Kernel Methods for Pattern Analysis Illustrated Edition

I never write reviews, but I was so impressed with this book that I couldn't keep quiet. It is a practical book with code examples so that you can really see what he is talking about. This is book is just at my level of math (being able to read matrix equation without counting fingers and toes). Some Hilbert spaces but not more theory than you need. The book separates the design of kernels from the design algorithms that use kernels (nice touch). I would have liked more practical examples or a game plan for designing kernels for specific tasks.I also have the book Learning with Kernels, (Scholkopf and Smola) but I found it harder to follow and fragmented in their presentation.

I am a graduate student who studies perceptual systems. My research interests are neuroscience, vision, statistics, classification, and machine learning.I recently discovered that kernel methods are valuable tools for solving classification problems in a nearly optimal way. Apparently they are also useful for regression.This is the third textbook that I purchased for the purpose of understanding kernel methods. I have scarcely encountered a more elegantly written text. It does a superb job of building intuition and is also mathematically rigorous. Such texts are rare.This is the first textbook that I rely on when it comes to kernel methods.

Well, at first I was petrified to find a book that sounded like it deeply explores the subject of kernel methods. But all in all, it did not quite achieve what I hoped for. As a practical approach, when it comes to implementation, it serves nicely as a reference. The deeper mathematical roots of kernels (especially when it comes to measure theory and functional analysis) are not dealt with at all or just scratched at the very surface. The notation is sometimes awkward, mentioning for example the representation of an object in a given vector space with respect to the basis. And: Too much copied and pasted from the former book about SVMs. Basically, reading papers of Carmeli, Aronszajn and others will give you a much deeper insight into the subject.

Satisfied

This work presents a coherent overview of an important field in machine learning. The unifying framework of kernel methods has proven state of the art results and the community has been waiting for a book like this to make both theory and practice of kernel methods accesssible for readers of all different backgrounds (researchers, students, practioners from both academia and industry, ...).It is theoretically well-founded, the resulting algorithms are well-explained and made accessible for practioners by providing pseudo-code and online, ready-to-use matlab code.This book nicely complements the previous, yellow book, written by the same authors. Indeed, after "getting into the field" by reading the accessible introduction to support vector machines (SVMs), it was clear to me that SVMs was only an example of a signifcantly larger framework, i.e., kernel methods. The blue book is the reference book about that larger framework I have been waiting for since then. I particularly like the way the book is set up, making clear the modular, flexible approach in kernel methods.

Sloppy language, sloppy definitions, sloppy proofs.Constant repetitions do not add any clarity either.

This book will, without doubt, become THE reference work in kernel methods for pattern recognition, and a must read for pattern recognition researchers and practitioners in general.It's built up in a nicely modular, accessible and didactive way, helping the reader understand thoroughly what kernel methods are all about and importantly, how to use them. This makes the book very useful as a cook book for practitioners, as well as a text book for students.The book covers all the relevant topics in the state of the art of kernel methods, a field of research in which the authors have been a driving force since the beginning. Even so, they managed to resist the temptation from squeezing in the(ir) latest (potentially still unstable) results, which greatly enhances the timelessness of the book.

The book is divided into 3 parts. The theory is all in part I,the rest of the book is a cook-book with plenty of matlab code.The website contains most of the same code + data online. Readable, complete.

++