Search for heavy resonances using deep neural network

Abstract

Extracting signals in a small cross-section compared to the large cross-section of the background processes has been helped by introducing machine learning (ML) techniques for classification purposes. Classification algorithms in machine learning is a type of super vised learning where the outputs are constrained only to a limited set of values or classes such as signals or backgrounds. This thesis presents the classification of the produced heavy resonances from pp collision at the Large Hadron Collider(LHC). The machine learning technique, deep neural network(DNN), is used to classify the resonances from the background processes. Heavy resonances such as Vector-like quark(VLQ), Tprime (T ′ ) at different mass points [600, 1200] GeV are used as the signal while Standard model Higgs(SMH) and Non-Resonant background(NRB) as the backgrounds for the training and testing of the DNN model. On the DNN output variable, the expected limit at 95% CL on T′ production processes have been has been extracted at each T′ mass in the range [600,1200] using the DNN based selection criteria, as well as for the present CMS analysis(BDT) with Higgs Combined Tools under the CMSSW environment. Preliminary studies show DNN can have a better/comparable results to other machine learning techniques used for the high energy physics analysis.