Using Deep Learning Techniques to Search for the MiniBooNE Low Energy Excess in MicroBooNE with > 3$\sigma$ Sensitivity

TL;DR

This paper presents a deep learning-based analysis for the MicroBooNE experiment, achieving over 3σ sensitivity in testing the MiniBooNE low energy excess hypothesis with a novel reconstruction approach.

Contribution

It introduces a novel deep learning and standard algorithmic analysis pipeline to identify electron neutrino appearance with high purity in MicroBooNE data.

Findings

Achieves 3.2σ sensitivity to exclude MiniBooNE anomaly

Develops a highly pure sample of neutrino interactions

Provides detailed data-to-simulation comparison at each analysis step

Abstract

This thesis describes an analysis developed for the MicroBooNE experiment to investigate an anomalous excess of electron-like events observed in the MiniBooNE detector. The hypothesis investigated here is that the MiniBooNE anomaly represents appearance of electron neutrinos. Using an amalgam of novel Deep Learning and standard algorithmic techniques this analysis reconstructs and identifies a highly pure sample of charged current quasi-elastic muon neutrino and electron neutrino interactions. This thesis describes the steps in the analysis chain and provides data-to-simulation comparisons for each step that establish confidence in the final prediction. When interpreted in the context of a $\nu e$ appearance like model, this analysis predicts a 3.2$\sigma$ sensitivity to exclude a standard model fluctuation which would appear as a MiniBooNE like anomaly using $7\times10^{20}$ protons on target of MicroBooNE Data.