next up previous contents
Next: Solving the Quadratic Programming Up: MPhil. in Computer Previous: SVM training and SRM

The Training of Support Vector Machines


In Chapter gif, the formulation of SVM concluded that SVM can be trained by solving Eqn. gif, subject to the constraints in Eqn. gif. This is standard constrained Quadratic Programming. In this chapter, the practical problems in solving this QP problem will be discussed. The effects of using different solvers are shown in Section gif. Section gif and gif show two major approaches that make the training of large classification problems tractable. In Section gif, the complexity and scalability of SVM training are discussed.

K.K. Chin
Thu Sep 10 11:05:30 BST 1998