Multivariable Function Interpolation using a GMDH Network

Abstract

A Group Method of Data Handling (GMDH) Network has been designed to perform multidimensional approximation by using multiple networks. The network is designed to interpolate a given multivariable function using data sets reduced in size. The ability of the GMDH network to fit the error of other subnetworks without overfitting its' own error is examined. This paper will demonstrate the GMDH network's ability to approximate various single-valued multi variable functions using multiple networks.