Generative adversarial nets for unsupervised outlier detection

Xusheng Du, Jiaying Chen, Jiong Yu, Shu Li, Qiyin Tan

Abstract: Highlights•Unsupervised outlier detection with GAN networks.•The GANs prefer to fit the normal object distributions to minimize error.•The fake data generated by the GAN are used to train the autoencoder.•A trained autoencoder can detect outliers via only one forward propagation.