AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

ExGRPO: Learning to Reason from Prior Successes

Runzhe Zhan, Yafu Li, Zhi Wang, Xiaoye Qu, Dongrui Liu, Jing Shao, Derek F. Wong, Yu Cheng

ICLR 2026 · Reinforcement learning › RL from feedback · approach: Scaling RL with verifiable rewards

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 experience replay with ExGRPO to on-policy RLVROn-policy GRPO · Qwen2.5-Math-7B reasoning avgaccuracy (%): 45.3 -> 48.3Improved
add ExGRPO on a weak backboneOn-policy RLVR · Llama3.1-8Bin-dist avg: 3.4 -> 6.1Improved
add ExGRPO on instruct modelOn-policy RLVR · Qwen2.5-7B-Instructin-dist avg: 35.2 -> 39.3Improved
add ExGRPO on small modelOn-policy RLVR · Qwen2.5-Math-1.5Bin-dist avg: 30.0 -> 35.3Improved
use continual ExGRPO instead of continual on-policy from LUFFYLUFFY · in-dist avgaccuracy (%): 48.7 -> 51.4Improved
extend ExGRPO to continuous rewardsGRPO · WildChat-IF chat benchmarksavg score: 20.0 -> 21.7Improved
use medium-difficulty Gaussian sampling (sigma=1) instead of narrower sigma=0.5 (ablation)ExGRPO · OOD avgaccuracy (%): 54.8 -> 58.3Improved
use sigma=1 instead of wider sigma=1.5 (ablation)ExGRPO · in-dist avgaccuracy (%): 46.2 -> 48.4Improved
use replay ratio 75% instead of 50% (ablation)ExGRPO · training performanceaccuracyHurt
add heuristic question selection instead of random selection (ablation)ExGRPO · validation scorescoreImproved
add policy shaping (ablation)ExGRPO · validation scorescoreImproved
apply shaping alone in on-policy RLVR (ablation)On-policy RLVR · validation scorescoreNo 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