Cross-Prediction-Powered Inference

TL;DR

This paper introduces cross-prediction, a machine learning-based method for valid statistical inference using both labeled and unlabeled data, improving power and stability over existing approaches.

Contribution

The paper proposes cross-prediction, a novel inference technique that leverages machine learning to incorporate unlabeled data, enhancing validity, power, and stability of statistical conclusions.

Findings

Cross-prediction achieves valid error control in inference.

It is more powerful than adapted prediction-powered inference.

It produces more stable confidence intervals with lower variability.

Abstract

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, which assumes that a good pre-trained 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 confidence intervals typically have significantly lower variability.