Particle Cloud Generation with Message Passing Generative Adversarial Networks
Authors: Raghav Kansal, Javier Duarte, Hao Su, Breno Orzari, Thiago Tomei, Maurizio Pierini, Mary Touranakou, jean-roch vlimant, Dimitrios Gunopulos
NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this work, we introduce a new particle cloud dataset (Jet Net), and apply to it existing point cloud GANs. Results are evaluated using (1) 1-Wasserstein distances between highand low-level feature distributions, (2) a newly developed Fréchet Particle Net 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. |
| Researcher Affiliation | Academia | Raghav Kansal, Javier Duarte, Hao Su University of California San Diego La Jolla, CA 92093, USA Breno Orzari, Thiago Tomei Universidade Estadual Paulista São Paulo/SP CEP 01049-010, Brazil Maurizio Pierini, Mary Touranakou European Organization for Nuclear Research (CERN) CH-1211 Geneva 23, Switzerland Jean-Roch Vlimant California Institute of Technology Pasadena, CA 91125, USA Dimitrios Gunopulos National and Kapodistrian University of Athens Athens 15772, Greece |
| Pseudocode | No | The paper describes the architecture and steps of the MPGAN but does not include a formal pseudocode block or algorithm listing. |
| Open Source Code | Yes | 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. ... Implementations for all metrics are provided in the JETNET package [3]. ... Training and implementation details for each can be found in App. D, and all code in Ref. [43]. [3] R. Kansal, JETNET Library . doi:10.5281/zenodo.5597893, https://github.com/jet-net/Jet Net. [43] R. Kansal, MPGAN Code . doi:10.5281/zenodo.5598407, https://github.com/rkansal47/MPGAN/tree/neurips21. |
| Open Datasets | Yes | We publish Jet Net [15] under the CC-BY 4.0 license, to facilitate and advance ML research in HEP, and to offer a new point-cloud-style dataset to experiment with. ... [15] R. Kansal et al., Jet Net , 05, 2021. doi:10.5281/zenodo.4834876. |
| Dataset Splits | Yes | The real samples are split 70/30 for training/evaluation. |
| Hardware Specification | Yes | We additionally perform a latency measurement and find, using an NVIDIA A100 GPU, that MPGAN generation requires 35.7 µs per jet. In comparison, the traditional generation process for Jet Net is measured on an 8-CPU machine as requiring 46ms per jet... |
| Software Dependencies | No | The paper mentions software tools like PYTHIA [5], HERWIG [6], and the JETNET package [3], along with Particle Net [10], but it does not specify version numbers for general software dependencies or machine learning frameworks (e.g., PyTorch version, Python version). |
| Experiment Setup | Yes | We choose model parameters which, during training, yield the lowest W M 1 score. ... We choose 2 MP layers for both networks. |