Research map / Image generation / Fast diffusion sampling
Improved Denoising Diffusion Probabilistic Models
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 |
|---|---|---|---|
| increase diffusion steps from 1K to 4K (ablation) | DDPM with linear schedule and L_simple · ImageNet 64×64 | NLL: 3.99 -> 3.77 | Helped |
| use the hybrid objective instead of L_simple (ablation) | DDPM with 4K steps and linear schedule · ImageNet 64×64 | NLL: 3.77 -> 3.66 | Helped |
| use cosine instead of linear noise schedule (ablation) | DDPM with 4K steps and L_simple · ImageNet 64×64 | FID: 31.3 -> 27.0 | Helped |
| use cosine instead of linear noise schedule (ablation) | DDPM with 4K steps and L_simple · ImageNet 64×64 | NLL: 3.77 -> 3.68 | Hurt |
| optimize L_vlb instead of L_hybrid (ablation) | DDPM with 4K steps and cosine schedule · ImageNet 64×64 | NLL: 3.62 -> 3.57 | Helped |
| optimize L_vlb instead of L_hybrid (ablation) | DDPM with 4K steps and cosine schedule · ImageNet 64×64 | FID: 28.0 -> 56.7 | Hurt |
| increase diffusion steps from 1K to 4K (ablation) | DDPM with linear schedule and L_simple · CIFAR-10 | NLL: 3.73 -> 3.37 | Helped |
| use cosine instead of linear noise schedule (ablation) | DDPM with 4K steps and L_simple · CIFAR-10 | NLL: 3.37 -> 3.26 | Helped |
| use cosine instead of linear noise schedule (ablation) | DDPM with 4K steps and L_simple · CIFAR-10 | FID: 2.90 -> 3.05 | Hurt |
| use L_hybrid instead of L_simple (ablation) | DDPM with 4K steps and cosine schedule · CIFAR-10 | NLL: 3.26 -> 3.17 | Helped |
| optimize L_vlb instead of L_hybrid (ablation) | DDPM with 4K steps and cosine schedule · CIFAR-10 | NLL: 3.17 -> 2.94 | Helped |
| optimize L_vlb instead of L_hybrid (ablation) | DDPM with 4K steps and cosine schedule · CIFAR-10 | FID: 3.19 -> 11.47 | Hurt |
Closest papers by meaning
- UDPM: Upsampling Diffusion Probabilistic Models (NeurIPS 2024 · similarity 0.94)
- O(d/T) Convergence Theory for Diffusion Probabilistic Models under Minimal Assumptions (ICLR 2025 · similarity 0.92)
- Denoising Diffusion Step-aware Models (ICLR 2024 · similarity 0.92)
- Assessing the quality of denoising diffusion models in Wasserstein distance: noisy score and optimal bounds (NeurIPS 2025 · similarity 0.92)
- Learning to Discretize Denoising Diffusion ODEs (ICLR 2025 · similarity 0.92)
- On Inference Stability for Diffusion Models (AAAI 2024 · similarity 0.92)
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.