Speaker
Description
We present NomAD (Nanosecond Anomaly Detection), an unsupervised machine learning algorithm developed for real-time anomaly detection in the ATLAS Level-1 Topological (L1Topo) trigger during Run 3. Combining a Variational Autoencoder with Decision Tree Regression, NomAD identifies rare and unconventional events in FPGA-based trigger hardware with low latency. Applied to dimuon events, the algorithm captures signals beyond standard selections, achieving up to a 21% increase in unique acceptance using B-Physics benchmarks. The anomaly detection trigger operates at a tunable rate, with around 1.8 kHz observed at a representative AD score threshold. This flexibility enables integration into existing trigger menus while maintaining sensitivity to new physics. This talk will cover the algorithm's design, performance, and its potential to enhance real-time event selection in high-energy physics.
