Research map / Reinforcement learning / RL from feedback
Reinforcement Learning-driven Diffusion Models: Language, Vision, and Beyond
Zhiyuan Ma, Jiabao Wei, Tianyu Wang, Hanqing Wang, Jiadong Pan, Haowen Xu, Liangliang Zhao, Ruixun Liu, Yuhan Wang, Yuwei Zhang, Liang Feng, Junhao Xiao and 12 more
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 Flow-GRPO RL alignment | Stable Diffusion 3.5-M · GenEval | GenEval score: 0.63 -> 0.95 | Improved |
| apply Flow-GRPO RL alignment | Stable Diffusion 3.5-M · OCR score | OCR score: 0.59 -> 0.92 | Improved |
| apply DanceGRPO | Stable Diffusion v1.4 · HPS-v2.1 | HPS-v2.1: 0.239 -> 0.365 | Improved |
| use DanceGRPO instead of DDPO | Stable Diffusion v1.4 · HPS-v2.1 | HPS-v2.1: 0.297 -> 0.365 | Improved |
| apply DanceGRPO with joint rewards | Stable Diffusion v1.4 · Pick-a-Pic | Pick-a-Pic: 0.202 -> 0.215 | Improved |
| apply DanceGRPO | FLUX.1-dev · Pick-a-Pic | Pick-a-Pic: 0.224 -> 0.228 | No clear effect |
| apply DanceGRPO | FLUX.1-dev · GenEval | GenEval score: 0.659 -> 0.687 | Improved |
| apply DPOK | Stable Diffusion v1.5 · DrawBench ImageReward | ImageReward: 0.13 -> 0.58 | Improved |
| apply Qwen-Image-2.0-RL post-training | Qwen-Image-2.0-Base · T2I Arena | Elo: 1115 -> 1193 | Improved |
| apply RL post-training (TraDo-8B-Thinking) to dLLM | LLaDA-8B-Instruct · MATH500 | accuracy: 36.2 -> 87.4 | Improved |
| apply d1 RL to dLLM | LLaDA-8B-Instruct · HumanEval | accuracy: 49.4 -> 39.0 | Hurt |
| apply wd1 RL to dLLM | LLaDA-8B-Instruct · Sudoku | accuracy: 11.7 -> 76.4 | Improved |
Closest papers by meaning
- Designing Reinforcement Learning for Diffusion Models: A Unified Path-Space View (NeurIPS 2026 · similarity 0.92)
- Masked Diffusion Language Models are Strong and Steerable Text-Based World Models for Agentic RL (NeurIPS 2026 · similarity 0.90)
- Task-Agnostic Pre-training and Task-Guided Fine-tuning for Versatile Diffusion Planner (ICML 2025 · similarity 0.90)
- Diffusion Guidance Is a Controllable Policy Improvement Operator (NeurIPS 2026 · similarity 0.90)
- Score as Action: Fine Tuning Diffusion Generative Models by Continuous-time Reinforcement Learning (ICML 2025 · similarity 0.90)
- Diffusion Spectral Representation for Reinforcement Learning (NeurIPS 2024 · 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.
