GraphVAE: Towards Generation of Small Graphs Using Variational Autoencoders

TL;DR

This paper introduces GraphVAE, a variational autoencoder designed to generate small graphs directly, addressing the challenge of graph generation without linearization, with applications demonstrated in molecule generation.

Contribution

It presents a novel graph generative model that outputs probabilistic graphs directly, bypassing linearization, and applies it to molecule generation tasks.

Findings

Successfully generates small graphs with probabilistic outputs

Demonstrates effectiveness in molecule generation

Advances graph generation methods beyond linearization approaches

Abstract

Deep learning on graphs has become a popular research topic with many applications. However, past work has concentrated on learning graph embedding tasks, which is in contrast with advances in generative models for images and text. Is it possible to transfer this progress to the domain of graphs? We propose to sidestep hurdles associated with linearization of such discrete structures by having a decoder output a probabilistic fully-connected graph of a predefined maximum size directly at once. Our method is formulated as a variational autoencoder. We evaluate on the challenging task of molecule generation.