The Scattered Data Interpolation Procedure (SDP) based on a modified Counterpropagation Neural Network (CPN) is presented. Matrix representation of SDIP facilitate the description of learning and working algorithms in comparison to a two-layered representation of CPN with competition and interpolation layers. The data matrix in SDP corresponds to the competition layer in CPN and the value matrix in SDIP corresponds to the interpolation layer in CPN. It is shown that during the working phase SDP, having an input vector X and output vector Y of size G, maps G scalar functions of vector argument Y_g = F_g(X), where g = 0,1, ..., G-1, rather than one vector function of vector argument Y = F(X). Two presented SDP learning algorithms allow compression of information stored in SDIP matrices during the learning phase, in dependence of a resolution parameter and thus allowing faster interpolation during the working phase. The results of a numerical example show that SDIP algorithms are superior to the weighted summation function used as an interpolation formula in CPN. The New Output Compression Algorithm (OCA), which evaluates the current efficiency of interpolation at each step of the learning process, is proposed.

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £52 +P&P)