Neural computations in control problems: aspects of computability and spatial-time characterization of cognitive functions

The thesis of the futurologist A. Clark that «sufficiently advanced technologies are indistinguishable from magic» is relevant nowadays as never before. Modern science, unlike magic, although it relies on logically consistent physical theories, formalisms of mathematics and the possibilities of computing, has not made the world less mysterious, especially with regard to the natural phenomena of consciousness, thinking and intelligence. From the point of view of computer science, a distinctive feature of these phenomena is their network-centric neuromorphic organization and the availability of complex memory resources - a non-casual computing mechanism for «transporting» processed data, the results of calculations and information interactions in time. Formalization of these features makes it possible to determine a model of cognitive processes on a multidimensional set of states, on which a strict relation between the time order and numerical equality is not introduced, therefore, contextually significant information, about past as well as current or forecast states, can be used for calculations at the same time states. For control systems operating under uncertainty, a distributed heterogeneous reconfigurable structure is proposed that implements the technology of «computation in memory» both for algorithms reflecting causal relationships and learning processes that require operational reconfiguration of the computation field.