Weighted Conformal Prediction

Overview

Weighted conformal prediction studies how to form prediction sets and intervals with reliable coverage when calibration and deployment data have different distributions. It reweights calibration scores using estimated density ratios, kernel matching, or local and temporal similarity, then applies weighted quantiles or conformal p-values. These methods are used for covariate and label shift, online and federated prediction, and settings such as molecular design, forecasting, and medical imaging. A central tension is that weighting can improve relevance to deployment data but unstable estimates, limited support overlap, and uncertain weights can weaken guarantees or make predictions wider; sharper conditional guarantees and efficient, robust calibration remain open challenges.

Latest papers

CardsList

33 papers2 papers in the last four weeks, down 71% on the four weeks before. 0.0% of all new papers.

  1. ConfHit: Conformal Generative Design with Oracle Free Guarantees

    Mar 7, 2026Siddhartha Laghuvarapu, Ying Jin, Jimeng SunMolecular GenerationWeighted Conformal Prediction

  2. Conformalized Neural Networks for Federated Uncertainty Quantification under Dual Heterogeneity

    Feb 26, 2026Quang-Huy Nguyen, Jiaqi Wang, Wei-Shinn KuFederated Conformal PredictionUncertainty Quantification

  3. Mask-Conditional Conformal Prediction: Valid Uncertainty For All Missing Data Mechanisms

    Dec 16, 2025Jiarong Fan, Juhyun Park. Thi Phuong Thuy Vo, Nicolas BrunelConformal PredictionRegression Prediction Intervals