Search for Exotic Higgs Boson Decays in Events with a Semi-Merged Photon Topology Using Graph Neural Networks at CMS

<p dir="ltr">This thesis presents a search for exotic decays of the 125GeV Higgs boson into pairs of light scalar or pseudoscalar particles of the form H → AA with A → γγ. The analysis targets a semi-merged final state with three reconstructed photons, where one of the pairs of photons from an A is reconstructed as two separate photons while the other pair of photons is reconstructed as a single photon, either due to instrumental merging or reconstruction thresholds. A novel graph neural network-based regression technique is developed to recover the invariant mass of the A particle from the single photon leg of the decay, providing a robust method of measuring the in variant mass of objects reconstructed as single photons. The search is performed using proton-proton collision data collected by the CMS detector during Run 2 of the LHC, corresponding to an integrated luminosity of 138fb⁻¹ at √s = 13TeV. No significant excess above the standard model expectation is observed, and upper limits are placed on the product of the Higgs production cross section and branching fraction, σ(pp → H) × B(H → AA → 4γ), ranging from 0.264 to 0.005pb at 95% confidence level for A masses m_A ∈ [1,15]GeV. In addition, a machine learning-based anomaly detection system is developed for real-time monitoring of the CMS electro magnetic calorimeter, demonstrating broader applications of deep learning techniques in experimental operations.</p>