We have reformulated the optimization problem in the ν-K-SVCR method as an affine BVIP
with a positive semi-definite matrix and proposed a regularized version of the nonsmooth Newton
method for solving the resulting problems. This method fully exploits the special feature
of the problem and the numerical experiments have shown that this method is considerably
faster than the standard routine. In particular, it also has shown good performance on large
datasets for which the standard routine may not be suitable.