Unlabeled Data Help in Graph-Based Semi-Supervised Learning: A Bayesian Nonparametrics Perspective

Abstract

In this paper we analyze the graph-based approach to semi-supervised learning under a manifold assumption. We adopt a Bayesian perspective and demonstrate that, for a suitable choice of prior constructed with sufficiently many unlabeled data, the posterior contracts around the truth at a rate that is minimax optimal up to a logarithmic factor. Our theory covers both regression and classification.

References in corpus (6)

• Rates of contraction of posterior distributions based on Gaussian process priors
• Reproducing kernel Hilbert spaces of Gaussian priors
• Error estimates for spectral convergence of the graph Laplacian on random geometric graphs towards the Laplace--Beltrami operator
• Local Regularization of Noisy Point Clouds: Improved Global Geometric Estimates and Data Analysis
• Posterior Consistency of Semi-Supervised Regression on Graphs
• Entropy of Sobolev's classes on Compact Homogeneous Riemannian Manifolds

Cited by in corpus (2)

• Graph-based Prior and Forward Models for Inverse Problems on Manifolds with Boundaries
• Finite Element Representations of Gaussian Processes: Balancing Numerical and Statistical Accuracy