Infrared safety of a neural-net top tagging algorithm

Suyong Choi,Seung J. Lee,Maxim Perelstein

doi:10.1007/jhep02(2019)132

Abstract

Neural network-based algorithms provide a promising approach to jet classification problems, such as boosted top jet tagging. To date, NN-based top taggers demonstrated excellent performance in Monte Carlo studies. In this paper, we construct a top-jet tagger based on a Convolutional Neural Network (CNN), and apply it to parton-level boosted top samples, with and without an additional gluon in the final state. We show that the jet observable defined by the CNN obeys the canonical definition of infrared safety: it is unaffected by the presence of the extra gluon, as long as it is soft or collinear with one of the quarks. Our results indicate that the CNN tagger is robust with respect to possible mis-modeling of soft and collinear final-state radiation by Monte Carlo generators.

Highlights

Tagger output makes it very difficult to identify the specific features in the jet substructure that the NN focuses on, let alone assess their robustness in the simulation
We show that the jet observable defined by the Convolutional Neural Network (CNN) obeys the canonical definition of infrared safety: it is unaffected by the presence of the extra gluon, as long as it is soft or collinear with one of the quarks
Starting with ref. [2], many studies have demonstrated the efficacy of Neural Networks for boosted top jet tagging, at the level of Monte Carlo (MC) simulations

Summary

Introduction

Tagger output makes it very difficult to identify the specific features in the jet substructure that the NN focuses on, let alone assess their robustness in the simulation To date, this issue has been addressed by cross-comparisons of NN taggers trained on samples produced by different MC generators, which employ different algorithms to model parton showers The training process appears to result in a network that largely disregards small-scale angular features in the energy distribution inside the jet, making the CNN tagger robust with respect to modeling such small-scale features in MC generators. Such robustness is a necessary pre-condition for practical applicability of MC-trained NN taggers, and it is highly reassuring that it is satisfied

Objectives

Results

Conclusion