ABSTRACTS OF ARTICLES OF THE JOURNAL "INFORMATION TECHNOLOGIES" N. 7, 2014

To the contents

V. I. Levin, Professor, e-mail: vilevin@mail.ru Penza State Technological University

Wiping Differential Calculus and its Application

A generalization of the classical differential calculus on the Newton-Leibniz function with interval uncertainty. In these functions, independent and dependent variables are defined as intervals of possible values. Construction of the new calculus algebra essentially uses interval numbers. Additionally, we use the concept of the limit interval function that is similar to the concept introduced limit normal function. The basic concept is the concept of the proposed calculation interval derivative. It is usual for the initial interval function as well as the concept of a classical derivative for normal function. However, the properties of interval derivative significantly different from those of the classical derivative. This is due to the laws of honors algebra interval number of laws of algebra of real numbers. It is proved that the existence of an interval derivative at some point it is necessary and sufficient that in some neighborhood of all the values of the independent variable initial interval function were non-degenerate intervals. An explicit expression is derived from the interval, interval function:

Introduced the concept of interval iteratively higher derivative (n-th) order as a derivative of the derivative of (n - 1) th order. In combination with the expression (1) the derivative of the 1st order is sequentially allows to obtain explicit expressions interval of the 2-nd, 3-rd and subsequent orders. Proved that interval derivatives 2-nd, 3-rd and all subsequent orders exist for the same necessary and sufficient condition that the interval derivative of the 1st order. Obtain estimates of interval derivatives of various orders in the form of intervals, lower and upper bounds are expressed through the lower and upper boundaries of the initial interval function and its independent variable. Possible applications of interval derivatives in economics, sociology and technology. Is an example.
Keywords: interval, interval function, function calculus, interval derivative, interval computing, nondeterministic differential
calculus

P. 3—10

A. A. Varfolomeeva, Student, V. V. Strijov, Researcher, Computer Centre PAS, e-mail: strijov@gmail.com

An Algorithm for Bibliographic Records Parsing Using Structure Learning Methods

The paper solves the application problem of structured texts segmentation, namely each segment of a bibliographic record must correspond to its filed type of the BibTeX format and each record must correspond to its bibliographic type.
This problem arises due to the existence of different standards for bibliographic records: an algorithm for determining the types of fields of bibliographic records, which is independent of the specific standards of their composition, should be proposed.
To solve the problem of determining the field type the method of constructing matrix "objects" and matrices "answers" is proposed. The authors offer an algorithm of a bibliography lists parsing using the structure regression method, and the optimization problem of regression model's parameters is also solved. According to the results of fields' segmentation bibliographic types of the records are clustered. The quality of the constructed model is investigated using a collection of non-parsed bibliography lists. In the paper it is shown the proposed algorithm has good quality of segmentation and clustering, if it has sufficient training sample.
Keywords: text parsing, structure learning, structure regression, segmentation, features selection, clustering

P. 11—15

A. K. Skuratov, Director, Directorate of State Scientific and Technical Programmes, D. E. Koshkin, Assistant, Moscow State University of Radio Engineering, Electronics and Automation, e-mail: minin89@mail.ru

Comparison of 12 Data Clustering Algorithms Applied to the Problem of Texts Clustering

In this paper made a comparative analysis of 12 data clustering algorithms applied to the problem of text clustering. Comparison based on the computational complexity of algorithms and their features and limitations. Article contents description of such algorithms as k-means, "Support Vector Machine" (SVM), Expectation-Minimization (EM-algorithm), CLIQUE, clustering with slope algorithm (CLOPE), parallel merging algorithm for adaptive finite intervals (pMAFIA), fuzzy c-means, the minimum spanning tree algorithm (MST), ROCK algorithm, CURE algorithm, WaveCluster algorithm, DBSCAN algorithm. For every algorithm was made little historic introduction, then described logic and idea of algorithm, and then gave features and limitations of algorithm.
At the end of the article was made comparison of algorithms and gave table with compute complexity, form of clusters, features and limitations of algorithms.
Keywords: clustering data, clustering algorithm, similarity metrics

P. 16—22

A. P. Karpaenko, Prof., M. K. Sakharov, Graduate Student, e-mail: max.sfn90@@gmail.com, Bauman Moscow State Technical University

Multi-Memes Global Optimization Based on the Algorithm of Mind Evolutionary Computation

A subject of the paper is the hybrid global optimization algorithms based on the concept of so-called meme meta-heuristic algorithms of search engine optimization. As the basic algorithm for multi-memes hybridization we use the algorithm of Mind Evolutionary Computation (MEC). Aim of the paper is to develop a multi-memes algorithm based on MEC algorithm for the class of loosely coupled distributed computing systems, as well as research the effectiveness of its consistent implementation on number of test problems of global optimization. We state the problem of global unconstrained optimization, present the algorithm MEC as the basic algorithm for hybridization, offer a hybrid algorithm HMEC, consider consistent software of this algorithm, present the results of a study of its effectiveness.
Keywords: global optimization, hybridization, meme, mind evolutionary computation

