Uncertainty propagation through radial basis function networks part II: classification networks

Radial basis function neural networks were trained using both partially supervised and fully supervised training techniques on a simple two-dimensional Gaussian data set. Forward uncertainty propagation through these networks was assessed using a technique of nested interval sets to form an information-gap model of classification performance of the networks. We demonstrate that the interval technique allows both the quantification of worst case and best case error performance of an individual network; and additionally provides an effective tool for optimal network selection in the presence of uncertainty.