Attacks Which Do Not Kill Training Make Adversarial Learning Stronger

Abstract

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.

arXiv commentThirty-seventh International Conference on Machine Learning (ICML 2020)

References in corpus (19)

• Ensemble Adversarial Training: Attacks and Defenses
• Wide Residual Networks
• Obfuscated Gradients Give a False Sense of Security: Circumventing Defenses to Adversarial Examples
• Theoretically Principled Trade-off between Robustness and Accuracy
• Provable defenses against adversarial examples via the convex outer adversarial polytope
• Certified Adversarial Robustness via Randomized Smoothing
• Fast is better than free: Revisiting adversarial training
• Formal Guarantees on the Robustness of a Classifier against Adversarial Manipulation
• Spatially Transformed Adversarial Examples
• Improving Adversarial Robustness via Promoting Ensemble Diversity
• On the Convergence and Robustness of Adversarial Training
• Are Labels Required for Improving Adversarial Robustness?
• Lipschitz-Margin Training: Scalable Certification of Perturbation Invariance for Deep Neural Networks
• Rademacher Complexity for Adversarially Robust Generalization
• Deep Defense: Training DNNs with Improved Adversarial Robustness
• Adversarial Robustness May Be at Odds With Simplicity
• A Simple Explanation for the Existence of Adversarial Examples with Small Hamming Distance
• Generalizable Adversarial Training via Spectral Normalization
• Curriculum Adversarial Training

Cited by in corpus (11)

• Uncertainty-Informed Deep Learning Models Enable High-Confidence Predictions for Digital Histopathology
• A Closer Look at Accuracy vs. Robustness
• Robust Pre-Training by Adversarial Contrastive Learning
• Recent Advances in Adversarial Training for Adversarial Robustness
• CIFS: Improving Adversarial Robustness of CNNs via Channel-wise Importance-based Feature Selection
• Data Quality Matters For Adversarial Training: An Empirical Study
• Amata: An Annealing Mechanism for Adversarial Training Acceleration
• RAIN: A Simple Approach for Robust and Accurate Image Classification Networks
• Robust Face Verification via Disentangled Representations
• Adversarial Augmentation Training Makes Action Recognition Models More Robust to Realistic Video Distribution Shifts
• Rethinking Uncertainty in Deep Learning: Whether and How it Improves Robustness