I. Introduction
Anomaly detection has been of practical interest to many fields of engineering, such as medical imaging and diagnosis [1], industry process fault detection [2] and cyber security [3]. Anomaly detection methods need a model given through historical data of a system in normal situations to describe its normality, and an anomaly scoring function obtained to estimate the anomaly of new data samples. Once the anomaly score of a sample is higher than a certain critical value, the sample is determined to be an abnormal sample.