AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

Reward and Guidance through Rubrics: Promoting Exploration to Improve Multi-Domain Reasoning

Baolong Bi, Shenghua Liu, Yiwei Wang, Siqian Tong, Lingrui Mei, Yuyao Ge, Yilong Xu, Jiafeng Guo, Xueqi Cheng

ICML 2026 · Reinforcement learning › RL from feedback · approach: Scaling RL with verifiable rewards · 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
use rubric reward (factual + process) instead of outcome-only rewardGRPO · multi-domain average, Qwen2.5-7Baverage accuracy (%): 52.3 -> 53.7Improved
use Likert reward instead of outcome-only rewardGRPO · multi-domain average, Qwen2.5-7Baverage accuracy (%): 52.3 -> 52.7Improved
add rubric-guided off-policy refinement to Rubric-GRPORubric-GRPO · multi-domain average, Qwen2.5-7Baverage accuracy (%): 53.7 -> 55.8Improved
add rubric-guided off-policy refinement to Rubric-GRPORubric-GRPO · multi-domain average, Qwen2.5-3Baverage accuracy (%): 43.6 -> 45.5Improved
add Exploration Assessment (EA) (ablation)RGR-GRPO · multi-domain average, Qwen2.5-7Baverage accuracy (%): 53.8 -> 55.8Improved
add off-policy shaping function (ablation)RGR-GRPO · multi-domain average, Qwen2.5-7Baverage accuracy (%): 54.5 -> 55.8Improved
add process criteria to factual criteria in rubrics (ablation)RGR-GRPO · multi-domain average, Qwen2.5-7Baverage accuracy (%): 55.2 -> 55.8Improved
add process criteria to factual criteria in on-policy rubric reward (ablation)Rubric-GRPO · multi-domain average, Qwen2.5-7Baverage accuracy (%): 53.5 -> 53.7No clear effect
use off-policy baselines Critique-GRPO or LUFFY instead of on-policy GRPOGRPO · multi-domain average, Qwen2.5-7Baverage accuracy (%): 52.3 -> 50.6 (Critique), 50.5 (LUFFY)Hurt
add rubric-guided self-refinement at test time (ablation)Outcome-GRPO checkpoints · WebInstruct-Verify 4K held-outoutcome score: gain of +16.9 to +18.5 across stagesImproved
add rubric-guided off-policy guidance for pass@kRGR-GRPO · SciBench Math pass@k (Qwen2.5-7B)pass@kImproved

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