CDEoH: Category-Driven Automatic Algorithm Design With Large Language Models

ArXi:2603.19284v1 Announce Type: cross With the rapid advancement of large language models (LLMs), LLM-based heuristic search methods have nstrated strong capabilities in automated algorithm generation. However, their evolutionary processes often suffer from instability and premature convergence. Existing approaches mainly address this issue through prompt engineering or by jointly evolving thought and code, while largely overlooking the critical role of algorithmic category diversity in maintaining evolutionary stability.