Implementace a testování SVM

Abstract

This bachelor thesis helps to understand problems related with the SVM, definition of primary optimalization task, and explain how it convert to convex dual optimalization task of quadratic programming. The thesis shows algorithms based on conjugate gradients method to solve convex dual problem and describes how to implement extensions like Soft margin and different kernel function. Due to this methods the SVM can classify non-linear datasets or classify error loaded datasets which can't be successfully separated by any function.