Measurement of the di-jet production at 13 TeV and use of adversarial neural-networks to identify $W$-jets

The measurement of the dijet production cross-section is measured in proton-proton collisions at a center-of-mass energy of 13 TeV. The data were collected by the ATLAS detector at the Large Hadron Collider at CERN during 2015. The measurement of the cross-sections were compared quantitatively with the Standard Model (SM) theoretical predictions at next-to-leading order (NLO) corrected for non-perturbative effects. These studies test the SM for possible evidence of new physics. In addition, jets production contains information on the distributions of partons inside the proton and also on the strength of their interaction $\alpha$$_{s}$, which allows to extend the knowledge of QCD to an unprecedented energy scale. The experimental results show agreement with the theoretical predictions, validating for the first time the SM dijet production at 13 TeV. In addition, this Thesis presents the development of machine learning techniques for identification of jets arising from boosted W bosons in a dominated QCD background, an essential task for heavy particles searches beyond SM or to improve the precision of Higgs properties measurements. The technique proposed is based on adversarial neural networks, which allows to build a non-correlated classifier with physical observables of interest, like the jet mass. Several searches in ATLAS are sensitive to undesirable effects introduced by these correlations, which results in a reduction of the statistical significance of the analysis. Studies based in Monte Carlo simulations showed significant improvements with respect to other analytical and multivariate techniques previously used in ATLAS, providing an encouraging first look for its application in future physics searches.