Digital Signal Processing with Kernel Methods

10Support Vector Machine and Kernel Classification Algorithms

10.1 Introduction

This chapter introduces the basics of SVM and other kernel classifiers for pattern recognition and detection. We start by introducing the main elements and concept underlying the successful binary SVM. We analyze the issue of regularization and model sparsity. The latter serves to introduce the ν‐SVM, a version of the SVM that allows us to control capacity of the classifier easily. A recurrent topic in classification settings is that of how to tackle problems involving several classes; for that we then introduce several available extensions to cope with multiclass and multilabel problems. Other kernel classifiers, such as the LSs SVM and the kernel Fisher’s discriminant (KFD) analysis, are also summarized and compared experimentally in this chapter.

After this first section aimed at introducing the basic classification structures, we introduce more advanced topics in SVM for classification, including large margin filtering (LMF), SSL, active learning, and large‐scale classification using SVMs. Section 10.4 has to do with large‐scale implementations of SVMs, either involving new efficient algorithmic extensions or parallelization techniques.

10.2 Support Vector Machine and Kernel Classifiers

10.2.1 Support Vector Machines

This section describes the perhaps more useful and impactful machine‐learning classifier presented in recent decades: the SVM (Boser et al. 1992; Cortes and Vapnik 1995; Gu ...