Tau Identification With Deep Neural Networks at the CMS Experiment

Abstract

The reconstruction and identification of tau leptons decaying to hadrons are crucial for new physics signatures and precision measurements with tau leptons in the final state at the LHC. The recently deployed tau identification algorithm using deep neural network (DNN) at the CMS experiment for the discrimination of hadronic tau decays from quark or gluon induced jets, electrons, or muons is an ideal example for the exploitation of modern deep learning neural network techniques in high energy physics. With this algorithm, significant suppression of tau misidentification rates has been achieved for the same identification efficiency compared to previous algorithms at the LHC, leading to considerable performance gains for physics studies with tau leptons. This new multiclass DNN-based tau identification algorithm at CMS and its performance are presented in this article.