Machine learning in top quark physics at ATLAS and CMS

TL;DR

The paper surveys machine-learning applications in top quark physics at ATLAS and CMS, focusing on reconstruction, analysis strategies, and statistical inference. It highlights reconstruction methods such as nu-flow and spanet for assigning decay products and inferring neutrino directions, as well as hypergraph-based approaches. It also covers background estimation strategies (ABCD, disco) and advanced inference techniques (likelihood-free inference and omnifold unfolding), with illustrative results from ATLAS and CMS. The HL-LHC outlook emphasizes reducing reliance on simulations through reweighting and pursuing higher-order accuracy, aiming for sustainable, high-precision top-quark measurements.

Abstract

This note presents an overview of current and potential future applications of machine-learning-based techniques in the study of the top quark. The research community has developed a diverse set of ideas and tools, including algorithms for the efficient reconstruction of recorded collision events and innovative methods for statistical inference. Recent applications of some techniques by the ATLAS and CMS collaborations are also highlighted.

Figures (3)

• Figure 1: Inference of the pseudorapidity of the neutrino originating from the leptonically decaying top quark in semileptonic $\mathrm{t}\bar{\mathrm{t}}$ events: (a) $\nu$-flow approach nuflow; (b) spanet approach spanet.
• Figure 2: Examples of unbinned, multidimensional unfolding with omnifoldomnifold: (a) average jet mass as a function of the jet $p_\mathrm{T}$ in Drell-Yan events atlas-omnifold by ATLAS; (b) number of charged jet constituents in minimum bias events cms-omnifold by CMS.
• Figure 3: Reweighting of simulated events using the dctr method dctr by CMS: (a) reweight to emulate a variation of the $h_\textrm{damp}$ parameter of the powheg event generator; (b) reweight to emulate NNLO accuracy from an NLO sample. Figures are taken from Ref. cms-dctr.