Graphite: Iterative Generative Modeling of Graphs

TL;DR

Graphite introduces an unsupervised deep generative model for graphs that leverages graph neural networks and an iterative refinement strategy, achieving superior performance on multiple graph tasks.

Contribution

It presents a novel framework combining VAEs with graph neural networks and an iterative decoding approach for large graph representation learning.

Findings

Outperforms existing methods on synthetic and benchmark datasets

Achieves better density estimation, link prediction, and node classification

Establishes a theoretical link between message passing and variational inference

Abstract

Graphs are a fundamental abstraction for modeling relational data. However, graphs are discrete and combinatorial in nature, and learning representations suitable for machine learning tasks poses statistical and computational challenges. In this work, we propose Graphite, an algorithmic framework for unsupervised learning of representations over nodes in large graphs using deep latent variable generative models. Our model parameterizes variational autoencoders (VAE) with graph neural networks, and uses a novel iterative graph refinement strategy inspired by low-rank approximations for decoding. On a wide variety of synthetic and benchmark datasets, Graphite outperforms competing approaches for the tasks of density estimation, link prediction, and node classification. Finally, we derive a theoretical connection between message passing in graph neural networks and mean-field variational inference.