Keywords: Adversarial examples, adversarial attacks
Abstract: Evading adversarial example detection defenses requires finding adversarial examples that must simultaneously (a) be misclassified by the model and (b) be detected as non-adversarial. We find that existing attacks that attempt to satisfy multiple simultaneous constraints often over-optimize against one constraint at the cost of satisfying another. We introduce Selective Projected Gradient Descent and Orthogonal Projected Gradient Descent, improved attack techniques to generate adversarial examples that avoid this problem by orthogonalizing the gradients when running standard gradient-based attacks. We use our technique to evade four state-of-the-art detection defenses, reducing their accuracy to 0% while maintaining a 0% detection rate.
One-sentence Summary: We break four defenses that detect adversarial examples by introducing an improved attack technique that modifies the gradient before applying PGD.