AI Paper Atlas

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

CVPR 2026 · Reinforcement learning › RL from feedback · approach: RL fine-tuning of generative models · 6 citations (OpenAlex)

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 changedWhereMeasuredEffect
increase group size from 2 to 4-8 (ablation)GDRO · OCR tasktraining stabilityImproved
increase group size from 4 to 6 or 8 (ablation)GDRO · OCR taskcorrected scoreNo clear effect
raise beta from 6 to 12 (ablation)GDRO · OCR taskcorrected scoreImproved
lower beta from 12 to 6 (ablation)GDRO · GenEval taskevaluation and corrected scoreImproved
use group-level rewards (k=6) instead of pairwise DPO (k=2) (ablation)GDRO · GenEval taskcorrected score: 0.4162 -> 0.5148Improved
use GDRO offline training instead of Flow-GRPO online RLFlow-GRPO · OCR taskcorrected score: 0.5482 -> 0.5701Improved
use GDRO instead of Flow-GRPO to reach equal rewardFlow-GRPO · OCR taskGPU hours: 59.73 -> 29.60Improved
use GDRO instead of Flow-GRPO to reach equal rewardFlow-GRPO · GenEval taskGPU hours: 250.27 -> 68.4Improved
train Flow-GRPO longer toward higher rewardFlow-GRPO · OCR taskcorrected score: 0.5482 -> 0.4810Hurt
add top-1 likelihood stabilization regularizer (ablation)GDRO · OCR taskimage qualityImproved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy