Reconstruction of ttH final states using advanced machine learning techniques

This thesis deploys a graph attention network to reconstruct the $t\overline{t}(H\to b\bar{b})$ process. Simulations of the LHC Run 2 and a simulated ATLAS detector are used to obtain data. Reconstruction involves predicting the node classes that correspond to jet flavour and the edge classes that correspond to edges between nodes originating from the same parent particle. These predictions are further refined using a logistic regression model to determine the final candidates for each edge class. Overall, this approach reconstructed 17.4% of all events completely correctly.