A Study of Physics-motivated Deep learning based algorithms for Jet tagging at the LHC

Abstract

In this thesis, we delve into the realm of particle physics with a focus on jet tagging using deep learning algorithms. Specifically, we explore the PELICAN (Permutation Equivariant Lorentz Invariant and Covariant Aggregator Network) architecture to identify jets originating from top quarks and bottom quarks. Jet tagging is crucial for reconstructing the properties of parent particles and probing new physics phenomena beyond the Standard Model. In this thesis, we study the PELICAN architecture, verifying the claims of Lorentz symmetry and Permutation symmetry preservation in the original paper. We put PELICAN to the test in a more realistic scenario by working with the ATLAS Open Dataset and confirm PELICAN's robustness. We utilize multiple datasets to research PELICAN's performance on various input quantities to test the features that bolster PELICAN's performance. We propose incorporating 4-vector momentum data and trajectory displacement information to enhance the accuracy of jet identification. We propose novel extensions to the PELICAN architecture, including the use of spacetime displacement 4-vectors and scalar particle identification labels, to improve the tagging of heavy-flavor jets. This work not only enhances the performance of existing jet tagging algorithms but also opens new avenues for future research in the field.