On the Clean Generalization and Robust Overfitting in Adversarial Training from Two Theoretical Views: Representation Complexity and Training Dynamics.

Binghui Li, Yuanzhi Li

Browse the full ICML paper archive.