Semi-Supervised Metric Learning: A Deep Resurrection

TL;DR

This paper introduces a deep semi-supervised metric learning method that propagates affinities from labeled to unlabeled data using a graph-based approach, improving the learning of discriminative embeddings with limited labels.

Contribution

It revamps semi-supervised distance metric learning with a deep, graph-based approach that propagates affinities and mines triplet constraints, incorporating orthogonality constraints for better performance.

Findings

Effective propagation of affinities improves metric learning.

Orthogonality constraints prevent model collapse.

Method outperforms classical linear approaches.

Abstract

Distance Metric Learning (DML) seeks to learn a discriminative embedding where similar examples are closer, and dissimilar examples are apart. In this paper, we address the problem of Semi-Supervised DML (SSDML) that tries to learn a metric using a few labeled examples, and abundantly available unlabeled examples. SSDML is important because it is infeasible to manually annotate all the examples present in a large dataset. Surprisingly, with the exception of a few classical approaches that learn a linear Mahalanobis metric, SSDML has not been studied in the recent years, and lacks approaches in the deep SSDML scenario. In this paper, we address this challenging problem, and revamp SSDML with respect to deep learning. In particular, we propose a stochastic, graph-based approach that first propagates the affinities between the pairs of examples from labeled data, to that of the unlabeled pairs. The propagated affinities are used to mine triplet based constraints for metric learning. We impose orthogonality constraint on the metric parameters, as it leads to a better performance by avoiding a model collapse.