Accurate surrogate amplitudes with calibrated uncertainties
Neural networks for LHC physics have to be accurate, reliable, and controlled. Using neural surrogates for the prediction of loop amplitudes as a use case, we first show how activation functions are systematically tested with Kolmogorov-Arnold Networks. Then, we train neural surrogates to simultaneously predict the target amplitude and an uncertainty for the prediction. We disentangle systematic uncertainties, learned by a well-defined likelihood loss, from statistical uncertainties, which require the introduction of Bayesian neural networks or repulsive ensembles. We test the coverage of the learned uncertainties using pull distributions to quantify the calibration of cutting-edge neural surrogates.
Authors
- Tilman Plehn (ORCID: https://orcid.org/0000-0001-5660-7790)
- Ramon Winterhalder (ORCID: https://orcid.org/0000-0002-9902-578X)
- Manuel Haußmann (ORCID: https://orcid.org/0000-0002-8967-0849)
- Luigi Favaro (ORCID: https://orcid.org/0000-0003-2421-7100)
- Nina Elmer (ORCID: https://orcid.org/0000-0001-9579-6010)
- Henning Bahl (ORCID: https://orcid.org/0000-0003-4212-8881)
Institutions
- University of Southern Denmark (DK)
- University of Milan (IT)
- Heidelberg University (DE)
- Istituto Nazionale di Fisica Nucleare, Sezione di Milano (IT)
- UCLouvain (BE)
Publication Details
- Journal
- SciPost Physics Core
- Published
- 2025-10-24
- DOI
- https://doi.org/10.21468/scipostphyscore.8.4.073
- Citations
- 4
- Primary Topic
- GNSS positioning and interference
- Type
- article
- Field-Weighted Citation Impact
- 19.97
Funders
- Baden-Württemberg Stiftung
- European Commission
- Deutsche Forschungsgemeinschaft
- Bundesministerium für Bildung und Forschung
- Fonds De La Recherche Scientifique - FNRS
- Israel Science Foundation
- International Max Planck Research School for Advanced Methods in Process and Systems Engineering