Anomaly Detection in the TDAQ System of ATLAS: Studying the HLT farm as proof of concep

Martes 16 de abril - 14:00 hs


Quedan todas/os invitadas/os al próximo seminario del Ciclo de Coloquios y Seminarios 2024 del IFLP, a realizarse el martes 16 de abril a las 14hs.

El mismo estará a cargo del Dr. Joaquín Hoya - Argonne National Laboratory.

Su charla será sobre: Anomaly Detection in the TDAQ System of ATLAS: Studying the HLT farm as proof of concept.
Se envía a continuación un breve resumen de la misma:

The ATLAS experiment at the Large Hadron Collider (LHC) explores fundamental particle physics through the analysis of collision data. At its core, the Trigger and Data Acquisition system (TDAQ) plays a central role in selecting and transferring the interesting events from frontend electronics to permanent storage. Given its responsibility for managingvast amounts of data in near-real-time, the TDAQ system occupies a uniquely privileged position suitable for early anomaly detection. This presentation introduces a proof of concept centered on leveraging the Data Collection Manager (DCM) rate to identify potential issues within the High-Level Trigger (HLT) farm. The HLT farm, a dynamically-configured network of computing resources responsible for the final stage of event selection, represents a critical component of any data taking session. Our approach involves a hybrid methodology incorporating both clustering, i.e. DBSCAN, and transformer-based outlier detection techniques. This combination enables the detection of subtle anomalies affecting specific features as well as long-term outliers, providing early alerts of potential system irregularities. We discuss ongoing efforts to enhance and expand our anomaly detection solution, including the incorporation of additional data sources. By extending our detection capabilities, and including other components of the system, we aim to establish a framework for anomaly detection throughout the entire data processing pipeline. The ultimate objective is to extend our focus beyond operational data, and include also the physics data, that would enable us to directly evaluate Data Quality from the TDAQ.

El seminario tendrá lugar en el auditorio "Prof. Dr. Luis N. Epele" del IFLP, sito en la diagonal 113 entre 63 y 64.

Esperamos contar con su participación.


+54 (0221) 6443202

icono ubicacion

Diagonal 113 entre 63 y 64

La Plata (1900) - Buenos Aires - Argentina