Polarization studies in same-sign W-boson pair production at the LHC with the ATLAS detector

The scattering of massive electroweak gauge bosons offers a great opportunity to study the Standard Model's innermost gauge symmetry structure and the electroweak sector. Especially the scattering of longitudinally polarized bosons is sensitive to the concrete realization of the electroweak symmetry breaking due to the unitarity violating energy dependency of the scattering amplitude without the Higgs boson. At the LHC, the pair production of same-charged $W^\pm$ bosons with the final state $\ell^\pm \nu \ell^\pm \nu jj$ is one of the most promising vector boson scattering channels due to the comparatively high cross-section of the electroweak production mode. In this thesis, this process is investigated for a center of mass energy $\sqrt{s} = 13$ TeV and an integrated luminosity of $139 \text{ fb}^{-1}$. As part of this study, a framework for automated and distributed hyperparameter optimization and input variable selection of artificial neural networks has been developed. It is applied to optimize multilayer perceptrons (MLPs) for the discrimination of longitudinally polarized $W^\pm$-boson pairs against the other polarizations and background processes. Maximum likelihood fits to Asimov data, including only statistical uncertainties, are performed, from which a discovery significance of $1.4 \sigma$ is obtained. A comparative study using gradient-boosted decision trees shows an 11\% improvement of the significance when using neural networks. An outlook for the High-Luminosity LHC (HL-LHC) results in an expected significance of $6.9 \sigma$ with the use of MLPs, compared to $3.2 \sigma$ obtained with the most sensitive single observable, indicating that observation is likely achievable with the HL-LHC.