Generalist: Decoupling Natural and Robust Generalization

Hongjun Wang, Yisen Wang

Deep neural networks obtained by standard training have been constantly plagued by adversarial examples. Although adversarial training demonstrates its capability to defend against…