Affinity guided Geometric Semi-Supervised Metric Learning

TL;DR

This paper introduces a deep semi-supervised metric learning approach using Riemannian geometry and affinity propagation, significantly improving over classical methods and addressing the lack of deep SSDML research.

Contribution

It extends classical SSDML to a deep learning framework with Riemannian optimization and proposes a novel affinity-based triplet mining strategy.

Findings

Outperforms existing SSDML methods in experiments.

Leverages Riemannian optimization for constrained metric learning.

Introduces a new affinity propagation triplet mining technique.

Abstract

In this paper, we revamp the forgotten classical Semi-Supervised Distance Metric Learning (SSDML) problem from a Riemannian geometric lens, to leverage stochastic optimization within a end-to-end deep framework. The motivation comes from the fact that apart from a few classical SSDML approaches learning a linear Mahalanobis metric, deep SSDML has not been studied. We first extend existing SSDML methods to their deep counterparts and then propose a new method to overcome their limitations. Due to the nature of constraints on our metric parameters, we leverage Riemannian optimization. Our deep SSDML method with a novel affinity propagation based triplet mining strategy outperforms its competitors.