Learning Diverse-Structured Networks for Adversarial Robustness

           Xuefeng Du * 1 2 Jingfeng Zhang * 3 Bo Han 1 Tongliang Liu 4 Yu Rong 5
                Gang Niu 3 Junzhou Huang 5 Masashi Sugiyama 3 6

            Abstract
                                Table 1. Performance misalignment for different NA in terms of ro-
   In adversarial training (AT), the main focus has      bustness (PGD-20) and standard accuracy. Models which perform
   been the objective and optimizer while the mod ...