<title>Optoelectronic neural network without global interneuron connections</title>

Abstract

A new neural network architecture with associative properties is proposed. Elements of every stored image obtain a special mark after local connections training. `All to all' connections are replaced by connections with a special Central Generator. It allows to unite all elements with the same mark into the whole image. The new architecture leads to a radical simplification of the connection system and therefore to the possibility of a large associative memory creation. Analog and digital optoelectronic versions are discussed.