Machine Learning and Track Reconstruction for the MOLLER Experiment

The purpose of this project was to develop a neural network to reconstruct particle trajectories in the MOLLER experiment. The MOLLER experiment is a collaboration at Jefferson Lab which plans on testing the Standard Model by measuring the parity-violating asymmetry in Moller scattering.  If the measurement disagrees with the theoretical value for the asymmetry, then it will provide evidence for physics beyond the Standard Model, and if it agrees, it will restrict many beyond-the-Standard-Model theories being developed. 

A necessary aspect of this experiment will be the reconstruction of particle trajectories. Neural networks are efficient pattern recognition tools and are equipped to handle large datasets, so they are a potential solution to this issue. Using data generated by the Geant 4 simulation for MOLLER, I trained a recurrent neural network which connects signals in a series of tracking detectors to signals in the main detector located downstream. It does so by predicting the position of a particle hit in the main detector from that particle's hit positions in each of the tracking planes.  The network predicts the main detector hit positions with better than 1% error.