Exploring Frequency Domain Interpretation of Convolutional Neural Networks

TL;DR

This paper investigates the frequency domain properties of CNN filters, particularly in the first layer, and demonstrates that controlling frequency proportions can significantly enhance model robustness to corruptions.

Contribution

It introduces a novel frequency domain interpretation of CNNs, analyzing filter frequency proportions and proposing a learnable modulation method to improve robustness.

Findings

Controlling frequency proportions affects robustness to corruptions.

Learnable frequency modulation improves robustness by 10.97%.

Minimal impact on natural accuracy.

Abstract

Many existing interpretation methods of convolutional neural networks (CNNs) mainly analyze in spatial domain, yet model interpretability in frequency domain has been rarely studied. To the best of our knowledge, there is no study on the interpretation of modern CNNs from the perspective of the frequency proportion of filters. In this work, we analyze the frequency properties of filters in the first layer as it is the entrance of information and relatively more convenient for analysis. By controlling the proportion of different frequency filters in the training stage, the network classification accuracy and model robustness is evaluated and our results reveal that it has a great impact on the robustness to common corruptions. Moreover, a learnable modulation of frequency proportion with perturbation in power spectrum is proposed from the perspective of frequency domain. Experiments on CIFAR-10-C show 10.97% average robustness gains for ResNet-18 with negligible natural accuracy degradation.