Anomaly detection with normalizing flows

Recent advances in machine learning have enabled new model-independent approaches to searching for physics beyond the Standard Model at the Large Hadron Collider. In this talk, I will present preliminary results of a study combining normalizing flows with the ABCD background estimation method for end-to-end out-of-distribution anomaly detection. I will discuss how normalizing flows can be used to construct two uncorrelated anomaly scores that define the ABCD plane, enabling both the detection of anomalous events and a data-driven estimation of their statistical significance.