Conformal Uncertainty Quantification Guarantees for Neural Operators

DOI10.48550/arXiv.2608.28515MaRDI QIDQ7285110arXivFDO

Authors Nicolas Boullé, Tom Stent

Publication date 28 August 2026

Published in arXiv

Tldr We develop a conformal framework to guarantee $1-γ$ coverage of neural operator predictions around true solutions for arbitrary probability spaces with calibrated uncertainty quantification and prove guarantees under mild assumptions on the data distribution. Our method reduces a normalized residual field to its $(1-γ)$-quantile, computing a scaling factor using a held-out calibration dataset and yielding tighter bands than existing corrections in Darcy flow and Navier--Stokes experiments.

Full work available at URL https://doi.org/10.48550/arXiv.2608.28515

arXiv classification

cs.AImath.NAmath.PR

Cites work

• A Mathematical Guide to Operator Learning
• A tutorial on conformal prediction
• Algorithmic Learning in a Random World
• Approximate Bayesian Neural Operators: Uncertainty Quantification for Parametric PDEs
• Choose a Transformer: Fourier or Galerkin
• Conditional validity of inductive conformal predictors
• Conformal Prediction: A Gentle Introduction
• Conformalized-DeepONet: a distribution-free framework for uncertainty quantification in deep operator networks
• Convergence Rates for Learning Linear Operators from Noisy Data
• Elliptic PDE learning is provably data-efficient
• Error estimates for DeepOnets: A deep learning framework in infinite dimensions
• Fourier Neural Operator for Parametric Partial Differential Equations
• GNOT: A General Neural Operator Transformer for Operator Learning
• MgNO: Efficient Parameterization of Linear Operators via Multigrid
• Multipole Graph Neural Operator for Parametric Partial Differential Equations
• Multiwavelet-based Operator Learning for Differential Equations
• On universal approximation and error bounds for Fourier Neural Operators
• Operator learning without the adjoint
• Uncertainty quantification in scientific machine learning: methods, metrics, and comparisons

This page was built for publication: Conformal Uncertainty Quantification Guarantees for Neural Operators

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q7285110)