Calibeating Prediction-Powered Inference

Lars van der Laan · 2026

Read Calibeating Prediction-Powered Inference by Lars van der Laan (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-21260.txt.

Introduction

Lars van der Laan and Mark Van Der Laan improve mean estimation from small labeled and large unlabeled samples when a black-box predictor may be off. Their calibrated prediction-powered inference tunes the prediction score on the labeled sample before estimation, without retraining. Isotonic tuning earns first-order gains, beating the raw score and simpler rules, while linear tuning matches PPI++. The work shows plain PPI is a special case of augmented inverse-probability weighting, and tuned estimators often beat PPI in tests.

Catalog record

CREATORLars van der Laan
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE181555
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.21260

Browse the full library · About the Tower of Babel