AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

Noise-corrected GRPO: From Noisy Rewards to Unbiased Gradients

Omar El mansouri, Fathinah Asma Izzati, Mohamed El Amine Seddik, Salem Lahlou

ICML 2026 · Reinforcement learning › RL from feedback · approach: GRPO stability and variants

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 noise correction to Dr.GRPO (synthetic noise 0.2,0.3)Dr.GRPO, Llama-3.2-1B · GSM8Kclean accuracy (%): 52.99 -> 68.51Improved
add noise correction to Dr.GRPO (synthetic noise 0.4,0.5)Dr.GRPO, Qwen3-0.6B · GSM8Kclean accuracy (%): 48.38 -> 56.34Improved
add noise correction under light noise (0.1,0.2)Dr.GRPO, Llama-3.2-1B · GSM8Kclean accuracy (%): 65.69 noiseless -> 71.04Improved
add noise correction across model scalesDr.GRPO, Qwen3-4B · MATH (synthetic noise 0.3,0.4)clean accuracy (%): 24.87 -> 31.58Improved
add noise correctionDr.GRPO, Llama-3.2-3B · MATH (synthetic noise 0.1,0.2)clean accuracy (%): 51.01 -> 51.70Improved
apply mean (Natarajan) correction only, without variance adjustment (ablation)GRPO, Qwen3-0.6B · GSM8K synthetic noise (0.2,0.1)clean accuracy (%): 75.73 -> 75.44No clear effect
apply mean correction only without variance adjustment (real reward model) (ablation)GRPO, Qwen3-0.6B · GSM8K (Llama3.1-8B-ORM-Mistral-Data)accuracy (%): 76.47 -> 72.82Hurt
add variance adjustment M=sqrt(Z) on top of mean correction (ablation)GRPO, Qwen3-0.6B · GSM8K (Llama3.1-8B-ORM-Mistral-Data)accuracy (%): 72.82 -> 79.12Improved
add noise correction with AceMath-7B-RM reward modelDr.GRPO, Llama-3.2-1B · GSM8Kaccuracy (%): 52.52 -> 59.21Improved
add noise correction with RM-Mistral-7B reward modelDr.GRPO, Qwen3-0.6B · APPSaccuracy (%): 9.01 -> 10.50Improved
add noise correction with RM-Mistral-7B reward modelDr.GRPO, Llama-3.2-1B · APPSaccuracy (%): 8.55 -> 8.68No clear effect
add gradient clipping instead of noise correction (ablation)GRPO-style training · noisy rewardsaccuracyNo clear effect

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