AI Paper Atlas

Research map / Reinforcement learning / Policy gradient and actor-critic

On Entropy Control in LLM-RL Algorithms

Han Shen

ICLR 2026 · Reinforcement learning › Policy gradient and actor-critic · approach: Regularized scalable deep value learning · 34 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 AEnt instead of GRPOQwen2.5-math-1.5b (MATH train) · AIME24test score: 0.192 -> 0.217Improved
use AEnt instead of GRPODeepSeek-R1-distilled-Qwen-1.5b (OpenR1-math train) · AMCtest score: 0.769 -> 0.825Improved
use AEnt instead of GRPOQwen2.5-math-1.5b (MATH train) · MATH-500test score: 0.756 -> 0.750Hurt
use AEnt instead of conventional entropy regularization (EntReg)DeepSeek-R1-distilled-Qwen-1.5b (OpenR1-math train) · MATH-500test score: 0.872 -> 0.882Improved
use EntReg instead of GRPOQwen2.5-math-1.5b (MATH train) · AIME24test score: 0.192 -> 0.167Hurt
use EntReg instead of GRPODeepSeek-R1-distilled-Qwen-1.5b (OpenR1-math train) · MATH-Hardtest score: 0.773 -> 0.808Improved
add adaptive entropy coefficient instead of constant coefficient (ablation)AEnt · test score and response lengthtest score: similar test score; more compact responsesNo clear effect
add adaptive entropy coefficient instead of constant coefficient (ablation)AEnt · policy entropy stabilityentropy / response lengthImproved
increase entropy clamping percentage p (ablation)AEnt · test scoretest score: performance drops beyond a pointMixed
vary clamping percentage p (ablation)AEnt · test score vs baselinestest scoreNo clear effect
use AEnt instead of GRPOQwen2.5-math-1.5b (MATH train) · training GPU hours to reach GRPO scoreGPU hours: 57 h -> 35 hImproved
use AEnt instead of GRPOQwen2.5-math-1.5b (MATH train) · training GPU hours to highest own scoreGPU hours: 57 h -> 69 hHurt

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