Your browser version is outdated. We recommend that you update your browser to the latest version.

Manuscript Title: Weights Adjustment of Two-Term Back-Propagation Network Using Adaptive and Fixed Learning Methods

Author : Citra Ramadhena, Ashraf Osman Ibrahim and Sarina Sulaiman



Artificial Neural Networks (ANNs) using the Back Propagation algorithm (BP) mainly depend on weights adjustment in training learning. The solutions can be faster by properly adjusting the magnitude and sign of the weights. Weight Changes is one of the solutions for a common problem faced by the two-term back-propagation network that suffers from slow convergence and trapping in a local minima. Many ways have offered to generate the proper weight magnitude and sign of the network. One of the solutions by adjusting a correct value of learning rate and momentum parameters to improve the performance of ANN by implementing an adaptive learning method. This paper implements both adaptive and fixed learning methods for two-term BP algorithms. Standard datasets of UCI machine learning are used with n-fold cross validation to train and test both methods to properly adjust and investigate the changes of weight sign accordingly. The results showed that the two-term BP using the adaptive learning method work better in producing proper weight changes with minimum time compared to two-term BP with fixed learning rate.

Keywords: Weight changes, adaptive learning, Two-Term BP, Mean square error, convergence rate, Supervised Learning.

 Vol 5 (2)