Recent Advances in Adversarial Training for Adversarial Robustness

Adversarial training is one of the most effective approaches defending against adversarial examples for deep learning models. Unlike other defenses that are limited to specific tasks, adversarial training is more general and can be extended easily. However, adversarial training is not perfect, many problems of which remain to be solved. During the last few years, adversarial training is being studied and discussed from various aspects, and many improvements and developments are proposed. In this survey, we systematically review the recent progress on adversarial training with novel taxonomy for the first time. Then we discuss the generalization problems in adversarial training from three perspectives. Finally, we highlight the challenges which are not fully solved and present potential future directions.