Rethinking Clustering for Robustness

Motasem Alfarra, Juan C. Pérez, Adel Bibi +3

This paper studies how encouraging semantically-aligned features during deep neural network training can increase network robustness. Recent works observed that Adversarial Trainin…