Karlsruhe 2024 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
T: Fachverband Teilchenphysik
T 43: Data, AI, Computing 3 (pointclouds & graphs)
T 43.2: Vortrag
Dienstag, 5. März 2024, 16:15–16:30, Geb. 30.33: MTI
Flow Matching Beyond Kinematics: Generating Jets with Particle-ID and Trajectory Displacement Information — •Joschka Birk1, Erik Buhmann1, Cedric Ewen1, Gregor Kasieczka1, and David Shih2 — 1Universität Hamburg — 2Rutgers University
Generative machine learning models are extensively researched in HEP for applications like anomaly detection and fast detector simulation. So far, the development of methods for jet generation has mainly focused on the JetNet dataset. However, as the complexity of generative models trained on the JetNet dataset increased, the lack of statistics in this dataset started to become a bottleneck. We present the first generative model trained on the more complex JetClass dataset, which was originally introduced with the ParT jet tagging algorithm. The JetClass dataset is much larger and contains more jet types as well as additional features that are not included in the JetNet dataset, which opens up new possibilities for jet generation development. Our model generates jets at the constituent level and is a permutation-equivariant continuous normalizing flow (CNF) trained with the flow-matching technique. It is conditioned on the jet type so that a single model can be used to generate the ten different jet types of the JetClass. For the first time, we also introduce a generative model that goes beyond the kinematic features of the jet components by including features such as the particle ID and the track impact parameter. We show that our CNF can accurately model these additional features as well, extending the versatility of existing methods for jet generation.
Keywords: Generative models; Machine learning; Jets