CEB Improves Model Robustness

Ian Fischer; Alexander A. Alemi

doi:10.3390/e22101081

CEB Improves Model Robustness

Machine Learning 2020-10-28 v1 Machine Learning

Authors: Ian Fischer , Alexander A. Alemi

View on arXiv ↗ PDF ↗ DOI ↗

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.

Keywords

adversarial robustness

Cite

@article{arxiv.2002.05380,
  title  = {CEB Improves Model Robustness},
  author = {Ian Fischer and Alexander A. Alemi},
  journal= {arXiv preprint arXiv:2002.05380},
  year   = {2020}
}

Related papers

View all related →