BYU

Abstract by Charles Lewis

Personal Infomation


Presenter's Name

Charles Lewis

Degree Level

Undergraduate

Abstract Infomation


Department

Physics and Astronomy

Faculty Advisor

John Colton

Title

Machine Learning Techniques to Approximate the Matrix Element Method for the \"ttH\" Process

Abstract

Artificial Neural Networks (NN) serve as a way to approximate the parton-level probability distribution function of the Matrix Element Method (MEM) of multivariate analysis. The MEM is a powerful tool for analysis of the "ttH" process, a process where a top quark pair is produced in conjuction with a Higgs boson. However this method is slow and computationally expensive. Using a NN to approximate the parton-level function could dramatically speed up the analysis. We initially trained NN models with success for simplified 3 and 4 particle final state systems, using Python, Keras, and Tensorflow. Fitting a NN to the integrand for the full 8 particle final state system proved more difficult. We found that training on the log of the integrand and supplying additional derived parameters to the network produced a significantly improved accuracy that allowed us to fit the NN for the full "ttH process".