AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

Alternating Reinforcement Learning for Rubric-Based Reward Modeling in Non-Verifiable LLM Post-Training

Ran Xu, Tianci Liu, Zihan Dong, Tony Yu, Ilgee Hong, Carl Yang, Linjun Zhang, Tuo Zhao, Haoyu Wang

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
update the judge first, then the rubric generator (ablation)Rubric-ARM · Reward-modeling benchmarks (avg of 9)average accuracy: 72.4 -> 74.8Improved
update judge first (voting@5) (ablation)Rubric-ARM · Reward-modeling benchmarks (avg of 9)average accuracy: 74.9 -> 76.2Improved
add format reward (ablation)Rubric-ARM · Reward-modeling benchmarks (avg of 9)average accuracy: 72.6 -> 74.8Improved
add format reward (voting@5) (ablation)Rubric-ARM · Reward-modeling benchmarks (avg of 9)average accuracy: 75.5 -> 76.2Improved
add format reward on RewardBench2 Precise IF (ablation)Rubric-ARM · RewardBench2 Precise IFaccuracy: 25.6 -> 41.9Improved
add format reward on RewardBench Chat (ablation)Rubric-ARM · RewardBench Chataccuracy: 89.8 -> 89.4No clear effect
alternating RL on rubric generator and judge instead of SFT-only (Rubric-RM)Rubric-ARM · Reward-modeling benchmarks (avg of 9)average accuracy: 70.1 -> 74.8Improved
majority voting over 5 judgesRubric-ARM · Reward-modeling benchmarks (avg of 9)average accuracy: 74.8 -> 76.2Improved
use Rubric-ARM instead of Rubric-RM as DPO reward for Qwen2.5-7B-InstructRubric-ARM · IFEvalaverage: 79.5 -> 80.4Improved
iterative DPO over single-round DPO with Rubric-ARMRubric-ARM · Arena-Hard + AlpacaEvalaverage: 51.7 -> 53.4Improved
iterative DPO vs DPO with Rubric-RM on WildBenchRubric-RM · WildBenchWB Score: 53.6 -> 54.0Mixed
use Rubric-ARM instead of RM-R1 as GRPO rewardRubric-ARM · Online RL avgaverage: 52.3 -> 55.4Improved

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