You are here

Particle Tracking Reconstruction with the ExaTrkX Pipeline

Dr. Xiangyang Ju, Lawrence Berkeley National Lab
Thursday, February 16, 2023 - 3:30pm
PAT C-421 (Zoom: https://washington.zoom.us/my/lubatti)

When a charged particle traverses the pixel detectors at the LHC, it leaves a list of hits (i.e. energy deposits) in the detector. The High-Luminosity LHC (HL-LHC) will create about 10,000 charged particles in one collision event, producing about 100,000 hits in the detector. Particle tracking is to reconstruct the 10,000 charged particles from the 100,000 hits. It is a challenging problem because of its combinatorial complexity. Existing algorithms based on Kalman Filtering method scale worse than linear against the number of charged particles in the event. Much work must be done to fine tune the Kalman Filtering method to meet the demands of HL-LHC. Instead, we developed a Graph Neural Network (GNN)-based track finding algorithm, the ExaTrkX pipeline. In this talk, we will present the ExaTrkX pipeline and its applications at different experiments. More important, we will discuss its potential for online and offline particle tracking.

Dr. Xiangyang Ju is a Computing System Engineer, working on Deep Learning methods for High Energy Physics problems and ATLAS core software. His activities include: exascale machine learning for particle physics tracking with the Exa.TrkX project, research on generative models, and GPU-accelerated inference frameworks. He obtained his Ph.D. degree from the University of Wisconsin-Madison in 2018 with a thesis titled "Observation of the Standard Model Higgs boson and search for an additional scalar in the four-lepton final state with the ATLAS detector.

Share