AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

Entropy Regularizing Activation: Boosting Continuous Control, Large Language Models, and Image Classification with Activation as Entropy Constraints

Zilin Kang, Chonghua Liao, Tingqiang Xu, Huazhe Xu

ICLR 2026 · Reinforcement learning › RL from feedback · approach: GRPO stability and variants · 4 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
add ERA activation to GRPOGRPO on Qwen2.5-Math-7B · Math reasoning average (6 benchmarks)accuracy: 45.7 -> 50.2Improved
add ERA activation to GRPOGRPO on Qwen2.5-Math-7B · AIME25accuracy: 12.3 -> 16.9Improved
add ERA activation to GRPOGRPO on Qwen2.5-Math-7B · AIME24accuracy: 34.4 -> 37.5Improved
use ERA instead of KL-Cov entropy controlQwen2.5-Math-7B · Math reasoning average (6 benchmarks)accuracy: 46.6 -> 50.2Improved
add ERA activation to GRPOGRPO on Qwen2.5-Math-1.5B · Math reasoning average (6 benchmarks)accuracy: 29.8 -> 34.0Improved
add ERA activation to GSPOGSPO on Qwen2.5-Math-7B · Math reasoning average (6 benchmarks)accuracy: 43.7 -> 46.7Improved
add ERA activation to GSPOGSPO on Qwen2.5-Math-7B · MATH500accuracy: 85.1 -> 84.3Hurt
add ERA activation to ResNet-50 (no augmentation)ResNet-50 · ImageNettop-1 accuracy: 74.75 -> 75.44Improved
add ERA activation to ResNet-50 (with augmentation)ResNet-50 · ImageNettop-1 accuracy: 76.93 -> 77.30Improved
add ERA activation to ResNet-50 (with augmentation)ResNet-50 · CIFAR-10top-1 accuracy: 93.53 -> 93.93Improved
add ERA activation to ResNet-50 (no augmentation)ResNet-50 · CIFAR-10top-1 accuracy: 93.61 -> 93.82Improved
add ERA activation to GRPOGRPO on Qwen2.5-Math-7B · OOD benchmarks (ARC-C, GPQA-Diamond, MMLU-Pro)accuracy: +16.9% on averageImproved

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