Support Vector Machines: Kernel SVM

This lecture covers the concept of non-linear SVM using kernels, where linear functions may not be suitable for data separation. It introduces the idea of transforming data into a higher-dimensional space for linear separability, defining a redescription space for linear separation. The lecture explains the general case of redescription spaces in Hilbert spaces and the importance of completeness and separability. It delves into the optimization problem of training SVM in the redescription space and the use of kernels to avoid explicit knowledge of the transformation. The lecture also discusses the definition and properties of kernels, including quadratic, polynomial, and radial basis function kernels.