Developing Machine Learning Models for Proton Computed Tomography and LHCb Particle Tracking

This thesis describes novel, proof-of-concept machine learning models for particle tracking in fundamental research and medicine. Proton computed tomography is a medical imaging technology with the potential to improve on current medical proton therapy treatment planning, but hampered by the computationally costly need to reconstruct individual proton tracks. With this in mind, we developed the Proton Path Neural Network, a neural network model capable of matching, and in some situations exceeding, the performance of the standard reconstruction method, with a significantly shorter execution time. Building on this experience, we turned to pattern recognition within track reconstruction at the LHCb experiment, one of the four major detector experiments located at CERN’s flagship LHC particle accelerator. Focusing on reconstruction within the VELO tracking subdetector, we developed a graph neural network approach with the capacity to draw inferences from all measurements made by the subdetector for a given event, which exhibited promising performance over existing trials.