Dr. Vitaly A. Yashchenko senior research worker of the Institute of the Mathematical Machines and System of the National Academy of Sciences of Ukraine is engage in the development of the intelligent systems and intelligent robots on the base of the new class neural-like growing networks. Yashchenko V. A. is a member of the Association of the intelligent system users, managing editor of the magazine "Mathematical Machines and Systems", scientific secretary of the section "Mathematical Machines and Systems" of the scientific seminar on the problem "Cybernetic" of the NAS of Ukraine, assistant professor of the chair MMEER of the Cybernetic faculty of T. Shevchenko National University.
BUSINESS ADDRESS
Institute of the Mathematical Machines and System of the National Academy of Sciences of Ukraine. 03187, Ukraine, Kiev-187, Academician Glushkov Prospect, 42, tel. (044) 5267045, fax (044) 5264091, Email: mis[at]immsp.kiev.ua
RESEACHES
On the basis of the analysis of the scientific ideas, reflecting the appropriateness in constructing and functioning the biological structures of brain, and also the analysis and syntheses of the knowledge worked out by different directions in Computer science, V.Yashchenko founder of the theory of the new class neural-like growing networks not having the analogy in the worldwide practice.
New class - neural-like growing networks consists of onelayer, multilayer, multidimensional onelayer, multidimensional multilayer neural-like growing networks (n-GN), and also onelayer, multilayer, multidimensional onelayer, multidimensional multilayer receptor-effector neural-like growing networks (ren-GN).
Neural-like growing networks give the possibility to form meaning (notions), as objects and connections between them as constructing the network itself. With such, every notion gets separate component of the network as tops connected with other networks. In general it corresponds the reflected in the brain structure, where every noting is presented by the particular structure and has its own marking symbol. Practically the network was free from limits on the number of neural elements in which it is necessary to set the corresponding information, i.e. to construct the network itself, representing the given subject field. Beside it, the network gets the higher semantical clearness owing to creation not only the connections between neural elements, but the elements themselves as such, i.e. there is not only simply constructing the network by setting the notion structures in midst of neural elements, but creation of midst itself, as equivalent to the memory midst. Such the neural-like growing networks are the most available apparatus for modeling mechanisms of the purposeful thought as performing particular psycho-physiological functions.
In neural-like growing networks the information is kept in consequence of its reflection in the structure of the network. Multidimensional neural-like growing networks is represented by multilevel, multidimensional structure, reflecting the structure of the described objects. Information about objects and their classes is represented by ensembles of the associatively interconnected tops distributed on the structure of the net. The input of the new information in the network causes the process of its structure constructing ( redistribution of the connections between already existing and again arising tops ) with simultaneous excitation of the neural elements. In the result of this process the inclusion of the described object into the class, to which it belongs, is going on, or the new class of the objects is formed. So the classification and choosing the common attributes of the objects is carried out. Algorithm of the network construction establishes automatically the associative connections between descriptions of the objects accordingly their attributes. The description of the object or the class of the objects is located in some part of the network that lets to carry out various operations of associative search effectively. Profitability of the information representation in n-GN is carried out owing to compression of the information on each its level and representation of the identical combinations of attributes of several objects by one common subset of tops of a network. The training of the network is carried out simultaneously with their construction according to rules of construction and functioning of a network.
The important property of the ren-GN is the opportunity to form managing actions on the external environment (i.e. to train network to develop managing signals in the effector zone), according to the knowledge acquired by a network in the result of accumulation, analysis, classification and integration of the information from the external worked (i.e. processing of the information in the receptor zone ren-GN). In case of the hardware realization ren-GN this property gets yet more meaning, especially under construction of robotechnical systems, owing to an opportunity of the parallel reception of the information on receptor field from perceiving bodies, spreading and distribution of the managing actions to the external world.
Beside this, that is no less impotent, the neural growing networks in consequence their principal difference from neuron networks are completely free from such lacks as problem of a local minimum, the temporary instability and paralysis of the network, inherent to structures according to ideology of neurons networks.
