Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

T: Fachverband Teilchenphysik

T 50: Data, AI, Computing, Electronics V

T 50.3: Vortrag

Mittwoch, 18. März 2026, 16:45–17:00, KH 00.024

Unbinned, High-dimensional Precision Measurements through the Lens of Deep Learning — •Jingjing Pan — Karlsruhe Institute of Technology, Karlsruhe, Germany

Unbinned, high-dimensional machine learning-based unfolding has rapidly progressed from a conceptual method to a practical analysis tool now deployed across multiple experiments, including but not limited to ATLAS, CMS, H1, LHCb, STAR and T2K. Building on the classifier-based framework of OmniFold, recent work has consolidated best practices for validation, calibration, uncertainty quantification, and data-release format, enabling robust unbinned measurements in the natural high-dimensional phase space of experimental data. Two recent analyses that highlight this progress are presented in this talk.

The recent H1 measurement performs the first OmniFold unfolding of all final-state particles in high-Q² events using a point-edge transformer to process variable-length event topologies. This full-phase-space result enables both re-measurements of classic DIS observables and new projections, such as simultaneous jet measurements in the laboratory and Breit frames from a single unfolded dataset. Meanwhile at the LHC, ATLAS has applied ML-assisted unfolding to extract jet track-function moments while circumventing binning artifacts that affect non-linear QCD evolution studies. These results demonstrate that modern ML-based unfolding delivers systematically controlled, fully differential data products that are broadly reusable for downstream physics.

Keywords: Simulation-based Inference; Deep Learning; Precision measurements; Higgs Boson and Electroweak Physics; QCD