Simulated model of auto-associative memory in the form of a neural network with a low level of activity

Information parameters of the neuron net performing the functions of autoassociative memory were investigated by imitation modelling method. The fully connected network with Hebb gradual synapses was studied. Its information capacity was shown to increase significantly with a decrease of the activity level of the stored patterns. This evidence agrees well with the analytical result obtained earlier by the replica method.