Search of Techniques to Improve Artificial Neural Networks Training Time

摘要

In the search of techniques to improve Artificial Neural Networks (ANN) training time, this work investigates the following approaches: parallel implementation of Kohonen's Self-Organizing Map (SOM) in a multiprocessing environment and utilization of an advanced training algorithm for Multilayer Perceptrons (MLP) networks. The parallel algorithm developed for the SOM was compared with its sequential analog and the training algorithm proposed for MLP networks was compared with the standard Backpropagation algorithm. The comparisons were realized using Remote Sensing data.