Projektowanie warstwowych sieci neuronowych z zastosowaniem programowania ewolucyjnego

Maciej Grzenda

Abstract

The purpose of the work was to prepare new methods of multilayer neural networks design. The primary assumption was to deploy new self-adaptive procedures making is possible to resign from time-consuming parameter selection. In addition, new methods are responsible for architecture design and weight adjustment. All the methods proposed in this work can be implemented as reinforcement learning methods and standard training methods, based on the error measure. Evolutionary programming has been used as a framework for the methods deployed. New self-adaptive procedure plays an important role in neural network design making it possible to automatically adjust the most important algorithm settings. The new methods, proposed in this work, have been illustrated by the results in game theory, pattern recognition and forecasting in Polish power system.
Diploma typeDoctor of Philosophy
Author Maciej Grzenda (FoEE)
Maciej Grzenda,,
- Faculty of Electrical Engineering
Title in PolishProjektowanie warstwowych sieci neuronowych z zastosowaniem programowania ewolucyjnego
Languagepl polski
Certifying UnitFaculty of Electrical Engineering (FoEE)
Disciplineinformation science / (technology domain) / (technological sciences)
Defense Date13-06-2001
Supervisor Bohdan Macukow (FoEE)
Bohdan Macukow,,
- Faculty of Electrical Engineering

Pages133
Keywords in English-
Abstract in EnglishThe purpose of the work was to prepare new methods of multilayer neural networks design. The primary assumption was to deploy new self-adaptive procedures making is possible to resign from time-consuming parameter selection. In addition, new methods are responsible for architecture design and weight adjustment. All the methods proposed in this work can be implemented as reinforcement learning methods and standard training methods, based on the error measure. Evolutionary programming has been used as a framework for the methods deployed. New self-adaptive procedure plays an important role in neural network design making it possible to automatically adjust the most important algorithm settings. The new methods, proposed in this work, have been illustrated by the results in game theory, pattern recognition and forecasting in Polish power system.
Thesis file
Grzenda_Maciej_projektowanie.pdf 5.07 MB
Citation count*4 (2015-07-23)

Get link to the record

Back
Confirmation
Are you sure?