Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

T: Fachverband Teilchenphysik

T 65: Trigger+DAQ 2

T 65.5: Vortrag

Mittwoch, 6. März 2024, 17:00–17:15, Geb. 30.23: 3/1

Implementation of Graph Neural Networks for online track reconstruction at the ATLAS experiment — •Poppy Hicks, Sebastian Dittmeier, and André Schöning — Physikalisches Institut, Universität Heidelberg, Heidelberg

The upcoming High Luminosity upgrade to the LHC poses several challenges, most notably the huge increase in data to process. This necessitates improvements to the Trigger and Data Acquisition systems at the ATLAS experiment, including to its final stage, the Event Filter. Significant effort is being invested into computing R&D for the Event Filter, to keep resources within capacity; one promising avenue is the use of algorithms based on graph neural networks (GNNs) for track reconstruction in the Inner Tracker detector. GNNs demonstrate exceptional capability at modelling complex relationships within graph-structured data. Here, a graph represents detector hits as nodes; edges connecting these nodes represent the possibility the hits belong to the same particle. A GNN is used to score these edges to quantify that probability. In this talk, an overview of the use of GNNs for track reconstruction will be summarized; the focus will be on optimizing graphs for subsets of the data, in the pursuit of minimizing GPU memory requirements and maximising throughput.

Keywords: ATLAS; TDAQ; Track Reconstruction; Graph Neural Networks