Research map / Graph learning / Graph self-supervised learning
Global and Local Topology-Aware Graph Generation via Dual Conditioning Diffusion
Yuhang Xie, Sinno Jialin Pan
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add self-conditioning (ablation) | DualDiff · ZINC250k | FCD: 0.65 -> 0.72 | Improved |
| add self-conditioning (ablation) | DualDiff · ZINC250k | KL: 0.82 -> 0.89 | Improved |
| replace self-conditioning with full dual conditioning (ablation) | DualDiff · ZINC250k | FCD: 0.72 -> 0.91 | Improved |
| replace self-conditioning with full dual conditioning (ablation) | DualDiff · ZINC250k | KL: 0.89 -> 0.98 | Improved |
| add global-to-local conditioning to local-to-global-only conditioning (full dual) (ablation) | DualDiff · ZINC250k | FCD: 0.75 -> 0.91 | Improved |
| add local-to-global conditioning to global-to-local-only conditioning (full dual) (ablation) | DualDiff · ZINC250k | FCD: 0.83 -> 0.91 | Improved |
| add dual conditioning (ablation) | DualDiff · QM9 sampling time for 10,000 graphs | sampling time (s): 32 -> 41 | Hurt |
| add dual conditioning (ablation) | DualDiff · ZINC250k sampling time for 10,000 graphs | sampling time (s): 169 -> 223 | Hurt |
| use DualDiff instead of GDSS | sampling · ZINC250k sampling time | sampling time (s): 1870 -> 223 | Improved |
| increase diffusion steps from 50 to 1000 (ablation) | DualDiff · Community-small | MMD: 0.098 -> 0.006 | Improved |
| use DualDiff instead of hierarchical PPGN | generic graph generation · Planar | Orbit MMD: 0.0006 -> 0.0002 | Improved |
| use DualDiff instead of hierarchical HiGen | generic graph generation · SBM | Orbit MMD: 0.0604 -> 0.0365 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- DiGress — DiGress: Discrete Denoising diffusion for graph generation (ICLR 2023) · better on 3 of 3 shared results
- GraphRNN — GraphRNN: Generating Realistic Graphs with Deep Auto-regressive Models (ICML 2018) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Graph Generation with Diffusion Mixture (ICML 2024 · similarity 0.92)
- Efficient and Scalable Graph Generation through Iterative Local Expansion (ICLR 2024 · similarity 0.91)
- Smooth Interpolation for Improved Discrete Graph Generative Models (ICML 2025 · similarity 0.90)
- Graph Generative Pre-trained Transformer (ICML 2025 · similarity 0.90)
- Discrete-state Continuous-time Diffusion for Graph Generation (NeurIPS 2024 · similarity 0.90)
- Unifying Generation and Prediction on Graphs with Latent Graph Diffusion (NeurIPS 2024 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
