OPTIMIZATION OF TAU IDENTIFICATION IN ATLAS EXPERIMENT USING MULTIVARIATE TOOLS

Abstract

Elementary particle physics experiments, which search for very rare processes, require the efficient analysis and selection algorithms able to separate a signal from the overwhelming background. Four learning machine algorithms have been applied to identify τ leptons in the ATLAS experiment: projective likelihood estimator (LL), Probability Density Estimator with Range Searches (PDE-RS), Neural Network, and the Support Vector Machine (SVM). All four methods have similar performance, which is significantly better than the baseline cut analysis. This indicates that the achieved background rejection is close to the maximal achievable performance.