Rachkovskij D. Methods of neural distributed representation of information for solution of Artificial Intelligence problems based on precedents and analogies

Dissertation is devoted to the development of theoretical and instrumental base of forming, processing and investigation of distributed representation of heterogeneous input information in the format of codevectors - binary sparse vectors reflecting similarity of information units by the value of dot product. The use of codevectors allowed for increasing computational efficiency and quality of solution of classification and information retrieval tasks. Unified approach to the solution of artificial intelligence problems, unified format of information representation, unified methods of its storage and processing allowed standardization and optimization of software and hardware means for the application systems. The developed methods provide widening of the application range of case-based and analogical reasoning, increasing efficiency of artificial intelligence systems and their scalability in performance and memory with increasing volumes and complexity of information.