Efficient Policy Transfer in Large-Scale Traffic Light Control via Multi-Agent Hierarchical Reinforcement Learning The paper is accepted by IEEE CASE 2023.