AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

On the Design of KL-Regularized Policy Gradient Algorithms for LLM Reasoning

Yifan Zhang, Yifeng Liu, Huizhuo Yuan, Yang Yuan, Quanquan Gu, Andrew C Yao

ICLR 2026 · Reinforcement learning › RL from feedback · approach: RL for LLM reasoning · 30 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 RPG-UFKL instead of DAPODAPO · AIME24 (8K, last)Mean@32 accuracy: 0.5438 -> 0.5938Improved
use RPG-REINFORCE-URKL instead of DAPODAPO · AIME25 (8K, last)Mean@32 accuracy: 0.4469 -> 0.5073Improved
use RPG-REINFORCE-UFKL instead of DAPODAPO · AIME25 (8K, best)Mean@32 accuracy: 0.4740 -> 0.5031Improved
use RPG-UFKL instead of GRPOGRPO · AMC23 (8K, last)Mean@32 accuracy: 0.9109 -> 0.9492Improved
use RPG-UFKL instead of GRPOGRPO · AIME24 (8K, last)Mean@32 accuracy: 0.3750 -> 0.5938Improved
use reverse KL (URKL) instead of forward KL (UFKL) in fully differentiable RPG (ablation)RPG · AIME24 (8K, last)Mean@32 accuracy: 0.5938 -> 0.4542Hurt
use reverse KL (URKL) instead of forward KL (UFKL) in fully differentiable RPG (ablation)RPG · AIME25 (8K, last)Mean@32 accuracy: 0.4698 -> 0.5261Improved
use REINFORCE-style loss instead of fully differentiable loss (URKL) (ablation)RPG-URKL · AIME24 (8K, last)Mean@32 accuracy: 0.4542 -> 0.5708Improved
use REINFORCE-style loss instead of fully differentiable loss (UFKL) (ablation)RPG-UFKL · AIME24 (8K, last)Mean@32 accuracy: 0.5938 -> 0.5906No clear effect
change clip ratios from (0.1,0.1) to (0.2,0.28) (ablation)REINFORCE-style RPG · Qwen3-4B trainingMean@32 accuracyNo clear effect
use DAPO clip-higher (ablation)RPG-REINFORCE · Qwen3-4B trainingMean@32 accuracyMixed

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