arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2512.19081cs.AI
收藏

Population-Evolve: 一种用于LLM数学推理的并行采样和进化方法

Population-Evolve: a Parallel Sampling and Evolutionary Method for LLM Math Reasoning

Academy of Mathematics and Systems Science, Chinese Academy of Sciences(中国科学院数学与系统科学研究院);University of Chinese Academy of Sciences(中国科学院大学);Zhongguancun Academy(中关村学院);Zhongguancun Institute of Artificial Intelligence(中关村人工智能研究院)

Yanzhi Zhang, Yitong Duan, Zhaoxi Zhang, Jiyan He, Shuxin Zheng

AI总结:

Population-Evolve通过并行推理和进化提示提升LLM数学推理能力,实现高准确性和低计算成本。

摘要:

测试时扩展已成为增强大型语言模型推理能力的一个有前途的方向。在本工作中,我们提出Population-Evolve,一种受遗传算法启发的无训练方法,用于优化LLM推理。我们的方法通过并行推理为每个问题维护一个动态的候选解群体。通过引入一个进化提示,LLM在所有迭代中自我进化其群体。在收敛后,最终答案通过多数投票得出。此外,我们建立了一个统一框架,通过遗传算法的视角解释现有的测试时扩展策略。实验结果表明,Population-Evolve在低性能方差和计算效率下实现了更高的准确性。我们的发现突显了进化策略在推理过程中解锁LLM推理能力的潜力。

英文摘要:

Test-time scaling has emerged as a promising direction for enhancing the reasoning capabilities of Large Language Models in last few years. In this work, we propose Population-Evolve, a training-free method inspired by Genetic Algorithms to optimize LLM reasoning. Our approach maintains a dynamic population of candidate solutions for each problem via parallel reasoning. By incorporating an evolve prompt, the LLM self-evolves its population in all iterations. Upon convergence, the final answer is derived via majority voting. Furthermore, we establish a unification framework that interprets existing test-time scaling strategies through the lens of genetic algorithms. Empirical results demonstrate that Population-Evolve achieves superior accuracy with low performance variance and computational efficiency. Our findings highlight the potential of evolutionary strategies to unlock the reasoning power of LLMs during inference.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