Research map / Image generation / Fast diffusion sampling
Beta Diffusion
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 beta diffusion instead of Gaussian VP-EDM | CIFAR-10 image generation · CIFAR-10 | FID: 1.97 -> 3.06 | Hurt |
| use beta diffusion instead of VDM | CIFAR-10 image generation · CIFAR-10 | FID: 4.00 -> 3.06 | Helped |
| swap the KLUB arguments, yielding the negative ELBO objective (ablation) | Beta Diffusion · CIFAR-10, NFE = 1000, η×10⁻⁴ = 1, B = 512 | FID: 3.38 -> 4.41 | Hurt |
| reduce η×10⁻⁴ from 1 to 0.1 (ablation) | Beta Diffusion with KLUB · CIFAR-10, NFE = 1000, B = 512 | FID: 3.38 -> 4.90 | Hurt |
| reduce η×10⁻⁴ from 1 to 0.3 (ablation) | Beta Diffusion with KLUB · CIFAR-10, NFE = 1000, B = 512 | FID: 3.38 -> 3.67 | Hurt |
| reduce batch size B from 512 to 288 (ablation) | Beta Diffusion with KLUB · CIFAR-10, NFE = 1000, η×10⁻⁴ = 1 | FID: 3.38 -> 3.08 | Helped |
| reduce η×10⁻⁴ from 1 to 0.8 (ablation) | Beta Diffusion with KLUB · CIFAR-10, NFE = 1000, B = 288 | FID: 3.08 -> 3.21 | Helped |
| increase η×10⁻⁴ from 1 to 1.2 (ablation) | Beta Diffusion with KLUB · CIFAR-10, NFE = 1000, B = 288 | FID: 3.08 -> 3.37 | Hurt |
| reduce batch size B from 512 to 128 (ablation) | Beta Diffusion with KLUB · CIFAR-10, NFE = 1000, η×10⁻⁴ = 1 | FID: 3.38 -> 3.40 | Hurt |
| increase η×10⁻⁴ from 1 to 2 (ablation) | Beta Diffusion with KLUB · CIFAR-10, NFE = 1000, B = 128 | FID: 3.40 -> 3.75 | Hurt |
| increase NFE from 10 to 1000 (ablation) | Beta Diffusion with KLUB · CIFAR-10, η×10⁻⁴ = 1, B = 288 | FID: 37.67 -> 3.08 | Helped |
Closest papers by meaning
- Denoising Diffusion Bridge Models (ICLR 2024 · similarity 0.90)
- EVODiff: Entropy-aware Variance Optimized Diffusion Inference (NeurIPS 2025 · similarity 0.90)
- O(d/T) Convergence Theory for Diffusion Probabilistic Models under Minimal Assumptions (ICLR 2025 · similarity 0.89)
- Towards Non-Asymptotic Convergence for Diffusion-Based Generative Models (ICLR 2024 · similarity 0.89)
- Fast Solvers for Discrete Diffusion Models: Theory and Applications of High-Order Algorithms (NeurIPS 2025 · similarity 0.89)
- Adapting to Unknown Low-Dimensional Structures in Score-Based Diffusion Models (NeurIPS 2024 · similarity 0.89)
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.