Multi-Jet Event classification with Convolutional neural network at Large Scale

Abstract

We present an application of Scalable Deep Learning to analyze simulation data of the LHC proton-proton collisions at 13 TeV. We built a Deep Learning model based on the Convolutional Neural Network (CNN) which utilizes detector responses as two-dimensional images reflecting the geometry of the Compact Muon Solenoid (CMS) detector. The model discriminates signal events of the R-parity violating Supersymmetry (RPV SUSY) from the background events with multiple jets due to the inelastic QCD scattering (QCD multi-jets). With the CNN model, we obtained x1.85 efficiency and x1.2 expected significance with respect to the traditional cut-based method. We demonstrated the scalability of the model at a Large Scale with the High-Performance Computing (HPC) resources at the Korea Institute of Science and Technology Information (KISTI) up to 1024 nodes.