A multivariate approach to heavy flavour tagging with cascade training

Abstract

This paper compares the performance of artificial neural networks and boosted decision trees, with and without cascade training, for tagging b-jets in a collider experiment. It is shown, using a Monte Carlo simulation of W H → lνqq events, that for a b-tagging efficiency of 50%, the light jet rejection power given by boosted decision trees without cascade training is about 55% higher than that given by artificial neural networks. The cascade training technique can improve the performance of boosted decision trees and artificial neural networks at this b-tagging efficiency level by about 35% and 80% respectively. We conclude that the cascade trained boosted decision trees method is the most promising technique for tagging heavy flavours at collider experiments.

Correlations in bottom quark pair production at the Fermilab Tevatron

“…As suggested in Ref. [67], the maximum track transverse and longitudinal impact parameter significances are included. In addition to the Generalized Jet Probability P b,LF (y jet = b|X = {x i }), the mean r Generalized Jet Probability is also quite important, on par with ∆Z(µ).…”

A multivariate training technique with event reweighting

“…Artificial Neural Networks (ANN) and Boosted Decision Trees (BDT) [1,2,3,4,5,6] are two important data analysis tools that have wide application in High Energy Physics experiments for particle identification and for event pattern recognition [7,8,9,10]. Both methods 'train' the 'networks' or the 'trees' based on a set of 'signal' and 'background' features (physical quantities) to obtain a powerful discriminant variable that distinguishes signal from background.…”

“…The main purpose of this paper is to compare the training performance with and without event reweighting. Performance comparisons between ANN and BDT can be found in the contexts of MiniBooNE neutrino oscillation analysis [2,4], D0 single top discovery [9] and B-tagging [10].…”