Kernel methods for statistical learning

Kernel machines are a class of algorithms for extracting information from data in a non-parametric framework. The interest aroused by these methods stems first and foremost from their excellent performance, particularly on large-scale problems. This good resistance to load is due to the parsimony of the solution and the low complexity of its computation. The interest of kernel machines also lies in their flexibility and rigor, an approach that holds great potential. This dossier aims to introduce kernel machines by focusing on the most popular one, the wide margin separator (SVM), taking stock of the various facets of its use. Emphasis is placed on the practical considerations involved in implementing this type of method.