How Does Frequency Bias Affect the Robustness of Neural Image Classifiers against Common Corruption and Adversarial Perturbations?

2022/05/09 by Chan, Alvin, Ong, Yew-Soon, Tan, Clement · 1 citation
#Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Machine Learning (cs.LG)

paper · doi:10.48550/arxiv.2205.04533

Abstract

Model robustness is vital for the reliable deployment of machine learning models in real-world applications. Recent studies have shown that data augmentation can result in model over-relying on features in the low-frequency domain, sacrificing performance against low-frequency corruptions, highlighting a connection between frequency and robustness. Here, we take one step further to more directly study the frequency bias of a model through the lens of its Jacobians and its implication to model robustness. To achieve this, we propose Jacobian frequency regularization for models' Jacobians to have a larger ratio of low-frequency components. Through experiments on four image datasets, we show that biasing classifiers towards low (high)-frequency components can bring performance gain against high (low)-frequency corruption and adversarial perturbation, albeit with a tradeoff in performance for low (high)-frequency corruption. Our approach elucidates a more direct connection between the frequency bias and robustness of deep learning models.

Cited by

• FastDINOv2: Frequency Based Curriculum Learning Improves Robustness and Training Speed

Related

• Towards Building More Robust Models with Frequency Bias
• Improving robustness against common corruptions with frequency biased models
• Intriguing Frequency Interpretation of Adversarial Robustness for CNNs and ViTs
• Adversarial amplitude swap towards robust image classifiers
• Benchmarking Neural Network Robustness to Common Corruptions and Perturbations