QDEvo: A Multi-Objective Quality-Diversity Framework for Automated Heuristic Design

arXiv cs.CL · 18 hours ago

Researchers developed QDEvo, a framework combining large language models with evolutionary computation to automatically design heuristics for solving combinatorial optimization problems while maintaining semantic diversity. The method uses pre-trained code embeddings and hierarchical self-reflection, achieving improvements in Hypervolume and Inverted Generational Distance metrics across benchmarks and industrial applications. This approach enables discovery of multiple high-performing, computationally efficient heuristics rather than converging to a single solution type.