DeepTreeGAN: Fast Generation of High Dimensional Point Clouds

EPJ Web of Conferences, vol. 295, pp. 09010

Abstract

In High Energy Physics, detailed and time-consuming simulations are used for particle interactions with detectors. To bypass these simulations with a generative model, the generation of large point clouds in a short time is required, while the complex dependencies between the particles must be correctly modelled. Particle showers are inherently tree-based processes, as each particle is produced by the decay or detector interaction of a particle of the previous generation. In this work, we present a novel Graph Neural Network model (DeepTreeGAN) that is able to generate such point clouds in a tree-based manner. We show that this model can reproduce complex distributions, and we evaluate its performance on the public JetNet dataset.

Authors 4

1. Moritz A.W. Scham Aachen III. Physikalisches Institut A

   RWTH Aachen University · Deutsches Elektronen-Synchrotron DESY · Jülich Supercomputing Centre

   Affiliation as printed

   Deutsches Elektronen-Synchrotron DESY, Germany

   Jülich Supercomputing Centre, Institute for Advanced Simulation, Germany

   RWTH Aachen University, III. Physikalisches Institut A, Germany

2. Dirk Krücker

   Deutsches Elektronen-Synchrotron DESY

   Affiliation as printed

   Deutsches Elektronen-Synchrotron DESY, Germany

3. Benno Käch

   Deutsches Elektronen-Synchrotron DESY · Universität Hamburg

   Affiliation as printed

   Deutsches Elektronen-Synchrotron DESY, Germany

   Universität Hamburg, Institut für Experimentalphysik, Germany

4. K. Borras Aachen III. Physikalisches Institut A

   RWTH Aachen University · Deutsches Elektronen-Synchrotron DESY

   Affiliation as printed

   Deutsches Elektronen-Synchrotron DESY, Germany

   RWTH Aachen University, III. Physikalisches Institut A, Germany