AI RESEARCH

Calibeating Prediction-Powered Inference

arXiv CS.LG

ArXi:2604.21260v1 Announce Type: cross We study semisupervised mean estimation with a small labeled sample, a large unlabeled sample, and a black-box prediction model whose output may be miscalibrated. A standard approach in this setting is augmented inverse-probability weighting (AIPW) [Robins, 1994], which protects against prediction-model misspecification but can be inefficient when the prediction score is poorly aligned with the outcome scale. We