This article presents MOCO, a learnable meta optimizer for combinatorial optimization problems, which are often NP-hard. Traditional methods use handcrafted heuristics, but advances in neural networks have motivated the development of methods to learn heuristics from data. MOCO learns a graph neural network that updates the solution construction procedure based on features extracted from the current search state. It can adapt to varying circumstances such as different computational budgets. Tests on the Traveling Salesman Problem and Maximum Independent Set show MOCO outperforms other approaches on MIS and is competitive on the TSP.

 

Publication date: 8 Feb 2024
Project Page: https://github.com/TimD3/Mocotion
Paper: https://arxiv.org/pdf/2402.04915