Speaker
Description
Anomaly detection expands on traditional model-focused search programs at colliders via data-driven machine learning (ML) selection algorithms. Deploying anomaly detection at filtering (trigger) level enables the acquisition of fundamentally novel events, probing below historically constraining energy thresholds. However, successful use of such ML algorithms in the trigger requires attention to algorithm resources, latency, and hardware implementation. We present the latest on the Generic Event-Level Anomaly Trigger Option (GELATO) for the ATLAS experiment. The algorithm will be described, namely its autoencoder-based structure, its deployment across the Level-1 and High Level Triggers, and its performance for selected beyond the Standard Model benchmarks. The introduction of GELATO to the ATLAS Run 3 menu will also be covered, including 2025 commissioning and a first look at 2026 data-taking, along with a discussion of future anomaly triggers for Run 4 of the LHC.
| I read the instructions above | Yes |
|---|