Robust training of radial basis networks

Robust training of radial-basis networks under non-normally distributed noise is considered. The simulation results show that multistep projection training algorithms minimizing various forms of module criteria are rather efficient in this case.