Development and Deployment of a Deep Neural Network based Flavor Tagger for Belle II

Sumitted to PubDB: 2021-06-14

Category: Phd Thesis, Visibility: Public

Tags: -

Authors	Florian Bernlochner, Jochen Gemmler
Non-Belle II authors	Michael Feindt
Date	Jan. 1, 2020
Belle II Number	BELLE2-PTHESIS-2021-005
Abstract	In this thesis a flavor tagging algorithm based on these Deep Learning methods is developed and investigated. It is essential for the usage of this algorithm to not only evaluate its performance on generated Monte Carlo data, but also its actual performance on experimental data. Potential differences have to be investigated and quantified to shed light on the accuracy and reliability of the algorithm. The algorithm is designed for the Belle II experiment, and is validated on existing Belle data. The effective tagging efficiency of the developed algorithm is determined on a selection of reconstructed B0 → D∗−π+ and B− → D0 π− decays and compared to the efficiency of an alternative, more traditional approach. Although the chosen hadronic decays have lower statistics than semi-leptonic decays, the reconstruction is much cleaner, and a lower cross-feed between signal side and tag side is expected.
Conference	Karlsruhe