Parameter Identification by Bayes Decision and Neural Networks

The problem of parameter identification by Bayes point estimation using neural networks is investigated. The network creates an estimate of the underlying distribution function of a measured data sequence, which by network recurrency is utilized for obtaining the Bayes estimate. The neural estimate is proven to be consistent on a mild basis of assumptions concerning the data generating process. Therefore the network is expected to be applicable to parameter identification within a broad class of practical problems. The network is characterized by very fast training, which is believed to make it especially useful where real time operation is required.