Oleg V. Favorov
University of North Carolina, Chapel Hill

Dan Ryder
University of British Columbia

SINBAD is a theory of the cerebral cortex that draws on psychological, functional, anatomical, and physiological considerations.

A neural network model based on the theory has the following principal capabilities:

1) Discovery of predictively useful variables that are only implicit in the network's input
Exposed to a set of inputs that vary in regular ways due to hidden influences, the network will infer the existence of those hidden influences. This is a simple variety of "explanatory inference," the same type of inference used by Mendel when he postulated the existence of genes to explain a set of complex regularities observed in pea plant crossings.
2) Acquisition of an internal model of the environment, including complex higher order regularities: Exposed to a set of regularities, the units in a SINBAD network come to serve as "stand-ins" for explicit and implicit environmental variables. The relations amongst the environmental variables are reflected in the relations amongst the network's stand-ins.
3) Use of the internal model to perform inferences:
Inference is accomplished by filling in missing information in the internal model. The range of inferences that can be accomodated is very large, since aspects of the internal model may be highly abstract.
4) Learning of behavioural contingencies:
When the network is given inputs about its own body, its internal model will come to include the effects of its own behaviour on the environment and its own internal state (e.g. its state of need and satisfaction). Such an internal model can be used to produce effective behaviour.
5) Use of behavioural contingencies to perform actions:
In order to maximize its state of satisfaction, a behaving SINBAD network will choose actions appropriate to its current needs and the state of its environment. It does this by inferring what the state of the environment ought to be, given high satisfaction and its current needs.