Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Fair Universe Higgs Uncertainty Challenge

Ragansu Chakkappai, Wahid Bhimji, Paolo Calafiura, Po-Wen Chang, Yuan-Tang Chou, Sascha Diefenbacher, Jordan Dudley, Steven Farrell, Aishik Ghosh, Isabelle Guyon, Chris Harris, Shih-Chieh Hsu, Elham E. Khoda, Benjamin Nachman, Peter Nugent, David Rousseau, Benjamin Thorne, Ihsan Ullah, Yulei Zhang

TL;DR

The Fair Universe Higgs Uncertainty Challenge targets robust uncertainty quantification for the $H \rightarrow \tau^+ \tau^-$ cross-section using ML techniques. It introduces a public benchmark with shifted data and nuisance-parameter variations, evaluated via pseudo-experiments to ensure correct $1\sigma$ coverage. The dataset, generated with Pythia8 and Delphes 3.5, includes 28 features and structured systematics, enabling rigorous uncertainty quantification. Winning approaches include unbinned inclusive cross-section measurements with machine-learned uncertainties and contrastive normalizing flows for uncertainty-aware parameter estimation, demonstrating complementary strengths and guiding future uncertainty-aware AI in HEP.

Abstract

This competition in high-energy physics (HEP) and machine learning was the first to strongly emphasise uncertainties in $(H \rightarrow τ^+ τ^-)$ cross-section measurement. Participants were tasked with developing advanced analysis techniques capable of dealing with uncertainties in the input training data and providing credible confidence intervals. The accuracy of these intervals was evaluated using pseudo-experiments to assess correct coverage. The dataset is now published in Zenodo, and the winning submissions are fully documented.

Fair Universe Higgs Uncertainty Challenge

TL;DR

The Fair Universe Higgs Uncertainty Challenge targets robust uncertainty quantification for the cross-section using ML techniques. It introduces a public benchmark with shifted data and nuisance-parameter variations, evaluated via pseudo-experiments to ensure correct coverage. The dataset, generated with Pythia8 and Delphes 3.5, includes 28 features and structured systematics, enabling rigorous uncertainty quantification. Winning approaches include unbinned inclusive cross-section measurements with machine-learned uncertainties and contrastive normalizing flows for uncertainty-aware parameter estimation, demonstrating complementary strengths and guiding future uncertainty-aware AI in HEP.

Abstract

This competition in high-energy physics (HEP) and machine learning was the first to strongly emphasise uncertainties in cross-section measurement. Participants were tasked with developing advanced analysis techniques capable of dealing with uncertainties in the input training data and providing credible confidence intervals. The accuracy of these intervals was evaluated using pseudo-experiments to assess correct coverage. The dataset is now published in Zenodo, and the winning submissions are fully documented.

Paper Structure

This paper contains 5 sections, 3 figures.

Table of Contents

  1. Introduction
  2. Challenge Setting
  3. Datasets and Systematics
  4. Evaluation and Scoring
  5. Competition results and best submissions

Figures (3)

  • Figure 1: Diagram of the particles in the final state chosen: one lepton, one tau hadron, up to two jets, and the missing transverse momentum vector
  • Figure 2: (\ref{['fig:mu_distribution']})Coverage plot: predicted intervals (blue lines) for each pseudo experiment generated for a given $\mu_{\rm true}$ (vertical dotted line). The coverage (here $70\pm5\%$) is determined by the fraction of horizontal blue lines intersected by the vertical line. The average width of the interval is here 1.068. (\ref{['fig:coverage_plot']}) Coverage penality: 1D function to penalise models with poor coverage. benato2025fairuniversehiggsmluncertainty
  • Figure 3: Comparative study of the three finalists (blue for Hzume, orange for HEPHY and green for Ibrahim's model) with 1000 trials of 100 pseudo-experiments. (\ref{['fig:coverage_vs_mu']}) the coverage from each trial, (\ref{['fig:interval_length_vs_mu']}) the average CI width and (\ref{['fig:quantile_score_vs_mu']}) the quantile score. benato2025fairuniversehiggsmluncertainty