Non-Linear Time-Series Prediction by Systematic Data Exploration on a Massively Parallel Computer

In this paper we describe the application of massively parallel processing (MPP) to the problem of time-series analysis. Time-series problems involve sequences of numbers (for example, the daily closing values of the stock market, EEG patterns of brainwave activity, or, as discussed in this paper, the temporal values from set of equations of motion). Often the problem of interest is the prediction of some future value(s) in the sequence using only past values. Taking advantage of the power of an MPP supercomputer, we describe techniques to perform exploratory data analysis on time-series problems in a quick and efficient manner. 1 Introduction Problems concerning the analysis of a time-series have existed for a very long time. Ever since the creation of the first financial markets, people have tried (and usually failed) to predict the future. However, there are many timeseries problems in science and engineering that are to some extent predictable (such as the occurrence of sun spot..