Approximately Bayes-Optimal Pseudo Label Selection

TL;DR

This paper introduces BPLS, a Bayesian pseudo-label selection framework that reduces confirmation bias in semi-supervised learning by using an analytically approximated posterior predictive criterion, improving performance on high-dimensional data.

Contribution

The paper develops a Bayesian criterion for pseudo-label selection, proving its Bayes optimality and providing an analytical approximation to enhance semi-supervised learning.

Findings

BPLS outperforms traditional methods on high-dimensional data.

The Bayesian criterion effectively mitigates confirmation bias.

Analytical approximations make the method computationally feasible.

Abstract

Semi-supervised learning by self-training heavily relies on pseudo-label selection (PLS). The selection often depends on the initial model fit on labeled data. Early overfitting might thus be propagated to the final model by selecting instances with overconfident but erroneous predictions, often referred to as confirmation bias. This paper introduces BPLS, a Bayesian framework for PLS that aims to mitigate this issue. At its core lies a criterion for selecting instances to label: an analytical approximation of the posterior predictive of pseudo-samples. We derive this selection criterion by proving Bayes optimality of the posterior predictive of pseudo-samples. We further overcome computational hurdles by approximating the criterion analytically. Its relation to the marginal likelihood allows us to come up with an approximation based on Laplace's method and the Gaussian integral. We empirically assess BPLS for parametric generalized linear and non-parametric generalized additive models on simulated and real-world data. When faced with high-dimensional data prone to overfitting, BPLS outperforms traditional PLS methods.