Unbinned multivariate observables for global SMEFT analyses from machine learning

The ML4EFT code is used in the following publication:

• Unbinned multivariate observables for global SMEFT analyses from machine learning, Raquel Gomez Ambrosio, Jaco ter Hoeve, Maeve Madigan, Juan Rojo and Veronica Sanz [Gomez Ambrosio et al., 2022]

We present the results of this paper here, including additional plots and animations not present in the paper, as well as links to the trained neural networks and associated training reports.

The results are divided into three sections, corresponding to the three physical processes under consideration: parton level top pair production, particle level top pair production in the dilepton channel and Higgs production in association with a Z boson in the \(\ell^{+} \ell^{-} b \bar{b}\) final state. Click on the links below for more details.

Contents:

• Inclusive top pair production at the parton level
  • Models and training reports
  • Results
    • Pseudo-data generation and benchmarking
    • Neural network training
      • Overview
      • Animations of the progression of the exact/ML model comparison with training
      • Animation of the training of the decision boundary function
    • Constraints on the SMEFT
• Inclusive top pair production at the particle level
  • Models
  • Results
    • Kinematic features used as inputs to the neural network
    • Results from the ML model vs binning in two features, \(O(\Lambda^{-2})\)
    • Results from the ML model trained on two features vs all features, \(O(\Lambda^{-2})\)
    • Results from the ML model vs binning in two features, \(O(\Lambda^{-4})\)
    • Results from the ML model training on two features vs all features, \(O(Λ^{-4})\)
    • Impact of methodological uncertainties
• Higgs production in association with a Z boson
  • Models
  • Results
    • Kinematic features used as inputs to the neural network
    • Results from the ML model vs binning in \(p_{T}^{Z}\), \(O(\Lambda^{-2})\), pair-wise fit
    • Results from the ML model vs binning in \(p_{T}^{Z}\), \(O(\Lambda^{-4})\), pair-wise fit
    • Results from the ML model vs binning in \(p_{T}^{Z}\), \(O(\Lambda^{-4})\), fully marginalised
    • Impact of methodological uncertainties