Representation Learning

Learning Adversarially Robust Representations via Worst-Case Mutual Information Maximization

We propose an unsupervised learning method for obtaining robust representations based on a notion of representation vulnerability.

Sicheng Zhu, Xiao Zhang, David Evans