Research map / Image generation / Fast diffusion sampling
Improved Immiscible Diffusion: Accelerating Diffusion Training by Reducing Miscibility
Yiheng Li, Feng Liang, Dan Kondratyuk, Masayoshi TOMIZUKA, Kurt Keutzer, Chenfeng Xu
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 immiscible diffusion with KNN noise selection | DDIM · CIFAR-10 | FID: 6.63 -> 4.45 | Improved |
| add immiscible diffusion with KNN noise selection | Flow Matching · CIFAR-10 | FID: 4.92 -> 3.56 | Improved |
| add immiscible diffusion with linear assignment | Flow Matching · CIFAR-10 | FID: 4.92 -> 5.02 | No clear effect |
| add immiscible diffusion with linear assignment | Consistency Models · CIFAR-10 | FID: 10.22 -> 8.54 | Improved |
| add immiscible diffusion on top of other acceleration methods | Flow Matching · CIFAR-10 | FID at 200k: 5.38 -> 3.93 | Improved |
| use immiscible KNN for SD inpainting | Stable Diffusion · ImageNet inpainting | FID: 18.35 -> 17.32 | Improved |
| use immiscible KNN for SD outpainting | Stable Diffusion · ImageNet outpainting | FID: 29.34 -> 27.57 | Improved |
| use immiscible KNN in diffusion policy | Diffusion Policy · PushT | average coverage (%): 79.56% -> 82.83% | Improved |
| replace linear assignment with KNN | Immiscible diffusion · execution time batch 1024 | ms: 22.8 -> 0.7 | Improved |
| vary KNN k from 8 to 32 (ablation) | KNN immiscible DDIM · CIFAR-10 | best FID: 4.45 -> 5.19 | Hurt |
| add immiscible diffusion to class-conditional SD | Stable Diffusion · ImageNet-1k | CLIPScore: 28.55 vs 28.55 | No clear effect |
Closest papers by meaning
- Immiscible Diffusion: Accelerating Diffusion Training with Noise Assignment (NeurIPS 2024 · similarity 0.90)
- ALTER: All-in-One Layer Pruning and Temporal Expert Routing for Efficient Diffusion Generation (NeurIPS 2025 · similarity 0.88)
- High-Fidelity Diffusion-Based Image Editing (AAAI 2024 · similarity 0.87)
- Faster Diffusion: Rethinking the Role of the Encoder for Diffusion Model Inference (NeurIPS 2024 · similarity 0.87)
- Early-Bird Diffusion: Investigating and Leveraging Timestep-Aware Early-Bird Tickets in Diffusion Models for Efficient Training (CVPR 2025 · similarity 0.87)
- Training-Free Adaptive Diffusion with Bounded Difference Approximation Strategy (NeurIPS 2024 · similarity 0.87)
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.
