AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

Consolidating Reinforcement Learning for Multimodal Discrete Diffusion Models

Tianren Ma, Mu Zhang, Yibing Wang, Qixiang Ye

ICLR 2026 · Reinforcement learning › RL from feedback · approach: RL fine-tuning of generative models · 10 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
apply MaskGRPO instead of diffu-GRPOLLaDA-8B-Instruct · GSM8K (len 256)pass@1 accuracy: 79.8 -> 84.2Improved
apply MaskGRPO instead of TraceRLLLaDA-8B-Instruct · MATH500 (len 512)pass@1 accuracy: 39.5 -> 41.5Improved
apply MaskGRPO instead of UniGRPOLLaDA-8B-Instruct · MBPP (len 256)pass@1 accuracy: 43.1 -> 45.4Improved
apply MaskGRPO instead of UniGRPOMMaDA-8B-Base · GenEvaloverall score: 0.63 -> 0.80Improved
add SFT on BLIP3-o data before MaskGRPOMMaDA-8B-Base · DPG-Benchoverall score: 75.81 -> 81.76Improved
add SFT on BLIP3-o data before MaskGRPOMMaDA-8B-Base · GenEvaloverall score: 0.80 -> 0.84Improved
use emerge sampler instead of MaskGIT sampler before RL (ablation)MMaDA-8B-Base · GenEval (before RL)overall score: 0.56 -> 0.51Hurt
use emerge sampler instead of MaskGIT sampler during RL (ablation)MMaDA-8B-Base · GenEval (after RL)overall score: 0.77 -> 0.84Improved
add MaskGRPO RL to MMaDAMMaDA-8B-Base · HPSv3 (DPG-Bench prompts)HPSv3: 8.81 -> 9.40Improved
truncate timestep range (gamma 0.6) for language tasks (ablation)MaskGRPO · GSM8Ktraining stabilityMixed
use high truncation gamma=0.8 instead of 0.4 for image RL (ablation)MaskGRPO · image generation (HPSv3 reward)training stabilityImproved
use small clip epsilon for visual tasks (ablation)MaskGRPO · image generation (HPSv3 reward)training dynamicsImproved

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