Machine-Learning Algorithms for Particle-Pattern Classification
Fri., Nov. 1, 2019 3:30 p.m. - Fri., Nov. 1, 2019 4:30 p.m.
Location: LB 235
Abstract: The GlueX Experiment will focus on the exploration of the light-quark domain, searching for exotic hybrid mesons produced in photon-proton collisions. A key subsystem of the GlueX detector is the barrel calorimeter (BCAL) - designed and built at Regina - whose geometry is a hollow cylinder. The BCAL is a 30-ton calorimeter, comprised of layers of lead sheets and scintillating fibres. It is positioned inside a superconducting solenoid and is instrumented with 3840 silicon photo sensors. High-energy charged (electrons, pion, protons) and neutral (photons and neutrons) particles strike the BCAL and develop ”hit" patterns of energy deposition in its 1536 electronic readout cells, called clusters, which are defined as topologically-connected groups of hit cells. The energy of a cluster is the sum of the energies measured in its cells. Proper accounting of cluster energies is constrained by energy and momentum conservation. Impinging charged particles benefit from tracking information in the GlueX drift chambers positioned inside the BCAL, and track-fitting matches their trajectories with “hits" in the BCAL. Neutral particles travel in straight lines, leave no signature in the chambers, but develop hadronic and electromagnetic showers inside the BCAL that generate different hit-cell patterns. In certain regions of the BCAL shower-particle associations are often misidentified using the current cut-based analysis. Supervised Machine Learning (ML) algorithms will be developed to optimize pattern recognition and energy assignment. ML algorithms exploit high-dimensional correlations and effectively function as dimensional-reduction methods.
The focus of the presentation will be on machine learning models used in the reconstruction of neutral pions and its potential use in calibration; along with the development of a model used to distinguish between photon and neutrons showers within the BCAL.
Speaker: James Giroux, University of Regina