Award Date

1-1-1997

Degree Type

Thesis

Degree Name

Master of Science (MS)

Department

Computer Science

Number of Pages

63

Abstract

Artificial neural networks have been studied for many years in the hope of achieving human like performance in the area of pattern recognition, speech synthesis and higher level of cognitive process. In the connectionist model there are several interconnected processing elements called the neurons that have limited processing capability. Even though the rate of information transmitted between these elements is limited, the complex interconnection and the cooperative interaction between these elements results in a vastly increased computing power; The neural network models are specified by an organized network topology of interconnected neurons. These networks have to be trained in order them to be used for a specific purpose. Backpropagation is one of the popular methods of training the neural networks. There has been a lot of improvement over the speed of convergence of standard backpropagation algorithm in the recent past. Herein we have presented a new technique for accelerating the existing backpropagation without modifying it. We have used the fourth order interpolation method for the dominant eigen values, by using these we change the slope of the activation function. And by doing so we increase the speed of convergence of the backpropagation algorithm; Our experiments have shown significant improvement in the convergence time for problems widely used in benchmarKing Three to ten fold decrease in convergence time is achieved. Convergence time decreases as the complexity of the problem increases. The technique adjusts the energy state of the system so as to escape from local minima.

Keywords

Acceleration; Algorithm; Artificial; Backpropagation; Intelligence; Networks; Neural Networks; Techniques

Controlled Subject

Computer science; Artificial intelligence

File Format

pdf

File Size

1392.64 KB

Degree Grantor

University of Nevada, Las Vegas

Language

English

Permissions

If you are the rightful copyright holder of this dissertation or thesis and wish to have the full text removed from Digital Scholarship@UNLV, please submit a request to digitalscholarship@unlv.edu and include clear identification of the work, preferably with URL.

Rights

IN COPYRIGHT. For more information about this rights statement, please visit http://rightsstatements.org/vocab/InC/1.0/

Download


Share

COinS