Research map / Reinforcement learning / RL from feedback
Do Less, Achieve More: Do We Need Every-Step Optimization for RL Fine-tuning of Diffusion Models?
Renye Yan, Jikang Cheng, Shikun Sun, Yi Sun, You Wu, Wei Peng, Zongwei Wang, Ling Liang, Junliang Xing, Yimao Cai
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 RL scope selection (AdaScope) plug-in | DDPO · Simple animals prompts | time per batch (min): 4.55 -> 2.65 | Improved |
| add adaptive RL scope selection (AdaScope) plug-in | DPOK · Simple animals prompts | time per batch (min): 5.63 -> 3.71 | Improved |
| add adaptive RL scope selection (AdaScope) plug-in | D3PO · Simple animals prompts | time per batch (min): 5.06 -> 3.19 | Improved |
| add adaptive RL scope selection (AdaScope) plug-in | TDPO · Simple animals prompts | time per batch (min): 6.37 -> 4.17 | Improved |
| add AdaScope plug-in | DDPO · Simple animals prompts (PickScore reaches 22) | LPIPS: 0.624 -> 0.679 | Improved |
| add AdaScope plug-in | TDPO · Simple animals prompts (PickScore reaches 22) | ImageReward: 0.16 -> 0.67 | Improved |
| add AdaScope plug-in | DPOK · Simple animals prompts (Aesthetic reaches 7) | LPIPS: 0.623 -> 0.674 | Improved |
| add AdaScope plug-in | DPOK · Simple animals prompts (PickScore reaches 22) | AES: 6.14 -> 6.09 | Hurt |
| add AdaScope plug-in | D3PO · Simple animals prompts (Aesthetic reaches 7) | ImageReward: 0.89 -> 0.76 | Hurt |
| use prompt-adaptive scope instead of fixed start/end scope ranges (ablation) | AdaScope · Simple animals / SDv15 | reward optimization | Improved |
| use AdaScope on the larger SDXL backbone | AdaScope · SDXL | reward optimization | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DPOK — DPOK: Reinforcement Learning for Fine-tuning Text-to-Image Diffusion Models (NeurIPS 2023) · better on 5 of 5 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Towards Better Alignment: Training Diffusion Models with Reinforcement Learning Against Sparse Rewards (CVPR 2025 · similarity 0.94)
- Using Human Feedback to Fine-tune Diffusion Models without Any Reward Model (CVPR 2024 · similarity 0.92)
- Training Diffusion Models with Reinforcement Learning (ICLR 2024 · similarity 0.92)
- Training Diffusion Models Towards Diverse Image Generation with Reinforcement Learning (CVPR 2024 · similarity 0.91)
- Directly Fine-Tuning Diffusion Models on Differentiable Rewards (ICLR 2024 · similarity 0.91)
- Learning to Sample From Diffusion Models via Inverse Reinforcement Learning (NeurIPS 2026 · similarity 0.90)
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.
