A Closer Look at the Adversarial Robustness of Information Bottleneck Models
This challenges prior claims about information bottlenecks enhancing adversarial robustness, which is important for researchers and practitioners in machine learning security.
The paper investigates the adversarial robustness of information bottleneck models for classification, finding that they are not a strong defense strategy and previous improvements were likely due to gradient obfuscation.
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.