Learning and structural design of feedforward neural networks by employing genetic algorithms

This work is an attempt to face learning and structural design problems simultaneously for feedforward neural networks by employing genetic algorithms and hybrid algorithm. In the learning process, the disadvantages of backpropagation as a learning algorithm can be avoided by using hybrid algorithm. On the other hand the ability of genetic algorithms to perform global search intelligently make this method as a robust learning algorithm, while in the same time design the structure. The proposed algorithm shows good performances where all of the trials of learning processes converge to the desired condition and most of structural design end with desired efficient structure.