AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

Hista and Numca: Estimate State Value Effectively for Large Language Model Reinforcement Learning

Zizhe Chen, Jiqian Dong, Yizhou Tian, Garry YANG, Yongqiang Chen, Zhitang Chen, James Cheng

ICML 2026 · Reinforcement learning › RL from feedback · approach: RL for LLM reasoning

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
train a PPO critic instead of using group-average baseline (evaluated on unseen data, PPO-1)PPO · SVEB-NumberMAE: 0.164 -> 0.161No clear effect
train PPO critic for another epoch (PPO-N) (ablation)PPO · SVEB-NumberMAE: 0.164 -> 0.152Improved
replace group-average baseline with Hista hidden-state weighted estimateGRPO baseline · SVEB-MathMAE: 0.208 -> 0.145Improved
use Numca numerical milestones instead of group-averageGRPO baseline · SVEB-NumberMAE: 0.175 -> 0.132Improved
use Numca on non-numeric domainsGRPO baseline · SVEB-ScienceMAE: 0.215 -> 0.217Hurt
add Numca to GRPO trainingGRPO · Math benchmark averageaccuracy: 0.541 -> 0.555Improved
add Hista to DAPODAPO · Qwen2.5 base 0.5B, Simple Math Datasetaverage accuracy: 19.6 -> 21.2Improved
add Hista to DAPODAPO · Qwen2.5-14B base, Simple Math Datasetaverage accuracy: 57.2 -> 58.3Improved
add Hista to GSPOGSPO · Qwen2.5-1.5B base, Simple Math Datasetaverage accuracy: 37.9 -> 37.7No clear effect
add Hista to CSIPOCSIPO · Qwen2.5-1.5B base, Simple Math Datasetaverage accuracy: 37.8 -> 38.7Improved
add Hista to DAPODAPO · Hybrid Reasoning Dataset, Qwen2.5-7B instruct, ALL averageaverage accuracy: 55.2 -> 56.3Improved
add Hista to DAPODAPO · Hard Math Dataset, R1-Distill-1.5Baverage accuracy: 60.3 -> 61.4Improved

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