Bregman–Hausdorff Divergence: Strengthening the Connections Between Computational Geometry and Machine Learning

The purpose of this paper is twofold. On a technical side, we propose an extension of the Hausdorff distance from metric spaces to spaces equipped with asymmetric distance measures. Specifically, we focus on extending it to the family of Bregman divergences, which includes the popular Kullback–Leibl...

Full description

Saved in:
Bibliographic Details
Main Authors: Tuyen Pham, Hana Dal Poz Kouřimská, Hubert Wagner
Format: Article
Language:English
Published: MDPI AG 2025-05-01
Series:Machine Learning and Knowledge Extraction
Subjects:
Online Access:https://www.mdpi.com/2504-4990/7/2/48
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849432006593609728
author Tuyen Pham
Hana Dal Poz Kouřimská
Hubert Wagner
author_facet Tuyen Pham
Hana Dal Poz Kouřimská
Hubert Wagner
author_sort Tuyen Pham
collection DOAJ
description The purpose of this paper is twofold. On a technical side, we propose an extension of the Hausdorff distance from metric spaces to spaces equipped with asymmetric distance measures. Specifically, we focus on extending it to the family of Bregman divergences, which includes the popular Kullback–Leibler divergence (also known as relative entropy). The resulting dissimilarity measure is called a Bregman–Hausdorff divergence and compares two collections of vectors—without assuming any pairing or alignment between their elements. We propose new algorithms for computing Bregman–Hausdorff divergences based on a recently developed Kd-tree data structure for nearest neighbor search with respect to Bregman divergences. The algorithms are surprisingly efficient even for large inputs with hundreds of dimensions. As a benchmark, we use the new divergence to compare two collections of probabilistic predictions produced by different machine learning models trained using the relative entropy loss. In addition to the introduction of this technical concept, we provide a survey. It outlines the basics of Bregman geometry, and motivated the Kullback–Leibler divergence using concepts from information theory. We also describe computational geometric algorithms that have been extended to this geometry, focusing on algorithms relevant for machine learning.
format Article
id doaj-art-c047930a098e4861b31d8ec7ab67d067
institution Kabale University
issn 2504-4990
language English
publishDate 2025-05-01
publisher MDPI AG
record_format Article
series Machine Learning and Knowledge Extraction
spelling doaj-art-c047930a098e4861b31d8ec7ab67d0672025-08-20T03:27:28ZengMDPI AGMachine Learning and Knowledge Extraction2504-49902025-05-01724810.3390/make7020048Bregman–Hausdorff Divergence: Strengthening the Connections Between Computational Geometry and Machine LearningTuyen Pham0Hana Dal Poz Kouřimská1Hubert Wagner2Department of Mathematics, University of Florida, Gainesville, FL 32611, USAApplied Geometry and Topology, University of Potsdam, Am Neuen Palais 10, 14469 Potsdam, GermanyDepartment of Mathematics, University of Florida, Gainesville, FL 32611, USAThe purpose of this paper is twofold. On a technical side, we propose an extension of the Hausdorff distance from metric spaces to spaces equipped with asymmetric distance measures. Specifically, we focus on extending it to the family of Bregman divergences, which includes the popular Kullback–Leibler divergence (also known as relative entropy). The resulting dissimilarity measure is called a Bregman–Hausdorff divergence and compares two collections of vectors—without assuming any pairing or alignment between their elements. We propose new algorithms for computing Bregman–Hausdorff divergences based on a recently developed Kd-tree data structure for nearest neighbor search with respect to Bregman divergences. The algorithms are surprisingly efficient even for large inputs with hundreds of dimensions. As a benchmark, we use the new divergence to compare two collections of probabilistic predictions produced by different machine learning models trained using the relative entropy loss. In addition to the introduction of this technical concept, we provide a survey. It outlines the basics of Bregman geometry, and motivated the Kullback–Leibler divergence using concepts from information theory. We also describe computational geometric algorithms that have been extended to this geometry, focusing on algorithms relevant for machine learning.https://www.mdpi.com/2504-4990/7/2/48computational geometrymachine learninginformation theoryShannon’s entropyrelative entropyBregman divergence
spellingShingle Tuyen Pham
Hana Dal Poz Kouřimská
Hubert Wagner
Bregman–Hausdorff Divergence: Strengthening the Connections Between Computational Geometry and Machine Learning
Machine Learning and Knowledge Extraction
computational geometry
machine learning
information theory
Shannon’s entropy
relative entropy
Bregman divergence
title Bregman–Hausdorff Divergence: Strengthening the Connections Between Computational Geometry and Machine Learning
title_full Bregman–Hausdorff Divergence: Strengthening the Connections Between Computational Geometry and Machine Learning
title_fullStr Bregman–Hausdorff Divergence: Strengthening the Connections Between Computational Geometry and Machine Learning
title_full_unstemmed Bregman–Hausdorff Divergence: Strengthening the Connections Between Computational Geometry and Machine Learning
title_short Bregman–Hausdorff Divergence: Strengthening the Connections Between Computational Geometry and Machine Learning
title_sort bregman hausdorff divergence strengthening the connections between computational geometry and machine learning
topic computational geometry
machine learning
information theory
Shannon’s entropy
relative entropy
Bregman divergence
url https://www.mdpi.com/2504-4990/7/2/48
work_keys_str_mv AT tuyenpham bregmanhausdorffdivergencestrengtheningtheconnectionsbetweencomputationalgeometryandmachinelearning
AT hanadalpozkourimska bregmanhausdorffdivergencestrengtheningtheconnectionsbetweencomputationalgeometryandmachinelearning
AT hubertwagner bregmanhausdorffdivergencestrengtheningtheconnectionsbetweencomputationalgeometryandmachinelearning