Research map / Image generation / Fast diffusion sampling
Adaptive Moments are Surprisingly Effective for Plug-and-Play Diffusion Sampling
Christian Belardi, Justin Lovelace, Kilian Q Weinberger, Carla P Gomes
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 |
|---|---|---|---|
| add adaptive moment estimation (Adam) to guidance (ablation) | DPS · ImageNet SR 16x | LPIPS: 0.30 -> 0.27 | Improved |
| add adaptive moment estimation (Adam) to guidance (ablation) | DPS · ImageNet inpainting 90% mask | LPIPS: 0.16 -> 0.12 | Improved |
| add adaptive moment estimation (Adam) to guidance (ablation) | DPS · CIFAR-10 class-conditional | accuracy (%): 42.8 -> 52.6 | Improved |
| add adaptive moment estimation (Adam) to guidance (ablation) | DPS · CIFAR-10 class-conditional | FID: 32.3 -> 58.0 | Hurt |
| add adaptive moment estimation (Adam) to guidance (ablation) | DPS · ImageNet class-conditional (standard classifier) | top-10 accuracy (%): 0.7 -> 10.5 | Improved |
| add adaptive moment estimation (Adam) to guidance (ablation) | CG · ImageNet time-aware classifier | accuracy (%): 61.0 -> 82.5 | Improved |
| add adaptive moment estimation (Adam) to guidance (ablation) | MPGD · ImageNet reconstruction | LPIPS: SR16x 0.35 -> 0.35; inpainting 0.45 -> 0.48 | No clear effect |
| add adaptive moment estimation (Adam) to guidance (ablation) | PiGDM · ImageNet inpainting 90% mask | LPIPS: 0.30 -> 0.11 | Improved |
| remove first moment (beta1=0) or second moment (beta2=0) (ablation) | AdamDPS · SR16x, deblur 12, inpainting 90% (Cats) | relative LPIPS improvement over DPS | Improved |
| increase task difficulty (SR from 4x to 16x) (ablation) | TFG vs DPS · ImageNet | LPIPS: TFG better than DPS at 4x (0.13 vs 0.19) but worse at 16x (0.32 vs 0.30) | Hurt |
| use TFG (recurrence, many components) instead of DPS under sparse conditioning (ablation) | DPS · ImageNet SR 16x | LPIPS: 0.30 -> 0.32 | Hurt |
| reduce sampling step budget (ablation) | AdamDPS vs DPS · ImageNet SR 16x (DDPM and DDIM) | relative LPIPS improvement over DPS | Improved |
Closest papers by meaning
- Multistep Distillation of Diffusion Models via Moment Matching (NeurIPS 2024 · similarity 0.90)
- A Simple Early Exiting Framework for Accelerated Sampling in Diffusion Models (ICML 2024 · similarity 0.89)
- CCS: Controllable and Constrained Sampling with Diffusion Models via Initial Noise Perturbation (NeurIPS 2025 · similarity 0.89)
- Diffusion Priors for Variational Likelihood Estimation and Image Denoising (NeurIPS 2024 · similarity 0.89)
- Diffusion Models With Learned Adaptive Noise (NeurIPS 2024 · similarity 0.89)
- Inductive Moment Matching (ICML 2025 · 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.
