Cross-prediction-powered inference
Proceedings of the National Academy of Sciences · 09.04.2024 · 121 (15)
DOI: 10.1073/pnas.2322083121 ↗Kurz gefasst
Fachartikel in der Zeitschrift Proceedings of the National Academy of Sciences, online erschienen am 03.04.2024. Verfasst von Tijana Zrnic und 1 weiteren Autor:innen. Beteiligte Einrichtung: Department of Statistics, Stanford University, Stanford, CA 94305. Die Arbeit wurde bisher 34-mal zitiert (Crossref) und nennt 60 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: DOD | USN | ONR | Office of Naval Research Global (ONRG), National Science Foundation (NSF), Simons Foundation (SF).
Abstract (Originalsprache)
While reliable data-driven decision-making hinges on high-quality labeled data, the acquisition of quality labels often involves laborious human annotations or slow and expensive scientific measurements. Machine learning is becoming an appealing alternative as sophisticated predictive techniques are being used to quickly and cheaply produce large amounts of predicted labels; e.g., predicted protein structures are used to supplement experimentally derived structures, predictions of socioeconomic indicators from satellite imagery are used to supplement accurate survey data, and so on. Since predictions are imperfect and potentially biased, this practice brings into question the validity of downstream inferences. We introduce cross-prediction: a method for valid inference powered by machine learning. With a small labeled dataset and a large unlabeled dataset, cross-prediction imputes the missing labels via machine learning and applies a form of debiasing to remedy the prediction inaccuracies. The resulting inferences achieve the desired error probability and are more powerful than those that only leverage the labeled data. Closely related is the recent proposal of prediction-powered inference [A. N. Angelopoulos, S. Bates, C. Fannjiang, M. I. Jordan, T. Zrnic, Science 382 , 669–674 (2023)], which assumes that a good pretrained model is already available. We show that cross-prediction is consistently more powerful than an adaptation of prediction-powered inference in which a fraction of the labeled data is split off and used to train the model. Finally, we observe that cross-prediction gives more stable conclusions than its competitors; its CIs typically have significantly lower variability.
Schlagwörter
Stichwörter der Autor:innen
Autor:innen und Institutionen
- Department of Statistics, Stanford University, Stanford, CA 94305.
- Stanford Data Science, Stanford University, Stanford, CA 94305.
- Department of Mathematics, Stanford University, Stanford, CA 94305.
ORCID-Profile: Tijana Zrnic
Förderung
- DOD | USN | ONR | Office of Naval Research Global (ONRG)
- National Science Foundation (NSF)
- Simons Foundation (SF)
- DOD | USA | AFC | CCDC | Army Research Office (ARO)
Zitieren
APA
Zrnic, T., & Candès, E. J. (2024). Cross-prediction-powered inference. Proceedings of the National Academy of Sciences, 121(15). https://doi.org/10.1073/pnas.2322083121
Vancouver
Zrnic T, Candès EJ. Cross-prediction-powered inference. Proceedings of the National Academy of Sciences. 2024;121(15). doi:10.1073/pnas.2322083121
BibTeX
@article{Zrnic2024,
author = {Zrnic, Tijana and Candès, Emmanuel J},
title = {Cross-prediction-powered inference},
journal = {Proceedings of the National Academy of Sciences},
year = {2024},
volume = {121},
number = {15},
doi = {10.1073/pnas.2322083121}
}
EUVIMED