Training Feed Forward Neural Networks Using Optimized Back Propagation Algorithm

V. Sowjanya; J. Sirisha; T. Kranthi kumar

Neural network (NN) is one of the most important data mining techniques. It is used with both supervised and unsupervised learning. Training NN is a complex task of great importance in problems of supervised learning. Most of NN training algorithms make use of gradient-based search. These methods have the advantage of the directed search, in that weights are always updated in such a way that minimizes the error, which called NN learning process. However, there are several negative aspects with these algorithm such as dependency to a learning rate parameter, network paralysis, slowing down by an order of magnitude for every extra (hidden) layer added and complex and multi-modal error space. The back-propagation algorithm is one of the most famous algorithm to train a feed forward network. Instead of its success rate the quest for development is observed through the various standard modifications in-order to meet the challenges of complex applications. This paper proposes an optimized back propagation algorithm to train feed forward artificial neural networks.

Select Volume / Issues:

• Volume 3 Issue 6, Nov. 2012

Year:

2012

Type of Publication:

Article

Keywords:

Error Rate; Neural Network; Stability; Supervised learning

Journal:

IJECCE

Volume:

3

Number:

6

Pages:

1338-1341

Month:

November