SERAPH: Semi-supervised Metric Learning Paradigm with Hyper Sparsity

TL;DR

Seraph introduces a semi-supervised metric learning framework that maximizes entropy on labeled data and minimizes it on unlabeled data, integrating information theory with low-rank regularization for improved metric learning.

Contribution

It presents a novel information-theoretic semi-supervised approach for metric learning that does not depend on the manifold assumption and employs an efficient EM-like optimization scheme.

Findings

Outperforms many existing metric learning methods

Effectively integrates labeled and unlabeled data

Employs low-rank regularization for better metric structure

Abstract

We propose a general information-theoretic approach called Seraph (SEmi-supervised metRic leArning Paradigm with Hyper-sparsity) for metric learning that does not rely upon the manifold assumption. Given the probability parameterized by a Mahalanobis distance, we maximize the entropy of that probability on labeled data and minimize it on unlabeled data following entropy regularization, which allows the supervised and unsupervised parts to be integrated in a natural and meaningful way. Furthermore, Seraph is regularized by encouraging a low-rank projection induced from the metric. The optimization of Seraph is solved efficiently and stably by an EM-like scheme with the analytical E-Step and convex M-Step. Experiments demonstrate that Seraph compares favorably with many well-known global and local metric learning methods.