Speaker
Description
Detector simulation and reconstruction are significant computational bottlenecks in particle physics. A state-of-the-art GenAI-based paradigm, Particle-flow Neural Assisted Simulations (PARNASSUS), has shown great promise for fast simulation in the context of CMS Open Data. Unlike conventional fast simulation models that target only simulation, PARNASSUS is an end-to-end approach that goes from generator events directly to reconstructed events. PARNASSUS uses the full set of detector-stable particles to conditionally generate a corresponding set of reconstructed particles via a flow matching objective. We train this algorithm on full-simulation ATLAS samples using particle flow objects as a training target. The results show that PARNASSUS models event-level and jet-level quantities accurately, often outperforming the existing fast simulation toolkit in ATLAS.