2.8.4 Effect of Model Selection by Cross-Validation: In realizing high gener alization ability of a support vector machine, selection of kernels and their parameter values, i. e, model selection, is very important. Here I discuss how cross-validation. which is one of the most well-used model selection methods, works to generate a support vector machine with high general- ization ability 3.4 I have deleted the section"Sophisticated Architecture"because it does not work 4.3 Sparse Support Vector Machines: Based on the idea of the empirical fea- ture space, sparse support vector machines, which realize smaller numbers of support vectors than those of support vector machines are discussed 4 Performance Comparison of Different Classifiers: Performance of some types of support vector machines is compared using benchmark data sets 4.8 Learning Using Privileged Information: Incorporating prior knowledge into support vector machines is very useful in improving the generalization ability. Here, one such approach proposed by Vapnik is explained 4.9 Semi-supervised Learning: I have explained the difference between semi- supervised learning and transductive learning 4.10 Multiple Classifier Systems: Committee machines in the first edition are renamed and new materials are added 4. 11 Multiple Kernel Learning: A weighted sum of kernels with positiv weights is also a kernel and is called a multiple kernel. A learning method of support vector machines with multiple kernels is discussed 5.6 Steepest Ascent Methods and Newton's Methods: Steepest ascent meth ods in the first edition are renamed as Newtons methods and steepest ascent methods are explained in Section 5.6.1 5.7 Batch Training by Exact Incremental Training: A batch training method based on incremental training is added 5.8 Active Set Training in Primal and Dual: Training methods in the primal or dual form by variable-size chunking are added. 5.9 Training of Linear Programming Support Vector Machines: Three de- composition techniques for linear programming support vector machines are 6 Kernel-Based Methods: Chapter 8 Kernel-Based Methods in the first edi- tion is placed just after Chapter 5 Training methods and kernel discrimi- nant analysis is added 11.5.3 Active Set Training: Active set training discussed in Section 5.8 extended to function approximation. 7 Variable Selection: Variable selection for support vector regressors isvi Preface 2.8.4 Effect of Model Selection by Cross-Validation: In realizing high generalization ability of a support vector machine, selection of kernels and their parameter values, i.e., model selection, is very important. Here I discuss how cross-validation, which is one of the most well-used model selection methods, works to generate a support vector machine with high generalization ability. 3.4 I have deleted the section “Sophisticated Architecture” because it does not work. 4.3 Sparse Support Vector Machines: Based on the idea of the empirical feature space, sparse support vector machines, which realize smaller numbers of support vectors than those of support vector machines are discussed. 4.4 Performance Comparison of Different Classifiers: Performance of some types of support vector machines is compared using benchmark data sets. 4.8 Learning Using Privileged Information: Incorporating prior knowledge into support vector machines is very useful in improving the generalization ability. Here, one such approach proposed by Vapnik is explained. 4.9 Semi-supervised Learning: I have explained the difference between semisupervised learning and transductive learning. 4.10 Multiple Classifier Systems: Committee machines in the first edition are renamed and new materials are added. 4.11 Multiple Kernel Learning: A weighted sum of kernels with positive weights is also a kernel and is called a multiple kernel. A learning method of support vector machines with multiple kernels is discussed. 5.6 Steepest Ascent Methods and Newton’s Methods: Steepest ascent methods in the first edition are renamed as Newton’s methods and steepest ascent methods are explained in Section 5.6.1. 5.7 Batch Training by Exact Incremental Training: A batch training method based on incremental training is added. 5.8 Active Set Training in Primal and Dual: Training methods in the primal or dual form by variable-size chunking are added. 5.9 Training of Linear Programming Support Vector Machines: Three decomposition techniques for linear programming support vector machines are discussed. 6 Kernel-Based Methods: Chapter 8 Kernel-Based Methods in the first edition is placed just after Chapter 5 Training methods and kernel discriminant analysis is added. 11.5.3 Active Set Training: Active set training discussed in Section 5.8 is extended to function approximation. 11.7 Variable Selection: Variable selection for support vector regressors is added