Entropy Regularizing Activation: Boosting Continuous Control, Large Language Models, and Image Classification with Activation as Entropy Constraints
Chonghua Liao
Abstract
ERA, a new paradigm using specially designed activations, enhances performance across LLMs, reinforcement learning, and image classification with minimal computational overhead.
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.
Community
🚀 Thrilled to introduce ERA: Entropy Regularizing Activation!
A new paradigm to control model entropy via output activations.
Delivering major performance boosts in:
🤖 LLM Reasoning
🤸 Continuous Control
🖼️ Image Classification
🤖For LLMs, ERA boosts Qwen2.5-Math-7B's AIME 2025 score by 37.4%! It mitigates entropy collapse, enhancing exploration and reasoning at different pass@k levels.
🤸In continuous control RL, ERA boosts performance of SAC, TD-MPC2, and PPO across various benchmarks.
🖼️ For image classification, ERA improves ResNet-50's ImageNet top-1 accuracy by 0.69%.
The best part? ERA is:
🔌 Plug-and-play: potentially applicable to a wide range of algorithms.
✍️ Provably effective: entropy bounds backed by theoretical analysis.
This is an automated message from the Librarian Bot. I found the following papers similar to this paper.
The following papers were recommended by the Semantic Scholar API
- ASPO: Asymmetric Importance Sampling Policy Optimization (2025)
- Asymmetric Proximal Policy Optimization: mini-critics boost LLM reasoning (2025)
- Risk-Sensitive RL for Alleviating Exploration Dilemmas in Large Language Models (2025)
- WeFT: Weighted Entropy-driven Fine-Tuning for dLLMs (2025)
- GTA: Supervised-Guided Reinforcement Learning for Text Classification with Large Language Models (2025)
- One-Token Rollout: Guiding Supervised Fine-Tuning of LLMs with Policy Gradient (2025)
- VCRL: Variance-based Curriculum Reinforcement Learning for Large Language Models (2025)
Please give a thumbs up to this comment if you found it helpful!
If you want recommendations for any Paper on Hugging Face checkout this Space
You can directly ask Librarian Bot for paper recommendations by tagging it in a comment:
@librarian-bot
recommend
Models citing this paper 0
No model linking this paper
Datasets citing this paper 0
No dataset linking this paper
Spaces citing this paper 0
No Space linking this paper
Collections including this paper 0
No Collection including this paper