Training and evaluation of neural networks for multi-variate time series processing
T.L. Fog, Jan Otto Larsen, Lars Kai Hansen · 2002
We study the training and generalization for multi-variate time series processing. It is suggested to used a quasi-maximum likelihood approach rather than the standard sum of squared errors, thus taking dependencies among the errors of the individual time series into account. This may lead to improved generalization performance. Further, we extend the optimal brain damage pruning technique to the multi-variate case. A key ingredient is an algebraic expression for the generalization ability of a multi-variate model. The variability of the suggested techniques are successfully demonstrated in a multi-variate scenario involving the prediction of the cylinder pressure in a marine engine.