a reliable approach to improve model robustness against adversarial attack. However, in this paper, we show that when trained on one t…