The role of information structures in game-theoretic multi-agent learning