2017/05/09 by Lee, Hyeungill, Han, Sungyeob, Lee, Jungwoo · 68 citations
Computer Science · #Advanced Neural Network Applications #Adversarial Robustness in Machine Learning #Adversarial system #Anomaly Detection Techniques and Applications #Artificial intelligence #Artificial neural network #Classifier (UML) #Computer science #Deep learning #FOS: Computer and information sciences #Generative Adversarial Networks and Image Synthesis #Generative adversarial network #Generative grammar #Generator (circuit theory) #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Machine learning #Overfitting #Pattern recognition (psychology)
paper · pdf · doi:10.48550/arxiv.1705.03387
published in arXiv (Cornell University) (Cornell University)
We propose a novel technique to make neural network robust to adversarial\nexamples using a generative adversarial network. We alternately train both\nclassifier and generator networks. The generator network generates an\nadversarial perturbation that can easily fool the classifier network by using a\ngradient of each image. Simultaneously, the classifier network is trained to\nclassify correctly both original and adversarial images generated by the\ngenerator. These procedures help the classifier network to become more robust\nto adversarial perturbations. Furthermore, our adversarial training framework\nefficiently reduces overfitting and outperforms other regularization methods\nsuch as Dropout. We applied our method to supervised learning for CIFAR\ndatasets, and experimantal results show that our method significantly lowers\nthe generalization error of the network. To the best of our knowledge, this is\nthe first method which uses GAN to improve supervised learning.\n