Semi-supervised Outlier Detection using Generative And Adversary Framework

In a conventional binary/multi-class classification task, the decision boundary is supported by data from two or more classes. However, in one-class classification task, only data from one class are available. To build an robust outlier detector using only data from a positive class, we propose a corrupted GAN(CorGAN), a deep convolutional Generative Adversary Network requiring no convergence during training. In the adversarial process of training CorGAN, the Generator is supposed to generate outlier samples for negative class, and the Discriminator as an one-class classifier is trained to distinguish data from training datasets (i.e. positive class) and generated data from the Generator (i.e. negative class). To improve the performance of the Discriminator (one-class classifier), we also propose a lot of techniques to improve the performance of the model. The proposed model outperforms the traditional method PCA + PSVM and the solution based on Autoencoder.