Attacks Which Do Not Kill Training Make Adversarial Learning Stronger
This work addresses a key problem in machine learning security by offering a method to enhance adversarial robustness without sacrificing model performance on clean data, though it is incremental as it builds on existing adversarial training techniques.
The paper tackles the trade-off between adversarial robustness and natural generalization in adversarial training by proposing friendly adversarial training (FAT), which uses less adversarial data to update models, and shows that this approach achieves adversarial robustness without compromising natural generalization.
Adversarial training based on the minimax formulation is necessary for obtaining adversarial robustness of trained models. However, it is conservative or even pessimistic so that it sometimes hurts the natural generalization. In this paper, we raise a fundamental question---do we have to trade off natural generalization for adversarial robustness? We argue that adversarial training is to employ confident adversarial data for updating the current model. We propose a novel approach of friendly adversarial training (FAT): rather than employing most adversarial data maximizing the loss, we search for least adversarial (i.e., friendly adversarial) data minimizing the loss, among the adversarial data that are confidently misclassified. Our novel formulation is easy to implement by just stopping the most adversarial data searching algorithms such as PGD (projected gradient descent) early, which we call early-stopped PGD. Theoretically, FAT is justified by an upper bound of the adversarial risk. Empirically, early-stopped PGD allows us to answer the earlier question negatively---adversarial robustness can indeed be achieved without compromising the natural generalization.