A Singular Value Perspective on Model Robustness

Abstract

Convolutional Neural Networks (CNNs) have made significant progress on several computer vision benchmarks, but are fraught with numerous non-human biases such as vulnerability to adversarial samples. Their lack of explainability makes identification and rectification of these biases difficult, and understanding their generalization behavior remains an open problem. In this work we explore the relationship between the generalization behavior of CNNs and the Singular Value Decomposition (SVD) of images. We show that naturally trained and adversarially robust CNNs exploit highly different features for the same dataset. We demonstrate that these features can be disentangled by SVD for ImageNet and CIFAR-10 trained networks. Finally, we propose Rank Integrated Gradients (RIG), the first rank-based feature attribution method to understand the dependence of CNNs on image rank.

References in corpus (8)

• Very Deep Convolutional Networks for Large-Scale Image Recognition
• Explaining and Harnessing Adversarial Examples
• Striving for Simplicity: The All Convolutional Net
• Theoretically Principled Trade-off between Robustness and Accuracy
• Linformer: Self-Attention with Linear Complexity
• Measuring the tendency of CNNs to Learn Surface Statistical Regularities
• Universal adversarial perturbations
• Towards Frequency-Based Explanation for Robust CNN