Robust Sensible Adversarial Learning of Deep Neural Networks for Image Classification
This work addresses the critical problem of adversarial robustness in image classification for AI safety, presenting an incremental improvement by synergizing robustness and accuracy through a novel learning approach.
The paper tackles the vulnerability of deep neural networks to adversarial attacks by introducing sensible adversarial learning, which theoretically establishes the Bayes classifier as the most robust under this framework and empirically shows the method maintains high natural accuracy while promoting robustness against various attacks on MNIST and CIFAR10 datasets.
The idea of robustness is central and critical to modern statistical analysis. However, despite the recent advances of deep neural networks (DNNs), many studies have shown that DNNs are vulnerable to adversarial attacks. Making imperceptible changes to an image can cause DNN models to make the wrong classification with high confidence, such as classifying a benign mole as a malignant tumor and a stop sign as a speed limit sign. The trade-off between robustness and standard accuracy is common for DNN models. In this paper, we introduce sensible adversarial learning and demonstrate the synergistic effect between pursuits of standard natural accuracy and robustness. Specifically, we define a sensible adversary which is useful for learning a robust model while keeping high natural accuracy. We theoretically establish that the Bayes classifier is the most robust multi-class classifier with the 0-1 loss under sensible adversarial learning. We propose a novel and efficient algorithm that trains a robust model using implicit loss truncation. We apply sensible adversarial learning for large-scale image classification to a handwritten digital image dataset called MNIST and an object recognition colored image dataset called CIFAR10. We have performed an extensive comparative study to compare our method with other competitive methods. Our experiments empirically demonstrate that our method is not sensitive to its hyperparameter and does not collapse even with a small model capacity while promoting robustness against various attacks and keeping high natural accuracy.