A Hybrid Method for Distance Metric Learning

TL;DR

This paper introduces a hybrid distance metric learning method that combines similarity ratings and class labels, improving retrieval accuracy in synthetic and medical image datasets.

Contribution

It presents a novel generative model that integrates class labels with similarity ratings to enhance distance metric learning.

Findings

Significant improvement in retrieval performance using the proposed method.

Effective on both synthetic and real medical image data.

Leverages class labels to capture additional information not in feature vectors.

Abstract

We consider the problem of learning a measure of distance among vectors in a feature space and propose a hybrid method that simultaneously learns from similarity ratings assigned to pairs of vectors and class labels assigned to individual vectors. Our method is based on a generative model in which class labels can provide information that is not encoded in feature vectors but yet relates to perceived similarity between objects. Experiments with synthetic data as well as a real medical image retrieval problem demonstrate that leveraging class labels through use of our method improves retrieval performance significantly.