Schemas and Neural Networks for Sixth Generation Computing. Invited Survey

Abstract The study of animal and human brains suggests overall architectural principles for “sixth generation computers.” Each such machine will comprise a network of more specialized devices, with many of these devices structured as highly parallel arrays of interacting neuron-like, possibly adaptive, components. We stress the interaction between computational neurobiology and neural engineering and note the two grains of analysis of schemas and neural networks, arguing that “schemas will be the programming language of the sixth generation.” After a brief introduction to the variety of neuron models now current, schemas are exemplified in a discussion of high-level vision. The integration of neural network and schema models is brought out in an integrated set of studies, called Rana computatrix , of mechanisms for visuomotor coordination in frog and toad. Whereas many articles on neural networks focus on learning and restrict themselves to a limited class of simple neurons, the present paper emphasizes the “domain-specific” structure of neural networks, as well as emphasizing that technologists have much to learn from the study of neurobiological systems. We close with a brief account of adaptation and the programming of sixth generation computers.