An Introduction to Support Vector Machines and Other Kernel-based Learning Methods

This textbook does an excellent job developing the mathematics and providing intuition behind SVMs. That being said, This textbook assumes you have a firm grasp on vector calculus and linear algebra. Chapters 1 and 2 provide a good overview of how machine learning can be used in classification and linear regression. Chapter 5 is an overview of what is covered in most introductory multivariate calculus classes. However, chapter 3 and chapter 4 are not easy. If you lack a background in real analysis you can read the first four or five pages of each chapter and get a basic idea of what is being discussed. Some of the proofs could use a bit more explanation, but at the same time they are mathematically rigorous and concise.Lastly, the author makes mention that you can read the book out of order. Here's the order that I read: chapter 1, chapter 2, chapter 5, chapter 3, chapter 6, chapter 4, chapter 8, and then chapter 7. I think this order gave me enough background information and motivated me to delve into the mathematical theory. I just want finish up by saying that if you're looking for a support vector implementation guide this is certainly not for you.

This book introduces the concepts of kernel-based methods and focuses specifically on Support Vector Machines (SVM). It is hard to read and a good background in mathematic is clearly needed. The book has a strong emphasis on SVM starting from the very first line of text. Concepts are well explained, although equations are not clear. The notation doesn't facilitate the reading at all. The book covers linear as well as kernel learning. The kernel trick is well described. It is easy to understand ideas behind SVM while reading the corresponding chapter. Finally a small chapter on SVM applications is proposed. Unfortunately, it only contains typical SVM applications (i.e. standard problems).I think this book is good if you: * Have a strong mathematical background * Work in the specific domain of SVM (or kernel-based methods in general) * Want to write a research paper about SVM and need the correct notationsHowever, this book is NOT intended for people who: * Don't like to read theorems, corollaries and remarks * Are not interested in reading hundreds of proofsThis is my personal opinion as a computer scientist: this book is definitely written for mathematicians.

I wrote my review of this book on the ai forum.You can see my write up there at the link below:[...]I liked the book overall.

I'm currently an undergraduate in computer science and math at a Cal State university, and I found this book to be both complete and accessible. In six weeks of independent study, I was able to implement an SVM and recreate the chessboard example from his book, and now I find myself reading research-level papers on more advanced kernels and datasets.My only real criticism of the book is that the authors sometimes resort to wacky matrix notation to get mathematical expressions to fit inline, when they really should be centered and displayed in full.I think the best addition would be a prologue entitled "How To Read This Book," because it seems that other reviewers are dissatisfied by the lack of hand-holding required for someone with none of the (prerequisite) comfort level with linear algebra notation and methods of proof. That being said, almost all of the proofs contained in the book are there for completeness alone. One does not need to reprove each proposition in order to understand how to implement an SVM or grasp the concepts behind kernel-based methods. For instance, the KKT conditions are essential as a measure of convergence in training an SVM, but the derivation is superfluous for an engineer.As an analogy, just because a book on algorithms presents bubble sort doesn't make bubble sort important, but omitting bubble sort (or some other introductory sorting algorithm) would make an introductory volume incomplete. The authors here provide the same foundations for support vector machines, so that the reader can actually understand why it works. This book is self-contained, and it's much better for it.In my search for a good book on SVMs, this one was by far superlative. I would recommend it to anyone in my position who is interested in mathematics and programming.

We incorporated a Support Vector Machine Classifier in our analysis software product. Although other texts and articles provided friendlier background and an easier introduction, when the time came to actually code a classifier, this was the book that offered the level of detail required to build something that ran. The math is heavy, the prose is terse, but it goes deep under the covers of what actually constitutes a kernel transformation, what function families qualify as kernels, as well as deep component-by-component algorithms.The biggest drawback of this book is that it does not meet the needs of the many non-mathematically inclined who are interested in SVM's. It uses the academic euphemism 'introduction' to mean 'brutally advanced, but if I called it that, no one would buy it'. One of the reviewers was expecting an actual introduction, and was disappointed.

Not exactly a walk through the park ... but so far the most comprehensive and broad overview of the subject that I have read. The underlying mathematical concepts are many: linear systems, analysis, statistics, set theory, orthogonal functions and function transforms. I have used Wikipedia to brush up and read up on some of the material that is assumed to be known. Even without full mathematical rigor the text provides insights at a deeper level, and as such is quite useful in getting a feel for the possibilities and limitations of kernels and SVMs. Which after all was the purpose of the book.