How similar is this to multi-task learning?