Goderdzi Lezhava

Inductive logic plays a fundamental role in the functioning of the human brain (and the brain in general). The notion of an induction output processor proved highly fruitful for modeling some functions of the human mind. The functional scheme of the induction output processor was developed based on the hypothesis of the basic operator of inductive logic. The processor allows the sensory information presented in the form of multidimensional vectors to be compared in a certain way to the information loaded into the processor's RAM matrix, which is the "knowledge" gained by the cybernetic system as a result of "previous experience" or learning. The result of the comparison is generated at the output of the processor in the form of distances defined in the sensor space, or "similarity measures". One of the modifications to the processor was actually implemented for a practical unmanned navigation system.

The results of the work, which were conducted at the Institute of Cybernetics, were reported. On the plantations of western Georgia, statistical material was collected in large numbers.On its basis, the statistical characteristics of the raw layer were built and a bank of images of shoots was created.A simulation research and demonstration program was created, which made it possible to transfer part of the research from the plantation to the laboratory. The concept of a tea-picking system of a new generation was developed. A prototype of the assembly system was created and tested at the stand.

A hypothesis about the basic operator of inductive logic is proposed. The processor that performs the basic operation of inductive logic is reviewed. A hybrid intelligent system is also considered, in which the role of an adaptive system is performed by an inductive inference processor. Intelligent procedures, the implementation of which became possible through such a hybrid system, are considered.

A hybrid intelligent system (GIS) is considered, which consists of two subsystems that are different in nature: a) a sensory information processing subsystem based on inductive inference methods, which by its nature represents a parallel information processing system, and b) a symbolic-sign information processing subsystem based on the method -dax deductive inference, which by its nature is a system of sequential information processing. GIS is designed to implement computationally intensive artificial intelligence algorithms in real time.

Unlike the methods of deductive inference, which are well developed and have powerful hardware for implementation in the form of mainframe computers and a developed art of programming, there is practically no theory or universal hardware for inductive inference systems. True, software implementation of inductive inference procedures using a computer is possible, but due to the sequential nature of deductive methods, this is associated with cumbersome calculations, in some cases it excludes the possibility of obtaining solutions in real time, which is unacceptable for artificial intelligence problems.

In the RAM of the inductive inference subsystem, a set of descriptions of the state of the environment is fixed, which forms the “a priori knowledge base” of the subsystem in the form of multidimensional vectors. When the inductive inference subsystem receives current sensory information in the form of a vector of the same dimension, interpreted as a description of the “current state of the environment”, the input vector is compared in parallel and almost instantly with all vectors of the “knowledge base” and the comparison results are calculated in in the form of quantities defined as a "measure of similarity" of the description of the current state of the environment with all descriptions presented in memory. The operation of determining the "measure of similarity" i.e. the comparison operation is the basic operation of inductive logic; with its help, you can build such artificial intelligence meta-procedures as description space clustering, concept formation, insight, pattern recognition, etc.

Combining inherently different inference systems into a single hybrid system is a promising area of research; it opens up opportunities for the development of the next generation of "intelligent systems".

The fact that the content side is completely ignored in Shannon's theory of information is one of the main reasons hindering progress towards the creation of artificial intelligent systems. In the proposed approach, information is considered as a process of cognitive interaction. The receiver consists of subsystems of inductive and deductive conclusions. The inductive inference subsystem, through the sensory system, perceives the physical impact corresponding to the “description of the environment”. An optoelectronic processor developed by us is used as an inductive output subsystem; as a subsystem of deductive inference - a conventional computer. The deductive inference subsystem operates with concepts formed by the inductive inference system and selects adequate responses to the current state of the environment from its own repertoire of possible actions. The joint work of the subsystems of inductive and deductive inference makes it possible to significantly approach the understanding of the structure of the human intellect; will come closer to solving the problem of creating a semantic theory of information. We have identified two types of information impact - two types of information: "current relevant information" and "information that forms the knowledge base". Methods have been developed to quantify these types of information impact. On the basis of large-scale experiments, the viability of this approach was shown.

