Talk: Efficient Deep Imitation and Reinforcement Learning in Multi-Agent Environments