Research map / Reinforcement learning / Offline RL
Training-Free Adaptation of Diffusion Models via Doob's $h$-Transform
Qijie Zhu, Zeqi Ye, Han Liu, Zhaoran Wang, Minshuo Chen
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 |
|---|---|---|---|
| apply DOIT-Practical correction to diffusion sampling | Stable Diffusion v1.5 · LAION aesthetic score | mean aesthetic score: 6.372 -> 6.726 | Improved |
| replace full-rollout DOIT-Proto with one-step surrogate (ablation) | DOIT · LAION aesthetic score | mean aesthetic score: 6.714 -> 6.726 | No clear effect |
| replace full-rollout DOIT-Proto with one-step surrogate (ablation) | DOIT · LAION aesthetic score | runtime per image (s): 39.584 -> 1.712 | Improved |
| combine DOIT-Practical with Best-of-K | Stable Diffusion v1.5 · ImageReward K=4 | ImageReward: 0.702 -> 0.875 | Improved |
| combine DOIT-Practical with Best-of-K | Stable Diffusion v1.5 · ImageReward K=8 | ImageReward: 0.896 -> 1.001 | Improved |
| combine DOIT-Practical with BFS | Stable Diffusion v1.5 · ImageReward K=4 | ImageReward: 0.882 -> 0.950 | Improved |
| combine DOIT-Practical with BFS | Stable Diffusion v1.5 · ImageReward K=8 | ImageReward: 1.087 -> 1.115 | Improved |
| use DOIT-Practical instead of TTS as inference-time method | D4RL diffusion policies · D4RL locomotion average | normalized score: 86.1 -> 87.6 | Improved |
| use DOIT-Practical instead of TTS | D4RL diffusion policies · Medium-Replay HalfCheetah | normalized score: 47.8 -> 46.9 | Hurt |
| use DOIT-Practical instead of TTS | D4RL diffusion policies · Medium-Replay Walker2d | normalized score: 79.3 -> 82.0 | Improved |
| lower temperature tau requires lower correction strength gamma (ablation) | DOIT-Practical · LAION aesthetic score | aesthetic score stability | Mixed |
Closest papers by meaning
- D-DOIT: Training-free Adaptation of Discrete Diffusion via Doob's h-Transform (NeurIPS 2026 · similarity 0.92)
- DEFT: Efficient Fine-tuning of Diffusion Models by Learning the Generalised $h$-transform (NeurIPS 2024 · similarity 0.90)
- Hierarchical Variational Policies for Reward-Guided Diffusion (NeurIPS 2026 · similarity 0.90)
- Test-time Alignment of Diffusion Models without Reward Over-optimization (ICLR 2025 · similarity 0.89)
- Adaptive Non-Uniform Timestep Sampling for Accelerating Diffusion Model Training (CVPR 2025 · similarity 0.88)
- Elucidating the Exposure Bias in Diffusion Models (ICLR 2024 · 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.
