Research map / Graph learning / GNN architectures
Neural Graph Generation from Graph Statistics
Kiarash Zahirnia, Yaochen Hu, Mark Coates, Oliver Schulte
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 |
|---|---|---|---|
| replace parametric statistics-based GGMs (BTER, Chung-Lu, Erdos-Renyi, SBM) with neural GenStat | statistics-based graph generation · 8 datasets | GNN-based MMD RBF: 12 out of 16 cases better | Improved |
| use GenStat instead of parametric baselines under Random/Pretrained-GNN | statistics-based graph generation · 8 datasets | GNN-based MMD RBF | Improved |
| use GenStat instead of parametric baselines on Protein | statistics-based graph generation · Protein | F1 PR: 83.72% vs. 71.08% (Random-GNN); 79.95% vs. 23.76% (Pretrained-GNN) | Improved |
| use GenStat instead of parametric baselines | statistics-based graph generation · most datasets | statistics-based MMDs: 1-2 orders of magnitude better | Improved |
| train on graph statistics only (GenStat) vs on full adjacency matrices (deep GGMs) | GenStat · real-world datasets | GNN-based MMD RBF: top two in 9 out of 10 scores | Improved |
| train on graph statistics only vs full adjacency matrices on synthetic regular graphs (Lobster, Grid, Triangle Grid) | GenStat · synthetic datasets | GNN-based MMD RBF: Grid Random-GNN 0.53 vs BiGG 0.14 | Hurt |
| train under edge epsilon-LDP (epsilon 0.1 to 4) with GenStat on statistics vs BiGG on perturbed adjacency | GenStat · MUTAG, Lobster, Grid | Random-GNN MMD RBF | Improved |
| use GenStat generated graphs instead of BiGG for GNN benchmarking on link prediction | GenStat · 8 datasets | benchmark effectiveness (correlation): at least 0.5046 Pearson correlation | Mixed |
| train on statistics instead of adjacency matrices | GenStat · training time | training time: up to two orders of magnitude lower than fastest auto-regressive model; up to 5 times lower than GraphVAE-MM | Improved |
| generate all-at-once (GenStat, GraphVAE-MM) instead of auto-regressively | GenStat · generation time | generation time | Improved |
Closest papers by meaning
- Graph Generative Model for Benchmarking Graph Neural Networks (ICML 2023 · similarity 0.91)
- Differentially Private Decoupled Graph Convolutions for Multigranular Topology Protection (NeurIPS 2023 · similarity 0.89)
- Can Graph Neural Networks Expose Training Data Properties? An Efficient Risk Assessment Approach (NeurIPS 2024 · similarity 0.86)
- GRAIN: Exact Graph Reconstruction from Gradients (ICLR 2025 · similarity 0.85)
- Generative Graph Pattern Machine (NeurIPS 2025 · similarity 0.85)
- Towards Addressing Frontiers in Graph Generation (AAAI 2025 · similarity 0.85)
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.
