A Closer Look at the Adversarial Robustness of Information Bottleneck Models
Abstract
We study the adversarial robustness of information bottleneck models for classification. Previous works showed that the robustness of models trained with information bottlenecks can improve upon adversarial training. Our evaluation under a diverse range of white-box $l_{\infty}$ attacks suggests that information bottlenecks alone are not a strong defense strategy, and that previous results were likely influenced by gradient obfuscation.
Cite
Text
Korshunova et al. "A Closer Look at the Adversarial Robustness of Information Bottleneck Models." ICML 2021 Workshops: AML, 2021.Markdown
[Korshunova et al. "A Closer Look at the Adversarial Robustness of Information Bottleneck Models." ICML 2021 Workshops: AML, 2021.](https://mlanthology.org/icmlw/2021/korshunova2021icmlw-closer/)BibTeX
@inproceedings{korshunova2021icmlw-closer,
title = {{A Closer Look at the Adversarial Robustness of Information Bottleneck Models}},
author = {Korshunova, Iryna and Stutz, David and Alemi, Alexander A and Wiles, Olivia and Gowal, Sven},
booktitle = {ICML 2021 Workshops: AML},
year = {2021},
url = {https://mlanthology.org/icmlw/2021/korshunova2021icmlw-closer/}
}