Parnassus: A GPU-enabled, Python-based Package for Fast Particle Detector Simulation and Reconstruction

by Eilam Gross (Weizmann Institute of Science (IL))

Wednesday 30 Sept 2026, 11:00 → 12:00 Europe/Zurich

503/1-001 - Council Chamber (CERN)

Abstract:
Generative AI models, including those behind image-generation tools, have demonstrated remarkable capabilities for transforming random inputs into coherent outputs. Inspired by these advancements, we've developed PARNASUS, a deep-learning model designed for particle physics. Parnassus processes point clouds representing particles interacting with a detector and outputs reconstructed particle data.  Parnassus accurately replicates the particle flow algorithm and generalizes well beyond its training set. 

PARNASUS shows that flow-matching AI can deliver full-simulation-quality detector reconstruction at fast-simulation speeds, paving the way for the next generation of HEP simulation. Its initial release includes two models of the CMS detector: one based on a flow-matching neural network architecture and one based on a PyTorch implementation of the Delphes CMS card (parametric bias and smearing). PyTorch versions of the ATLAS and ALEPH Delphes cards are also available. PARNASSUS is a paradise for experimentalists as well as phenomenologists who want to test their models in a real experimental environment.

Bio: 

Professor Eilam Gross is an experimental particle physicist at the Weizmann Institute of Science and a long-standing member of the ATLAS Collaboration. As co-convener of the ATLAS Higgs Physics Group, he played a leading role in the searches that culminated in the discovery of the Higgs boson in 2012. He also helped establish the statistical framework now widely used for searches and measurements at the LHC, notably through seminal work on likelihood-based tests and the look-elsewhere effect. Over the past decade, his research has increasingly focused on deep-learning applications in particle physics, including particle reconstruction, anomaly detection, and generative approaches to detector simulation.

 

Coffee served at 10h30