ABSTRACTS OF ARTICLES OF THE JOURNAL "INFORMATION TECHNOLOGIES".
No. 12. Vol. 23. 2017
To the contents
A. A. Razorenov, Ph. D. Student, e-mail: email@example.com, National Research University Higher School of Ecomonics (HSE), Moscow, Russia
Mathematical Foundations of Applying the Theory of K-Representations to Developing Algorithms of Executing Natural Language Instructions by Computer Systems
The paper continues the author's publication in Informational Technologies 2017, No. 10 describing the principles of applying the theory of K-representations to developing methodological foundations of designing computer systems processing natural language (NL) instructions. The goal of this paper is to formally define a binary relation of a new kind on semantic structures used as the ground of a new method of executing NL instructions by applied systems. The paper introduces a mathematical model describing the transformations of semantic representations (SRs) into another SRs. Besides, the paper proposes a mathematical model of forming the scripts/programs in the language of the controlled system for executing instructions expressed in the form of SRs. The properties of the introduced relations are investigated, and an experience of applying the developed method to designing a NL interface of a file manager is set forth.
Keywords: natural language processing, semantic parsing, theory of K-representations, SK-language, K-string, natural language interface, pattern matching relation, semantic-command dictionary, Natural Language Commander
Yu. A. Zack, Dokt.-Ing., E-mail: firstname.lastname@example.org, http: optimorum.de
About One Algorithm of Fuzzy-Regression Analysis
Methods for solving the Fuzzy regression analysis problem are proposed under conditions when the input, output variable and free term of the regression equation are represented by fuzzy sets of the most general kind, and the regression coefficients are real numbers. The sum of the squares of the average weighted coordinates of the minimum and maximum abscissas of different sections of the fuzzy-set membership functions of the output variable and their estimations by the Fuzzy regression model, and also the coordinates of the centers of gravity of the membership functions of these Fuzzy sets are used as the approximation criteria. Deterministic equivalents of this class of fuzzy regression models, estimates of the adequacy of the constructed regression equations, and a computational scheme of the algorithm for solving the problem are obtained.
Keywords: Fuzzy-regression analysis, fuzzy sets, non-numeric statistics, linguistic and Boolean variables, least-squares method
A. L. Stempkovskiy, Academician, Director, e-mail: email@example.com, N. N. Levchenko, Ph. D., Head of the Department, e-mail: firstname.lastname@example.org, A. S. Okunev, Ph. D., Chief Researcher, e-mail: email@example.com, A. V. Klimov, Senior Researcher, e-mail: firstname.lastname@example.org, D. N. Zmejev, Researcher, e-mail: email@example.com, Institute for Design Problems in Microelectronics (IPPM RAS), Moscow, 124365, Russia
Programming of the Molecular Dynamics Task in the Dataflow Computing Model
The scaling of HPC is often constrained due to excessive use of global barriers, especially when the load is non-uniform and varying, not to mention the costs to the barrier itself. The proposed in the article the approach for solving this problem relies on the dataflow computing paradigm. This approach is described by the example of an algorithm for the molecular dynamics task. In this task, instead of barriers, counting of nearby particles is introduced, which occurs asynchronously and in parallel with the main work. Synchronization relies only on local interactions between computational cores. The article also describes ways to increase the efficiency of molecular dynamics task execution on the parallel dataflow computing system (Buran) using various optimization techniques.
The approach to the solution of the molecular dynamics task considered in the article actually uses only local communications between neighboring computational cores (cuboids). Moreover, different particles are processed, basically, independently and asynchronously. The only synchronization is related to the counting of the exported and imported particles. Therefore, it can be expect that this algorithm has no obstacles for unlimited scalability in solving the molecular dynamics task on the PDCS.
During the experiments, results were obtained that attest to the high degree of scalability of the molecular dynamics task on the cycle-accurate model of PDCS, and also on the emulator of the system on the "Lomonosov" supercomputer. For the PDCS, the optimal number of particles processed per computational core, at which scaling does not decrease, is about 103 particles, which is actually two orders of magnitude greater than the results obtained on traditional cluster systems. However, in this case, the PDCS remains universal.
On the basis of the universal parallel dataflow computing system "Buran", by changing the architecture of computing elements, it can be possible to proceed to the creation of special computers for the molecular dynamics task.
