CHARACTER-RECOGNITION WITH THE RANDOM NEURAL NETWORK

We study the random neural network model (Gelenbe, 1989, 199 1) as an auto-associative memory and present new storage and recognition methods which we apply to digit recognition. Very good recognition performance is obtained whatever the correlation between the stored patterns and the pattern to be recognized. Experiments show that as the network parameter choice becomes more precise, the recognition performance improves. However, the learning and recognition methods become time consuming. Here, we also present very fast methods which can give a good recognition.