Sparse Proximal SVM via Convex Minimization

摘要

We propose Sparse Proximal SVM (SPSVM), a binary classifier that determines the plane by solving a quadratic programming problem (QPP), which is by far smaller than in a standard SVM. The SPSVM, that suppresses input space features, is in the spirit of Proximal SVM. The method does not require the solution of a dual formation. This proposed method can handle classification problems with a large number of data points. For kernel-based classifiers, the method also reduces the number of kernel functions.