SMuK 2023 – wissenschaftliches Programm

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

T: Fachverband Teilchenphysik

T 151: Exp. Methods IV

T 151.2: Vortrag

Donnerstag, 23. März 2023, 17:45–18:00, WIL/C129

Prospects for machine-learning based unfolding techniques with a focus on the measurement of differential Higgs boson production cross sections — Johannes Erdmann, •David Kavtaradze, and Jan Lukas Späh — III. Physikalisches Institut A, RWTH Aachen University

In high-energy physics experiments, measured distributions are the result of Poissonian fluctuations around expectation values that are obtained from folding the underlying distribution with detector effects. The inference of the underlying distribution from the measurement in cases where no parametric form is available is known as "unfolding".

Traditional unfolding methods rely on a categorisation of events in a certain binning scheme. This limits the flexibility of the unfolding and does not allow for a simultaneous deconvolution of multiple observables.

An alternative approach, termed "Omnifold" in the literature, does not have these restrictions and benefits from machine-learning to take into account the whole information from each event. This approach is contrasted with the traditional approaches using a physically motivated example from a measurement of differential Higgs boson production cross sections in the diphoton decay channel.