This report is the result from a project at the IT-university in Gothenburg, Sweden. The project is a part of the course Adaptive Algorithms that was running in the beginning of Autumn 2002. The part of the course that this project is about was Neural Networks, in particular Back Propagation and viewing multi dimensional data sets. For the multidimensional viewing of data sets, the project described in this report is using the Kohonen's SOFM.
A short introduction to Back Propagation, Kohonen's SOFM and the Principal Component Analysis is to be found in the respective section of the report.
The problem was divided into two parts. The first was to examine some data
sets, the second was to train a feed forward network with two of the data sets
using the back propagation algorithm. The first part was intended to visualize
the problem, so that the difficulty of one particular data set could be
evaluated. Then two data sets, one easy and one difficult, were to be chosen.
Each of those two data sets should then be used to train a feed forward network.
Both networks should be trained using the back propagation algorithm.
[ toc | next |
up ]
The organization of this report is the following. Chapter one, is this short introduction.
Chapter two and three presents the data analyze stage, where the
Kohonen's SOFM and the Principal Component Analysis is described. The resulting program, its procedure
and the results are described in detail as well. Chapter four and five present the Feed
Forward network that was trained using the back propagation algorithm. Chapter
six and seven present the conclusion, what modifications this program need
to be usable in the future and possible applications for the program. The report
is ended with a small glossary that explains the technical terms.
[ toc | next |
up ]