CREDO: Epistemic-Aware Conformalized Credal Envelopes for Regression

This collection of arXiv stat-ML papers (Mar 10 2026) covers theoretical and applied advances in uncertainty quantification, fairness and causality, robustness/transfer, sequential decision-making, and probabilistic forecasting. Highlights include: CREDO — a ‘‘credal-then-conformalize’’ recipe that builds epistemic-aware credal envelopes then applies split conformal calibration to retain marginal coverage while making interval width interpretable (aleatoric vs epistemic vs calibration slack); a full-data private conformal framework that leverages differential privacy stability to avoid data-sp