FormuEvo: LLM-Guided Evolution for Discovering Solver-Efficient Mixed-Integer Programming Formulations
FormuEvo reframes Mixed-Integer Programming (MIP) formulation design as an evolutionary optimization process using Large Language Models (LLMs). It iteratively generates and refines MIP formulations by analyzing solver performance statistics to guide improvements. This method discovers computationally efficient formulations that significantly outperform both expert designs and existing LLM approaches, achieving up to 5.5x speedup.