Evaluating the Effects of Size and Precision of Training Data on ANN Training Performance for the Prediction of Chaotic Time Series Patterns

In this research, artificial neural networks (ANN) with various architectures are trained to generate the chaotic time series patterns of the Lorenz attractor. The ANN training performance is evaluated based on the size and precision of the training data. The nonlinear Auto-Regressive (NAR) model is trained in open loop mode first. The trained model is then used with closed loop feedback to predict the chaotic time series outputs. The research goal is to use the designed NAR ANN model for the simulation and analysis of Electroencephalogram (EEG) signals in order to study brain activities. A simple ANN topology with a single hidden layer of 3 to 16 neurons and 1 to 4 input delays is used. The training performance is measured by averaged mean square error. It is found that the training performance cannot be improved by solely increasing the training data size. However, the training performance can be improved by increasing the precision of the training data. This provides useful knowledge towards reducing the number of EEG data samples and corresponding acquisition time for prediction.

Generalized Relational Tensors For Irregularly Sampled Time Series: Methods To Store, Re-generate, And Predict

The paper deals with a generalized relational tensor, a novel discrete structure to store information about a time series, and algorithms (1) to fill the structure, (2) to generate a time series from the structure, and (3) to predict a time series, for both regularly and irregularly sampled time series. The algorithms combine the concept of generalized z-vectors with ant colony optimization techniques. In order to estimate quality of the storing/re-generating procedure, a difference between characteristics of the initial and regenerated time series is used. The structure allows working with a multivariate time series, with an irregularly sampled time series, and with a number of series as well. For chaotic time series, a difference between characteristics of the initial time series (the highest Lyapunov exponent, the auto-correlation function) and those of the time series re-generated from a structure is used to assess the effectiveness of the algorithms in question. The approach has shown fairly good results for periodic and benchmark chaotic time series and satisfactory results for real-world chaotic data.