Causal Information Bottleneck Boosts Adversarial Robustness of Deep Neural Network

TL;DR

This paper introduces a causal information bottleneck approach that enhances the adversarial robustness of deep neural networks by separating robust and non-robust features using causal inference, demonstrating improved resistance to attacks.

Contribution

It integrates causal inference into the information bottleneck framework to better distinguish and mitigate non-robust features, improving adversarial robustness.

Findings

Significant robustness gains against multiple adversarial attacks.

Effective separation of content and style features via causal inference.

Validated on MNIST, FashionMNIST, and CIFAR-10 datasets.

Abstract

The information bottleneck (IB) method is a feasible defense solution against adversarial attacks in deep learning. However, this method suffers from the spurious correlation, which leads to the limitation of its further improvement of adversarial robustness. In this paper, we incorporate the causal inference into the IB framework to alleviate such a problem. Specifically, we divide the features obtained by the IB method into robust features (content information) and non-robust features (style information) via the instrumental variables to estimate the causal effects. With the utilization of such a framework, the influence of non-robust features could be mitigated to strengthen the adversarial robustness. We make an analysis of the effectiveness of our proposed method. The extensive experiments in MNIST, FashionMNIST, and CIFAR-10 show that our method exhibits the considerable robustness against multiple adversarial attacks. Our code would be released.