Boosting Multiagent Reinforcement Learning via Permutation Invariant and Permutation Equivariant Networks.