Optimization of event selection in search for doubly-charged Higgs bosons at ATLAS using machine learning techniques

The analysis of proton-proton collision data recorded by ATLAS during Run 2 of the LHC identified an excess of data over the Standard Model prediction in both the $W^{\pm}Z$ and $W^{\pm}W^{\pm}$ vector boson scattering processes. These excesses could be attributed to resonances of the singly-charged ($H_5^{\pm}$) and doubly-charged ($H_5^{\pm\pm}$) Higgs bosons, which are hypothesized by the Georgi-Machacek (GM) model. To investigate this excess and assess its compatibility with the GM model, a dedicated search is being performed for the $H_5^{\pm}$ and $H_5^{\pm\pm}$ bosons where they are produced by vector boson fusion and decay to $W^{\pm}Z$ and $W^{\pm}W^{\pm}$ respectively. In this thesis, the selection of the $H_5^{\pm\pm}$ signal region is optimized by training a neural network to discriminate signal events from background events. The characteristics of the $H_5^{\pm\pm}$ events vary significantly with mass, which leads to undesired behaviour when training a single network for a large mass range. A number of strategies are devised to address this problem; the best solution is to modify the weighting of different simulated masses during training. The neural network is used to define a new $W^{\pm}W^{\pm}$ signal region which has a greater sensitivity to the GM model compared to a cuts-based approach.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.