Approximation of inverse maps through RBF neural networks

A general framework to obtain approximated solutions to ill-posed inverse problems in terms of Radial Basis Function (RBF) neural networks is proposed. The possibility of implementing RBFs in hardware in a network fashion makes this approach particularly appealing for real time applications, when the solution is needed on line in order to react with certain actions. An applicative example is reported in the field of acoustic remote sensing, where Gaussian RBF networks are employed to estimate a set of geophysical parameters of the seafloor from the measurement of the acoustic field in the water column