A complexity and efficiency of adaptive momentum on back-propagation algorithmt
Alaa Ali Hameed and Wisam Husam
Selcuk University, Turkey
Yıldız Technical University, Turkey
: J Comput Eng Inf Technol
Abstract
The use of artificial neural networks algorithm based on adaptive momentum techniques has been derived to improve the speed of convergence, and minimizing the error misadjustment to obtain high accuracy in short time of process. However, these techniques suffer from computational complexity. The recently back-propagation with adaptive momentum (PBPAM) algorithm has demonstrated superiority performance of various proposed adaptive momentums of back-propagation algorithm versions in terms of convergence rate, sum of squared error (SSE), and accuracy. In this paper, we will compare the computational complexity of PBPAM algorithm with other BP versions. The PBPAM algorithm is characterized by its simplicity, because it does not need much CPU processing, and it obtains good results in a short period of time. Simulation results have shown that the PBPAM algorithm provides a faster convergence, lower SSE, higher accuracy and lower computational complexity comparing to other BP algorithms using different dataset from UCI and KEEL repositories.
Biography
Email: dr.alaa85@yahoo.com
