Towards an intelligent automatic event interpretation at LHCb

The main aim of the LHCb experiment at the Large Hadron Collider at CERN is to gather high-precision measurements of heavy hadron decays. The large amount of data produced can not entirely be stored, necessitating the use of a trigger system. This system performs fast analysis to reject parts of the data that are not the focus of LHCb research, storing only the remaining data for further offline analysis. Future upgrades at LHC will drastically increase the amount of data produced, meaning that triggers will need to select fewer data. In this context, a novel approach is proposed: The Deep Full Event Interpretation (DFEI), which is based on graph neural networks (GNN). The graphs represent the particles in a particular event and their relationships with each other. The DFEI algorithm suppresses the nodes and edges that don’t relate to heavy hadron decays, leaving only those decays remaining. This thesis proposes an enhancement of this algorithm. Currently, it has two main shortcomings: charmed hadrons are disregarded and only the charged part of the decays is selected. The primary goal of this work is to extend the capabilities of this algorithm to include neutral particles. The use of Boosted Decision Trees, using the reconstructed decays that the DFEI algorithm has selected, is explored. The encouraging results which are discussed constitute the ground work for future enhancements and for the implementation of this extension.