The Creation of a Neural Computer with Parallel Extensible Architecture to Solute Intelligent Problems of Identification in Real Time
Tech Area / Field
- INF-COM/High Performance Computing and Networking/Information and Communications
3 Approved without Funding
Belarussian State University of Informatics and Radioelectronics, Belarus, Minsk
- National Academy of Sciences of the Republic of Belarus / Institute of Informatics Problems, Belarus, Minsk
- Fraunhofer Institute Zerstörungsfreie Prüfverfahren / Branch Lab Dresden, Germany, Dresden\nFraunhofer Institut Nachrichtentechnik. Heinrich-Hertz-Institut, Germany, Berlin\nSlovak University of Technology / Faculty of Materials Science and Technology, Slovakia, Trnava
Project summaryProject purpose: Developing and constructing a neural supercomputer prototype intended for solving problems of identification in the process of searching for objects images in large databases.
Target setting. There is a number of typical problems connected with processing of great volumes fuzzy, incomplete, or statistical data. These problems cannot be effectively solved in the frames of traditional universal computer architecture and traditional software DBMS. Consider an example. Increasing threats of terrorism for the society demand a person to be controlled and monitored fast and efficiently. Modern level of IT developments allows to identify an inpidual using biometric data received in the process of analysis his/her face images, fingerprints, eye irises. Searching for the most similar images in database involves multiple repetitions of the procedure of identification. The problem is that one procedure of identification and drawing final conclusion (at intelligent processing, for example, to 100 signs) for modern computers equals 1-10 milliseconds. Obviously, the search for 1,000,000 items of information in the database will take 1,000 – 10,000 seconds. It’s inadmissible long for the applications where a central node of data storage with the remote authorized access is used. Therefore, the problem of designing parallel neural computers to provide fast access to the necessary biometric data information is urgent.
Influence of the offered project on progress in the development field. The present project is based on the project of the participants original theory which allows to develop a neural processor, and on its basis to create a neural supercomputer with high efficiency degree of solving the problems of identification, with high level of the computer unification to various applied problems. The neural supercomputer that is being developed will not have industrial analogues. It will compete with perspective samples by the criterion of price-quality. Objective testing of efficiency of the above mentioned supercomputer will be organized during the realization of the project. Universality and a rather low cost of the future neural computer prototype will allow to have a wide market for introduction, for example, in criminalistics, in medicine, in biology, in electronic libraries etc. to create of electronic catalogues and provide fast access to the necessary information. Taken separately neural processors can be used as inexpensive, but effective intellectual decision-making modules for embedding to various measuring devices.
Project Participants. The project will be carried out at the Belarusian State University of Informatics and Radioelectronics (BSUIR) - 8 persons and the United Institute of Informatics Problems of the National Academy of Sciences of Belarus (UIIP NASB) — 3 persons. In BSUIR sufficient experience in developing hardware-software complexes for solving problems of identification and recognition in real time is accumulated. The project participants from UIIP NASB are experts in the field of new information technology development of computer processing, analysis, recognition and identification of cartographical images for decision-making in guidance and management systems.
Expected results. Development category — applied researches, technology development.
In the process of four problems solving the following results will be received:
- The original scientific concept of neural supercomputers construction, including mathematical models of neural solutions, neural computer architecture, test databases and techniques of correct comparative estimations of neural computer efficiency.
- The neural processor experimental sample in the form of a hardware module produced on the basis of FPGA, with the software of the bottom level. A pilot batch of 10 same modules capable of being combined into a cluster to provide the set productivity by parallel calculations will be made.
- The neural supercomputer prototype in a kind of neural processors cluster, connected to the server computer, with the system software of top level. Experimental system of person identification and fast search of images in biometric databases.
- An experimental system of person identification and fast searching for the biometric images in databases. In the process of solution of the given problem comparative experimental estimations of efficiency and productivity of the neural supercomputer and the personal supercomputer "SKIF-TRIADA" developed at incorporated UIIP NASB will be received.
Thus, as a result of the project the new technology for the neural computers construction, applicable at an industrial scale with the possibility to be used in various subject domains will be developed. Also the neural supercomputer prototype on the basis of which the intelligent system of identification is realized will be developed.
Project results application.
Theoretical impact of the project results — creation of original methodology for the neural processors construction, for the effective solution of intelligent problems of identification, clasterization, classifications, recognition, decision-making, fuzzy management.
