Robust Learning with Frequency Domain Regularization
This addresses robustness issues in computer vision models for applications requiring stability against frequency-based variations, though it appears incremental as a regularization technique.
The paper tackles the problem of convolutional neural networks being biased toward low-frequency components and sensitive to high-frequency perturbations by introducing a frequency domain regularization method that constrains filter spectra. The method demonstrated effectiveness in defending against adversarial perturbations, reducing generalization gaps across architectures, and improving transfer learning performance without fine-tuning.
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.