Towards Expressive Graph Representations for Graph Neural Networks

Chengsheng Mao, Liang Yao, Yuan Luo

TL;DR: graph representation, graph neural network, set representation, expressive power

Abstract: Graph Neural Network (GNN) shows its powerful capability for graph representation learning in various application areas. However, most existing GNN variants learn the graph representations in a non-injective or non-continuous fashion, both reducing the model expressive power. In this paper, we present a theoretical framework to improve the expressive power of GNN by taking both injectivity and continuity into account. Accordingly, we develop \textit{Injective Continuous Graph Neural Network} (ICGNN) that learns the graph and node representations in an injective and continuous fashion, so that it can map similar nodes or graphs to similar embeddings, and non-equivalent nodes or non-isomorphic graphs to different embeddings. We validate the proposed ICGNN model for graph classification and node classification on multiple benchmark datasets including both simple graphs and attributed graphs. The experimental results demonstrate that our model achieves state-of-the-art performances on most of the benchmarks.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

Submission Guidelines: Yes

Please Choose The Closest Area That Your Submission Falls Into: Deep Learning and representational learning