LLM-Driven Co-Evolutionary Automated Heuristic Design for Bi-Component Coupled Combinatorial Optimization
Researchers introduce CoEvo-AHD, an LLM-driven framework that co-evolves paired operator populations to solve coupled combinatorial optimization problems like the Traveling Thief Problem. Unlike previous automated heuristic design methods that treat operators in isolation, this approach captures interactions between decision components, achieving competitive results with traditional heuristics.
CoEvo-AHD represents a meaningful advancement in how artificial intelligence can tackle complex optimization problems that contain tightly interdependent decision structures. Traditional automated heuristic design methods treat optimization operators independently, missing opportunities to exploit synergies between different problem components. This new framework recognizes that real-world problems often feature coupled subsystems where route planning and item selection, for example, directly influence each other's success.
The advancement builds on growing interest in leveraging LLMs for algorithm design beyond their traditional language processing capabilities. Previous work demonstrated that LLMs could generate competitive heuristics, but the single-operator approach proved limiting for problems requiring coordinated decision-making across multiple domains. CoEvo-AHD addresses this by introducing a dual-population co-evolutionary mechanism where route and selection operators evolve together, with evaluation specifically measuring their cooperative performance rather than individual merit.
The practical implementation includes a tool library that standardizes problem-specific operations into callable functions, allowing LLM-generated code to avoid inefficient custom loops while maintaining flexibility. This infrastructure approach makes the framework more scalable and reproducible compared to purely generative alternatives.
For the optimization and operations research community, this work demonstrates that LLM-guided design can discover non-obvious operator combinations that outperform manually-crafted heuristics. The methodology could extend beyond TTP and TPP to other coupled optimization domains in supply chain, manufacturing, and resource allocation. The competitive performance against established baselines suggests LLM-driven design is maturing into a viable alternative to traditional algorithm engineering.
- →LLM-driven co-evolution enables discovery of complementary heuristic pairs that work better together than in isolation
- →Tool-invocation libraries allow LLMs to leverage standardized problem operations, reducing implementation errors and improving efficiency
- →Framework achieves competitive solution quality on coupled optimization problems without hand-designed heuristics
- →Dual-population approach captures decision subsystem interactions that single-operator methods systematically miss
- →Methodology extends automated heuristic design beyond isolated operators to multi-component problems