Deep Learning in High Energy Physics

The study of the neutrino is currently one of most promising ways to search beyond Standard Model Physics. Already, the unexpected discovery that neutrinos can transform from one type to another through a process known as neutrino oscillations is beyond the Standard Model effect. Through further measurements of neutrino oscillations, the field aims to answer additional open questions such as “do neutrinos and anti-neutrinos behave the same way?” and “how many types of neutrinos are there?”. The answer to these questions will impact not only our understanding of particle physics but also will influence our models of how the universe evolved. However, to get to these answers, neutrino experiments will need to move into an era of high-precision measurements. One way to achieve this is through the use of a detector known as a liquid argon time-projection chamber, or LArTPC. These detectors are capable of producing high-resolution images of neutrino interactions that can be used to reject backgrounds more effectively than past experiments, through a method called Deep Learning, in particular convolution neural networks. In this talk, Wongjirad will focus on one effort to use convolutional neural networks (CNNs) to reconstruct and select neutrino events. CNNs, a type of machine learning algorithm, are often the state-of-the-art approach in many computer vision tasks. Wongjirad will describe first steps in applying CNNs to the task of analyzing neutrino events in LArTPCs and discuss possible future directions.