Robust Learning with Frequency Domain Regularization

Abstract

Convolution neural networks have achieved remarkable performance in many tasks of computing vision. However, CNN tends to bias to low frequency components. They prioritize capturing low frequency patterns which lead them fail when suffering from application scenario transformation. While adversarial example implies the model is very sensitive to high frequency perturbations. In this paper, we introduce a new regularization method by constraining the frequency spectra of the filter of the model. Different from band-limit training, our method considers the valid frequency range probably entangles in different layers rather than continuous and trains the valid frequency range end-to-end by backpropagation. We demonstrate the effectiveness of our regularization by (1) defensing to adversarial perturbations; (2) reducing the generalization gap in different architecture; (3) improving the generalization ability in transfer learning scenario without fine-tune.

Preface: celebrating the 60th anniversary of the University of Science and Technology of China

“…Then a mask is generated to enhance the domain invariant frequency components and suppress the components that are not conducive to generalization. Guo and Ouyang [12] learn an effective frequency range for the features of each convolution layer to improve the convergence and robustness of CNN. Our method decomposes images from the frequency domain and aims at learning them all rather than supposing that there are frequency components with good generalization.…”