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

T: Fachverband Teilchenphysik

T 71: Data analysis, Information technology III

T 71.8: Vortrag

Mittwoch, 17. März 2021, 17:45–18:00, Tu

Treating Uncertainties with Bayesian Neural Networks in a ttH Measurement — •Nikita Shadskiy and Ulrich Husemann — Institut für Experimentelle Teilchenphysik (ETP), Karlsruher Institut für Technologie (KIT)

In the Standard Model of particle physics, fermions couple to the Higgs boson via a Yukawa-type coupling with a strength proportional to their mass. The top quark is the heaviest known fermion and, therefore, has the strongest coupling to the Higgs boson.
One of the processes to investigate this coupling is the associated ttH production in which the Higgs boson decays into a bb pair. This signal process has a much smaller cross section than the challenging background processes like tt+jets production. Especially tt+bb events are very signal-like. A common approach to separating this signal from the backgrounds is to use artificial neural networks.
Neural networks usually do not take into account uncertainties. In contrast, Bayesian neural networks use weight distributions instead of single weight values. This not only prevents overfitting but also allows to obtain an uncertainty estimate on the predictions of the neural network. In this talk it is investigated how this feature of Bayesian neural networks can be used in a ttH(bb) measurement.