A Multiscale Graph Convolutional Network Using Hierarchical Clustering

TL;DR

This paper introduces a multiscale graph convolutional network that leverages hierarchical clustering to better utilize the hierarchical structure in networks, improving learning across multiple scales.

Contribution

It presents a novel architecture combining hierarchical clustering with graph convolutional layers to capture multiscale network representations.

Findings

Achieved competitive performance on a citation network benchmark.

Demonstrated the method's potential for hierarchical network analysis.

Applicable to various domains like molecular and protein networks.

Abstract

The information contained in hierarchical topology, intrinsic to many networks, is currently underutilised. A novel architecture is explored which exploits this information through a multiscale decomposition. A dendrogram is produced by a Girvan-Newman hierarchical clustering algorithm. It is segmented and fed through graph convolutional layers, allowing the architecture to learn multiple scale latent space representations of the network, from fine to coarse grained. The architecture is tested on a benchmark citation network, demonstrating competitive performance. Given the abundance of hierarchical networks, possible applications include quantum molecular property prediction, protein interface prediction and multiscale computational substrates for partial differential equations.