English

GraphDF: A Discrete Flow Model for Molecular Graph Generation

Machine Learning 2021-06-03 v2 Artificial Intelligence

Abstract

We consider the problem of molecular graph generation using deep models. While graphs are discrete, most existing methods use continuous latent variables, resulting in inaccurate modeling of discrete graph structures. In this work, we propose GraphDF, a novel discrete latent variable model for molecular graph generation based on normalizing flow methods. GraphDF uses invertible modulo shift transforms to map discrete latent variables to graph nodes and edges. We show that the use of discrete latent variables reduces computational costs and eliminates the negative effect of dequantization. Comprehensive experimental results show that GraphDF outperforms prior methods on random generation, property optimization, and constrained optimization tasks.

Keywords

Cite

@article{arxiv.2102.01189,
  title  = {GraphDF: A Discrete Flow Model for Molecular Graph Generation},
  author = {Youzhi Luo and Keqiang Yan and Shuiwang Ji},
  journal= {arXiv preprint arXiv:2102.01189},
  year   = {2021}
}

Comments

Accepted by ICML 2021

R2 v1 2026-06-23T22:44:40.418Z