NomAD: Unsupervised Machine Learning for Real-Time Anomaly Detection in the ATLAS Level-1 Trigger

Unsupervised machine learning models are a growing tool being deployed at colliders to identify rare signals in the first-level trigger system. In this presentation, we discuss the training and deployment of the NomAD (Nanosecond Anomaly Detection) in the ATLAS Level-1 Topological trigger. The algorithm is trained on level-1 muon information. The first phase uses a Variational Autoencoder and the second phase reduces this model using a BDT for implementation on an FPGA. We present results of the model using Run 3 data collected in 2026.