Research map / Image generation / Fast diffusion sampling
Generalised Flow Maps for Few-Step Generative Modelling on Riemannian Manifolds
Oscar Davis, Michael Samuel Albergo, Nicholas Matthew Boffi, Michael M. Bronstein, Joey Bose
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 GFM (G-LSD) one-step sampling instead of RFM | RFM · protein torsion angles General (2D) | MMD at 1 NFE: 0.45 -> 0.02 | Improved |
| use G-LSD instead of RFM at 1 NFE | RFM · RNA (7D) | MMD at 1 NFE: 0.68 -> 0.08 | Improved |
| use G-ESD instead of G-LSD at 1 NFE (ablation) | G-LSD · protein torsion angles | MMD at 1 NFE: 0.02 -> 0.29 (General) | Hurt |
| use GFM objectives instead of RFM for likelihood | RFM · protein side chain datasets | test NLL: 1.01 -> 0.95 (General, G-PSD) | Improved |
| use GFM objectives instead of RFM for likelihood | RFM · RNA (7D) torus | test NLL: -5.20 -> -4.61 (G-ESD) | Hurt |
| use GFM instead of RFM for likelihood | RFM · Earth datasets | test NLL: Earthquake -0.28 -> -1.08; Volcano -7.93 -> -4.96 | Mixed |
| use GFM instead of RFM on SO(3) | RFM · SO(3) synthetic | test NLL | No clear effect |
| use GFM instead of RFM on SO(3) | RFM · SO(3) synthetic | MMD | Improved |
| use G-MF or G-ESD instead of RFM at high NFE | RFM · hyperbolic Poincare ball | MMD | Hurt |
| add self-distillation loss (G-ESD instead of RFM training) | RFM · Volcano dataset | training throughput: 0.60 -> 0.29 epochs/min | Hurt |
Closest papers by meaning
- Rectifying Categorical Flows on Statistical Manifolds for One-Step Generation (NeurIPS 2026 · similarity 0.91)
- Learning Distributions on Manifolds with Free-Form Flows (NeurIPS 2024 · similarity 0.89)
- Generative Modeling on Manifolds Through Mixture of Riemannian Diffusion Processes (ICML 2024 · similarity 0.89)
- Riemannian Consistency Model (NeurIPS 2025 · similarity 0.89)
- How to build a consistency model: Learning flow maps via self-distillation (NeurIPS 2025 · similarity 0.88)
- Flow Matching on General Geometries (ICLR 2024 · similarity 0.88)
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.
