Naslov
Accelerated gradient learning algorithm for neural network weights update

Autori
Hocenski, Željko ; Antunović, Mladen ; Filko, Damir

Izvornik
Neural computing & applications (0941-0643) 19 (2010), 2; 219-225

Vrsta, podvrsta i kategorija rada
Radovi u časopisima, članak, znanstveni

Ključne riječi
neural network; weights update; gradient learning method; parallel processing.

Sažetak
This work proposes decomposition of square approximation algorithm for neural network weights update. Suggested improvement results in alternative method that converge in less iteration and is inherently parallel. Decomposition enables parallel execution convenient for implementation on computer grid. Improvements are reflected in accelerated learning rate which may be essential for time critical decision processes. Proposed solution is tested and verified on MLP neural network case study, varying a wide range of parameters, such as number of inputs/outputs, length of input/output data, number of neurons and layers. Experimental results show time savings up to 40% in multiple thread execution.

Izvorni jezik
Engleski

Znanstvena područja
Elektrotehnika, Računarstvo

Napomena
Special Issue - KES2008 ; Rad je kao predavanje prezentiran na skupu 12th International Conference Knowledge-Based Intelligent Information and Engineering System (KES 2008), održanom od 03.-05.09.2008., Zagreb, Hrvatska ; u cjelosti je uz međunarodnu recenziju objavljen u Proceedings, Part I. ; Ignac Lovrek, Robert J. Howlett, Lakhmi C. Jain (ur.) ; Lecture Notes in Computer Science. Vol. 5177 ; Berlin : Springer, 2008. ; ISBN 978-3-540-85562-0 ; str. 49-56.

POVEZANOST RADA