History of development and prospects for system engineering of neural architectures

In this article typical neural network architectures are classified and key historical stages in their development are examined. Fundamental functional properties of neural architectures are identified and method for multi-criteria evaluation of their potential effectiveness is presented. Several approaches to system engineering of neural networks are also proposed. It is shown that, along with well-established principles of multilayering, hierarchy and recurrence, new issues are becoming increasingly important such as modularity, structural adaptability, and computational efficiency of neural networks.