Research Of Support Vector Machine Learning Algorithms

Support Vector Machine (SVM) has some advantages, such as simple structure and good generalization, which is one implementation in Statistical Learning Theory. It has been applied to many pattern recognition and machine learning fields successfully and become a popular ongoing research area. However, its need for complexity of computation and storage requirements is the bottle-neck to deal with large-scale data. To overcome the disadvantages of SVM in training speed and precision, some researches are carried out aimed at fast learning and incremental learning based on SVM.The researches included in the thesis can be summarized as follows:The basic concepts of Statistical Learning Theory and SVM are summarized firstly. Some algorithms of SVM are described and compared, which are the groundwork of next research works.A heuristic fast learning algorithm of SVM is presented. Heuristic rules about how to select samples as support vector are presented according to geometric characters on the basis of SVM algorithms and related combined methods. By using the ideas of active learning, samples are trained with heuristic rules and classified with inner product matrix decomposing algorithm. Heuristic rules make proposed algorithm select the most advantageous samples to train and improve the training speed and accuracy, while the inner product matrix decomposing algorithm can improve the classifying speed without decreasing classifying accuracy.A gradual incremental SVM learning algorithm is presented. Some incremental learning methods and strategies are discussed. Advantages and disadvantages of several representative algorithms which are based on chunking method and KKT condition are analyzed. The incremental samples which are not satisfied KKT condition defined by initial classifier are selected to train in proposed gradual incremental learning algorithm. This decreases the training time...