Slipchenko S. Numeric and symbolic information processing based on distributed representations in Artificial Intelligence tasks.

The thesis is devoted to the development and research of methods for distributed representation and processing of symbolic and numeric information. Methods of building binary representations for analogical access, mapping and inference have been developed. For the task of analogical access, the proposed methods increased recall on 11-22% and precision by 3-4 times compared to the best existing symbolic methods. The proposed methods of mapping and inference demonstrated results corresponding to the results of psychological tests and existing methods. In the task of predicting existence of chemical compounds those methods gave results at the level of the best available systems (94,8%-99,91%). Analytical characteristics for a known coarse coding method of numeric vectors of R. Prager have been obtained, i.e. code density, overlap, resolution, and others. Those characteristics allow an efficient selection of coding parameters for application problems. An experimental investigation on artificial and real data has been fulfilled.