A MACHINE LEARNING APPROACH TO ATLAS PARTICLE ENERGY CALIBRATIONS

Abstract

This project investigates a machine learning-based approach to improving energy measurements of signals in ATLAS, a particle detection experiment at the CERN Large Hadron Collider (LHC). Accurate energy calibrations are integral to making precision measurements of objects such as jets and to search for new, beyond-the-Standard Model particles. ATLAS currently calibrates the energy of its detector signals, both electromagnetic and hadronic, by a lengthy series of applied corrections called local cell weighting (LCW) [1]. For this research, several iterations of a deep neural net (DNN) were trained, using features from the calorimeter signals to predict their true energy deposits. Final models successfully outperformed the existing method employed by ATLAS, built upon the work done by prior groups in similar fields [2] [3] [4]. This project provides concrete evidence for the improvements machine learning can bring to calorimeter energy calibrations.