An approximation theory approach to learning with \(\ell^1\) regularization (Q1944318)

Kernel based regularization schemes with an \(l^1\)-regularizer and a general loss function are studied. Assuming that the input data space \(X\) satisfies an interior cone condition and some regularity conditions on the marginal distribution and target functions as well as an appropriate smoothness condition on the kernel, an error analysis is carried out by means of a local polynomial reproduction formula of approximation theory. An error bound is proved and learning rates are improved which are independent of the dimension of \(X\).