Reinforcement learning for combinatorial optimization: A survey

Many traditional algorithms for solving combinatorial optimization problems involve using hand-crafted heuristics that sequentially construct a solution. Such are designed by domain experts and may often be suboptimal due to the hard nature of problems. Reinforcement learning (RL) proposes good alternative automate search these training an agent in supervised or self-supervised manner. In this survey, we explore recent advancements applying RL frameworks Our survey provides necessary background operations research machine communities showcases works moving field forward. We juxtapose recently proposed methods, laying out timeline improvements each problem, as well make comparison with algorithms, indicating models can become promising direction