The problems of creating an optical supercomputer and artificial intelligence are among the most pressing problems of recent decades. They have been defining the nature of the current research at the Institute of Cybernetics of the Georgian Academy of Sciences for years. It is known that the optical signal as a transmitter of information has significant advantages over the electrical one. However, despite great efforts, an optical computer could not be built. Nevertheless, optical means turned out to be extremely effective in computationally intensive tasks as nodes of preliminary - pre-computer processing of information. There was an in-depth reason for this. Electronics is effective for deductive systems, which has a consequential character and ineffective for inductive system, as they do not allow the realization of the basic advantage of such conclusion –greater potential of parallel of operation.

Considering the above, an inductive inference processor was developed. The processor made it possible to develop a new approach to such a fundamental problem of cybernetics as the problem of quantitative assessments of semantic information.

The inductive inference processor fixes in memory descriptions of the environment that are important for the stability of the system.marking and clustering of descriptions take place in parallel. These operations can be treated as training.Some environment descriptions trigger a system response, but have little or no effect on the system's continued operation. The information contained in such descriptions is what we have called "currently up-to-date". The information contained in the descriptions, the fixation of which significantly affects the further operation of the processor, we called: "Information that forms the knowledge base".

In the nineties, the opinion matured that the artificial intelligence system, similar to the human brain, should consist of two subsystems: an inductive system, which by its nature is a parallel action system, and a deductive inference system, which by its nature is a sequential action system. The role of the latter is well performed by a conventional computer. The situation is worse with the inductive inference subsystem. Attempts to use neural networks have not yielded practical results. As is known, the Institute of Cybernetics has considerable experience in creating practical real-time intelligent systems. In these systems, the role of pre-computer processing subsystems is usually performed by special purpose parallel optical devices. Based on the accumulated experience, a general purpose induction inference processor was developed for artificial intelligence systems.

The development of computer technologies has made it possible to solve a number of problems of artificial intelligence. At the same time, a large set of important problems remained unsolved, requiring real-time solution, the solution of which requires the use of inductive methods. Intellectual activity is carried out through two systems that are different in nature – through the joint coordinated work of systems of inductive and deductive conclusions. Therefore, any artificial intelligent system must contain two information processing subsystems: a) a sensory information processing subsystem based on inductive inference methods, which by its nature is a parallel information processing system, and b) a symbolic-sign information processing subsystem based on deductive inference methods. , which by its nature is a sequential information processing system. To solve a specific practical problem, we have developed a processor that imitates the operation of inductive inference. The processor made it possible to see a number of problems in a new way and, in particular, the problem of quantifying semantic information. In the RAM of the processor, the “learning material” is represented in the form of multidimensional vectors. “The learning material is a set of concepts and their corresponding probabilities. In practice, we have a statistical ensemble; the uncertainty of the ensemble changes when new learning material arrives. This change in uncertainty can be used as a measure of the semantic information contained in the incoming new learning material. 

The inductive inference processor allows fixing individual concepts in its memory in the form of delimited clusters.These clusters, together with the corresponding probabilities, form a statistical ensemble, with some uncertainty. New information changes ensemble; changes and its indefiniteness. This concept-related change in uncertainty can be used as a measure of the incoming (new) semantic information.

Doctoral Thesis Referee

Master Theses Supervisor

Doctoral Thesis Supervisor/Co-supervisor

Scientific editor of monographs in foreign languages

Scientific editor of a monograph in Georgian

Editor-in-Chief of a peer-reviewed or professional journal / proceedings

Review of a scientific professional journal / proceedings

Member of the editorial board of a peer-reviewed scientific or professional journal / proceedings

Participation in a project / grant funded by an international organization

Participation in a project / grant funded from the state budget

Patent authorship

Membership of the Georgian National Academy of Science or Georgian Academy of Agricultural Sciences

Membership of an international professional organization

Membership of the Conference Organizing / Program Committee

National Award / Sectoral Award, Order, Medal, etc.

Honorary title

Monograph

Handbook

Research articles in high impact factor and local Scientific Journals

Publication in Scientific Conference Proceedings Indexed in Web of Science and Scopus