ML Coffee: Unbinned inclusive cross-section measurements with machine-learned systematic uncertainties (aka Gollum goes Higgs Uncertainty Challenge)

by Dr Claudius Krause (Austrian Academy of Sciences)

Friday 4 Jul 2025, 10:00 → 11:00 Europe/Paris

Salle 9

I present "Team HEPHY" 's approach to the FAIR Universe Higgs Uncertainty Challenge:  a novel methodology for addressing systematic uncertainties in unbinned inclusive cross-section measurements and related collider-based inference problems. Our approach “Guaranteed Optimal Log-Likelihood-based Unbinned Method” (GOLLUM) incorporates known analytic dependencies on parameters of interest, including signal strengths and nuisance parameters. When these dependencies are unknown, as is frequently the case for systematic uncertainties, dedicated neural network parametrizations provide an approximation that is trained on simulated data. The resulting machine-learned surrogate captures the complete parameter dependence of the likelihood ratio, providing a near-optimal test statistic. We perform a first-principles inclusive cross-section measurement of H → ττ in the single-lepton channel, utilizing simulated data from the FAIR Universe Higgs Uncertainty Challenge.

Relevant paper (ordered by relevance):
The talk will be based on 2505.05544. The FAIR Universe Higgs Uncertainty Challenge is introduced at 2410.02867. The initial idea (and some more details) are given in 2406.19076, which focuses on the SMEFT case.
 

To join by Zoom: https://uw-edu-pl.zoom.us/j/91603361707

Meeting ID: 916 0336 1707

 

Gollum.C.Krause.pdf