Multiple-Prediction-Powered Inference
Charlie Cowen-Breen, Alekh Agarwal, Stephen Bates, William W. Cohen, Jacob Eisenstein, Amir Globerson, Adam Fisch · 2026
Read Multiple-Prediction-Powered Inference by Charlie Cowen-Breen, Alekh Agarwal, Stephen Bates, William W. Cohen, Jacob Eisenstein, Amir Globerson, Adam Fisch (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-2603-27414.txt.
Description
Cowen-Breen and co-authors tackle a common data problem in statistics. Good measurements cost a lot, while cheap proxy scores abound. Their MultiPPI framework blends many proxy sources with a few gold labels. It learns the cost and overlap of each source. Then it splits the budget for the best gain. The team proves strong guarantees for the estimator. Tests on three language model scoring tasks beat prior methods. Errors stay lower at the same total cost.
Catalog record
| CREATOR | Charlie Cowen-Breen, Alekh Agarwal, Stephen Bates, William W. Cohen, Jacob Eisenstein, Amir Globerson, Adam Fisch |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 126684 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.27414