Particle Cloud Generation with Message Passing Generative Adversarial Networks

dc.contributor.authorKansal, Raghav
dc.contributor.authorDuarte, Javier
dc.contributor.authorSu, Hao
dc.contributor.authorOrzari, Breno [UNESP]
dc.contributor.authorTomei, Thiago [UNESP]
dc.contributor.authorPierini, Maurizio
dc.contributor.authorTouranakou, Mary
dc.contributor.authorVlimant, Jean-Roch
dc.contributor.authorGunopulos, Dimitrios
dc.contributor.institutionUniversity of California San Diego
dc.contributor.institutionUniversidade Estadual Paulista (UNESP)
dc.contributor.institutionEuropean Organization for Nuclear Research (CERN)
dc.contributor.institutionCalifornia Institute of Technology
dc.contributor.institutionUniversity of Athens
dc.description.abstractIn high energy physics (HEP), jets are collections of correlated particles produced ubiquitously in particle collisions such as those at the CERN Large Hadron Collider (LHC). Machine learning (ML)-based generative models, such as generative adversarial networks (GANs), have the potential to significantly accelerate LHC jet simulations. However, despite jets having a natural representation as a set of particles in momentum-space, a.k.a. a particle cloud, there exist no generative models applied to such a dataset. In this work, we introduce a new particle cloud dataset (JetNet), and apply to it existing point cloud GANs. Results are evaluated using (1) 1-Wasserstein distances between high- and low-level feature distributions, (2) a newly developed Fréchet ParticleNet Distance, and (3) the coverage and (4) minimum matching distance metrics. Existing GANs are found to be inadequate for physics applications, hence we develop a new message passing GAN (MPGAN), which outperforms existing point cloud GANs on virtually every metric and shows promise for use in HEP. We propose JetNet as a novel point-cloud-style dataset for the ML community to experiment with, and set MPGAN as a benchmark to improve upon for future generative models. Additionally, to facilitate research and improve accessibility and reproducibility in this area, we release the open-source JETNET Python package with interfaces for particle cloud datasets, implementations for evaluation and loss metrics, and more tools for ML in HEP development.en
dc.description.affiliationUniversity of California San Diego, La Jolla
dc.description.affiliationUniversidade Estadual Paulista, SP
dc.description.affiliationEuropean Organization for Nuclear Research (CERN)
dc.description.affiliationCalifornia Institute of Technology
dc.description.affiliationNational and Kapodistrian University of Athens
dc.description.affiliationUnespUniversidade Estadual Paulista, SP
dc.identifier.citationAdvances in Neural Information Processing Systems, v. 29, p. 23858-23871.
dc.relation.ispartofAdvances in Neural Information Processing Systems
dc.titleParticle Cloud Generation with Message Passing Generative Adversarial Networksen
dc.typeTrabalho apresentado em evento