Although support vector machines (SVMs) are a relatively new topic in the field of machine learning, there have been many papers and books published that report on the progress of both theoretical research and practical developments. This new book is a collection of papers that discuss recent advances and extensions of SVMs, together with a series of interesting and sound applications in a broad class of scientific domains.

Chapter 1 introduces the model augmented-SVM (A-SVM) in the framework of the theory of dynamical systems, as an extension of the standard SVM that resulted by combining gradient observations with observations taken on function values. The behavior of the proposed machine corresponds to a multi-stable dynamical system with attractors at the desired locations, each of the classified regions being invariant with respect to the learned dynamical system. Following the required theoretical arguments presented in the first sections, several applications are presented in the final part of the chapter.

A new extension of SVMs to the multi-class classification, referred to as simplified multi-class SVM (SimMSVM), is presented in chapter 2. Following a brief presentation of a series of already proposed SVM extensions to the multi-class classification problem, together with their classification as an indirect or direct approach, the model SimMSVM is introduced and extensively investigated. SimMSVM provides a consistent reduction of the dual variables of the quadratic problem corresponding to the design of the SVM, with numerous tests on real-world examples pointing out significant improvements from both points of view: a quicker training process and classification accuracy.

Chapter 3 is devoted to novel transfer approaches based on SVM, namely the knowledge-leverage-based TSK fuzzy system (KL-TSK-FS) for inductive transfer learning and the domain adaptation kernelized support vector machine (DAKSVM) for transductive transfer learning. The theoretical foundations of the KL-TSK-FS and DAKSVM models, together with their corresponding learning algorithms and computational complexity analyses, are presented in the first sections. In the final sections, an extensive comparative analysis is developed on an experimental basis against similar approaches, enabling interesting conclusions about the proposed models. In computer security applications, SVMs may be attacked through patterns that can evade detection (evasion), mislead the learning algorithm (poisoning), or gain information about the internal parameters or training data (privacy violation).

The main contributions of chapter 4 are the proposal of a formal general framework for the empirical evaluation of the security of machine-learning systems and demonstration of the feasibility of evasion, poisoning, and privacy attacks against SVMs in real-world security problems. The contributions can be summarized as follows: an evasion algorithm against SVMs with differentiable kernels, an algorithm that allows the adversary to find an attack pattern whose addition to the training sat maximally decreases the SVM’s classification accuracy, and an SVM learning algorithm that preserves differential privacy.

In chapter 5, the authors explore an application of SVMs to the bag-of-features model. Recently, the bag-of-features model achieved real success in a wide range of applications in the field of visual recognition. Given that the performance of SVM classifiers is strongly dependent on the particular types of kernels used, the developments presented in this chapter are focused on the use of representative kernels in cases of images represented by histograms and point sets.

Chapter 6 reports the results of using SVMs for specific tasks in neuroimage analysis. The main contributions include an SVM-based feature selection method to detect a discriminative subset of features that can optimize performance, interpretations obtained from SVM decision boundary that encode the class discrimination, and generative models for representation based on the parameters of the learned SVM.

The authors investigate the effects of imbalanced datasets on the performance of one-class and binary SVMs in chapter 7. The imbalanced data problem arises quite frequently in real-world problems when the majority class is compactly clustered and the minority class is scattered in the input space. The fundamentals of discriminative two-class SVMs and recognition-based one-class SVMs, and considerations concerning the effects of imbalanced training datasets on the performance of the resulted classifier, are briefly presented in the first sections of the chapter. Next, a hybrid kernel ensemble machine (HKME) whose architecture combines two different types of SVMs, a binary support vector classifier (BSVC) and a non-discriminative recognition-based model (vSVC), are introduced in order to compensate for the effects of imbalanced training data, together with a set of fusion rules for their integration. The performance of the HKME is evaluated in the final sections of the chapter, proving the benefit of using this ensemble model to handle imbalanced datasets. Finally, chapter 8 presents SVM applications to soft biometrics recognition in face images.

The book brings substantial contributions to the field of SVMs from both theoretical and practical points of view. The concepts and methods are presented in a clear and accessible way, and the illustrative examples and applications provide a valuable source of inspiration for similar developments. Each chapter provides comments and a list of references citing the most representative works related to the respective topics. This book is of considerable value to researchers in the fields of machine learning, data mining, and statistical pattern recognition.