Research map / Reinforcement learning / RL from feedback
GDRO: Group-level Reward Post-training Suitable for Diffusion Models
Yiyang Wang, Xi Chen, Xiaogang Xu, Yu Liu, Hengshuang Zhao
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 |
|---|---|---|---|
| increase group size from 2 to 4-8 (ablation) | GDRO · OCR task | training stability | Improved |
| increase group size from 4 to 6 or 8 (ablation) | GDRO · OCR task | corrected score | No clear effect |
| raise beta from 6 to 12 (ablation) | GDRO · OCR task | corrected score | Improved |
| lower beta from 12 to 6 (ablation) | GDRO · GenEval task | evaluation and corrected score | Improved |
| use group-level rewards (k=6) instead of pairwise DPO (k=2) (ablation) | GDRO · GenEval task | corrected score: 0.4162 -> 0.5148 | Improved |
| use GDRO offline training instead of Flow-GRPO online RL | Flow-GRPO · OCR task | corrected score: 0.5482 -> 0.5701 | Improved |
| use GDRO instead of Flow-GRPO to reach equal reward | Flow-GRPO · OCR task | GPU hours: 59.73 -> 29.60 | Improved |
| use GDRO instead of Flow-GRPO to reach equal reward | Flow-GRPO · GenEval task | GPU hours: 250.27 -> 68.4 | Improved |
| train Flow-GRPO longer toward higher reward | Flow-GRPO · OCR task | corrected score: 0.5482 -> 0.4810 | Hurt |
| add top-1 likelihood stabilization regularizer (ablation) | GDRO · OCR task | image quality | Improved |
Closest papers by meaning
- GARDO: Reinforcing Diffusion Models without Reward Hacking (NeurIPS 2026 · similarity 0.89)
- Reward Score Matching: Unifying Reward-based Fine-tuning for Flow and Diffusion Models (NeurIPS 2026 · similarity 0.88)
- PRDP: Proximal Reward Difference Prediction for Large-Scale Reward Finetuning of Diffusion Models (CVPR 2024 · similarity 0.88)
- Flow-GRPO: Training Flow Matching Models via Online RL (NeurIPS 2025 · similarity 0.88)
- Reward Fine-Tuning Two-Step Diffusion Models via Learning Differentiable Latent-Space Surrogate Reward (CVPR 2025 · similarity 0.88)
- Towards Better Alignment: Training Diffusion Models with Reinforcement Learning Against Sparse Rewards (CVPR 2025 · 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.
