no code implementations • ICCV 2015 • Borislav Antic, Bjorn Ommer
Nevertheless, the common approach to this problem in feature selection and kernel methods is to down-weight or eliminate entire training samples or the same dimensions of all samples.