Skip to content

Uncertainty quantification & calibration

Section status

Contributor: Unclaimed — this section has no named author yet. See Contribute.

Last reviewed: —

Status: Stub — a starting point only. The curated list is not yet written.

Overview

A classifier that is confident and wrong is worse than useless in a measurement. Particle physics has an unusually demanding relationship with uncertainty: results carry systematic uncertainties that must be defensible to a collaboration and a referee, and "the network said so" is not an error budget. Two distinct problems live here. The first is estimating uncertainty on a network's output — the usual Bayesian, ensemble and calibration machinery. The second, more particular to this field, is what to do when the systematic is a property of the simulation the network trained on, so a mismodelled input yields a confidently biased output. Inference-aware training, which optimises the final measurement rather than a proxy classification loss, is the main response.

  • Dealing with Nuisance Parameters using Machine Learning in High Energy Physics: a Review, Dorigo et al. (2020) (arXiv:2007.09121) — on handling nuisance parameters with machine learning, which is the version of the problem that usually dominates
  • Solving Simulation Systematics in and with AI/ML, Viren et al. (2022) (arXiv:2203.06112) — on solving simulation systematics in and with ML
  • Uncertainty in Physics and AI: Taxonomy, Quantification, and Validation, Haussmann et al. (2026) (arXiv:2605.10378) — the orienting reference: sorts out what the various uncertainties actually are before getting to methods, and treats validation as a first-class problem rather than an afterthought. Read this first
  • Deep-Learning Jets with Uncertainties and More, Bollweg et al. (2020) (arXiv:1904.10004) — Bayesian networks for jet tagging: the reference demonstration that a HEP classifier can report a per-event uncertainty, and that the number means something
  • Learning to Pivot with Adversarial Networks, Louppe et al. (2017) (arXiv:1611.01046) — training a classifier to be independent of a nuisance parameter by adversarially removing the information. The founding method for the second problem in the Overview, and still the baseline everything else is measured against
  • INFERNO: Inference-Aware Neural Optimisation, De Castro & Dorigo (2019) (arXiv:1806.04743) — optimize the expected uncertainty on the final measurement rather than a classification loss — the paper that made "inference-aware" a category
  • Uncertainty Aware Learning for High Energy Physics, Ghosh et al. (2021) (arXiv:2105.08742) — a systematic comparison of the ways systematics can be folded into training, and what each actually buys

Curated paper list

This list is a seed, not a curated selection

A few landmark papers are listed to give the section a starting shape. A proper curated list — thematically grouped, with an annotation on every entry — is what this section still needs. See Contribute.

  • Development of systematic uncertainty-aware neural network trainings for binned-likelihood analyses at the LHC, CMS Collaboration (2025) (arXiv:2502.13047) — systematic-uncertainty-aware network training for binned analyses, from CMS
  • Precision calibration of calorimeter signals in the ATLAS experiment using an uncertainty-aware neural network, ATLAS Collaboration (2024) (arXiv:2412.04370) — calorimeter calibration in ATLAS with uncertainty handled explicitly
  • FAIR Universe HiggsML Uncertainty Challenge Competition, Bhimji et al. (2024) (arXiv:2410.02867) — the FAIR Universe HiggsML uncertainty challenge, which poses the problem as a shared task

Benchmarks, datasets & software

Not yet compiled for this section.

Open questions

What is settled here, what is contested, and what remains unsolved? This is the part a bibliography structurally cannot provide, and often the most useful paragraph on the page.

Further reading

Relevant work that is not an entry point — too specialized, too recent, or simply not where a newcomer should start. Suggestions that do not fit the curated list above belong here rather than being turned away.

Nothing listed yet.

Cross-references