Outlier Generation

We propose ARN, a semisupervised anomaly detection and generation method based on adversarial reconstruction. ARN exploits a regularized autoencoder to optimize the reconstruction of variants of normal examples with minimal differences, that are …