Research map / Image generation / Fast diffusion sampling
Scaling Riemannian Diffusion Models
Aaron Lou, Minkai Xu, Adam Farris, Stefano Ermon
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 accurate heat kernel in denoising score matching instead of 50-eigenfunction approximation | denoising score matching · Volcano (S^2) | negative log likelihood: -1.28 -> -4.69 | Improved |
| use accurate heat kernel instead of inaccurate approximation | denoising score matching · Earthquake (S^2) | negative log likelihood: 0.13 -> -0.27 | Improved |
| use accurate heat kernel instead of inaccurate approximation | denoising score matching · Flood (S^2) | negative log likelihood: 0.73 -> 0.44 | Improved |
| use accurate heat kernel instead of inaccurate approximation | denoising score matching · Fire (S^2) | negative log likelihood: -0.60 -> -1.51 | Improved |
| accurate denoising score matching versus sliced score matching | Riemannian diffusion · Volcano (S^2) | negative log likelihood: -4.92 vs -4.69 | No clear effect |
| accurate denoising score matching versus sliced score matching | Riemannian diffusion · Fire (S^2) | negative log likelihood: -1.33 vs -1.51 | No clear effect |
| add Riemannian diffusion on hyperspherical embeddings | CIDER (hypersphere) · CIFAR-100 OOD: SVHN | false positive rate: 93.53 -> 28.95 | Improved |
| add Riemannian diffusion on hyperspherical embeddings | CIDER (hypersphere) · CIFAR-100 OOD: LSUN | false positive rate: 93.68 -> 35.78 | Improved |
| add Riemannian diffusion on hyperspherical embeddings | CIDER (hypersphere) · CIFAR-100 OOD: Texture | false positive rate: 92.41 -> 62.87 | Improved |
| diffusion on hypersphere versus penultimate layer features | CIDER · CIFAR-100 OOD: LSUN | false positive rate: 16.16 -> 35.78 | Hurt |
| diffusion on hypersphere versus penultimate layer features | CIDER · CIFAR-100 OOD: Places365 | false positive rate: 79.63 -> 76.54 | Improved |
| Riemannian score matching versus Riemannian flow matching on torus checkerboards | Riemannian diffusion · torus checkerboard | density quality | Improved |
Closest papers by meaning
- Efficient Diffusion Models for Symmetric Manifolds (ICML 2025 · similarity 0.92)
- Generative Modeling on Manifolds Through Mixture of Riemannian Diffusion Processes (ICML 2024 · similarity 0.90)
- Toward a Unified Geometry Understanding : Riemannian Diffusion Framework for Graph Generation and Prediction (NeurIPS 2025 · similarity 0.88)
- Manifold Diffusion Fields (ICLR 2024 · similarity 0.87)
- Improving the Euclidean Diffusion Generation of Manifold Data by Mitigating Score Function Singularity (NeurIPS 2025 · similarity 0.87)
- Manifolds, Random Matrices and Spectral Gaps: The geometric phases of generative diffusion (ICLR 2025 · 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.
