Real-time Anomaly Detection at the CMS Experiment

Isobel Ojalvo, Princeton

During Run-3 at the LHC, the search for new physics has significantly expanded. Development of advanced algorithms for the Level-1 trigger, encompassing techniques such as Long-Lived Particles, Data Scouting, and Muon showers to ensure broad data selection at the CMS experiment. Here, we introduce CICADA (Calorimeter Image Convolutional Anomaly Detection Algorithm), a state-of-the-art, fully autonomous AI algorithm engineered to process LHC event data in real-time and trigger on anomalous topologies. Operating directly on the rawest recorded data - the calorimeter energy deposits - CICADA demonstrates sensitivity to Standard Model processes. It unveils potential for Beyond the Standard Model Physics, including the exploration of previously untapped phase space at the LHC.