Fermi Gamma-ray Space Telescope

LAT Particle Discrimination Using Deep Learning

B. Anderson
J. Conrad

Abstract:

Machine learning methods including one or more hidden layers have recently become a feasible solution for a wide range of problems, including particle identification in detectors. Given a set of training data, methods of so-called ?deep learning? can model both low and high-level features to accurately classify events. We present a basic implementation of the multilayer perceptron algorithm to classify protons and photons in the Fermi Large Area Telescope (LAT) and compare it with the successful current system of boosted decision trees. We also include a preliminary investigation of the method?s usefulness in a more challenging case ? discriminating events which interact solely in the LAT calorimeter.