Deep anomaly detection on set data: Survey and comparison

Michaela Masková, Matej Zorek, Tomás Pevný, Václav Smídl

Abstract: Highlights•Identified building blocks of anomaly detectors for set data.•Modular variants of anomaly scores are formulated and analyzed for sensitivity to set cardinality.•A publicly available dataset repository is created for comparative study.•Fair comparison of deep anomaly detectors including sensitivity to hyper-parameters.