Speaker
Description
During Run 3 of the LHC, the CMS experiment introduced real-time anomaly detection algorithms for online event selection for the first time. The deployment of AXOL1TL and CICADA demonstrated the feasibility of signal-agnostic autoencoder models for ultra-low-latency inference in Level-1 trigger hardware. Building on AXOL1TL, a second anomaly-detection layer has been explored for the software-based High-Level Trigger, aiming to improve the purity of the anomaly stream while enabling the development of particle-level transformer models. For Phase-2 CMS at the HL-LHC, the focus shifts toward transformer-based architectures that learn physics-informed embeddings of Level-1 trigger-reconstructed events. This reflects a shift from task-specific models toward structured latent representations that can be leveraged across multiple trigger applications, including anomaly detection, and can represent a first step toward foundation models for trigger systems. The talk reviews recent anomaly detection results in CMS and presents ongoing developments in embedding-based inference, highlighting the evolution toward increasingly sophisticated approaches in the Phase-2 CMS trigger system.
| Do you plan to submit a 4-page extended abstract on OpenReview (only for Presentations/Posters)? | Yes |
|---|