My research has combined machine learning techniques with neutrino physics. I am part of the MicroBooNE collaboration. Specifically I am on a team using machine learning techniques to analyze the data and search for events where a low energy neutrinos interact with the liquid argon in the detector to produce one electron and one proton (1e1p). My current work is focused on one of the major backgrounds to these types of events where a neutral pion is produced and decays into two photons. For my all of my work, I look at data points which shows the charge deposited by neutrino events in the detector at a specific time and location and use this to reconstruct the neutrino event that caused the charge deposition.

Projects:

Infill Neural Network: The Infill network is a generative convolutional neural network which is part of the larger deep-learning MicroBooNE analysis effort. The main issue this project is attempting to address is dead channels in the MicroBooNE detector. These often cause failures in simple line-following algorithms. The goal of the infill network is to fill in the dead channels with realistic charge values to improve the performance of both line-following algorithms and deep-learning pattern recognition algorithms later in the analysis chain.

Shower Reconstruction Algorithm: This algorithm does not use deep-learning itself, but uses the results of a network. Electrons and photons in MicroBooNE creating shower-like patterns. This work takes the result of a neural net which classifies data points as being track (proton, muon, charged pion) or shower. The shower pixels are clustered and the charge is used to reconstruct 3D kinematic variables of the particle.

Neutral Pion Event Selection: Neutral pion events are one of the major backgrounds to 1e1p events. They decay into two photons and a photon is often confused for an electron in the reconstruction. I have worked on isolating a sample of neutral pions. This sample has allowed for further understanding of the 1e1p background. It also has provided useful validation of the shower reconstruction algorithm described above.

To be continued...!