CEB Improves Model Robustness
2020-02-13Code Available0· sign in to hype
Ian Fischer, Alexander A. Alemi
Code Available — Be the first to reproduce this paper.
ReproduceCode
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, ImageNet-A, and PGD attacks.