Learning in RBF networks