Information-Theoretic Representation Learning for Positive-Unlabeled Classification

TL;DR

This paper introduces an information-theoretic representation learning method for positive-unlabeled classification that improves class-prior estimation and classification accuracy without needing prior probability estimation.

Contribution

It proposes a novel representation learning approach based on the information-maximization principle that bypasses the need for class-prior estimation in PU classification.

Findings

Improves class-prior estimation accuracy with deep neural networks

Achieves state-of-the-art PU classification performance

Provides a preprocessing method that enhances existing PU classifiers

Abstract

Recent advances in weakly supervised classification allow us to train a classifier only from positive and unlabeled (PU) data. However, existing PU classification methods typically require an accurate estimate of the class-prior probability, which is a critical bottleneck particularly for high-dimensional data. This problem has been commonly addressed by applying principal component analysis in advance, but such unsupervised dimension reduction can collapse underlying class structure. In this paper, we propose a novel representation learning method from PU data based on the information-maximization principle. Our method does not require class-prior estimation and thus can be used as a preprocessing method for PU classification. Through experiments, we demonstrate that our method combined with deep neural networks highly improves the accuracy of PU class-prior estimation, leading to state-of-the-art PU classification performance.