Learning multi-agent state space representations