Gaussian Mixture Models as Automated Particle Classifiers for Fast Neutron Detectors

• Activity Number: 175 - Contributed Poster Presentations: Section on Statistical Learning and Data Science
• Type: Contributed
• Date/Time: Monday, July 31, 2017 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Learning and Data Science
• Author(s): Brenton Blair* and Ron Wurtz
• Companies: Lawrence Livermore National Laboratory and Lawrence Livermore National Laboratory
• Keywords: clustering ; mixture model ; nuclear detection

Abstract

Pulse Shape Discrimination (PSD) is the task of classifying pulse shapes for different particle types (e.g. gamma rays vs. fast neutrons). For decades, this field has neglected to adopt standard techniques found in the statistical learning community. Instead, methods initially employed in the 1960's persist in the current PSD literature. Despite vast amounts of potential data that can be collected at low energy levels, traditional PSD methods are unable to discriminate particles below a certain threshold. In this poster, I will demonstrate how Gaussian Mixture Models (GMMs) can be used as a clustering technique for fast neutron detection in the absence of labeled data. GMMs yield improvements spanning the energy spectrum in a desirably efficient, unsupervised fashion. An extension to the Dirichlet Process GMM will also be discussed.