Projetando rede neurais RBF através de informações estruturais dos dados e cobertura de conjuntos

Abstract

The use of RBF (Radial Basis Function) in classification problems of samples has been a very discussed approach in the literature. Its diverse applications allow you to deal with problems of different complexities. One of the challenges in using this type of function is to find the appropriate parameters to avoid sub or oversize in the solution of the problem and in this context to difine radius normally given by the constant sigma is another issue. This work proposes a methodology based on well-known mathematical postulates to find the sigma, for a given set of samples, generated in an autonomous way after using the Gaussians in the samples in a similar way to the other known methods.