Research map / Reinforcement learning / RL from feedback
Reward Sharpness-Aware Fine-Tuning for Diffusion Models
Kwanyoung Kim, Byeongsu Sim
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 RSA-FT (input and weight perturbation) to AlignProp | AlignProp (SD1.5) · HPD | HPSv2.1 / ImageReward: HPSv2.1 24.93 -> 32.02; ImageReward 0.032 -> 0.528 | Improved |
| add RSA-FT to Draft-LV | Draft-LV (SD1.5) · DrawBench | HPSv2.1 / PickScore: HPSv2.1 25.59 -> 26.67; PickScore 20.96 -> 21.12 | Improved |
| add RSA-FT to ReFL | ReFL (SD1.5) · HPD | HPSv2.1: 34.95 -> 35.81 | Improved |
| add RSA-FT to DRTune | DRTune (SD1.5) · HPD | PickScore: 21.91 -> 21.92 | No clear effect |
| add RSA-FT to ReFL on SDXL | ReFL (SDXL) · HPD | HPSv2.1: 31.21 -> 32.66 | Improved |
| add RSA-FT to DRTune on SDXL | DRTune (SDXL) · HPD | HPSv2.1: 32.36 -> 33.74 | Improved |
| add RSA-FT to ReFL on SD3 | ReFL (SD3) · DrawBench | HPSv2.1: 29.45 -> 29.81 | Improved |
| add RSA-FT on SD3 with DRTune | DRTune (SD3) · HPD | PickScore: 22.33 -> 22.41 | No clear effect |
| combine image-space and parameter-space flattening versus each alone (ablation) | RSA-FT · SD1.5 RDRL | reward metrics | Improved |
| higher reward sharpness during fine-tuning | DRaFT-K SD1.5 with HPSv2 · training dynamics | PickScore / ImageReward correlation: Pearson r = -0.802 (PickScore), -0.669 (ImageReward) | Hurt |
Closest papers by meaning
- Directly Fine-Tuning Diffusion Models on Differentiable Rewards (ICLR 2024 · similarity 0.92)
- Using Human Feedback to Fine-tune Diffusion Models without Any Reward Model (CVPR 2024 · similarity 0.91)
- Reward Score Matching: Unifying Reward-based Fine-tuning for Flow and Diffusion Models (NeurIPS 2026 · similarity 0.90)
- Confronting Reward Overoptimization for Diffusion Models: A Perspective of Inductive and Primacy Biases (ICML 2024 · similarity 0.89)
- PRDP: Proximal Reward Difference Prediction for Large-Scale Reward Finetuning of Diffusion Models (CVPR 2024 · similarity 0.89)
- Towards Better Alignment: Training Diffusion Models with Reinforcement Learning Against Sparse Rewards (CVPR 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.
