Sun Dec 8th through Sat the 14th, 2019 at Vancouver Convention Center
This paper first shows that additional unlabeled data can significantly lower the generalization gap of robust classifiers under a simple Gaussian model. Then, it presents two unsupervised adversarial training (UAT) methods for learning complex deep networks. Empirical improvements on various datasets are significant. However, this paper heavily overlaps with another paper "Unlabeled Data Improves Adversarial Robustness". As a condition to accepting and including the paper in the proceedings, put the following disclaimer in the footnote on the first page: "The authors declare that the present paper is independent of "Unlabeled Data Improves Adversarial Robustness"."