<ul><li>Machine learning (ML) has shown potential in solving combinatorial optimization (CO) problems, but practical effectiveness on large-scale datasets remains uncertain.</li><li>The introduction of FrontierCO benchmark aims to address limitations by evaluating 16 ML-based solvers on eight CO problem types with challenging instances from real-world applications.</li><li>The benchmark includes graph neural networks and large language model (LLM) agents, providing insights into current ML methods' strengths and limitations for CO problems.</li><li>FrontierCO dataset is available for further research and to guide advancements in ML for combinatorial optimization. More information can be found at https://huggingface.co/datasets/CO-Bench/FrontierCO.</li></ul>

A Comprehensive Evaluation of Contemporary ML-Based Solvers for Combinatorial Optimization

Discover more