Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Diffusion-based Graph Generative Methods

Hongyang Chen, Can Xu, Lingyu Zheng, Qiang Zhang, Xuemin Lin

TL;DR

This work surveys diffusion-based graph generative methods, detailing three core paradigms—DDPMs, SGMs, and Score SDEs—and their adaptations to graph data. It offers a structured taxonomy of diffusion backbones, enhancement strategies for discrete and non-Euclidean graphs, and a comprehensive account of applications in AI for scientific discovery, computer vision, and generic graph generation. The review also compiles datasets and evaluation metrics across tasks, and discusses limitations and future directions, arguing that it is among the most exhaustive analyses to date. Overall, the paper provides practical guidance for selecting diffusion frameworks, tailoring them to graph structures, and benchmarking across standard domains to advance graph-oriented generation research.

Abstract

Being the most cutting-edge generative methods, diffusion methods have shown great advances in wide generation tasks. Among them, graph generation attracts significant research attention for its broad application in real life. In our survey, we systematically and comprehensively review on diffusion-based graph generative methods. We first make a review on three mainstream paradigms of diffusion methods, which are denoising diffusion probabilistic models, score-based genrative models, and stochastic differential equations. Then we further categorize and introduce the latest applications of diffusion models on graphs. In the end, we point out some limitations of current studies and future directions of future explorations. The summary of existing methods metioned in this survey is in https://github.com/zhejiangzhuque/Diffusion-based-Graph-Generative-Methods.

Diffusion-based Graph Generative Methods

TL;DR

This work surveys diffusion-based graph generative methods, detailing three core paradigms—DDPMs, SGMs, and Score SDEs—and their adaptations to graph data. It offers a structured taxonomy of diffusion backbones, enhancement strategies for discrete and non-Euclidean graphs, and a comprehensive account of applications in AI for scientific discovery, computer vision, and generic graph generation. The review also compiles datasets and evaluation metrics across tasks, and discusses limitations and future directions, arguing that it is among the most exhaustive analyses to date. Overall, the paper provides practical guidance for selecting diffusion frameworks, tailoring them to graph structures, and benchmarking across standard domains to advance graph-oriented generation research.

Abstract

Being the most cutting-edge generative methods, diffusion methods have shown great advances in wide generation tasks. Among them, graph generation attracts significant research attention for its broad application in real life. In our survey, we systematically and comprehensively review on diffusion-based graph generative methods. We first make a review on three mainstream paradigms of diffusion methods, which are denoising diffusion probabilistic models, score-based genrative models, and stochastic differential equations. Then we further categorize and introduce the latest applications of diffusion models on graphs. In the end, we point out some limitations of current studies and future directions of future explorations. The summary of existing methods metioned in this survey is in https://github.com/zhejiangzhuque/Diffusion-based-Graph-Generative-Methods.
Paper Structure (34 sections, 18 equations, 4 figures, 2 tables)

This paper contains 34 sections, 18 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Problem formulation
  3. Denoising diffusion probabilistic models (DDPMs)
  4. Score-Based Generative Models (SGMs)
  5. Stochastic Differential Equations (Score SDEs)
  6. Brief Summary of Three Paradigms of Diffusion Models
  7. Enhancement of Adaptability of Diffusion on Graphs
  8. Comparisons and Connections with other generative methods
  9. Applications
  10. AI for Scientific Discovery
  11. De novo molecule design
  12. Conformation generation and optimization
  13. De novo ligand design
  14. Ligand docking
  15. Protein design
  16. ...and 19 more sections

Figures (4)

  • Figure 1: An illustration of diffusion models on images and graphs scoresde_21_songgsdm_22_luo. The right arrow points at the direction of diffusion process, where noises are injected to ground-truth data. The left one indicates the sampling phase where samples are generated.
  • Figure 2: An illustration of diffusion-based molecule generation targetdiff_23_guan
  • Figure 3: An illustration of diffusion-based motion generation modiff_23_zhao
  • Figure 4: Illustrations of HouseDiffusionhousediffusion_23_shabani and NAPnap_23_lei. The upper one is the overall structure of HouseDiffusion while the lower one is the overview of NAP.