Keywords: molecular dynamics, scaling, asynchronous computing, barrier synchronization, parallel computing, dataflow computation model
P. V. Kazakov, Lecturer, e-mail: firstname.lastname@example.org, Bryansk State Technical University, Bryansk, 241035, Russia
The Model of Parallel Computing Based on CUDA Technology for Multi-Objective Genetic Algorithms
There is the effective way for improving of multi-objective genetic algorithms (MOGA). It is using the technologies of parallel computing. Theses technologies allow to reduce time of the big population processing, the nondominated solution search, the calculation of criterion values, etc. There are three most known approaches for the implementation of the parallel computing in multi-objective genetic algorithms: the master-slave model, the island model and the massive parallel model. These models allow to obtain different goals in parallel MOGA, for example, the master-slave model reduces the computing time for criterion values calculation, the island model efforts an opportunity for independent population processing on different computers and the last model allows to obtain the extreme performance in MOGA running on multiprocessors systems. The most perspective is integrated the advantages of all models ofparallel MOGA in the hybrid model. The some problems appear with software/hardware platforms for a such hybrid model. One of solutions there is to use CUDA technology for high parallel computing on graphic processing unit. In article the original hybrid parallel model based on CUDA for MOGA are considered. This model differs from other that operations of MOGA are distributed between CPU, GPU for its loading balance, crossover and mutation operators implemented on some population topologies: ring, grid, hypercube. The experiments for evaluation hybrid CUDA-model were conducted on the benchmark problems DTLZ with different number of criterions. The sequential and parallel versions of multi-objective genetic algorithms SPEA2, NSGA-II were compared. The result demonstrated that speedups of parallel MOGA range from 5,6 to 11,7 depending on number of criterions. Also it is found the use of hybrid parallel CUDA-model for MOGA allowed to improve the quality of Pareto-set on some indicators. For future work the influence of different population sizes on the search time and the quality of solutions will be studied.
Keywords: Multi-objective optimization, Pareto's principles, Pareto front, multi-objective genetic algorithms, parallel computing, CUDA technology
V. V. Kureichik, Professor, e-mail: email@example.com, V. V. Bova, Assistant Professor, e-mail: firstname.lastname@example.org, D. V. Leshchanov, Student, e-mail: email@example.com, Southern Federal University
Semantic Search Model for Knowledge Management Systems Based on Genetic Procedures
Nowadays, one of the main functions of knowledge management in modern intelligent information systems is the semantic search for knowledge elements that have a distributed character of representation. The article presents the innovative approach to the implementation of the semantic search model. Such approach is applicable to large volumes of distributed sources of knowledge. Authors propose a model of semantic search and evaluation of the proximity of knowledge elements in the ontology of the subject domain. Particularly, the ontology is defined as a semantic network. The relevance of knowledge is assessed in the way of the proximity to some metric of the similarity estimation of knowledge objects. The objects are included in the meta-descriptions of ontology elements in intelligent information systems. On the one hand, semantic meta-descriptions are considered as the sets of triplets "subject-predicate-object".
On the other hand, it is presented in terms of the ontology of the subject area of problem-oriented information systems, and in the terms of the search image of the user's query as well. Authors developed a complex model for calculating semantic proximity for calculating measures of semantic similarity and coherence of knowledge. The model uses a set of interpreted metrics of taxonomic and associative dependencies, objects in the meta-descriptions of the search query. The algorithm for evaluating semantic proximity is based on evolutionary procedures and operators of the genetic search for optimal solutions. Such procedures allow excluding non-informative or insignificant descriptions of knowledge elements from consideration, as well as control the speed of learning by setting the magnitude of the proximity threshold. The article considers a genetic algorithm based on the use of analogs with evolutionary reproduction processes, crossing-over, mutation, and natural selection. A series of experiments have been performed for analyzing of the developed algorithm. The obtained data have confirmed the theoretical significance and prospects of the proposed approach application and allowed to determine the optimal values of the parameters of the algorithm. The proposed approach is useful in the following tasks: development of advanced semantic search technologies, ontological design, and integration of knowledge to support the task of forming semantically concentrated knowledge in knowledge management systems in the conditions of information distribution.
Keywords: semantic similarity, ontology, semantic network, semantic meta-model, genetic algorithms, genetic operators, knowledge management systems
E. N. Reshetova, Teacher, e-mail: firstname.lastname@example.org, National Research University Higher School of Economics (HSE)
Teaching Experience the Academic Discipline "Group Dynamics and Communication in Professional Software Engineering Practice"
The article is devoted to defining goals, objectives and content of training on the professional-oriented discipline "Group dynamics and communication in professional software engineering practice", which is included in the curriculum of preparation of bachelors in the direction "Software engineering". We propose an innovative approach to teaching the discipline, which was based on team development of software projects.
