Optimizing Neural Network Classifiers with ROOT on a Rocks Linux Cluster

摘要

We present a study to optimize multi-layer perceptron (MLP) classification power with a Rocks Linux cluster. Simulated data from a future high energy physics experiment at the Large Hadron Collider (LHC) is used to teach a neural network to separate the Higgs particle signal from a dominant background. The MLP classifiers have been implemented using the ROOT data analysis framework. Our aim is to reach a stable physics signal recognition for new physics and a well understood background rejection. We report on the physics performance of new neural classifiers developed in this study. We have used the benchmarking capabilities of ROOT and of the Parallel ROOT facility (PROOF) to compare the performance of the Linux clusters at our campus.