Performance improvement of back propagation neural network learning algorithms by introducing gain variation of activation function

Mohd Nawi, Nazri and Ghazali, Rozaida and Mohd Salleh, Mohd Najib Performance improvement of back propagation neural network learning algorithms by introducing gain variation of activation function. Universiti Tun Hussein Onn Malaysia.

[img]
Preview
PDF
1035Kb

Abstract

Ye propose a method for ? pnopose a nethod for inprovtng the of the back prcpagation algarithm i.'' ifr{toduehg gah voiation of ,lre rctivation fwtion. In '_ a feed forwud' algoritla tIrc slop of the xtiwtion ii:fiorction is diretly Wwted by a parnw ,eM i' ,ta as 'gain'. In this pqer, the idluarce of the wriation 'of 'galn' on the lc.arning eility $ a nanral rctwark is onlysd. tfrilti lalter fed forawd neural naworb ,raw been awessed. Pbsiel intapretatiat 6 t rc relatiottship between tlre gsin valtte and the lernhg rde erd *elght wfus is given. Instcd of a wstant 'gain' vslue, try Wpose an algaritlm u clwtge the gain wlrc futfvegfor ereh n&. ne e$furcy olthe p@ nethd is wr$€d by n e6u:s of sinutaion (m a Wtty problem and classification prcblem. The results show that the praposd netld considerably improws tlu leaning Wed of tle general back prqngation algorithm.

Item Type:Other
Uncontrolled Keywords:Back-propgation Nearal rVenror&s, Gain. Act iv at ion fu n c t ion, Le am i n g ro t e, Tra in i n g E_{li c i e nc 1-
Subjects:T Technology > T Technology (General)
ID Code:2592
Deposited By:M.Iqbal Zainal A
Deposited On:28 Jun 2012 08:30
Last Modified:28 Jun 2012 08:30

Repository Staff Only: item control page