Efficient Graph-Based Active Learning with Probit Likelihood via Gaussian Approximations

TL;DR

This paper introduces a new method for active learning in graph-based semi-supervised learning that uses Gaussian approximations to handle non-Gaussian Bayesian models, enabling efficient acquisition functions.

Contribution

It develops Gaussian approximations for non-Gaussian models, introduces a rank-one update for look-ahead methods, and proposes a novel model change acquisition function.

Findings

Efficient approximation of non-Gaussian distributions for active learning.

Development of a rank-one update for model retraining.

Introduction of a new model change acquisition function.

Abstract

We present a novel adaptation of active learning to graph-based semi-supervised learning (SSL) under non-Gaussian Bayesian models. We present an approximation of non-Gaussian distributions to adapt previously Gaussian-based acquisition functions to these more general cases. We develop an efficient rank-one update for applying "look-ahead" based methods as well as model retraining. We also introduce a novel "model change" acquisition function based on these approximations that further expands the available collection of active learning acquisition functions for such methods.