Interpolation results in the split inversion neural network algorithm

It is shown that the split inversion neural network algorithm can provide accurate interpolation for nonlinear situations that would be difficult to obtain with a look-up table. Neural networks trained with the algorithm are shown to provide complex interpolation when used as associative memories. T...

Full description

Saved in:
Bibliographic Details
Published inProceedings of the IEEE National Aerospace and Electronics Conference pp. 695 - 697 vol.2
Main Authors McAulay, A.D., Ravula, R.
Format Conference Proceeding
LanguageEnglish
Published IEEE 1989
Subjects
Associative memory
Backpropagation algorithms
Computer networks
Computer science
Design engineering
Equations
Intelligent networks
Interpolation
Neural networks
Table lookup
Online AccessGet full text

Cover

More Information
Summary:It is shown that the split inversion neural network algorithm can provide accurate interpolation for nonlinear situations that would be difficult to obtain with a look-up table. Neural networks trained with the algorithm are shown to provide complex interpolation when used as associative memories. The nonlinear relations need never be known; the network learns these from training with representative examples. The discretization of the parameters used in obtaining training samples and the nature of the underlying equations determine the accuracy of the interpolation. This approach has been shown to be valuable for engineering design where expertise is often dependent on association with previous designs that on following a set of rules. The neural network has been applied to the design of a beam-truss-spring structure.< >
DOI:10.1109/NAECON.1989.40286