An architecture for an adaptive associative memory system based on autonomous reactions between images

To construct a "thinking-like" processing system, an architecture of an adaptive associative memory system, which treats "images" as basic units of information, is proposed. This memory system adapts to an environment of the external world based on the autonomous reactions between the images and generated prototype images of the environment. It is a kind of neural-network made up of nodes and links, called a locally spreading-activation network. Each node holds one image locally. Images in high-activity nodes interact autonomously and generate new images and links. By this reaction between images, various forms of images are generated under constraints of links with adjacent nodes in the associative memory system.