Practical impact — creation of the operating samples presented in the form of original neural processor modules, a neural supercomputer prototype, system and applied software. The whole hardware-software complex will allow to demonstrate the efficiency of the developed technology and also to offer it to potential customers for introduction.
Spheres of application of the developed technology — the developed neural supercomputer will be applied in the state organizations which are responsible for safety of citizens to struggle against criminality and terrorism, to create systems of operative access to biometric data information files: by a photograph, by a soundtrack, by fingerprints, by an eye iris, by signatures, etc. Neural supercomputers can be employed in the research centers and universities to carry out researches connected with modeling of intellectual systems and neural networks of large dimension where supercomputers economically inexpedient. Neural processors as separate modules can be used in various subject domains where substantial increase in productivity as compared with realization on a universal PC, a microcontroller or a digital Signal Processor (DSP) is required. For example, the new neural processors can be applied in chemical analyzers, in devices of non-destroying control (as an intellectual kernel for decision-making in real time), in systems of identification of authenticity of passports, currency banknotes, etc.
The project compliance with the purposes and problems ISTC. The present project
- will give possibilities to participants of the project, connected with the weapon, to redirect their abilities at researches aimed at struggle against criminality and terrorism;
- will allow to integrate participants of the project into an international scientific community by means of close interaction with collaborators, joint publications, participation in international scientific and technical conferences;
- will contribute to solving national and international problems in the field of designing new computer architectures to support of intellectual technologies;
- will help to carry out transition for project participants to independent development in the conditions of the market, by creating new competitive technology, suitable for industrial introduction and commercial distribution.
Data about the volume of work.
Duration of the project – 27 months, with total amount of 2800 man-days. During the work four problems will be solved. The first, the second and the third will be solved by BSUIR. The fourth – jointly, by BSUIR and UIIP NASB.
Foreign collaborators role. Neural processors and a neural supercomputer Efficiency testing. Solving their own problems of intellectual processing on the basis of the neural supercomputer. Joint advancement of the project to the market of information technologies. Carrying out joint scientific seminars and publication of results.
Technical approach and methodology.
- Participants of the project will develop the generalized model of the classifier capable of adaptation in a wide range of functionality: from a usual formal neuron to a block of an indistinct logic conclusion. On the basis of the generalized model the architecture of a neural processor with flexible adaptation to a wide class of concrete applied problems of recognition will be created, a set of instructions and a programmed control principle of neural computer will be developed.
- In the process of working out of the structure of the neural processor, the speed of data input, the number of processor elements and the capacity of memory for storage of adjusting factors neurons will be optimized. By means of programming neural processor it will be possible to flexibly adjust neurons functions, their number and the topology of connections in a neural network. Thus, it will be possible to increase the number of processed informative signs beyond all bounds. At desire, it is possible to solve any arbitrary number of problems of identification in a parallel-serial mode on one module neural processor.
- Construction of a neural supercomputer means: assemblage of K same neural processor modules in a cluster, connection of the cluster on the standard interface to the server computer, and working out the system software for a server. The server will accept and operate inquiries about solving the problems of identification, to send the entrance data in the form of vectors of informative signs to the cluster, to receive and interpret the results of identification (or the results of search through a database) and to send them to the clients.
- The final stage of the neurocomputer efficiency in intellectual problems solving will be the realization of the experimental system of the person biometric data identification. To this effect the method of identification by analysis of distances between anthropometrical points on the portrait was chosen as a basic one. As a result of full-scale experiments it will be demonstrated that the neural supercomputer efficiency will increase linearly in the process of identification problems solving subject to the number of neuroprocessors in the cluster. Productivity of 5 modules of the neural supercomputer will already surpass productivity of the distributed network of 10 personal computers and will be comparable with the productivity reached on a personal supercomputer "SKIF-TRIADA" (with the configuration of 4 nuclear cluster, 100-500 GFlops).
The International Science and Technology Center (ISTC) is an intergovernmental organization connecting scientists from Kazakhstan, Armenia, Tajikistan, Kyrgyzstan, and Georgia with their peers and research organizations in the EU, Japan, Republic of Korea, Norway and the United States.
ISTC facilitates international science projects and assists the global scientific and business community to source and engage with CIS and Georgian institutes that develop or possess an excellence of scientific know-how.