Search for New Physics Using Deep Learning in MicroBooNE

by KAZUHIRO TERAO (C)

Tuesday 7 Mar 2017, 12:30 → 13:30 US/Pacific

Madrone (SLAC)

Since the discovery of neutrino oscillation, experiments have measured all mixing angles and squared-mass splittings associated with a three-neutrino-flavor model. Despite revolutionary advancements of this field, however, there are still remaining physics questions that can be answered from the oscillation measurements. These include the neutrino mass hierarchy, CP violation in lepton sector, and the possible presence of sterile neutrinos. These searches require high-precision detector-technology with good energy resolution, high signal efficiency and excellent background rejection capability. A promising new technology is the Liquid Argon Time Projection Chamber (LArTPC). This is the proposed detector for the Short Baseline Neutrino (SBN) and Deep Underground Neutrino Experiment (DUNE) programs. MicroBooNE, a part of the SBN program, employs the first large scale (> 100 ton) LArTPC detector in the U.S. The goal is to perform a definitive study of the observed event excess at low energy by the MiniBooNE experiment, which could indicate the presence of sterile neutrinos. The current challenge of the experiment is efficient and effective event reconstruction to identify any possible event excess signal, which is a also part of necessary R&D for the future experiments. In this talk, I describe the use of the machine learning technique called Deep Learning to these problems. Deep Learning is making revolutionary advancements in the field of artificial intelligence and computer vision. We demonstrate that Convolutional Neural Networks (CNNs), a type of Deep Learning algorithm, can also be used for event reconstruction using LArTPC data. I will discuss the results of our recently published study, as well as the current status of the application of this technique for the MicroBooNE oscillation analysis.

SLACSeminar_Mar2017.pdf