Learning by networked agents under partial information