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
10/9/2025
cs.LG

No official code found. Best implementation: nothingbutbut/era (17 stars, last commit 2025-11-26).

Abstract

We propose ERA, a new paradigm that constrains the sampling entropy above given thresholds by applying specially designed activations to the outputs of models. Our approach demonstrates broad effectiveness across different domains: 1) for large language models(LLMs), boosting the AIME 2025 score for Qwen2.5-Math-7B by 37.4%; 2) for continuous control reinforcement learning agents, improving performance by more than 30% over strong baselines such as SAC on the challenging HumanoidBench; 3) for image classification, enhancing ImageNet top-1 accuracy by 0.69% for ResNet-50. These gains are achieved with a computational overhead of less than 7%. Our work validates output activation as a powerful tool for entropy control, opening a new direction for designing simpler and more robust algorithms.

View on arXivView PDF

Code Implementations(1)

nothingbutbut/eraCommunity73%

An official code repo of paper Entropy Regularizing Activation: Boosting Continuous Control, Large Language Models, and Image Classification with Activation as Entropy Constraints.

174MDX, RoffOct 1, 202510 months ago

Similar Papers

Rethinking Entropy Regularization in Large Reasoning Models

86% match1 repo322025

SEE-DPO: Self Entropy Enhanced Direct Preference Optimization

86% match2024

Flexible Entropy Control in RLVR with Gradient-Preserving Perspective

86% match2026

On Entropy Control in LLM-RL Algorithms

85% match1 repo32025

Revisiting Entropy in Reinforcement Learning for Large Reasoning Models

85% match2025

Entropy-Preserving Reinforcement Learning

85% match2026

Cite this paper

@article{kang2025entropy,
  title  = {Entropy Regularizing Activation: Boosting Continuous Control, Large Language Models, and Image Classification with Activation as Entropy Constraints},
  author = {Zilin Kang and Chonghua Liao and Tingqiang Xu and Huazhe Xu},
  year   = {2025},
  eprint = {2510.08549},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2510.08549v2}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2510.08549.svg)](https://codeofpaper.com/paper/2510.08549)

Discussion