Tagging of Leptonically Decaying Boosted Top Quarks Using Machine Learning

The study of boosted top quarks is very important for probing a wide variety of new physics models. When considering the leptonic channel decay mode of highly boosted top quarks, the axis of the jet resulting from the bottom quark gets very close in distance to the lepton. The effectiveness of conventional techniques of identifying boosted top quark jets in anevent deteriorates in such scenarios as such techniques heavily rely on reconstructing and identifying an isolated lepton, which becomes challenging when the angular separation between the bottom quark and the lepton becomes very small. The use of machine learning techniques for tagging leptonically decaying boosted top quarks has not yet been explored as ex-tensively as the hadronic decay channel. In this project, we compare the performance of two Convolutional Neural Network (CNN) based taggers with a particle-cloud-based tagger, namely ParticleNet, for tagging highly boosted leptonically decaying top quarks. The ParticleNet classifier showed a significantly improved performance over the CNN-based taggers. Theconventional method for reconstructing top quark jets was compared to the ParticleNet-based method in a search for the heavy W′ boson, where the W′ decays to a top and a bottom quark, and the top quark decays leptonically. Studying the signal-to-background ratios and the expected discovery signifi- cances for both methods showed that the particle-cloud-based tagger outper-forms the conventional method