On Graph Classification Networks, Datasets and Baselines
This work addresses the problem of evaluating graph classification methods for researchers, highlighting that incremental improvements may be overstated due to baseline issues.
The paper investigates graph classification networks and finds that their performance is highly sensitive to initialization and jumping-knowledge structures, while showing that simpler models like MLPs and single-layer GCNs achieve competitive results, proposing these as new baselines.
Graph classification receives a great deal of attention from the non-Euclidean machine learning community. Recent advances in graph coarsening have enabled the training of deeper networks and produced new state-of-the-art results in many benchmark tasks. We examine how these architectures train and find that performance is highly-sensitive to initialisation and depends strongly on jumping-knowledge structures. We then show that, despite the great complexity of these models, competitive performance is achieved by the simplest of models -- structure-blind MLP, single-layer GCN and fixed-weight GCN -- and propose these be included as baselines in future.