Machine Learning based Anomaly Detection for Online Data Quality Monitoring of the CMS Electromagnetic Calorimeter

Online Data Quality Monitoring (DQM) of the CMS electromagnetic calorimeter (ECAL) is a vital operational tool that allows ECAL experts to quickly identify, localize, and diagnose a broad range of detector issues that would otherwise prevent physics-quality data taking. Although the ECAL DQM system has been in operation since the start of the LHC and continuously updated to respond to new problems, it is challenging to anticipate anomalies in different shapes and sizes that had not been observed before. With the need for a more robust anomaly detection system, a real-time unsupervised machine learning based method is developed that can catch ECAL anomalies unseen in past data. After accounting for spatio-temporal deviations in the ECAL response, the Auto-Encoder based online DQM system is able to detect and localize anomalies with an estimated false discovery rate of 10^{-2} to 10^{-4} at 99% anomaly detection rate. We present anomaly detection results from the ECAL Barrel and Endcap regions, including new results from early LHC Run3 collision data.