📄 中文摘要
随着大型语言模型（LLMs）的快速发展，基于LLM的启发式搜索方法在自动化算法生成方面展现出强大的能力。然而，它们的进化过程往往面临不稳定性和过早收敛的问题。现有方法主要通过提示工程或共同进化思维与代码来解决这一问题，但在维护进化稳定性方面，算法类别多样性的重要性却被忽视。为此，提出了基于类别驱动的大型语言模型自动算法设计（CDEoH），该方法明确建模算法类别，并在种群管理中平衡性能与类别多样性，从而实现并行探索。
CDEoH: 基于类别驱动的大型语言模型自动算法设计
📄 English Summary
CDEoH: Category-Driven Automatic Algorithm Design With Large Language Models
The rapid advancement of large language models (LLMs) has led to the emergence of LLM-based heuristic search methods that exhibit strong capabilities in automated algorithm generation. However, these evolutionary processes often suffer from instability and premature convergence. Existing approaches primarily address these challenges through prompt engineering or by jointly evolving thought and code, largely overlooking the critical role of algorithmic category diversity in maintaining evolutionary stability. To address this gap, Category-Driven Automatic Algorithm Design with Large Language Models (CDEoH) is proposed, which explicitly models algorithm categories and balances performance with category diversity in population management, enabling parallel exploration.