2025SAT_Competition 新技术学习笔记

 

	A Self-Optimizing Framework for SAT Solvers via Population Evolution and Large Language Model Collaboration Hang Ding, Mao Luo, Chu-Min Li, Shunwei Li, Runyao Chen, Caiquan Xiong, and Xinyun Wu   Abstract—This paper proposes the AE framework, an automated optimization approach that integrates collective learning and genetic evolutionary principles. Unlike conventional practices relying on single models or manual tuning, AE autonomously enhances solver performance through large-model interaction,eliminating the need for extensive human annotation or complex training processes. 本文提出了一种AE框架，这是一种自动优化方法，集成了集体学习和遗传进化原理。与依赖单一模型或手动调整的传统做法不同，AE通过大规模模型交互自主提升求解器性能，消除了对大量人工标注或复杂训练过程的需求。     I. Introduction A. Hybrid of Experts (MoE) framework AE采用专家混合（MoE）框架，具体集成了DeepSeek-R1模型用于策略分析，Claude3.7模型用于代码实现，以及ChatGPT-4.5模型用于策略生成。   B. Collaborative Learning     C. Genetic Evolution   II. Main work Utilizing the aforementioned framework, we enhanced both the latest version of Kissat (version 4.0.2) and its modified variant incorporating the MAB strategy, thereby developing three improved SAT solving frameworks: AE kissat2025 bump, AE issat2025 rescale, and AE kissat2025 MAB.   A. AE kissat2025 bump The AE kissat2025 bump framework implements mod ifications to the kissat bump score increment() function. 通过多轮迭代框架优化，我们确定了一种有效的增强方法，该方法引入了动态混合分段策略与补偿因子机制相结合。   B. AE kissat2025 rescale The AE kissat2025 rescale framework implements mod ifications to the kissat rescale scores() . 通过多轮迭代框架优化，我们确定了一种有效的增强方法，该方法引入了三种不同的分数缩放策略，并结合了动态切换的伪随机选择机制，从而优化了分数调整过程。具体来说，基于当前增量（scinc）的伪随机数生成器确定策略选择：   这种混合策略动态适应不同的分数分布，同时保持计算效率。   C. AE kissat2025 MAB The AE kissat2025 MAB framework primarily modi f ies the restart mab() function. 通过多轮迭代框架优化，确定了一种有效的增强方法，引入了动态动量调控的自适应探测机组，构建了增益跟踪反馈系统。       References [1] M. S. Cherif, D. Habet, and C. Terrioux, ”Combining VSIDS and CHB Using Restarts in SAT,” in *Proc. 27th Int. Conf. Principles and Practice of Constraint Programming*, 2021. [2] Y. Sun, F. Ye, X. Zhang, S. Huang, B. Zhang, K. Wei, and S. Cai, “AutoSAT: Automatically Optimize SAT Solvers via Large Language Models,” arXiv preprint arXiv:2402.10705, 2024