[2603.19284] CDEoH: Category-Driven Automatic Algorithm Design With Large Language Models

Computer Science > Neural and Evolutionary Computing arXiv:2603.19284 (cs) [Submitted on 8 Mar 2026] Title:CDEoH: Category-Driven Automatic Algorithm Design With Large Language Models Authors:Yu-Nian Wang, Shen-Huan Lyu, Ning Chen, Jia-Le Xu, Baoliu Ye, Qingfu Zhang View a PDF of the paper titled CDEoH: Category-Driven Automatic Algorithm Design With Large Language Models, by Yu-Nian Wang and 5 other authors View PDF HTML (experimental) Abstract:With the rapid advancement of large language models (LLMs), LLM-based heuristic search methods have demonstrated 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. To this end, we propose Category Driven Automatic Algorithm Design with Large Language Models (CDEoH), which explicitly models algorithm categories and jointly balances performance and category diversity in population management, enabling parallel exploration across multiple algorithmic paradigms. Extensive experiments on representative combinatorial optimization problems across multiple scales demonstrate that CDEoH effectively mitigates convergence toward a single evolutionary direction, significantly enhancing evolutionary stability and achievi...