A local–global mixed kernel with reproducing property

摘要: A wide variety of kernel-based methods have been developed with great successes in many fields, but very little research has focused on the reproducing kernel function Reproducing Kernel Hilbert Space (RKHS). In this paper, we propose a novel method which call local-global mixed property (LGMKRP) to successfully perform range classification tasks RKHS rather than more conventionally used space. The LGMKRP proposed paper consists two major components. First, find basic solution generalized differential operator by delta function, and prove that is new specific called local H-reproducing (LHRK) RKHS. This good properties, including odd order vanishing moment, fast dilation attenuation. Second, RKHS, LHRK satisfies condition Mercer's theorem, it typical polynomial global property, also possesses property. Furthermore, (i.e., LGMKRP) based these different properties. Experimental results demonstrate approximation regularization performance kernel, can enhance generalization ability methods. We operator, kernel.We Mercer property.We define named evaluate our standard UCI datasets.We effectiveness kernel.