Resonant anomaly detection with multiple reference datasets

A bstract An important class of techniques for resonant anomaly detection in high energy physics builds models that can distinguish between reference and target datasets, where only the latter has appreciable signal. Such techniques, including Classification Without Labels (CWoLa) Simulation Assisted Likelihood-free Anomaly Detection (SALAD) rely on a single dataset. They cannot take advantage commonly-available multiple datasets thus fully exploit available information. In this work, we propose generalizations CWoLa SALAD settings are available, building weak supervision techniques. We demonstrate improved performance number with realistic synthetic data. As an added benefit, our enable us to provide finite-sample guarantees, improving existing asymptotic analyses.