Various Aspects Of Artificial And Biological Neural Networks

The field of neural networks is a wide and diverse field which spans a variety of interests, modelling methods, and potential uses. In my thesis, I discuss two different topics. The first is the use of feed-forward networks for time-series analysis and prediction. A new criterion for the quality of predictions of neural networks is presented. The criterion is based on the performance of networks for long term predictions, where one compares the distribution of the predicted values with that of the original data. Using this criterion as a weak constraint, one is able to choose the best networks from a population of trained networks that are similar in their short term predictions. This criterion may be also used to create a new predictor in which the various networks are weighted according to their long term behavior. In addition, a method for improving both short and long term predictions is proposed. Since quite often the consecutive errors of prediction are correlated in a systematic..