It is worked out the architecture of the multiprocessor system with homogenous matrix multidimensional neural growing network. The system is intended for the decision of the complex tasks, requiring the fulfillment of the sizable volume of the parallel calculations, processing of the big information files, forming of the attribute space and description of the classes, image recognition, testing and diagnostics of the technical facilities, knowledge base forming, creation of the intelligent systems with powerful hardware supporting in the real time from different areas of the human activity: biology, medicine, military business, meteorology, geology, nuclear physics, criminal law, production managing, ecology and others, and also for creation of the intelligent robots.
THE RESEARH PROJECTS
The researches were holding in the limits of the competitive state scientific-technical program 6-03-01 "Competitive computer facilities for the decision of Ukrainian problems". Highly productive computers and problem oriented general purpose complexes of State Committee on Science and Technology of Ukraine on them 6.03.03/027-93 "Development and creation of multiprocessor computer facilities with homogeneous matrix neuroensemble structure" and also in the limits of scientific technical program of Ukrainian Academy of Sciences on the theme "Researches and development of the theoretical found dations of the creation of the high intelligent computers with using new neural growing networks".
PEDAGOGICAL ACTIVITY
Reading of the lectures on special course "Artificial intellect and intellectualization of computers on neural growing networks" for the 4-th and 5-th courses of the faculty Cybernetic in T.Shevchenko National University.
PUBLICATIONS
More than 35 publication on the them Artificial Intellect, including 12 certificates to invention.
The list of main publications:
V.A. Yashchenko Neural growing networks as a model of an intelligent multimicroprocessor system with a neuron-ensemble structure, Cybernetics and Systems and Analysis, Vol. 30. No. 3, 1994. С. 348 - 364
V.A. Yashchenko Multidimensional neural growing networks and computer intelligence // Cybernetics and Systems and Analysis, Vol. 30. No. 4, 1994. С. 505 - 517
V.A. Yashchenko Dual architecture of an intelligent multimicroprocessor system with a multidimensional neural-ensemble structure, Cybernetics and Systems and Analysis, Vol. 30. No. 5, 1994. С. 657 - 665
V.A. Yashchenko Receptor-effector neural growing networks as an efficient tool for intelligence modeling. I Cybernetics and Systems and Analysis, Vol. 31. No. 4, 1995. С. 524 - 530
V.A. Yashchenko Receptor-effector neural growing networks as an efficient tool for intelligence modeling. II Cybernetics and Systems and Analysis, Vol. 31. No. 5, 1995. С. 711 - 717
Yashchenko V.A., ShagunV.A. Receptor-effector neurosimilar growing networks //International Journalon Information Theories & Application1995, Vol.3, No.7 p. 36 - 41
V.P.Klimenko and V.A. Yashchenko V.M.Glushkov and new-generation computers Cybernetics and Systems and Analysis, Vol. 32. No. 4, 1996. С. 488 - 492
Z.L. Rabinovich and V.A. Yashchenko Approach to modeling thought processes by neural growing nets Cybernetics and Systems and Analysis, Vol. 32. No. 5, 1996. С. 615 - 624
Yashchenko V., Shagyn V. Representation of the Knowledge of Receptor Effector Neural Growing Networks.//Conference INFORMATION THEORIES & APPLICATIONS ITA'96 Fourth International May, 12-22 1996 Troyan & Sofia, Bulgaria
Yashchenko V.А., Vishnevsky V.V., Morozov А.А. Use of neural-like growing networks for the recognition and categorizations of tests an оncotest. Cybernetics and Systems and Analysis, Vol.33. No 2, 1997. С.23-30.
V.A. Yashchenko Bionic approach to knowledge representation in intelligent systems. I Cybernetics and Systems and Analysis, Vol. 34. No. 1, 1998. С. 1 - 11
Vitaly Yashchenko. Neural-like growing networks - new class of the neural networks. In Proceedings of the International Conference on Neural Networks and Brain Proceeding, pages 455 -458, Beijing, China, Oct. 27-30' 98.
Morozov A.A., Yashchenko V.A. Intellectualization of the COMPUTER on the basis of a new class neural-like of growing networks. Kiev, ГКПП "Circulation", 1997, с.125.
