Robustifying ${\ell }_{\infty }$ Adversarial Training to the Union of Perturbation Models

Ameya D. Patil, Michael Tuttle, Alexander G. Schwing +1

Classical adversarial training (AT) frameworks are designed to achieve high adversarial accuracy against a single attack type, typically ${\ell }_{\infty }$ norm-bounded perturbations. R…