An Active Set Method for Solving Certain Support Vector Machine Problems
Amal Al-Saket, Duaa Arman · 2019 IEEE Jordan International Joint Conference on Electrical Engineering and Information Technology (JEEIT) · 2019
We propose an active set method to solve the dual of the convex quadratic programming problem which is the core of the support vector machine (SVM) training. The method stems from the more general method developed by the first author. By using the special handling of certain quadratic programming problems where the Hessian matrix in the objective function is given as a product of a matrix and its transpose, and by simplifying the solution of the linear system arising at each iteration of the method, we were able to produce an implementation for certain SVMs. The results of an experiment using MATLAB are reported.