PyTorch Neural Networks and Track Analysis for Top Quark Tagging

The identification of top quarks is motivated by their high mass and strong coupling to the Higgs mechanism. Boosted top quarks also allow for improved measurements of the Standard Model in the high momentum tails of event feature distributions. Neural networks have been proven as an effective method for distinguishing top quarks from Quantum ChromoDynamic (QCD) events using jet constituent features from the ATLAS and CMS calorimeters. In this project Deep Neural Networks (DNN’s) and Long Short-Term Memory (LSTM) networks were built in PyTorch to compare their performances to previously tested Keras models. After applying similar preprocessing and optimization techniques, the performance of the PyTorch models was found to be highly comparable to the Keras models. Track features from the inner tracker offer promising new information to improve the performance of top tagging neural networks by utilizing information typically used in b-jet identification. Track features were analyzed and incorporated in processing scripts used to prepare the data for input to neural networks. It was found that keeping 100 "p_T" ordered tracks with "p_T" greater than 1 GeV could retain relevant information for jet classification while minimizing noise and computing time.