P. 23—30

A. N. Rodionov, Leading Researcher, Computer Centre Of Far-Eastern Branch of RAS, e-mail: ran@newmail.ru

Modeling and Implementation of "Is Part of Relation at a Set of Databases Composite Entities

Any domain of interest incorporates entities which consist of other entities. The letter may be both simple, indivisible entities and composite entities. The "is part of (IPO)" relation is always arisen on the set which integrates such entities. The implementation of given relation is become compulsory if it needs for the application domain under consideration.
In this article we research IPO relation to find out the set of the functional dependencies FDc, which exist between the members of the set above-mentioned. For this purpose it utilizes a graph as a most suitable formal construction. In data models the structure of composite entities are often presented by means of the objects referred as week entities. The relations, which correspond to a week entities in a relational model, comprises the owned functional dependencies FDw. Comparison of FDc with FDw shows that FDc I FDw In this case the difference of FDw and FDc sets may consider as a new integrity constraint class for relational models. These constraints must be fulfilled for all relations where accommodate data about the structure of composite types. Finally, we propose the algorithm for checking of derived constraints and formulate the new problems in this avenue of research.
Keywords: entity, composite type, relation, integrity constraints, conceptual and logical models

P. 31—36

Yu. V. Polishuk, PhD, Associate Professor of Computer Security mathematical software and information systems, e-mail: youra_polishuk@bk.ru, T. A. Chernyh, PhD, Associate Professor of Computer Science, Orenburg State University

On the Methods of Implementing the Concept of a Single Source

The most widespread methods of implementing the concept of a single source are considered. The advantages and disadvantages of these methods are listed. Provides a method of implementing the concept of a single source using a data warehouse based on work with semistructured information content of documents. The mathematical description and graphical presentation of a semistructured content model document are considered. As an example, the model describes the "Manual workstation" from the document package documentation. The structure of this document was developed in accordance with GOST 19.505—79 "Operator's Manual. Requirements for content and design". Proposed in the implementation of the concept of a single source combines the advantages of traditional methods for constructing systems of this type and provides additional control of correctness of factual content of documents due to restrictions imposed by the document models
Keywords: single source publishing, semistructured content

P. 37—41

L. A. Kozlova, Senior Lecturer, N. K. Trubochkina, Professor, e-mail: ntrubochkina@hse.ru, Higher School of Economics, Moscow

Information Technology in English Linguistics — Visualization of Grammar Rules

The article describes the method of presenting and studying English grammar using information technology. Sequential memorizing of large blocks of text describing the rules of verb tense formation is replaced by remembering a simple image, ie visualization of grammar rules.
The authors presented a dictionary, as well as the language of symbols, that facilitate and accelerate the memorization of English grammar rules. The "graphic" information is in the table, where each tense gets its finished graphic image and becomes easy to understand and memorize quickly. The method of studying English grammar for non-English speaking students was developed and tested in an Intermediate level group. The results showed high efficiency of storing large amounts of information. The table of integrated visualized rules can be recommended as a supplementary teaching material in the course of English grammar for non-English speaking students. Further development of the approach of optimal visualization of text blocks in English will improve the computer translation system by reducing time in addition to the educational effect.
Keywords: interdisciplinary study, information technology, English grammar, grammar rules, visualization, essentially-temporal processes, vocabulary and language of symbols, compression of information

P. 42—49

B. D. Zaleshchanskiy, Prof., A. P. Sviridov, Prof., O. A. Pavlova, Graduate Student, E. A. Shalobina, Graduate Student

Probabilistic and Statistical Strategy of Quality Ensurance of Personnel Training in Sotsio-Tehnical Systems by Means of Full and Partial Testing Optimisation

Periodical strategies of vocational training quality management are examined with consideration of forgetting infectivity and economical parameters (training costs, material losses due to personal errors). Strategies are based optimal planning of repetitions and preventive knowledge renewal. There initial data: forgetting infectivity, costs of full and partial testing, minimal probability of correct answer to the random chosen test. Aplicability of this strategies for full and partial testing optimization is shown.
Examples of strategies application are given for periodicity optimization of full and partial students knowledge checking in of Pascal (TP) and C programming course. With this purpose forgetting intensity is obtained for each of 44 task in TP and 51 tasks in C. On this base forgetting intensity for partial checking is calculated and summary forgetting intensity of all tasks for full checking.
Keywords: strategies, personnel training, forgetting infectivity, test costs

P. 50—54

V. Yu. Osipov, Leading Researcher, e-mail: osipov_vasiliy@mail.ru, Saint-Petersburg Institute for Informatics and Automation of RAS

Recurrent Neural Network with Structure of Layers in the Form of the Double Spiral

