Decomposition Technique For Support Vector Machines

 

This software implents a binary support vector machine classifier [1,2].

A decomposition technique [3] is used to split the whole quadratic programming (QP) problem into a sequence of smaller QP sub-problems; the dimension NC of the subproblems can be chosen by the user and range from 10 to 300. To solve the quadratic programming (QP) sub-problem a non-monotone projected gradient is used [4].

In this algorithm, a different SVM formulation is adopted [5]; the dual of this formulation becomes a simpler  bound-constrined problem.

A caching strategies is used to avoid the re-computation of the kernel function. The user can define the cache dimension in MB.

 

References:

1. Cristianini, N., Shawe-Taylor, J: "An Introduction to Support Vector Machines", Cambridge University Press, Cambridge, UK. 2000. http://www.support-vector.net

2. Kecman, V: "Learning and Soft Computing",  MIT Press, Cambridge, MA. 2001.

3. J.C. Platt: "A Fast Algorithm for Training  Support Vector Machines",  Advances in Kernel Methods - Support Vector Learning, MIT Press, 1998.

4. M. Antonelli, A. Rizzi: "A Non-Monotone Optimization Algorithm for IIR Filter Design", Univ. of Rome La Sapienza, Rome.

5. CHIH-WEI HSU CHIH-JEN LIN: "Simple Decomposition Method for Support Vector Machines", Machine Learning, 46, 291–314, 2002 Kluwer Academic Publishers.