This article is devoted to learning functional networks. After a short intr
oduction and motivation of functional networks using a CAD problem, four st
eps used in learning functional networks are described: (1) selection of th
e initial topology of the network, which is derived from the physical prope
rties of the problem being modeled, (2) simplification of this topology, us
ing functional equations, (3) estimation of the parameters or weights, usin
g feast squares and minimax methods, and (4) selection of the subset of bas
ic functions lending to the best fit to the available data, using the minim
um-description-length principle. Several examples are presented to illustra
te the learning procedure, including the use of a separable functional netw
ork to recover the missing data of the significant wave height records in t
wo different locations, based on a complete record from a third location wh
ere the record is complete.