![]() | Only 14 pages are availabe for public view |
Abstract In this thesis, generic methods to construct feed forward neural networks will be developed. Each net is designed through determining the proper configuration with a set of weights to accomplish a defined objective in two phases: Phase one: includes the determination of the number of layers and the neurons count in each layer. Phase two: which is the learning phase that means finding the suitable values of the link weights between different cells. Although there are several methods to accomplish these two phases separately and others to accomplish them concurrently, both methods are contained some problems such as the resulting network could be incoherent or uni-layered or the method is very slow even reach an optimal net. |