Attacking Adversarial Attacks as A Defense

Boxi Wu, Heng Pan, Li Shen +6

It is well known that adversarial attacks can fool deep neural networks with imperceptible perturbations. Although adversarial training significantly improves model robustness, fai…