| Summary: | Since the pioneering work of Kolmogoroff and Wiener the use of computing devices to solve problems concerned with - the prediction of future states of a time series has stimulated a large amount of research. Despite all this, however the results have been disappointing. If significant progress were to be made in this field it would lead not only to the possibility of forecasting economic events, the weather, earthquakes, epidemics, and so on, but also -to the possibility of simulating these system. Approaches involving the programming of a computer to carry out this task run into the difficulty of defining the variables involved in a precise enough manner, whereas using the computer -to investigate all the past events of that time series requires a large processing time and an enormous memory store. This thesis examines an approach to this problem which involves the use of a device for processing information in a parallel manner. The system envisaged consists of a holographic recognition device controlled by a digital computer -a combination of analogue and digital techniques. The principle of this device is that developed by Gabor and others, and allows the system to learn to predict the future of a time series. The system learns to predict the future of a time series by using a past length of time series as a training set. Using this training set it attempts to predict the next values of the time series, which can then be compared with the actual time series. The system, then, attempts to optimize its prediction by minimizing the error between the predicted and actual values.
|
|---|
