A Closer Look at the Adversarial Robustness of Information Bottleneck Models

Iryna Korshunova, David Stutz, Alexander A Alemi, Olivia Wiles, Sven Gowal

Keywords: Information Bottlenecks, Adversarial Robustness

TL;DR: Information bottleneck models are less robust to adversarial attacks than previously thought

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.