Particle Cloud Generation with Message Passing Generative Adversarial Networks

In high energy physics (HEP), jets are collections of correlated particles\nproduced ubiquitously in particle collisions such as those at the CERN Large\nHadron Collider (LHC). Machine learning (ML)-based generative models, such as\ngenerative adversarial networks (GANs), have the potential to significantly\naccelerate LHC jet simulations. However, despite jets having a natural\nrepresentation as a set of particles in momentum-space, a.k.a. a particle\ncloud, there exist no generative models applied to such a dataset. In this\nwork, we introduce a new particle cloud dataset (JetNet), and apply to it\nexisting point cloud GANs. Results are evaluated using (1) 1-Wasserstein\ndistances between high- and low-level feature distributions, (2) a newly\ndeveloped Fr\\'{e}chet ParticleNet Distance, and (3) the coverage and (4)\nminimum matching distance metrics. Existing GANs are found to be inadequate for\nphysics applications, hence we develop a new message passing GAN (MPGAN), which\noutperforms existing point cloud GANs on virtually every metric and shows\npromise for use in HEP. We propose JetNet as a novel point-cloud-style dataset\nfor the ML community to experiment with, and set MPGAN as a benchmark to\nimprove upon for future generative models. Additionally, to facilitate research\nand improve accessibility and reproducibility in this area, we release the\nopen-source JetNet Python package with interfaces for particle cloud datasets,\nimplementations for evaluation and loss metrics, and more tools for ML in HEP\ndevelopment.\n