Date Thesis Awarded


Access Type

Honors Thesis -- Access Restricted On-Campus Only

Degree Name

Bachelors of Science (BS)




Justin Stevens

Committee Members

David S. Armstrong

Daniel Vasiliu


The objective of this study is to investigate Convolutional Neural Networks
(CNNs) as a means of improving particle identification (PID) capabilities in the GlueX Detector at Jefferson Lab (JLab). The GlueX experiment aims to map the spectrum of exotic hybrid mesons through light-quark meson spectroscopy. Lattice Quantum Chromodynamics predicts a wide veriety of hybrid mesons produced by gluonic excitations which decay to kaon final states, making kaon identification extremely important. The GlueX detector currently uses Detection of Internally Reflected Cherenkov Radiation (DIRC) with the goal of facilitating particle identification up to particle momenta near 4 GeV. The DIRC detector utilizes an array of fused silica bars that yield Cherenkov radiation created by charged kaons and pions to produce image-like data. These data lend themselves nicely to common machine learning image classification methods, particularly CNNs. We therefore look to use CNNs as a method of expanding the momentum range where PID is possible. Following a previous study of CNN-based PID, this study aims to optimize the hyperparameters of one of the more successful models using an expanded data set of kaon and pion data with a momentum of 3 GeV.

On-Campus Access Only