CEB Improves Model RobustnessDownload PDF

25 Sep 2019 (modified: 24 Dec 2019)ICLR 2020 Conference Blind SubmissionReaders: Everyone
  • Original Pdf: pdf
  • Keywords: Information Theory, Adversarial Robustness
  • TL;DR: Training your model with the Conditional Entropy Bottleneck is easy and makes your model more robust.
  • Abstract: We demonstrate that the Conditional Entropy Bottleneck (CEB) can improve model robustness. CEB is an easy strategy to implement and works in tandem with data augmentation procedures. We report results of a large scale adversarial robustness study on CIFAR-10, as well as the IMAGENET-C Common Corruptions Benchmark.
9 Replies