Associative memory without global inter-neuron connections

A new architecture for a neural network with associative properties is proposed. Elements of each stored image receive a special mark following local connection training. “All to all” connections are replaced with connections to a special “central generator”. This makes possible the unification of all elements with the same mark into a single image as directed by the central generator. The architecture leads to a radical simplification of the connection system and allows construction of a large associative memory. Analog and digital optoelectronic versions are discussed