Dortmund 2021 – wissenschaftliches Programm

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

T: Fachverband Teilchenphysik

T 71: Data analysis, Information technology III

T 71.7: Vortrag

Mittwoch, 17. März 2021, 17:30–17:45, Tu

Training of an extended b-tagging algorithm with deep neural networks. — •Thea Engler, Manuel Guth, Gregor Herten, and Andrea Knue — Uni Freiburg, Deutschland

The search for the tt H(H → b b) signal provides direct access to the top-Higgs Yukawa coupling. This channel has four b-jets in the final state and is suffering from large physics background, which makes b-tagging a crucial tool for this analysis. The irreducible tt + bb background has the same final-state particles as the signal process. In this background process, a radiated gluon splits into a b-quark pair. If these b-hadrons are close to each other, they can be reconstructed as one single jet (bb-jet). The irreducible background tt + bb can be better rejected if these bb-jets can be classified. Accordingly, an extended b-tagging algorithm is prepared, based on the ATLAS recommended b-tagger, with an additional classification category for bb-jets. To prepare the extended b-tagging algorithm, the importance of balanced input classes in the training of deep neural networks is studied. The studied approaches are implemented in the extended bb-tagger and a first training is presented.