Learning with the Optimized Data-Dependent Kernel

This paper presents a method of optimizing a data-dependent kernel by maximizing a measure of class separability in the empirical feature space, an Euclidean space in which the training data are embedded in such a way that the geometrical structure of the data in the feature space is preserved. An effective algorithm is derived to perform the optimization. The optimized kernel show more adaptive to the data and leads to a substantial improvement in the performance of the kernel machines. Simulations are carried out to demonstrate this improvement.