A scalable dual approach to semidefinite metric learning | IEEE Conference Publication | IEEE Xplore