Optimization of the ttH Selection Including Systematics Using Machine Learning with ATLAS Data

This research is centered on the goal of accurately selecting events produced from the. interaction of two gluons in a proton-proton collision at the LHC, resulting in a top-antitop pair and a Higgs boson, a process known as ttH. With data recorded by the detector, the study aims to distinguish these ttH events from those generated by other processes. To this end, we employ a deep learning approach, specifically a FT-Transformer architecture, for the event selection. The use of such a machine learning method enhances our ability to identify ttH events accurately, leading to an improved signal-to-noise ratio and statistical significance, thereby contributing to our understanding of the Higgs boson’s properties. Aside of exploring more advanced NN architectures, we improve previous work, by. exploring the use of an extended training set, which allows us to dramatically increase the training statistic and thus achieve a much better performance. An integral part of this research is the evaluation of both statistical and systematic un- certainties associated with this event selection process. The findings and methodologies presented in this thesis offer promising advancements in particle physics event selection, contributing to the Collaboration’s ongoing endeavors to probe the fundamental properties of the universe.