Conformal Prediction for Time-Dependent PDEs
Abstract
Uncertainty quantification is crucial in scientific machine learning, where models inform safety-critical tasks such as flood forecasting, financial risk management, and thermal control in machines. Conformal prediction provides distribution-free coverage guarantees, but in time-dependent settings common to physics and engineering, these guarantees can break down, leading to systematic undercoverage. We study this problem in the context of surrogate models for time-dependent physical systems described by partial differential equations. We prove that in a function space setting, distributions at arbitrarily close times can be mutually singular, making exact coverage guarantees impossible. We further show that in discretized settings, the total variation distance between distributions often saturates exponentially fast with the grid resolution, mandating restrictions on the spatial discretization to ensure coverage. Finally, for certain settings, we show how to use principled weighted conformal prediction to obtain finite-sample coverage guarantees over growing time horizons.
Related
- Conformal prediction regions for dependent censored lifetimes
- Conformal Prediction for Spatially Dependent Data via Sequential Whitening
- Auditing Conformal Prediction for Language Models Under Inference-Time Model Shift
- Conformal Prediction for Time Series with Deep Sequence Models
- Efficient conformal prediction intervals for time series: Online PID-Expert aggregation
BibTeX
@misc{stiller2026conformal,
title = {{Conformal Prediction for Time-Dependent PDEs}},
author = {Joshua Stiller and Annika Schneider and Eyke Hüllermeier},
howpublished = {arXiv},
year = {2026},
month = nov,
doi = {10.48550/arxiv.2611.08409},
eprint = {2611.08409},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2611.08409},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Conformal Prediction for Time-Dependent PDEs
AU - Joshua Stiller
AU - Annika Schneider
AU - Eyke Hüllermeier
PY - 2026
DA - 2026/11/05
PB - arXiv
DO - 10.48550/arxiv.2611.08409
UR - https://arxiv.org/abs/2611.08409
AB - Uncertainty quantification is crucial in scientific machine learning, where models inform safety-critical tasks such as flood forecasting, financial risk management, and thermal control in machines. Conformal prediction provides distribution-free coverage guarantees, but in time-dependent settings common to physics and engineering, these guarantees can break down, leading to systematic undercoverage. We study this problem in the context of surrogate models for time-dependent physical systems described by partial differential equations. We prove that in a function space setting, distributions at arbitrarily close times can be mutually singular, making exact coverage guarantees impossible. We further show that in discretized settings, the total variation distance between distributions often saturates exponentially fast with the grid resolution, mandating restrictions on the spatial discretization to ensure coverage. Finally, for certain settings, we show how to use principled weighted conformal prediction to obtain finite-sample coverage guarantees over growing time horizons.
ER -