Learning to Defense by Learning to Attack

Zhehui Chen, Haoming Jiang, Yuyang Shi, Bo Dai, Tuo Zhao

Abstract: Adversarial training provides a principled approach for training robust neural networks. From an optimization perspective, the adversarial training is essentially solving a minimax robust optimization problem. The outer minimization is trying to learn a robust classifier, while the inner maximization is trying to generate adversarial samples. Unfortunately, such a minimax problem is very difficult to solve due to the lack of convex-concave structure. This work proposes a new adversarial training method based on a generic learning-to-learn (L2L) framework. Specifically, instead of applying the existing hand-designed algorithms for the inner problem, we learn an optimizer, which is parametrized as a convolutional neural network. At the same time, a robust classifier is learned to defense the adversarial attack generated by the learned optimizer. Our experiments over CIFAR-10 and CIFAR-100 datasets demonstrate that the L2L outperforms existing adversarial training methods in both classification accuracy and computational efficiency. Moreover, our L2L framework can be extended to the generative adversarial imitation learning and stabilize the training.

Community Implementations: [ 2 code implementations](https://www.catalyzex.com/paper/arxiv:1811.01213/code)