MicroBooNE Investigation of Low-Energy Excess Using Deep Learning Algorithms

TL;DR

This paper reports on the use of deep learning, specifically convolutional neural networks, combined with traditional methods to investigate low-energy electron-neutrino interactions in the MicroBooNE experiment, aiming to clarify the low-energy excess observed in previous experiments.

Contribution

It introduces a hybrid analysis approach combining CNNs with traditional reconstruction methods for neutrino event identification in MicroBooNE.

Findings

Development of CNN-based event classification techniques

Addressing domain adaptation issues between simulated and real data

Progress in identifying low-energy electron-neutrino interactions

Abstract

MicroBooNE is a neutrino experiment based at Fermilab which consists of a liquid argon time-projection chamber in the Booster Neutrino Beam (BNB). The experiment aims to investigate the excess of electron-neutrino-like events seen by the MiniBooNE experiment, also located in the BNB, which is potential evidence for new non-Standard Model physics such as sterile neutrinos. I discuss the status of a search for low-energy electron-neutrino interactions within the MicroBooNE detector. This analysis features a hybrid approach of traditional reconstruction methods along with the use of convolutional neural networks (CNNs), a type of deep learning algorithm highly adept at pattern recognition. I describe the identification of events and the ways in which the CNNs are used. I also outline the ways that we are addressing issues related to applying CNNs, which are trained on simulated data, to data from the detector.