Performance of a Geometric Deep Learning Pipeline for HL-LHC Particle\n Tracking

Abstract

The Exa.TrkX project has applied geometric learning concepts such as metric\nlearning and graph neural networks to HEP particle tracking. Exa.TrkX's\ntracking pipeline groups detector measurements to form track candidates and\nfilters them. The pipeline, originally developed using the TrackML dataset (a\nsimulation of an LHC-inspired tracking detector), has been demonstrated on\nother detectors, including DUNE Liquid Argon TPC and CMS High-Granularity\nCalorimeter. This paper documents new developments needed to study the physics\nand computing performance of the Exa.TrkX pipeline on the full TrackML dataset,\na first step towards validating the pipeline using ATLAS and CMS data. The\npipeline achieves tracking efficiency and purity similar to production tracking\nalgorithms. Crucially for future HEP applications, the pipeline benefits\nsignificantly from GPU acceleration, and its computational requirements scale\nclose to linearly with the number of particles in the event.\n