Aug
13
Generalized Representer Theorem
Scholkopf, Herbrich, Smola, “Generalized Representer Theorem”.
The theorem is quite general: for an arbitrary loss function L(.) that depends on data x_1,…,x_m, and regularizer in the form of g(||f||) with g(.) as strictly monotonic increasing, we have the usual expansion in the form of f(.) = \sum_{i=1}^m a_i k(.,x_i) .
Also the Remark 6, biased regularization, is quite interesting. I thought people had not considered it before!
