Prediction Uncertainty Controlled Computer Model Calibration

TL;DR

This paper proposes a two-layer framework for computer model calibration that controls predictive error. The framework uses a conditional flow-matching model and conformal risk control to achieve nominal miscoverage. It outperforms existing methods in synthetic and real-world benchmarks.

Why It Matters

The proposed framework has practical significance in applications where predictive accuracy and uncertainty quantification are crucial, such as epidemiology and climate modeling. It provides sharper and more reliable uncertainty estimates, which can inform decision-making. This can lead to better resource allocation and policy decisions.

Key Findings

• The proposed framework achieves nominal miscoverage in finite samples
• It outperforms KOH-MCMC in synthetic and real-world benchmarks
• It exposes structural inadequacy of mechanistic models