Learning from graphs with structural variation

TL;DR

This paper investigates how structural variations and errors in graph data affect the performance of graph kernels, proposing a noise-robust adaptation and analyzing dataset-dependent effects.

Contribution

It introduces a noise-robust version of the GraphHopper kernel and evaluates the impact of structural errors on the Weisfeiler-Lehman kernel across datasets.

Findings

The noise-robust GraphHopper kernel shows modest performance improvements.

Structural errors impact kernel performance differently depending on the dataset.

Performance degradation varies with the type and extent of structural errors.

Abstract

We study the effect of structural variation in graph data on the predictive performance of graph kernels. To this end, we introduce a novel, noise-robust adaptation of the GraphHopper kernel and validate it on benchmark data, obtaining modestly improved predictive performance on a range of datasets. Next, we investigate the performance of the state-of-the-art Weisfeiler-Lehman graph kernel under increasing synthetic structural errors and find that the effect of introducing errors depends strongly on the dataset.