Neurex, an expert network for the autonomous design of artificial neural networks

Artificial neural networks (ANNs) have proved to be increasingly useful for complex problems that are difficult to solve with conventional methods. With their learning abilities, they avoid the need to develop a mathematical model or to acquire any particular knowledge to solve the problem. But the difficulty now lies in the ANN design process. Many choices must be made in designing an ANN, and there are no available design rules to make these choices directly for a particular problem. Therefore the design of an ANN demands a certain number of iterations, guided mainly by the expertise and intuition of the designer. To solve this difficulty, we have designed NEUREX, an expert network composed of an expert system and an ANN simulator. NEUREX guides the iterative design process of ANNs autonomously. Its structure tries to reproduce the expertise brought to bear by a human expert in conceiving ANNs. It serves as a shell to implement this expertise for different learning paradigms. This article presents the general characteristics of the system and its use for designing ANNs with standard backpropagation learning.