Machine Learning Techniques for Tagging Heavy Flavor Jets at RHIC

Abstract The properties of the Quark Gluon Plasma (QGP), a hot and dense medium made up of deconfined quarks and gluons (partons), can be studied through ultrarelativistic heavy-ion collisions. In the early stages of the collisions, high energy partons are created, which fragment into collimated sprays of hadrons, called jets. Jets are used to probe the entire evolution of the QGP that they traverse. Classifying jets based on the flavor of the parton that initiated them as heavy or light is a fundamental tool for studying the properties of the QGP as different flavors interact differently with the medium. Jets resulting from heavy ion collisions are compared to those resulting from proton-proton collisions to study their modification in the medium. Therefore, as a first step, we use jets resulting from simulated proton-proton events for the identifying their flavors, which could be extended to identifying jets resulting from heavy-ion events once heavy flavor features are added to heavy-ion event generators. We utilize different deep learning techniques and employ different strategies to minimize the misidentification probability while maintaining the efficiency of tagging heavy flavor jets at RHIC. Similar analysis has been done for LHC ? experiments but never at RHIC energies ( rates of heavy flavor jets to light flavor jets is very low. In this paper, we compare and contrast the performances of the different models we have developed for tagging heavy flavor jets at RHIC energies.