Research map / Graph learning / Graph self-supervised learning

A Large-scale Training Paradigm for Graph Generative Models

ICLR 2025 · Graph learning › Graph self-supervised learning · approach: Diffusion-based graph generation

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes a large-scale training paradigm for graph generative models, pretraining on diverse graph data to enable generation across domains.

Problem

Prior graph generative models (GraphRNN, MDVAE, MoFlow, GDSS, DiGress) are trained on small single-domain datasets, lacking the diversity and generalization benefits of large generative models.

Approach

Adopts the large-scale training paradigm of LGMs (GPT, Stable Diffusion, Sora) for graph generation, training on large, diverse graph data rather than per-dataset models.

Experimental setup

Baselines: GraphRNN MDVAE MoFlow GDSS DiGress

Builds on: GraphRNN MDVAE MoFlow GDSS DiGress

Closest papers by meaning

• SBGD: Improving Graph Diffusion Generative Model via Stochastic Block Diffusion (ICML 2025 · similarity 0.89)
• Generative Graph Pattern Machine (NeurIPS 2025 · similarity 0.89)
• Towards Addressing Frontiers in Graph Generation (AAAI 2025 · similarity 0.89)
• GLBench: A Comprehensive Benchmark for Graph with Large Language Models (NeurIPS 2024 · similarity 0.89)
• GOFA: A Generative One-For-All Model for Joint Graph Language Modeling (ICLR 2025 · similarity 0.88)
• Quality Measures for Dynamic Graph Generative Models (ICLR 2025 · similarity 0.88)