Beyond Homophily with Graph Echo State Networks

TL;DR

This paper evaluates Graph Echo State Networks (GESN) on node classification tasks across various homophily levels, demonstrating their competitive accuracy and efficiency compared to traditional deep models, and analyzing the impact of reservoir parameters.

Contribution

It is the first to assess GESN on diverse homophily levels in node classification, highlighting their effectiveness and efficiency relative to fully trained deep models.

Findings

Reservoir models achieve comparable or better accuracy than deep models.

GESN performs well across different degrees of graph homophily.

Reservoir radius impacts model performance.

Abstract

Graph Echo State Networks (GESN) have already demonstrated their efficacy and efficiency in graph classification tasks. However, semi-supervised node classification brought out the problem of over-smoothing in end-to-end trained deep models, which causes a bias towards high homophily graphs. We evaluate for the first time GESN on node classification tasks with different degrees of homophily, analyzing also the impact of the reservoir radius. Our experiments show that reservoir models are able to achieve better or comparable accuracy with respect to fully trained deep models that implement ad hoc variations in the architectural bias, with a gain in terms of efficiency.