A Smoothing Function for 1-norm Support Vector Machines
Ruopeng Wang, Hongmin Xu · 2009
In this paper, a novel smoothing function method for the 1-norm support vector regression (SVR for short) is proposed and an attempt to overcome some drawbacks of former method which are complex, subtle, and sometimes difficult to implement. The model of smoothing support vector machine (SVM) based on 1-norm is provided from the optimization problem, yet it is discrete programming. With the smoothing technique and optimality knowledge, the discrete programming is changed into a continuous programming. Experimental results show that the algorithm is easy to implement and this method is fast and insensitive to initial point. Theory analysis illustrate that smoothing function method for 1-norm SVM are feasible and effective.