Improvement of the NOvA Near Detector Event Reconstruction and Primary Vertexing through the Application of Machine Learning Methods

Abstract

The purpose of this work is to examine the application of a deep learning model in event reconstruction of neutrino interactions. The challenges faced in event reconstruction include the placement of an accurate primary neutrino interaction vertex which is used to support the particle track and prong algorithms. The result of accurate primary vertex ensures all particles involved in a neutrino interaction are included. We propose a regression-based Convolutional Neural Network (CNN) method to predict the primary vertex of a particle interaction. We show that with raw two-dimensional pixel map views as input, the regression-based CNN can predict the primary vertex in all three coordinates. This work is applied as part of the NOvA (NuMI Off-axis $\nu_e$ Appearance) near detector reconstruction efforts. The primary vertex predicted by the regression-based CNN model shows promising results for future applications. This deep learning method can be extended to secondary vertexing through a Kernel Density Estimate algorithm discussed in this work.