Speaker
Description
The production of Higgs boson pairs through vector boson fusion (VBF) offers a powerful probe of the electroweak structure of the Higgs sector and sensitivity to possible deviations from the Standard Model. Event topologies featuring boosted Higgs bosons are particularly interesting, as they exhibit distinctive kinematic features that can be exploited to reduce background contributions. Among the available decay channels, HH → 4b has the largest branching fraction, but poses significant experimental challenges due to the dominant multijet background and the fully hadronic nature of the final state. This work presents a study of boosted Higgs boson pair production via vector boson fusion in the HH → 4b final state using proton–proton collision data collected by the ATLAS experiment during Run-II and early Run-III of the Large Hadron Collider. The analysis focuses on events containing two high-transverse-momentum Higgs boson candidates reconstructed as large-radius jets, together with the characteristic VBF topology. To improve the discrimination between signal and background, machine learning techniques are investigated. A traditional boosted decision tree (BDT) approach is explored and compared with a more advanced method based on Graph Neural Networks (GNNs). The GNN framework models reconstructed physics objects and their kinematic and geometric relationships, enabling the exploitation of correlations between jet substructure and VBF event features that are difficult to capture with conventional techniques.