📄 中文摘要
大型语言模型（LLMs）在组合优化（CO）中的自动启发式设计（AHD）方面取得了显著进展。然而，现有的发现流程往往需要大量的手动试错或依赖领域专业知识，以适应新的或复杂的问题。这种情况源于内部机制的紧密耦合，限制了基于LLM的设计过程的系统性改进。为了解决这一挑战，提出了一种结构化框架，明确将启发式发现过程分解为模块化阶段：前向评估、后向分析反馈和程序优化更新步骤。这种分离为迭代优化提供了清晰的抽象。
启发式搜索作为语言引导的程序优化
出处: Heuristic Search as Language-Guided Program Optimization
发布: 2026年2月19日
📄 English Summary
Heuristic Search as Language-Guided Program Optimization
Large Language Models (LLMs) have made significant advancements in Automated Heuristic Design (AHD) for combinatorial optimization (CO). However, existing discovery pipelines often require extensive manual trial-and-error or reliance on domain expertise to adapt to new or complex problems. This limitation arises from tightly coupled internal mechanisms that hinder systematic improvements in the LLM-driven design process. To address this issue, a structured framework is proposed that explicitly decomposes the heuristic discovery process into modular stages: a forward pass for evaluation, a backward pass for analytical feedback, and an update step for program refinement. This separation provides a clear abstraction for iterative refinement.