Pileup Cluster Identification in the ATLAS Calorimeter Using Deep Neural Networks

The ATLAS calorimeter measures the energy of particles produced in proton-proton collisions at the Large Hadron Collider (LHC). The presence of pileup, or additional simultaneous collisions with a hard scatter, complicates the accurate energy measurements of jets, which are the physics objects of interest. This presentation explores the use of deep neural networks (DNNs) to classify calorimeter topological clusters as originating from either pileup or non-pileup collisions. By analyzing the response of simulated single-particle interactions with the calorimeter, both with and without pileup, suitable target definitions for the DNN are determined. Using simulated dijet events that include pileup, key features such as cluster position and spatial distributions are identified that help distinguish between these two types of topological clusters. Result on the pileup cluster identification will be presented, and future steps for applying these classifications to correct jet energy measurements will be discussed.