Actor-Attention-Critic for Multi-Agent Reinforcement Learning https://arxiv.org/abs/1810.02912