Research map / Graph learning / Graph self-supervised learning
FLAGG: Flexible Autoregressive Graph Generation
Samuel Cognolato, Alessandro Sperduti, Luciano Serafini
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 |
|---|---|---|---|
| use DFS node ordering instead of BFS (ablation) | Vanilla FLAGG {1,4} · QM9 | FCD: 0.144 -> 0.153 | Hurt |
| use random node ordering instead of BFS (ablation) | Vanilla FLAGG {1,4} · QM9 | FCD: 0.144 -> 0.253 | Hurt |
| use random ordering instead of BFS (ablation) | Vanilla FLAGG {1,4} · QM9 | Validity (%): 99.69 -> 99.72 | No clear effect |
| replace learned halting and insertion with empirical node-count sampling (ablation) | FLAGG {1,3} · Enzymes | Degree MMD: 0.048 -> 0.050 | Mixed |
| replace learned halting and insertion with empirical node-count sampling (ablation) | FLAGG {1,2,8} · Enzymes | GIN MMD: 0.048 -> 0.037 | Mixed |
| add property-based edge weights (P-FLAGG) to FLAGG | FLAGG seq-small · Ego | GIN MMD: 0.075 -> 0.037 | Improved |
| add property-based edge weights (P-FLAGG) to FLAGG | FLAGG seq-small · Ego-small | GIN MMD: 0.031 -> 0.034 | Hurt |
| use large block sizes {1,4} instead of {1,2} | FLAGG · QM9 | FCD: 0.212 -> 0.144 | Improved |
| use FLAGG improvements instead of IFH with intermediate sequentiality | IFH {1,4,8} · ZINC250k | Validity (%): 65.68 -> 97.81 | Improved |
| increase block size from seq-small to seq-big | FLAGG · Ego | Sampling time (minutes): 283.899 -> 67.213 | Improved |
| use properties FLAGG instead of seq-small FLAGG for sampling speed | FLAGG · Ego | Sampling time (minutes): 283.899 -> 1.259 | Improved |
| reduce DiGress denoising steps by factor J=10 (ablation) | FLAGG {1,4} · QM9 | Validity: validity over 99% | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DiGress — DiGress: Discrete Denoising diffusion for graph generation (ICLR 2023) · better on 1 of 1 shared result
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
- A Deep Probabilistic Framework for Continuous Time Dynamic Graph Generation (AAAI 2025 · similarity 0.88)
- Pard: Permutation-Invariant Autoregressive Diffusion for Graph Generation (NeurIPS 2024 · similarity 0.87)
- SBGD: Improving Graph Diffusion Generative Model via Stochastic Block Diffusion (ICML 2025 · similarity 0.86)
- LayerDAG: A Layerwise Autoregressive Diffusion Model for Directed Acyclic Graph Generation (ICLR 2025 · similarity 0.86)
- Efficient and Scalable Graph Generation through Iterative Local Expansion (ICLR 2024 · similarity 0.86)
- Discrete-state Continuous-time Diffusion for Graph Generation (NeurIPS 2024 · similarity 0.86)
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.
