Ge. Peterson et al., USING TAGUCHI METHOD OF EXPERIMENTAL-DESIGN TO CONTROL ERRORS IN LAYERED PERCEPTRONS, IEEE transactions on neural networks, 6(4), 1995, pp. 949-961
A significant problem in the design and construction of an artificial
neural network for function approximation is limiting the magnitude an
d the variance of errors when the network is used in the field, Networ
k errors can occur when the training data does not faithfully represen
t the required function due to noise or low sampling rates, when the n
etwork's flexibility does not match the variability of the data, or wh
en the input data to the resultant network is noisy, This paper report
s on several experiments whose purpose was to rank the relative signif
icance of these error sources and thereby find neural network design p
rinciples or limiting the magnitude and variance of network errors.