Research map / Image generation / Fast diffusion sampling
Spontaneous symmetry breaking in generative diffusion models
Gabriel Raya, Luca Ambrogioni
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 Gaussian late-start initialization with DDPM | DDPM · MNIST, 10 denoising steps | FID: 6.75 -> 4.21 | Improved |
| Use Gaussian late-start initialization with DDPM | DDPM · CIFAR10, 10 denoising steps | FID: 43.35 -> 28.77 | Improved |
| Use Gaussian late-start initialization with DDPM | DDPM · CelebA32, 10 denoising steps | FID: 26.79 -> 11.05 | Improved |
| Use Gaussian late-start initialization with DDPM | DDPM · Imagenet64, 10 denoising steps | FID: 65.68 -> 57.31 | Improved |
| Use Gaussian late-start initialization with DDPM | DDPM · CelebA64, 10 denoising steps | FID: 36.66 -> 23.79 | Improved |
| Use Gaussian late-start initialization with DDIM | DDIM · MNIST, 10 denoising steps | FID: 4.46 -> 2.44 | Improved |
| Use Gaussian late-start initialization with DDIM | DDIM · CIFAR10, 10 denoising steps | FID: 19.79 -> 15.98 | Improved |
| Use Gaussian late-start initialization with DDIM | DDIM · CelebA32, 10 denoising steps | FID: 11.37 -> 7.27 | Improved |
| Use Gaussian late-start initialization with DDIM | DDIM · Imagenet64, 10 denoising steps | FID: 38.21 -> 36.25 | Improved |
| Use Gaussian late-start initialization with DDIM | DDIM · CelebA64, 10 denoising steps | FID: 19.37 -> 15.82 | Improved |
| Use Gaussian late-start initialization with PNDM | PNDM · MNIST, 10 denoising steps | FID: 14.36 -> 5.02 | Improved |
| Use Gaussian late-start initialization with PNDM | PNDM · CIFAR10, 10 denoising steps | FID: 8.35 -> 5.90 | Improved |
Closest papers by meaning
- Manifolds, Random Matrices and Spectral Gaps: The geometric phases of generative diffusion (ICLR 2025 · similarity 0.89)
- Tracing the Roots: Leveraging Temporal Dynamics in Diffusion Trajectories for Origin Attribution (NeurIPS 2025 · similarity 0.88)
- Towards Non-Asymptotic Convergence for Diffusion-Based Generative Models (ICLR 2024 · similarity 0.88)
- CCS: Controllable and Constrained Sampling with Diffusion Models via Initial Noise Perturbation (NeurIPS 2025 · similarity 0.88)
- Critical windows: non-asymptotic theory for feature emergence in diffusion models (ICML 2024 · similarity 0.88)
- A solvable model of learning generative diffusion: theory and insights (NeurIPS 2025 · 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.
