Non-parametric targeted Bayesian estimation of class proportions in unlabeled data

TL;DR

This paper presents a new Bayesian method for estimating class proportions in unlabeled data, focusing on low-dimensional parameters, with theoretical guarantees and practical applications demonstrated through simulations and a medical case study.

Contribution

It introduces a targeted Bayesian estimator that focuses on specific parameters, providing asymptotic efficiency and robustness, unlike traditional full-data Bayesian approaches.

Findings

Posterior distribution converges to an efficient, Gaussian limit.

Method is doubly robust and asymptotically normal.

Numerical studies confirm favorable frequentist properties.

Abstract

We introduce a novel Bayesian estimator for the class proportion in an unlabeled dataset, based on the targeted learning framework. Our procedure requires the specification of a prior (and outputs a posterior) only for the target of inference, instead of the prior (and posterior) on the full-data distribution employed by classical non-parametric Bayesian methods .When the scientific question can be characterized by a low-dimensional parameter functional, focus on such a prior and posterior distributions is more aligned with Bayesian subjectivism, compared to focus on entire data distributions. We prove a Bernstein-von Mises-type result for our proposed Bayesian procedure, which guarantees that the posterior distribution converges to the distribution of an efficient, asymptotically linear estimator. In particular, the posterior is Gaussian, doubly robust, and efficient in the limit, under the only assumption that certain nuisance parameters are estimated at slow rates. We perform numerical studies illustrating the frequentist properties of the method. We also illustrate their use in a motivating application to estimate the proportion of embolic strokes of undetermined source arising from occult cardiac sources or large-artery atherosclerotic lesions. Though we focus on the motivating example of the proportion of cases in an unlabeled dataset, the procedure is general and can be adapted to estimate any pathwise differentiable parameter in a non-parametric model.