A Framework for an Automatic Generation of Neural Networks

The automatic generation of neural network architecture is a useful concept as in many applications while the optimal architecture is not a priori known. Often trial and error is done before a satisfactory architecture is found. Construction deconstruction algorithms can be used as an approach but they have several drawbacks. Sometimes an evolutionary computation and evolutionary algorithms are used but the idea in this paper is reserved for a special kind of evolutionary algorithms. So in this paper we proposed framework for neural networks which try to get best solution for problems by automatic generation technique. The obtained results are promising, suggesting many other research directions..