A Wasserstein Graph Distance Based on Distributions of Probabilistic Node Embeddings

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Scholkemper, Michael, Kühn, Damin, Nabbefeld, Gerion, Musall, Simon, Kampa, Björn, Schaub, Michael T.
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914747490762752
author Scholkemper, Michael
Kühn, Damin
Nabbefeld, Gerion
Musall, Simon
Kampa, Björn
Schaub, Michael T.
author_facet Scholkemper, Michael
Kühn, Damin
Nabbefeld, Gerion
Musall, Simon
Kampa, Björn
Schaub, Michael T.
contents Distance measures between graphs are important primitives for a variety of learning tasks. In this work, we describe an unsupervised, optimal transport based approach to define a distance between graphs. Our idea is to derive representations of graphs as Gaussian mixture models, fitted to distributions of sampled node embeddings over the same space. The Wasserstein distance between these Gaussian mixture distributions then yields an interpretable and easily computable distance measure, which can further be tailored for the comparison at hand by choosing appropriate embeddings. We propose two embeddings for this framework and show that under certain assumptions about the shape of the resulting Gaussian mixture components, further computational improvements of this Wasserstein distance can be achieved. An empirical validation of our findings on synthetic data and real-world Functional Brain Connectivity networks shows promising performance compared to existing embedding methods.
format Preprint
id arxiv_https___arxiv_org_abs_2401_03913
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle A Wasserstein Graph Distance Based on Distributions of Probabilistic Node Embeddings
Scholkemper, Michael
Kühn, Damin
Nabbefeld, Gerion
Musall, Simon
Kampa, Björn
Schaub, Michael T.
Computational Engineering, Finance, and Science
Social and Information Networks
Distance measures between graphs are important primitives for a variety of learning tasks. In this work, we describe an unsupervised, optimal transport based approach to define a distance between graphs. Our idea is to derive representations of graphs as Gaussian mixture models, fitted to distributions of sampled node embeddings over the same space. The Wasserstein distance between these Gaussian mixture distributions then yields an interpretable and easily computable distance measure, which can further be tailored for the comparison at hand by choosing appropriate embeddings. We propose two embeddings for this framework and show that under certain assumptions about the shape of the resulting Gaussian mixture components, further computational improvements of this Wasserstein distance can be achieved. An empirical validation of our findings on synthetic data and real-world Functional Brain Connectivity networks shows promising performance compared to existing embedding methods.
title A Wasserstein Graph Distance Based on Distributions of Probabilistic Node Embeddings
topic Computational Engineering, Finance, and Science
Social and Information Networks
url https://arxiv.org/abs/2401.03913