Adversarial Robustness Comparison of Vision Transformer and MLP-Mixer to CNNs

TL;DR

This paper empirically compares the adversarial robustness of Vision Transformers and MLP-Mixers to CNNs, finding that ViT is generally more robust and that frequency features influence robustness, with MLP-Mixer being highly vulnerable to universal attacks.

Contribution

It provides the first comprehensive empirical evaluation of adversarial robustness across ViT, MLP-Mixer, and CNN architectures, highlighting their differences and underlying factors.

Findings

ViT is more robust than CNNs against adversarial attacks.

MLP-Mixer is extremely vulnerable to universal adversarial perturbations.

Low-frequency features contribute to ViT's robustness.

Abstract

Convolutional Neural Networks (CNNs) have become the de facto gold standard in computer vision applications in the past years. Recently, however, new model architectures have been proposed challenging the status quo. The Vision Transformer (ViT) relies solely on attention modules, while the MLP-Mixer architecture substitutes the self-attention modules with Multi-Layer Perceptrons (MLPs). Despite their great success, CNNs have been widely known to be vulnerable to adversarial attacks, causing serious concerns for security-sensitive applications. Thus, it is critical for the community to know whether the newly proposed ViT and MLP-Mixer are also vulnerable to adversarial attacks. To this end, we empirically evaluate their adversarial robustness under several adversarial attack setups and benchmark them against the widely used CNNs. Overall, we find that the two architectures, especially ViT, are more robust than their CNN models. Using a toy example, we also provide empirical evidence that the lower adversarial robustness of CNNs can be partially attributed to their shift-invariant property. Our frequency analysis suggests that the most robust ViT architectures tend to rely more on low-frequency features compared with CNNs. Additionally, we have an intriguing finding that MLP-Mixer is extremely vulnerable to universal adversarial perturbations.