Research map / Image generation / Fast diffusion sampling
Breaking the Curse of Dimensionality: Diffusion Models Efficiently Learn Low-Dimensional Distributions
Peng Wang, Huijie Zhang, Zekai Zhang, Siyi Chen, Yi Ma, Qing Qu
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 MoLRG mixture of low-rank Gaussians parameterization instead of a single full-rank Gaussian | theoretical network parameterization · FashionMNIST | distance to U-Net samples: 72.62 -> 57.56 | Improved |
| use MoLRG mixture of low-rank Gaussians parameterization instead of a single full-rank Gaussian | theoretical network parameterization · MNIST | distance to U-Net samples: 69.12 -> 62.53 | Improved |
| use MoLRG mixture of low-rank Gaussians parameterization instead of a single full-rank Gaussian | theoretical network parameterization · CIFAR-10 | distance to U-Net samples: 33.55 -> 31.29 | Improved |
| use MoLRG mixture of low-rank Gaussians parameterization instead of a single full-rank Gaussian | theoretical network parameterization · FFHQ | distance to U-Net samples: 36.75 -> 35.78 | Improved |
| normalize training sample count by intrinsic dimension N/ID instead of N/ID^2 as predictor of generalization (ablation) | U-Net diffusion on MoLRG synthetic data · synthetic MoLRG (n=48, K=2) | GL score fit to sigmoid | Improved |
| normalize training sample count by intrinsic dimension N/ID instead of N/ID^2 as predictor of generalization (ablation) | U-Net diffusion on real image datasets · CIFAR-10, CelebA, FFHQ, AFHQ | GL score collapse across datasets | Improved |
| increase number of training samples N past the intrinsic dimension | diffusion models learning MoLRG · synthetic MoLRG phase transition (n=48, d 2-8, N 2-15) | subspace recovery success | Improved |
| steer along leading singular vectors of subspace basis instead of a random direction | pretrained DDPM on MetFaces · image editing | semantic edit visibility | Improved |
Closest papers by meaning
- Shallow diffusion networks provably learn hidden low-dimensional structure (ICLR 2025 · similarity 0.89)
- Residual Learning in Diffusion Models (CVPR 2024 · similarity 0.88)
- Adapting to Unknown Low-Dimensional Structures in Score-Based Diffusion Models (NeurIPS 2024 · similarity 0.87)
- Understanding Representation Dynamics of Diffusion Models via Low-Dimensional Modeling (NeurIPS 2025 · similarity 0.86)
- Sparsely Supervised Diffusion (NeurIPS 2026 · similarity 0.86)
- A solvable model of learning generative diffusion: theory and insights (NeurIPS 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.
