Department of Information Technology
Information processing research lab
Goal of this project is to analyze if genetic algorithms, especially differential evolution, can be used to train feed-forward neural networks. Disadvatanges and advantages of evolutionary methods are compared to traditional network training methods, mostly based on gradient information.
Project is finished. All the goals were met and results will be published in a scientific journal.
Contact authors for reprints.
Matlab(tm) source code (for scientific and non-profitable use only)
|Differential evolution training algorithm for Neural-Network Toolbox (tm)||traindiffevol.m|
|Script for demonstration of the TRAINDIFFEVOL||demo01.m|
To download data used in our experiments, explore data repositories from links section.
|Differential evolution bibliographies by professor Jouni Lampinen|
|Local links by professor Jouni Lampinen|
|Differential Evolution page by Rainer Storn|
|Data repository: University of California|
|Benchmark group on pattern recognition|
[ML97] Masters, Timothy and Land, Walker (1997). A new training algorithm for the general regression neural network. 1997 IEEE International Conference on Systems, Man, and Cybernetics, Computational Cybernetics and Simulation., Volume: 3, pp. 1990-1994.
|Jouni Lampinen||WWW||Professor, responsible author, Lappeenranta University of Technology|
|Joni Kamarainen||WWW||Researcher, Lappeenranta University of Technology|
|Jarmo Ilonen||WWW||Research assistant, Lappeenranta University of Technology|