Learning from ambiguously labeled images | IEEE Conference Publication | IEEE Xplore