A fuzzy linear interpolating network (FLIN) has been developed that is
a local processing neural network. Local processing advantageously fu
rnishes a traceable mechanism of inference and a bounty of diagonistic
information in the variable scores and observation loadings of the pr
ocessing units. FLIN is a two layer network for which the first layer
accomplishes data driven model selection and the second layer provides
linear predictive models. A new method of training is presented that
enhances the relations between unsupervised and supervised layers of t
his network. The advantages of FLIN are demonstrated with a spectropho
tometic titration of litmus.