Calibration of a Generic Event-Level Anomaly Detection Trigger (GELATO) for the ATLAS Experiment

Modern collider experiments face increasing data rates that challenge traditional, threshold-based trigger systems. This report explores an alternative, machine-learning–driven approach: the GELATO anomaly trigger, based on a VAE–GAN architecture designed to identify unusual event topologies without relying on predefined selections. We evaluate its performance across several simulated Standard Model processes, studying correlations between anomaly score and reconstructed kinematic features. Distinct behaviors, such as enhanced sensitivity to subleading muons in the endcap region, demonstrate the trigger’s ability to exploit multi-object correlations beyond conventional trigger logic.