Top Quark Reconstruction with Generative Models for Searches for Heavy Higgs Bosons and Top-Antitop Quark Bound States with the CMS Experiment

The discovery of the Higgs boson at the Large Hadron Collider at CERN confirmed the existence of an elementary particle with zero spin. This particle is a crucial component of the Standard Model (SM) of particle physics, which currently represents our best understanding of the microscopic world of elementary particles. While the SM predicts only one such particle, many extensions propose additional ones. They might act as mediators between SM particles and dark matter, providing a link to this still unexplained phenomenon.
In this project, we investigate an intriguing excess observed in the production of top-antitop quark pairs (tt) initially identified during our DASHH search for heavy Higgs bosons with the CMS experiment. The aim is to determine whether this deviation arises from physics beyond the SM or if it can be attributed to a tt bound state. While bound states have been detected for all five lighter quark types, such states, called toponium, have yet to be observed for the heaviest of all elementary particles, the top quark.
To distinguish heavy Higgs particles from toponium or background fluctuations, advanced machine learning methods using regression algorithms are essential for accurately reconstructing top quarks and their kinematics in data. We aim to develop and investigate generative models, such as generative adversarial networks or normalizing flows and graph neural networks, and combine them with disambiguation-free partial label learning techniques. We expect that such an approach will allow us to improve the sensitivity of the analysis and increase the regression accuracy.