Search papers, labs, and topics across Lattice.
This paper introduces RefineEvo, an innovative evolutionary framework that enhances Automatic Heuristic Design (AHD) by integrating a dynamic Planner and a Reflector to optimize the use of historical search experience. By transitioning from static evolutionary operators to a planning-guided approach, RefineEvo adapts its search strategies in response to the evolving complexity of combinatorial optimization problems. Experimental results on classic benchmarks reveal that RefineEvo not only outperforms existing methods in solution quality but also improves token efficiency, paving the way for more autonomous heuristic design.
RefineEvo transforms heuristic design from static trial-and-error into a dynamic, experience-driven process that significantly boosts solution quality and efficiency.
Automatic Heuristic Design (AHD) has emerged as a transformative approach for solving combinatorial optimization problems. While recent Large Language Model (LLM)-based methods have shown promise, they predominantly rely on fixed evolutionary operators and struggle to effectively accumulate and reuse historical search experience. This paper proposes RefineEvo, a novel evolutionary framework that transforms AHD from a static trial-and-error process into a planning-guided, experience-driven system. RefineEvo introduces a Planner to dynamically schedule evolutionary operators and trigger refinement based on the current search state, and a Reflector to distill valuable lessons into a Bidirectional Experience Pool containing both positive insights and negative pitfalls. This synergistic framework enables the system to adapt its search tools to the evolving complexity of the problem and leverage trajectory-aware, situation-conditioned insights to guide generation. Experiments on several classic combinatorial optimization benchmarks demonstrate that RefineEvo consistently outperforms strong baselines. In particular, RefineEvo delivers superior solution quality while improving token efficiency, enabling more efficient and autonomous heuristic design.
