C++ Neural Networks and Fuzzy Logic: Preface

C++ Neural Networks and Fuzzy Logic

by Valluru B. Rao

MTBooks, IDG Books Worldwide, Inc.

ISBN: 1558515526   Pub Date: 06/01/95

Previous Table of Contents Next

Chapter 13

Backpropagation II

Enhancing the Simulator

In Chapter 7, you developed a backpropagation simulator. In this chapter, you will put it to use with examples

and also add some new features to the simulator: a term called momentum, and the capability of adding noise

to the inputs during simulation. There are many variations of the algorithm that try to alleviate two problems

with backpropagation. First, like other neural networks, there is a strong possibility that the solution found

with backpropagation is not a global error minimum, but a local one. You may need to shake the weights a

little by some means to get out of the local minimum, and possibly arrive at a lower minimum. The second

problem with backpropagation is speed. The algorithm is very slow at learning. There are many proposals for

speeding up the search process. Neural networks are inherently parallel processing architectures and are suited

for simulation on parallel processing hardware. While there are a few plug−in neural net or digital signal

processing boards available in the market, the low−cost simulation platform of choice remains the personal

computer. Speed enhancements to the training algorithm are therefore very buffernecessary.