A Closer Look at the Adversarial Robustness of Information Bottleneck Models
arXiv:2107.05712
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 attacks suggests that information bottlenecks alone are not a strong defense strategy, and that previous results were likely influenced by gradient obfuscation.
References in corpus (6)
- Auto-Encoding Variational Bayes
- On Evaluating Adversarial Robustness
- Adversarial Robustness through Local Linearization
- An Alternative Surrogate Loss for PGD-based Adversarial Testing
- Unpacking Information Bottlenecks: Unifying Information-Theoretic Objectives in Deep Learning
- Diagnosing Vulnerability of Variational Auto-Encoders to Adversarial Attacks