Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Boosting Sensitivity to $HH\to b\bar{b} γγ$ with Graph Neural Networks and XGBoost

Mohamed Belfkir, Mohamed Amin Loualidi, Salah Nasri

TL;DR

This work addresses the challenge of constraining the Higgs self-coupling through non-resonant di-Higgs production in the HH → bbγγ channel at the LHC. It compares two ML approaches—a tree-based XGBoost classifier and a geometry-aware graph neural network (GNN) that encodes event topology—and finds the GNN provides a substantial sensitivity gain, including about a 28% improvement in the expected 95% CL upper limit on the HH cross-section. Using MC simulations at 13.6 TeV with 168 fb^-1, the study also observes tighter constraints on the self-coupling modifier and competitive performance relative to ATLAS Run-2/Run-3 results. Overall, the results demonstrate that graph-based learning can significantly enhance searches for rare processes like double Higgs production at the LHC and likely benefit future analyses in this domain.

Abstract

In this paper, we explore the use of advanced machine learning (ML) techniques to enhance the sensitivity of double Higgs boson searches in the \( HH \to b\bar{b}γγ\) decay channel at $\sqrt{s} = $ 13.6 TeV. Two ML models are implemented and compared: a tree-based classifier using XGBoost, and a geometrical-based graph neural network classifier (GNN). We show that the geometrical model outperform the traditional XGBoost classifier improving the expected 95\% CL upper limit on the double Higgs boson production cross-section by 28\%. Our results are compared to the latest ATLAS experiment results, showing significant improvement of both upper limit and Higgs boson self-coupling ($κ_λ$) constraints.

Boosting Sensitivity to $HH\to b\bar{b} γγ$ with Graph Neural Networks and XGBoost

TL;DR

This work addresses the challenge of constraining the Higgs self-coupling through non-resonant di-Higgs production in the HH → bbγγ channel at the LHC. It compares two ML approaches—a tree-based XGBoost classifier and a geometry-aware graph neural network (GNN) that encodes event topology—and finds the GNN provides a substantial sensitivity gain, including about a 28% improvement in the expected 95% CL upper limit on the HH cross-section. Using MC simulations at 13.6 TeV with 168 fb^-1, the study also observes tighter constraints on the self-coupling modifier and competitive performance relative to ATLAS Run-2/Run-3 results. Overall, the results demonstrate that graph-based learning can significantly enhance searches for rare processes like double Higgs production at the LHC and likely benefit future analyses in this domain.

Abstract

In this paper, we explore the use of advanced machine learning (ML) techniques to enhance the sensitivity of double Higgs boson searches in the decay channel at 13.6 TeV. Two ML models are implemented and compared: a tree-based classifier using XGBoost, and a geometrical-based graph neural network classifier (GNN). We show that the geometrical model outperform the traditional XGBoost classifier improving the expected 95\% CL upper limit on the double Higgs boson production cross-section by 28\%. Our results are compared to the latest ATLAS experiment results, showing significant improvement of both upper limit and Higgs boson self-coupling () constraints.

Paper Structure

This paper contains 12 sections, 5 equations, 10 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Double Higgs production at the LHC
  3. Monte Carlo event generation
  4. Object definitions and event reconstruction
  5. Object definitions
  6. Event selections
  7. ML event categorization
  8. Event pre-processing
  9. Tree-based classifier
  10. Graph-based classifier
  11. Results and discussion
  12. Conclusion

Figures (10)

  • Figure 1: Leading-order Feynman diagrams for the dominant ggF production (a and b), and VBF production (c-e) of double Higgs boson.
  • Figure 2: Normalized distributions of (a) $m_{\gamma\gamma}$ and (b) $m_{bb}$ of selected events for SM signal, single Higgs background and the continuum $\gamma\gamma$ + jets. Distributions are normalized to unity.
  • Figure 3: XGBoost feature importance ranking.
  • Figure 4: Graphical representation of the event graph for (a) the SM ggF $HH\to b\bar{b}\gamma\gamma$, (b) the SM VBF $HH\to b\bar{b}\gamma\gamma$ and (c) the continuum $\gamma\gamma$ + jets.
  • Figure 5: Overview of the geometry-aware GNN model.
  • ...and 5 more figures