High accuracy data representation via sequence of neural networks

Sequence of neural networks has been applied to high accuracy regression in 3D as data representation in form z = f(x,y). The first term of this series of networks estimates the values of the dependent variable as it is usual, while the second term estimates the error of the first network, the third term estimates the error of the second network and so on. Assuming that the relative error of every network in this sequence is less than 100 %, the sum of the estimated values converges to the values to be estimated, therefore the estimation error can be reduced very significantly and effectively. To illustrate this method the geoid of Hungary was estimated via RBF type network. The computations were carried out with the symbolic - numeric integrated system Mathematica.