A parallel design and implementation for backpropagation neural network using MIMD architecture

Author

Fathy, Sherif Kassem ; Syiam, Mostafa Mahmoud

Author_Institution

Dept. of Comput. Eng. & Sci., MTC, Cairo, Egypt

Volume

3

fYear

1996

fDate

13-16 May 1996

Firstpage

1472

Abstract

The paper illustrates a parallel backpropagation algorithm that depends upon a transputer network. Since the backpropagation algorithm is regarded as being communication intensive, a mechanism is developed to decrease the communication time overhead between different related processor elements. The developed parallel backpropagation algorithm is applied to the problem of printed Arabic character recognition. The experimental results revealed that the speed up approaches the maximum value of 15.2 when a network of 16-transputers is utilized

Keywords

backpropagation; neural net architecture; optical character recognition; parallel algorithms; parallel architectures; performance evaluation; transputer systems; MIMD architecture; backpropagation neural network; communication intensive; communication time overhead; experimental results; parallel backpropagation algorithm; parallel design; printed Arabic character recognition; transputer network; Algorithm design and analysis; Backpropagation algorithms; Character recognition; Computer architecture; Computer networks; Convergence; Neural networks; Neurons; Parallel algorithms; Training data;

fLanguage

English

Publisher

ieee

Conference_Titel

Electrotechnical Conference, 1996. MELECON '96., 8th Mediterranean

Conference_Location

Bari

Print_ISBN

0-7803-3109-5

Type

conf

DOI

10.1109/MELCON.1996.551228

Filename

551228