ABSTRACT
Historically, all the necessary elements that form the theory and algorithm of SVMs have been known since the early 1970s. But it took about 25 years before the concept of SVMs was developed and the spirit of SVMs was systematically elucidated in a formal way in the two fundamental monographs: Statistical Learning Theory and The Nature of Statistical Learning Theory [1, 2]. As pointed out in the two books, in contrast to traditional learning methods where dimension reduction is performed in order to control the generalization performance of the model, the SVMs dramatically increase the dimensionality of the data and then build an optimal separating hyperplane in the high-dimensional feature space relying on the so-called margin maximizing technique. It’s surprising but expected that very excellent performance is observed when SVMs are applied to practical problems such as handwriting recognition.
