Group-Conditional Conformal Prediction

Overview

Conformal prediction uses calibration data to build prediction sets or intervals with coverage guarantees, while group-conditional methods seek reliability for groups such as classes, languages, or missing-data patterns. Common approaches include Mondrian calibration with separate group thresholds, hierarchical or learned subgrouping, and local or online calibration. These methods are used in settings from image classification and clinical prediction to language-model deferral and forecasting, where pooled coverage can hide failures for particular groups. A central tension is that finer-grained guarantees can require more calibration data and produce larger prediction sets, while distribution shifts can undermine group coverage.

Latest papers

CardsList

32 papers6 papers in the last four weeks, down 25% on the four weeks before. 0.0% of all new papers.

  1. Efficient Federated Conformal Prediction with Group-Conditional Guarantee

    Mar 15, 2026Haifeng Wen, Osvaldo Simeone, Hong XingFederated Conformal PredictionGroup-Conditional Conformal Prediction

  2. Beyond Procedure: Substantive Fairness in Conformal Prediction

    Feb 18, 2026Pengqi Liu, Zijun Yu, Mouloud Belbahri +3Algorithmic FairnessConformal Prediction