Keywords: software engineering, team project, group dynamics, team development
R. A. Korotchenko, Ph. D., Senior Researcher, e-mail: email@example.com, V. I. Il'ichev Pacific Oceanological Institute, Vladivostok
Application of Singular Spectral Analysis to Detection of Acoustic Signals of Gray Whales
The paper presents a methodology and an algorithm for processing hydroacoustic monitoring data that are used for automated detection of gray whale signals. Using singular spectrum analysis in acoustic signals processing makes it possible to efficiently separate the nonstationary impulse components that are typical for gray whale signals from the ambient noise. A number of examples are given for real signals processing. The possibility of using the methodology to detect other localized signals is shown.
Keywords: singular spectral analysis, hydroacoustic signals, monitoring, gray whales
B. I. Filippov, Ph. D., Associate Professor, e-mail: firstname.lastname@example.org,
FSFEI of HPE Novosibirsk State Technical University, Novosibirsk, 630073, Russia
Research and Development of the Device of Protection against Mistakes for System of Transfer of Images on the Hydroacoustic Communication Channel
The most effective remedy of providing a high noise stability of difficult system is entering of the redundancy necessary for the detection and error correction arising during the work of system and its elements. Theoretical base of effective use of the entered redundancy is the theory of noiseproof coding. As show pilot studies, statistical characteristics of hydroacoustic communication
channels (HACC) have the analogs in short-wave, VHF and other radio channels with variable parameters. Therefore the principles and methods of protection against mistakes developed for these channels can be used and in the systems of information transfer using GAKS is final taking into account specific properties of distribution of acoustic signals in the water environment. Therefore the subject of work is represented urgent. For reasons for the choice of parameters of the developed device of protection against mistakes (DPM) researches of statistical characteristics of a flow of mistakes in HACC of the existing equipment of the hydroacoustic system of communication (HASC) by handling of the provided records (files) of samples of a personnel of experimental transfer of images with use of this equipment were executed. By comparison of samples of the transferred and accepted personnel of the image massifs of mistakes in the channel of the existing GASS equipment were received. In total four files with massifs of mistakes were received: errMass 1.txt, errMass 2.txt, errMass 3. txt, errMass 4. txt. Each of them contains about 3-105 binary symbols. These files were used for determination of statistical characteristics of a flow of mistakes in a real hydroacoustic communication channel, and also for statistical testing of the developed DPM. In work are considered: the principles and algorithms of increase in reliability of digital transmission of images in HASC; the algorithm of implementation of DPM in the conditions of hindrances in a HASC equipment communication channel is developed; the state-of-the-art review of methods of increase in reliability of transfer of blocks of digital information on HACC is made. Results of testing of DPM for digital transmission of images in HASC give the grounds to draw the following conclusions: offered DPM provides digital data transmission from the module of the autonomous underwater robot on the module of the serving vessel with required reliability; the received dependences of probability of mistakes at the DPM module exit from quality of a discrete communication channel allow to determine requirements to the relation signal/noise on an
entrance of a reception path of a ship part and, respectively, the requirement to the size of capacity of an acoustic signal of the sending device depending on depth of its stay.
Keywords: hydroacoustic communication channel, hydroacoustic system of communication, digital data transmission, increase in reliability, noiseproof coding, statistical characteristics of a stream of mistakes, device of protection against mistakes
E. A. Andreeva, Professor, e-mail: email@example.com, I. S. Khramov, Graduate Student, e-mail: firstname.lastname@example.org, Tver State University
Optimization of the Artificial Neural Network with Allowing for Delay
This article discusses the discrete optimal control problem with delay in state and control vectors for artificial neural networks, the algorithm for constructing an optimal solution for this problem. In the process of work, a neural network model is considered, in which there is a lag in the state vector, and the control functions are the weights of the neural network. To construct an approximate optimal solution, it is proposed to use the gradient projection method. The described methods of modeling and learning the artificial neural network of the general topology described by the system of recurrence relations with delay are universal, which allows them to be used in various fields to solve a wide range of applied problems.
Keywords: artificial neural networks, optimization of management, necessary conditions for optimization
To the contents