Improving Robustness of Deep Convolutional Neural Networks via Multiresolution Learning
This addresses robustness issues in deep learning for signal and image prediction, offering a novel approach that avoids the typical trade-off between accuracy and robustness.
The paper tackles the problem of improving robustness in deep convolutional neural networks by introducing multiresolution learning, showing it significantly enhances noise and adversarial robustness and works well with small datasets without sacrificing standard accuracy.
The current learning process of deep learning, regardless of any deep neural network (DNN) architecture and/or learning algorithm used, is essentially a single resolution training. We explore multiresolution learning and show that multiresolution learning can significantly improve robustness of DNN models for both 1D signal and 2D signal (image) prediction problems. We demonstrate this improvement in terms of both noise and adversarial robustness as well as with small training dataset size. Our results also suggest that it may not be necessary to trade standard accuracy for robustness with multiresolution learning, which is, interestingly, contrary to the observation obtained from the traditional single resolution learning setting.