Purpose. Search approaches to eliminate excess storage and to empower recognition of dynamic signals in recurrent neural networks. Methods: The proposed approach is based on well-known models and methods of information processing in recurrent neural networks (RNN) with operated synapses. To justify the proposed approach used method of mathematical modeling. Results: improved method of processing information in a bilayer RNN operated synapses. Recommended by changing the attenuation functions of synapses endow the network layer structure in the form of a double spiral. Features of the implementation of a neural network with the structure disclosed. Its capabilities compared with known solutions. It is shown that the recurrent neural network with the structure of the layers in the form of a double spiral has the benefits of intellectual processing dynamic signals. Practical relevance: empowering recurrent neural network proposed structure can significantly reduce the redundancy of storing information. Recognition results of signals are better expressed in space and time through a change in the network settings. The developed method is useful if you create associative perspective of intelligent machines and systems.
Keywords: recurrent neural network, structure, double spiral, information processing

P. 56—60

O. V. Mandrikova^{1, 2}, Prof., Head of Laboratory, Yu. A. Polozov^{1, 2}, Researcher, e-mail: up_agent@mail.ru,
¹ Institute of Cosmophysical Researches and Radio Wave Propagation FEB RAS
² Kamchatka State Technical University

Approximation and Analysis of Ionospheric Parameters Based on a Combination of Wavelet Transformation and Neural Networks Groups

The paper suggests a method of approximation and analysis of ionospheric parameter time variation based on the combination of multi-scale wavelet decompositions and neural network groups. Calculation solutions to determine ionospheric parameter time series components, to form neural networks and to combine them into groups are described. In order to approximate a series smoothed component, a three-layer neural network of signal direct transmission was developed; approximation of detailing components is carried out on the basis of neural network groups. Analyzing approximation errors in ionospheric parameters, anomalies are detected.
Testing of the method was carried out on ionospheric critical frequency fOF2 data for 1969—2009, obtained at "Paratunka" observatory (IKIR FEB RAS, Paratunka, Kamchatskiy krai). On the basis of estimation of approximation errors in the recorded ionosphere critical frequency time series, anomalies were detected during increased solar activity as well as before and at the time of strong earthquakes in Kamchatka. The suggested method of ionospheric data analysis combined with other methods will allow us to increase the quality of evaluation and control of ionosphere state and detection of anomalies occurring during ionospheric disturbances.
Keywords: wavelet transformation, neural networks, the critical frequency of the ionosphere, anomalies, earthquakes

P. 61—65

A. I. Galushkin, Head of Laboratory, e-mail: neurocomputer@yandex.ru, Moscow Institute of Physics & Technology

The Back Propagation Error Method and Russian Works on Neural Networks Theory

In this paper, the significant role of works by Paul Verbos and other American writers in the development of algorithms of readjustment of the multi-layer neural network weight coefficients is described. Gist and role of Russian works in this field are noted in this paper, as well as comparison of Russian methods and back propagation (back propagation error), and the prospects of both directions for promising neural computers using memristors. The structure of approaches to the synthesis of multilayer neural networks developed in the 60s of the last century is noted for multilayer neural networks with consistent cross and feedback with such features as:

- Continuum of the number of classes;
- Continuum of number of solutions;
- Continuum number of features.
- Continuum of the number of neurons in the layer.

Developed in Russia approaches' focus on specific hardware implementations of neural computers based on actual restrictions on the weights of neural networks is marked.
It is also noted that methods of multilayer neural networks readjustment developed in Russia in the 60th years of the last century correspond to various operating modes:

- In case of unequal probabilities for different classes of appearance;
- For classification and clustering modes;
- Training with the teacher who has the final qualification, that allows making algorithms for the classification and clustering unified;
- For non-stationary images;
- For controlled errors first and second types in decision-making;
- For non-diagonal matrices in allocating losses to the images of a particular class.

Formulated and partially solved the problem of the choice of initial conditions in the procedure of neural network coefficients' readjustment for some practical problems.
Outlined ways of development of readjustment methods of multilayer neural networks for advanced neural computers using memristors.
Keywords: multilayer neural networks, the theory of neural networks, readjustment of the neural network weight coefficients, neural computers, memristors, back propagation

P. 66—76

D. A. Boronnikov, Head of Departament, D. V. Pantiukhin, Assistant, S. V. Danko, Student

Neural Networks Algorithm of Spatial Relief Data Organization

Neural networks algorithm of spatial relief data organization and its implementation in MATLAB language are developed. Experimental studies on the data on terrain of the Ozernyi mining and processing plant showed that the neural network has successfully memorized and generalize input information about the terrain (110 149 spatial points), with an error less than 0,5 meters. Compressing ratio on the original data is about 12 times.
Keywords: neural networks, spatial data organization, GIS, subsoil use

P. 77—80

To the contents