Sumitted to PubDB: 2022-02-02

Category: Master Thesis, Visibility: Public

Tags: -

Authors	Alizee Brouillard, Giulio Dujany
Date	Jan. 1, 2021
Belle II Number	BELLE2-MTHESIS-2022-002
Abstract	To find new physics beyond Standard Model, the Belle II experiment performs precision measurements to bring to light deviations between theory and experimental data. SuperKEKB collider, located in Tsukuba in Japan, collides electron and positron beams with the energies 7 GeV and 4 GeV. The annihilation of an electron and a positron produces the Υ(4S) resonance which decays to a pair of charged mesons B+B− or neural mesons B0B0. These mesons decay to intermediate particles, and so on, up to the so-called final state particles. The purpose of this end-of-course internship is to implement an algorithm which assigns each particle to its native B meson, with the matrix of relations. We implemented a graph neural network with the library PyTorch of Python because we work with fully connected graph and disintegration tree, which are structured data. We showed the efficiency of our algorithm working on disintegration modelled by the phasespace library. Each disintegration is a topology. We gradually increased the complexity of the data and highlighted the factors which affect the most the efficiency. We compared the performances obtained with the neural graph network and the Thrust method, which is traditionally used to solve this problem. This end-of-course internship for the Telecom Physique Strasbourg engineering degree took place at the Institut Pluridisciplinaire Hubert-Curien in Strasbourg. This work was conducted between March first 2021 and July 16th 2021.
Conference	Strasbourg

• BELLE2-MTHESIS-2022-002.pdf (versions: 1)