Selecting informative actions improves cooperative multiagent learning