Universiti Teknologi Malaysia Institutional Repository

A new fuzzy back-propagation learning method based on derivation of min-max function tuned with genetic algorithms

Mashinchi, Mohammad Hadi (2007) A new fuzzy back-propagation learning method based on derivation of min-max function tuned with genetic algorithms. Masters thesis, Universiti Teknologi Malaysia, Faculty of Computer Science and Information System.

[img] PDF
Restricted to Repository staff only

485Kb
[img] PDF (Abstract)
63Kb
[img] PDF (Table Of Content)
95Kb
[img] PDF (1st Chapter)
123Kb

Abstract

Nowadays, environments are covered by lots of qualitative (inexact) data. Making proper decisions according to this qualitative data is an ultimate aim. Artificial neural networks can be adapted to the quantitative environments, since they have ability of learning. On the other hand, fuzzy logic has the power of dealing with inexact data. To get benefit of advantageous of artificial neural networks and fuzzy logic, fuzzy artificial neural networks are proposed. This hybrid soft-computing technique has the ability of learning in qualitative environments. Thus, fuzzy artificial neural networks can make qualitative decisions according to inexact data which are fed to it. Learning process of fuzzy artificial neural networks is one of the most important issues. Thus, many learning methods for feed forward fuzzy artificial neural networks are proposed. Low speed of convergence and accuracy of training have made fuzzy artificial methods inapplicable in most of problems. Thus, efficient learning method for fuzzy artificial neural networks is demandable. In this study a “genetically tuned fuzzy back propagation method based on derivation of min-max function� as new learning method has been proposed by author. The proposed learning method has resolved some of the previous shortcomings. Importance of the proposed method is that, three main benefits are reached simultaneously; it can learn from any kind of convex fuzzy numbers, accuracy of training is higher since error function is more realistic comparing to gradient based learning methods, and convergence speed is acceptable.

Item Type:Thesis (Masters)
Additional Information:Thesis (Master of Science (Computer Science)) - Universiti Teknologi Malaysia, 2007; Supervisor : Assoc. Prof. Dr. Siti Mariyam Hj. Shamsuddin
Uncontrolled Keywords:fuzzy artificial neural networks, learning methods, qualitative data, hybrid soft-computing technique
Subjects:Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Divisions:Computer Science and Information System (Formerly known)
ID Code:6796
Deposited By: Ms Zalinda Shuratman
Deposited On:25 Nov 2008 02:56
Last Modified:24 Oct 2012 08:02

Repository Staff Only: item control page