Detail of the student project

Topic:Graph Neural Network Trackster Linking in CMS HGCal at CERN
Department:Katedra kybernetiky
Supervisor:Felice Pantaleo
Announce as:Diplomová práce, Semestrální projekt
Description:The CERN High-Luminosity Large Hadron Collider (HL-LHC) project aims to significantly increase the number of collisions in the LHC, posing a large data volume and complexity challenge to the current reconstruction algorithms. The aim of the project is to explore Graph Neural Network (GNN) approaches scalable to large data volumes within the framework of TICL [1] to improve the reconstruction performance of particle showers in the novel endcap High-Granularity Calorimeter (HGCal) [2] which is designed by the CMS Collaboration to sustain the harsher conditions of the HL-LHC.

Tracksters are three-dimensional structures formed by energy deposits in the calorimeter. Trackers formed by the same particle are frequently fragmented due to gaps between detector layers, noise, physical processes (such as Bremsstrahlung), and particle overlaps. The objective of the project is to reconnect the incomplete fragments generated by the same particle into well-formed tracksters and assess the scalability of the solutions to large datasets.
Bibliography:[1] Pantaleo, Felice, and Marco Rovere. “The Iterative Clustering framework for the CMS HGCAL Reconstruction”. No. CMS-CR-2022-037. 2022.

[2] HGCAL website. Retrieved September 27, 2022, from

[3] Qasim, Shah Rukh, et al. "Learning representations of irregular particle-detector geometry with distance-weighted graph networks." The European Physical Journal C 79.7 (2019): 1-11.

[4] Di Pilato, Antonio, et al. "Reconstruction in an imaging calorimeter for HL-LHC." Journal of Instrumentation 15.06 (2020): C06023.
Responsible person: Petr Pošík