Prototype Entropy Alignment: Reinforcing Structured Uncertainty in LLM Reasoning

Zhengyuan Pan; Yanhao Chen; Zhongquan Jian; Wanru Zhao; Haonan Ma; Meihong Wang; Qingqiang Wu

AAAI 2026technical0 citations

Prototype Entropy Alignment: Reinforcing Structured Uncertainty in LLM Reasoning

Zhengyuan Pan, Yanhao Chen, Zhongquan Jian, Wanru Zhao, Haonan Ma, Meihong Wang, Qingqiang Wu

Abstract

Recent research reveals that a minority of high-entropy tokens significantly influence the reasoning quality of large language models (LLMs). Inspired by this, we propose Prototype Entropy Alignment (PEA), a reinforcement learning framework that models effective reasoning not as a single path but as a collection of learnable "entropy signatures." PEA identifies these signatures by clustering expert trajectories

BibTeX

@inproceedings{aaai2026_prototypeentropy,
  title = {Prototype Entropy Alignment: Reinforcing Structured Uncertainty in LLM Reasoning},
  author = {Zhengyuan Pan and Yanhao Chen and Zhongquan Jian and Wanru Zhao and Haonan Ma and Meihong Wang and Qingqiang Wu},
  booktitle = {AAAI 2026},
  year = {2026}
}