OPTIMIZATION OF A DEEP GRAPH NEURAL NETWORK TO RECONSTRUCT GENERIC B DECAYS AT BELLE II

Sumitted to PubDB: 2024-03-11

Category: Master Thesis, Visibility: Public

Tags: -

Authors	Jacopo Cerasoli, Giulio Dujany, Corentin Santos
Date	Jan. 1, 2023
Belle II Number	BELLE2-MTHESIS-2024-002
Abstract	This work presents the results of a fourteen weeks internship on the optimization of a deep graph neural network in the context of the search for New Physics (NP) at the Belle II experiment. The neural network is the core of an algorithm named graFEI to reconstruct B decays. A powerful probe for the search of NP is the B to K nu nubar decay channel because of its rarity and the precision of its branching ratio prediction in the Standard Model (SM). The Belle II experiment is the only experiment capable of detecting such a decay because of its cleanness and hermeticity. Moreover, the production of a partner B during e+e− collisions at the SuperKEKB collider allows to recover the missing energy from invisible particles in the final state, a technique name Full Event Reconstruction (FER), specific to Belle II. During this internship, starting from a 15% B reconstruction efficiency, the multiple optimizations performed on the graFEI get the performances to 21%. This is about 6 times the reconstruction efficiency of the main reconstruction algorithm currently used at Belle II, the Full Event Interpretation (FEI).
Conference	Strasbourg