Feature Purification: How Adversarial Training Performs Robust Deep Learning.

Zeyuan Allen-Zhu, Yuanzhi Li

Browse the full FOCS paper archive.