Virtual presentation / poster accept
Squeeze Training for Adversarial Robustness
Qizhang Li · Yiwen Guo · Wangmeng Zuo · Hao Chen
Keywords: [ Model Robustness ] [ adversarial examples ] [ adversarial training ] [ Deep Learning and representational learning ]
The vulnerability of deep neural networks (DNNs) to adversarial examples has attracted great attention in the machine learning community. The problem is related to non-flatness and non-smoothness of normally obtained loss landscapes. Training augmented with adversarial examples (a.k.a., adversarial training) is considered as an effective remedy. In this paper, we highlight that some collaborative examples, nearly perceptually indistinguishable from both adversarial and benign examples yet show extremely lower prediction loss, can be utilized to enhance adversarial training. A novel method is therefore proposed to achieve new state-of-the-arts in adversarial robustness. Code: https://github.com/qizhangli/ST-AT.