A failure and warranty cost model is gained from a failure database. The model is a combination of statistical components with a multi-layer perceptron and a cross-entropy based learning rule. The model is used for forecasting warranty costs in alternative warranty condition scenarios. The estimate of forecast variance considers both the individual vehicle risk and the overall manufacturing quality fluctuation risk.

Utility of product variants is a nonlinear function of product features. Such a utility function can be represented by a multi-layer perceptron and embedded into the classical logit demand function. However, the utility (which is the output of the multi-layer perceptron to be learned) is not explicitly known. This is why the backpropagation learning rule has been extended to fit the demand function directly to observed market shares. Forecasts of market shares on the German automobile market with the help of a perceptron-based and classical logit model are compared. The perceptron-based model leads to a significant improvement of the forecast quality.

Simplified critic based neurocontrol algorithms are analyzed for expected number of samples to convergence. It is shown that there is a fundamental difference in the complexity behavior between the batch and the incremental algorithm, and between the algorithm with and without an explicit plant model. The batch algorithm using a plant model is superior to other variants.

A modular architecture for supervised learning is presented. Three ways to modularize the learning are investigated: (1) preprocessing the input by a self-organizing subnetwork to extract strong features from the data, (2) supervised feature extraction, and (3) an iterative learning cycle in which only one layer learns at a time, with the output-layer weights learned by an exact method. With this modular architecture, only a small fraction of connection weights is determined by the gradient-descent method. A series of computational experiments shows the superiority of the modular model in learning quality and speed. The author begins by considering the decomposition of a feedforward network into a feature-discovery and a supervised-learning modules, then he introduces supervised feature discovery, and finally he describes a modular backpropagation algorithm

In a certain two-layer network architecture, the delta learning rule leads to a self-organization of connection weights. A formal analysis of this learning rule shows that the weight vectors ofnsecond-layer nodes converge to a rotation of the firstnprincipal components. Therefore, the delta-rule-based self-organization performs optimal encoding and decoding of data in the sense of the principal component analysis. This property of the delta learning rule has been verified by a series of computational experiments, which also showed good convergence stability of the rule. For data compression tasks, it performs substantially better than a three-layer autoassociative perceptron with linear or nonlinear hidden units