Cluster Purging: Efficient Outlier Detection based on Rate-Distortion Theory