Sara Elisabeth Dahl Pinholt

Title: Improving Z and Higgs selections in ATLAS using Machine Learning

Abstract:

This thesis presents a novel approach to search for the decay. Previously, only cut-based methods has been explored. Here, a machine learning-based approach is introduced, using tree-based methods in the Higgs boson search. 

The analysis presented is based on the decay products of the decay; muons, electrons---their combination into a $Z$ boson---and photons. The decay is utilized as a control channel, as it shares decay channels with the Higgs boson, allowing for testing in data from the ATLAS detector.

All $Z$ and Higgs models are evaluated against the selection in on Monte Carlo data (MC) and real data.

The model increase the signal with 40-106% (17-30%) in MC (data) while the model shows increase in signal of 12-44% (6-48%) in MC (data) for same amount of background as the ATLAS selection. The Higgs trained models are evaluated in MC only, where an increase of 44-53% (29-32%) is seen for the decay for muons (electrons).                              

As a large decrease is seen for the MC-trained model when applied to data, expected increase for the decay is 16% while for the model, 22% more signal is expected compared with the selection in [1]. 

In summary, the results of this thesis strongly suggests that adoption of a tree-based approach would drastically improve the signal selection of the decay compared to previous cut-based methods.

Please notice that the MSc thesis defense will take place virtually via Zoom. To receive a link to join the MSc thesis defense, please send an e-mail in advance to Troels C. Petersen: petersen@nbi.dk