Scalability in neural network design: an add-on scheme

The authors investigate how a neural network architecture can be expanded with add-on neural networks to deal with a scale-up problem. It is shown that if the new scale-up associative memory design problem is solvable, then the authors´ proposed scheme is also valid. A scale-up methodology is developed which allows an increase in dimension for each stored pattern in a dynamic neural network by augmenting additional neural subnetworks, while preserving the interconnection architecture of the original neural network. The proposed scale-up approach eliminates the need for redesigning an overall new neural network architecture. The ease of design is illustrated by the stability requirement of the neural network. The add-on feature proves to be efficient with respect to the design of the interconnection weights