Learning using privileged information