[C13] Cascade Adversarial Machine Learning Regularized with a Unified Embedding


Injecting adversarial inputs during training, known as adversarial training, can improve robustness against one-step attacks, but not for unknown iterative attacks. To address this challenge, (1) we first show iteratively generated adversarial images easily transfer between networks trained with the same strategy. Inspired by this observation, (2) we propose cascade adversarial training, which transfers the knowledge of the end results of adversarial training. We train a network from scratch by injecting iteratively generated adversarial images crafted from already defended networks in addition to one-step adversarial images from the network being trained. (3) We also propose to utilize embedding space for both classification and low-level (pixel-level) similarity learning to ignore unknown pixel level perturbation. During training, we inject adversarial images without replacing their corresponding clean images and penalize the distance between the two embeddings (clean and adversarial). Experimental results show that cascade adversarial training together with our proposed low-level similarity learning efficiently enhance the robustness against iterative attacks under both white box attack and black box attack scenarios.

31th Conference on Neural Information Processing Systems (NIPS) Workshop on Machine Learning and Computer Security