Intriguing Frequency Interpretation of Adversarial Robustness for CNNs and ViTs
Preprint
Abstract
Adversarial examples have attracted significant attention over the years, yet understanding their frequency-based characteristics remains insufficient. In this paper, we investigate the intriguing properties of adversarial examples in the frequency domain for the image classification task, with the following key findings. (1) As the high-frequency components increase, the performance gap between adversarial and natural examples becomes increasingly pronounced. (2) The model performance against filtered adversarial examples initially increases to a peak and declines to its inherent robustness. (3) In Convolutional Neural Networks, mid- and high-frequency components of adversarial examples exhibit their attack capabilities, while in Transformers, low- and mid-frequency components of adversarial examples are particularly effective. These results suggest that different network architectures have different frequency preferences and that differences in frequency components between adversarial and natural examples may directly influence model robustness. Based on our findings, we further conclude with three useful proposals that serve as a valuable reference to the AI model security community.Keywords
All Related Versions
This publication has 12 references indexed in Scilit:
- Swin Transformer: Hierarchical Vision Transformer using Shifted Windows2021
- Focal Frequency Loss for Image Reconstruction and Synthesis2021
- SpectralDefense: Detecting Adversarial Attacks on CNNs in the Fourier Domain2021
- High-Frequency Component Helps Explain the Generalization of Convolutional Neural Networks2020
- Adversarial Sensor Attack on LiDAR-based Perception in Autonomous Driving2019
- BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding2019
- Defense Against Adversarial Attacks Using High-Level Representation Guided Denoiser2018
- Towards Evaluating the Robustness of Neural Networks2017
- Deep Residual Learning for Image Recognition2016
- Distillation as a Defense to Adversarial Perturbations Against Deep Neural Networks2016