A neural network model for large-sized implementation has been developed. In this model neural interconnections for N neurons are imposed of $N^2$ fixed global interconnections and 2N adaptive interconnections. For given number of adaptive elements this model allows us to implement neural networks with larger number of neurons. Storage capacity and error correction performance of this new model are evaluated by computer simulation and compared to those of perceptron model. A simple application for deformed character recognition is also included.
