Graph Normalizing Flows

TL;DR

This paper introduces graph normalizing flows, a reversible graph neural network model that enables scalable prediction and generation of graphs with reduced memory usage, and demonstrates competitive generative capabilities.

Contribution

The paper presents a novel reversible graph neural network model called graph normalizing flows, combining them with a graph auto-encoder for efficient, permutation-invariant graph generation.

Findings

Performs comparably to message passing neural networks on supervised tasks.

Reduces memory footprint, enabling larger graph processing.

Achieves competitive results with state-of-the-art auto-regressive models in graph generation.

Abstract

We introduce graph normalizing flows: a new, reversible graph neural network model for prediction and generation. On supervised tasks, graph normalizing flows perform similarly to message passing neural networks, but at a significantly reduced memory footprint, allowing them to scale to larger graphs. In the unsupervised case, we combine graph normalizing flows with a novel graph auto-encoder to create a generative model of graph structures. Our model is permutation-invariant, generating entire graphs with a single feed-forward pass, and achieves competitive results with the state-of-the art auto-regressive models, while being better suited to parallel computing architectures.