Learning Implicit Credit Assignment for Cooperative Multi-Agent Reinforcement Learning