Timezone: »
The study on improving the robustness of deep neural networks against adversarial examples grows rapidly in recent years. Among them, adversarial training is the most promising one, which flattens the \textit{input loss landscape} (loss change with respect to input) via training on adversarially perturbed examples. However, how the widely used \textit{weight loss landscape} (loss change with respect to weight) performs in adversarial training is rarely explored. In this paper, we investigate the weight loss landscape from a new perspective, and identify a clear correlation between the flatness of weight loss landscape and robust generalization gap. Several well-recognized adversarial training improvements, such as early stopping, designing new objective functions, or leveraging unlabeled data, all implicitly flatten the weight loss landscape. Based on these observations, we propose a simple yet effective \textit{Adversarial Weight Perturbation (AWP)} to explicitly regularize the flatness of weight loss landscape, forming a \textit{double-perturbation} mechanism in the adversarial training framework that adversarially perturbs both inputs and weights. Extensive experiments demonstrate that AWP indeed brings flatter weight loss landscape and can be easily incorporated into various existing adversarial training methods to further boost their adversarial robustness.
Author Information
Dongxian Wu (Tsinghua University)
Shu-Tao Xia (Tsinghua University)
Yisen Wang (Peking University)
More from the Same Authors
-
2021 Spotlight: Clustering Effect of Adversarial Robust Models »
Yang Bai · Xin Yan · Yong Jiang · Shu-Tao Xia · Yisen Wang -
2021 Poster: Clustering Effect of Adversarial Robust Models »
Yang Bai · Xin Yan · Yong Jiang · Shu-Tao Xia · Yisen Wang -
2020 Poster: Stochastic Deep Gaussian Processes over Graphs »
Naiqi Li · Wenjie Li · Jifeng Sun · Yinghua Gao · Yong Jiang · Shu-Tao Xia -
2018 Poster: BML: A High-performance, Low-cost Gradient Synchronization Algorithm for DML Training »
Songtao Wang · Dan Li · Yang Cheng · Jinkun Geng · Yanshu Wang · Shuai Wang · Shu-Tao Xia · Jianping Wu -
2017 Poster: Accelerated Stochastic Greedy Coordinate Descent by Soft Thresholding Projection onto Simplex »
Chaobing Song · Shaobo Cui · Yong Jiang · Shu-Tao Xia -
2017 Spotlight: Accelerated Stochastic Greedy Coordinate Descent by Soft Thresholding Projection onto Simplex »
Chaobing Song · Shaobo Cui · Yong Jiang · Shu-Tao Xia