Using Neural Networks to Probe the Parameter Space of a 3HDM with a U(1) times Z2 Flavor Symmetry

Sammanfattning: Probing the physical regions in large parameter spaces of typical Standard Model (SM) extensions can be a very difficult computational task. In this thesis project, a new framework has been developed that utilises well-known Machine Learning (ML) techniques in the form of neural networks trained by a genetic algorithm. This framework is rather generic and designed to explore new physics model parameter spaces with a large number of dimensions, implementing a given set of theoretical and experimental constraints, in a time-efficient and smart way. The ML framework has been applied for analysis of a large parameter space in a recently proposed Three Higgs Doublet Model (3HDM) with a U(1)times Z_2 flavor symmetry implementing theoretical constraints on tree-level unitarity and boundedness from below, as well as the experimental bounds on oblique corrections. We have developed an inversion procedure that enables us to use the scalar boson masses, mixing angles and off-alignment parameters as inputs in our ML framework. This lets us use measured values of the SM-like Higgs boson mass (within errors) and couplings in the near-alignment regime, as well as possible experimental bounds on masses of additional scalar bosons as inputs.A similar inversion algorithm has also been implemented in the quark sector, enabling us to take the measured values of quark masses and mixing angles (within errors) as inputs randomized within the experimental uncertainties. Our ML implementation makes an important step towards an efficient and detailed exploration of large parameter spaces of new physics models highly constrained by precision experimental bounds.