Preview

PDF, English Download (2MB) | Terms of use

Abstract

Uncertainties are crucial in particle physics, affecting experimental data and theoretical predictions. This thesis investigates the impact of uncertainties on global analyses and the estimation of uncertainties using machine learning architectures. In the first part of this thesis, we perform global analyses using effective field theory approaches. We start with the Standard Model effective field theory in the top, Higgs, and electroweak sectors, including public experimental likelihoods. In particular, we focus on the role of theory uncertainties and their interplay with correlations. Next, we perform the first global electric dipole moment analysis constraining model parameters from the hadronic- and weak-scale Lagrangians while exploring the impact of theory uncertainties on the parameter constraints. The second part discusses machine-learning methods that have become increasingly important with the growing data from future LHC runs. Thus, we study the calibration of systematic and statistical uncertainties and the precision and reliability of machine learning architectures for amplitude surrogate models. We compare Bayesian neural networks and repulsive ensembles as uncertainty estimators regarding their precision and use Kolmogorov-Arnold networks to explore the impact of activation functions. Overall, this work emphasizes the importance of reducing theory uncertainties and paves new ways of uncertainty estimation using machine learning models in particle physics and global analyses.