Graph Neural Networks with Precomputed Node Features

TL;DR

This paper explores augmenting Graph Neural Networks with precomputed node features like positional embeddings and IDs to overcome their limitations in graph discrimination, demonstrating improved performance and sample efficiency.

Contribution

Introduces and evaluates several node feature augmentations for GNNs, showing their effectiveness in theoretical and empirical tasks.

Findings

Positional embeddings outperform other augmentations in subgraph detection.

Augmentations improve GNN performance on benchmarks.

Positional embeddings outperform ground truth positions in some cases.

Abstract

Most Graph Neural Networks (GNNs) cannot distinguish some graphs or indeed some pairs of nodes within a graph. This makes it impossible to solve certain classification tasks. However, adding additional node features to these models can resolve this problem. We introduce several such augmentations, including (i) positional node embeddings, (ii) canonical node IDs, and (iii) random features. These extensions are motivated by theoretical results and corroborated by extensive testing on synthetic subgraph detection tasks. We find that positional embeddings significantly outperform other extensions in these tasks. Moreover, positional embeddings have better sample efficiency, perform well on different graph distributions and even outperform learning with ground truth node positions. Finally, we show that the different augmentations perform competitively on established GNN benchmarks, and advise on when to use them.