Pattern Recognition, 4th Edition

4.19. Beyond the SVM Paradigm

One of the most attractive properties of the support vector machines, which has contributed to their popularity, is that their computational structure allows for the use of a kernel function, as discussed in the previous section. Sometimes this is also known as the kernel trick. This powerful tool makes the design of a linear classifier in the high-dimensional space independent of the dimensionality of this space. Moreover, due to the implicit nonlinear mapping, dictated by the adopted kernel function, the designed classifier is a nonlinear one in the original space. The success of the SVMs in practice has inspired a research effort to extend a number of linear classifiers to nonlinear ones, by embedding the kernel ...