<ul><li>Mixed-Integer Linear Programming (MILP) is essential for solving complex decision-making problems and has led to various instance generation methods.</li><li>MILP instance generation methods have outpaced standardized evaluation techniques, leading to challenges in assessing synthetic MILP instances' fidelity and utility.</li><li>A new benchmark framework has been introduced to evaluate MILP instance generation methods systematically and objectively, focusing on mathematical validity, structural similarity, computational hardness, and utility in machine learning tasks.</li><li>The framework includes analysis of solver-internal features to compare solver outputs such as root node gap, heuristic success rates, and cut plane usage, aiming to encourage robust comparisons among different generation techniques and enhance the reliability of research using synthetic MILP data.</li></ul>

EVA-MILP: Towards Standardized Evaluation of MILP Instance Generation

Discover more