Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Transformer Neural Networks in the Measurement of $t\bar{t}H$ Production in the $H\,{\to}\,b\bar{b}$ Decay Channel with ATLAS

Chris Scheulen

TL;DR

This work demonstrates a transformer-based multi-class analysis to measure $ttH$ production in the $H\to b\bar{b}$ channel using ATLAS Run 2 data at $\sqrt{s}=13$ TeV. The approach leverages permutation-invariant transformer networks for event classification and Higgs reconstruction, enabling simultaneous constraints on the $tt+jets$ background and the $ttH$ signal across six STXS bins. The analysis achieves a significant improvement in signal acceptance and precision, reporting $\sigma_{ttH}^{obs} = 411^{+101}_{-92}$ fb (stat) $^{+85}_{-75}$ fb (syst) with an observed $4.6$-sigma excess over background and SM consistency within uncertainties. The results provide the most precise single-channel measurement of $ttH$ to date and highlight the impact of advanced neural architectures on complex collider analyses. Overall, the study strengthens the direct probe of the top Yukawa coupling via $ttH$ production and demonstrates the utility of transformer networks in high-energy physics event classification and reconstruction.

Abstract

A measurement of Higgs boson production in association with a top quark pair in the bottom anti-bottom Higgs boson decay channel and leptonic final states is presented. The analysis uses $140\,\mathrm{fb}^{-1}$ of $13\,\mathrm{TeV}$ proton proton collision data collected by the ATLAS detector at the Large Hadron Collider. A particular focus is placed on the role played by transformer neural networks in discriminating signal and background processes via multi-class discriminants and in reconstructing the Higgs boson transverse momentum. These powerful multi-variate analysis techniques significantly improve the analysis over a previous measurement using the same dataset. Overall, an excess of 4.6 (5.4) standard deviations over the background-only hypothesis was observed (expected).

Transformer Neural Networks in the Measurement of $t\bar{t}H$ Production in the $H\,{\to}\,b\bar{b}$ Decay Channel with ATLAS

TL;DR

This work demonstrates a transformer-based multi-class analysis to measure production in the channel using ATLAS Run 2 data at TeV. The approach leverages permutation-invariant transformer networks for event classification and Higgs reconstruction, enabling simultaneous constraints on the background and the signal across six STXS bins. The analysis achieves a significant improvement in signal acceptance and precision, reporting fb (stat) fb (syst) with an observed -sigma excess over background and SM consistency within uncertainties. The results provide the most precise single-channel measurement of to date and highlight the impact of advanced neural architectures on complex collider analyses. Overall, the study strengthens the direct probe of the top Yukawa coupling via production and demonstrates the utility of transformer networks in high-energy physics event classification and reconstruction.

Abstract

A measurement of Higgs boson production in association with a top quark pair in the bottom anti-bottom Higgs boson decay channel and leptonic final states is presented. The analysis uses of proton proton collision data collected by the ATLAS detector at the Large Hadron Collider. A particular focus is placed on the role played by transformer neural networks in discriminating signal and background processes via multi-class discriminants and in reconstructing the Higgs boson transverse momentum. These powerful multi-variate analysis techniques significantly improve the analysis over a previous measurement using the same dataset. Overall, an excess of 4.6 (5.4) standard deviations over the background-only hypothesis was observed (expected).

Paper Structure

This paper contains 5 sections, 2 equations, 3 figures.

Table of Contents

  1. Introduction
  2. Analysis Strategy
  3. Transformer Network Performance
  4. Analysis Results
  5. Conclusion

Figures (3)

  • Figure 1: Flowchart depicting the multi-variate analysis setup used to assign events into signal and control regions via separate neural networks for event classification and Higgs boson reconstruction. The Figure is taken from Ref. ATLAS-HIGG-2020-24.
  • Figure 2: Confusion matrices of the Higgs boson transverse momentum ($pT\xspace^H$) reconstruction in the signal regions of (\ref{['subfig:reco-confusion:lj']}) the single-lepton resolved, (\ref{['subfig:reco-confusion:boost']}) the single-lepton boosted, and (\ref{['subfig:reco-confusion:dil']}) the dilepton channel. For each truth $pT\xspace^H$ range, the fraction of events assigned to each signal region $pT\xspace^H$ bin is given. The Figures are taken from Ref. ATLAS-HIGG-2020-24.
  • Figure 3: Measurements of the ttH cross-section in bins of the truth Higgs boson pT and inclusively. The Higgs boson rapidity is restricted to $|y_H| \leq 2.5$ for the differential measurement in line with the STXS prescription. Separate uncertainties for measurement and prediction are included. The Figure is taken from Ref. ATLAS-HIGG-2020-24.