Multi-agent reinforcement learning by the actor-critic model with an attention interface.