The study of the Standard Model of Particle Physics (SM) requires the analysis of enormous amounts of data that has been measured by particle detectors. It is therefore essential to build tools that can handle such data efficiently. An important step in the data processing is the reconstruction of events that are produced in high-energy collisions. This thesis presents Machine Learning based approaches for the processing of data from the ATLAS detector at the LHC at CERN. An important part of the event reconstruction is the identification of the hadron of origin for sprays of particles, called jets. The knowledge about such jets is a crucial ingredient for physics analyses since they provide information about the underlying hard scattering process. An algorithm for the reconstruction of the decay chain of heavy hadrons is proposed. The decay chain reconstruction provides additional information about the jet's origin which might be of help for the identification of jets containing heavy hadrons.

Secondly, the reconstruction of neutrinos is studied. Neutrinos cannot be measured directly which is why a Machine Learning based approach is proposed to estimate their momenta. A Probabilistic Generative Model is presented that is able to sample the momentum distribution of a varying number of neutrinos conditioned on the reconstructed objects that form an event. The information gained by this can be used in further applications. In this thesis, its utility in the classification of events with four top quarks from processes predicted by the SM as well as those that go beyond it is studied.