AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

Stabilizing Reinforcement Learning for Diffusion Language Models

Jianyuan Zhong, Kaibo Wang, Ding Ding, Zijin Feng, Haoli Bai, Yang Xiang, Jiacheng Sun, Qiang Xu

ICML 2026 · Reinforcement learning › RL from feedback · approach: GRPO stability and variants · 0 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
add unconditional clipping (vs self-normalization only) (ablation)StableDRL · Countdown/GSM8K training rewardtraining stabilityImproved
add self-normalization (vs clipping only) (ablation)StableDRL · Countdown/GSM8K training rewardtraining stabilityImproved
clipping threshold epsilon 5 instead of 1 (ablation)StableDRL · GSM8K training rewardconvergence and final rewardImproved
clipping threshold epsilon 100 or 1000 instead of 5 (ablation)StableDRL · GSM8K training rewardtraining stabilityHurt
replace SPG with StableDRL at length 256LLaDA-8B-Instruct · Countdownpass@1 accuracy: 70.7 -> 84.4Improved
replace SPG with StableDRL at length 256LLaDA-8B-Instruct · MATH500pass@1 accuracy: 40.0 -> 45.2Improved
replace ESPO with StableDRLLLaDA-8B-Instruct · GSM8K avgpass@1 accuracy: 82.0 -> 84.2Improved
replace ESPO with StableDRLLLaDA-8B-Instruct · Sudoku avgpass@1 accuracy: 86.0 -> 91.5Improved
replace SPG with StableDRLLLaDA-8B-Instruct · Sudoku avgpass@1 accuracy: 90.0 -> 91.5Improved
replace base SDAR-8B with StableDRL RL (static)SDAR-8B · AIME 24pass@1 accuracy: 11.8 -> 16.7Improved
replace Trado with StableDRL (static)SDAR-8B · AIME 24pass@1 accuracy: 13.3 -> 16.7Improved
replace Trado with StableDRL (dynamic)SDAR-8B · AIME 24pass@1 accuracy: 11.0 -> 13.3Improved

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