Nonlinear Proximal Support Vector Machine Classifiers Aiming At Large Scale Classification Problems
Xiaoming Xu, Ning Ye, Qiaolin Ye · 2008
In [1], Fung et al, had constructed by a very fast algorithm: PSVM classifier, which mainly makes use of the Sherman-Morrison-Woodbury (SWM) identity [1,7,8].However, for one thing, when handling nonlinear problems, the matrix H in (1) always is of dimension m m , such that the SWM identity is of no use.For another, for large scale classification problems, its inversion is not feasible and it is not stored.Aiming at the orientation problems, proposed in this paper is new fast algorithm.Experimental results also show LPSVM is fast and feasible to solve large scale classification problems.