What Do GNNs Actually Learn? Towards Understanding their Representations

Giannis Nikolentzos, Michail Chatzianastasis, Michalis Vazirgiannis

Keywords: learned representations, walks, message passing neural networks

Abstract: In recent years, graph neural networks (GNNs) have achieved great success in the field of graph representation learning. Although prior work has shed light on the expressiveness of those models (i.e., whether they can distinguish pairs of non-isomorphic graphs), it is still not clear what structural information is encoded into the node representations that are learned by those models. In this paper, we address this gap by studying the node representations learned by four standard GNN models. We find that some models produce identical representations for all nodes, while the representations learned by other models are linked to some notion of walks of specific length that start from the nodes. We establish Lipschitz bounds for these models with respect to the number of (normalized) walks. Additionally, we investigate the influence of node features on the learned representations. We find that if the initial representations of all nodes point in the same direction, the representations learned at the k-th layer of the models are also related to the initial features of nodes that can be reached in exactly k steps. We also apply our findings to understand the phenomenon of oversquashing that occurs in GNNs. Our theoretical analysis is validated through experiments on synthetic and real-world datasets.

Submission Type: Full paper proceedings track submission (max 9 main pages).

Software: https://github.com/giannisnik/gnn-representations

Submission Number: 167