To search, Click below search items. |
| All Published Papers Search Service |
|
|
|
Title |
Applying Genetic Algorithm in Architecture and Neural Network Training |
|
Author |
Mahshid Kaviani Seyed Majid MirRokni |
| Citation |
Vol. 17 No. 6 pp. 118-124 |
|
Abstract |
The structure of Artificial Neural Network with ANN symbol is not a predefined program. Therefore, architecture and proper training of ANN have significant effect on the results. ANN architecture of consists determination of a number of optimal neurons in hidden layers creating a challenging puzzle that makes us use the trial and error method. ANN training is an optimization process to determine the optimal values of weights and biases. The classic optimization algorithms provides us with the optimal local values, wherein calculation time is increased in multivariable problems and the rate of convergence is reduced to optimal value. Applying Genetic Algorithm with symbol GA offers an appropriate solution for these challenges. GA is optimized through random search technique. It simultaneously considers various responses and reduces chance of convergence to a local optimum. Therefore, by an intelligent search method values close to the optimum point are found. The objective of using convergence in this case study is to predict the daily average temperature in 2009 which is obtained through some parameters such as pressure, vapor pressure and relative humidity. GA is used in this study, to avoid the trial and error methods in the architecture of ANN and to achieve absolute optimum values in a short time, for determining the number of hidden layer neurons as well as the optimal values of weights and biases in ANN training. The results show that the GA approach can be replaced with trial and error methods in determining optimal state of ANN architecture, and to increase speed, accuracy and efficiency of the training. |
|
Keywords |
Artificial Neural Network, Genetic Algorithm, optimization algorithms, trial and error, ANN architecture. |
|
URL |
