CONSTRUCTION PRINCIPLES OF COMPUTING UNITS FOR INTELLECTUAL DATA PROCESSING

Abstract

An approach to the development of problem-oriented parallel computers for a wide range of tasks of intelligent data processing is described. A high level architecture is shown, where Data Mining algorithms are considered as functional elements and a general or specialized co-processor as an accelerator for computer performance. An accent has been made on the necessity to provide compatibility of the librarian algorithms and the architecture of the parallel co-processor. The technique for correct comparative evaluation of the alternative hardware platforms in the framework of the proposed approach is presented.