Learning from graphs with structural variation
This work addresses the robustness of graph kernels for machine learning practitioners, but it is incremental as it adapts existing methods with limited gains.
The paper tackled the problem of structural variation in graph data affecting graph kernel performance by introducing a noise-robust adaptation of the GraphHopper kernel, resulting in modestly improved predictive performance on benchmark datasets, and found that the impact of synthetic structural errors on the Weisfeiler-Lehman kernel varies significantly by dataset.
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.