Taming modeling uncertainties with Mass Unspecific Supervised Tagging

TL;DR

This paper investigates the robustness of jet taggers built with Mass Unspecific Supervised Tagging against different modeling schemes, finding minimal dependence and supporting their use in new physics searches.

Contribution

It demonstrates that jet taggers are relatively insensitive to variations in parton showering and hadronisation models, enhancing their reliability for physics analyses.

Findings

Model dependence of taggers is small despite substructure differences.

Results support the robustness of generic supervised taggers.

Findings improve confidence in new physics search methods.

Abstract

We address the modeling dependence of jet taggers built using the method of Mass Unspecific Supervised Tagging, by using two different parton showering and hadronisation schemes. We find that the modeling dependence of the results - estimated by using different schemes in the design of the taggers and applying them to the same type of data - is rather small, even if the jet substructure varies significantly between the two schemes. These results add great value to the use of generic supervised taggers for new physics searches.