Research map / Image generation / Fast diffusion sampling
Learning Discrete Diffusion on Graphs via Free-Energy Gradient Flows
Dario Rancati, Jan Maas, Francesco Locatello
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 learned free-energy gradient flow (W_K JKO) instead of pretrained OpenFIM | discrete diffusion forecasting · Synthetic graphs, beta=0.01 | Hellinger distance: 0.142 -> 0.066 | Improved |
| use learned free-energy gradient flow instead of OpenFIM | discrete diffusion forecasting · Synthetic graphs, beta=0.10 | Hellinger distance: 0.150 -> 0.059 | Improved |
| use learned free-energy gradient flow instead of OpenFIM | discrete diffusion forecasting · Synthetic graphs, beta=0.20 | Hellinger distance: 0.159 -> 0.069 | Improved |
| increase graph size N from 10 to 50 (ablation) | free-energy learner · Synthetic graphs | Hellinger distance: approximately linear growth in N over N in {10, 20, 30, 40, 50} | Hurt |
| reduce number of samples to 500 (from 10000) (ablation) | free-energy learner · Synthetic graphs, N=10 | Hellinger distance: slight degradation at 500 samples | Hurt |
| vary samples among 1000, 2000, 5000, 10000 (ablation) | free-energy learner · Synthetic graphs, N=10 | Hellinger distance | No clear effect |
Closest papers by meaning
- Scalable Discrete Diffusion Samplers: Combinatorial Optimization and Statistical Physics (ICLR 2025 · similarity 0.89)
- Learning diffusion at lightspeed (NeurIPS 2024 · similarity 0.89)
- Graph Generation with Diffusion Mixture (ICML 2024 · similarity 0.88)
- Wasserstein Gradient Flows and Forward-Only Diffusion Are Not Enough for Multimodal Sampling (NeurIPS 2026 · similarity 0.88)
- DiPhon: Diffusion on Graphons for Scalable Graph Generation (NeurIPS 2026 · similarity 0.87)
- Test-time Scaling of Diffusions with Flow Maps (NeurIPS 2026 · similarity 0.87)
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.
