Journal Menu
Related Events
International Journal of Engineering & Technology Sciences
IJETS 2015, 3(6), 382-392
Modification of multilayer back-propagation neural networks using variable learning rate algorithm and automaton theory and determination of optimal learning rate
Structures Engineering , University of Ilam, Iran
ABSTRACT
Multilayer feed-forward neural networks have been extensively studied by many researchers. Although these networks have been successful notably, they have their own limitations and disadvantages. For instance, training time is relatively long and the network may not be trained. This is mainly due to the wrong selection of network parameters. The used method to determine the parameters of weight and network bios is energy gradient function. It is apparent that network error function has rough surface and is likely to be entrapped. In order to increase training speed and avoid entrapping in local optimum, adaptive variable learning rate (AVLR) and automaton algorithm are employed, respectively. Also, optimal learning rate for different networks may be obtained through using above-mentioned methods.