On learning with imperfect representations | IEEE Conference Publication | IEEE Xplore