Институт проблем информатики Российской Академии наук
Институт проблем информатики Российской Академии наук
Российская Академия наук

Институт проблем информатики Российской Академии наук




«INFORMATICS AND APPLICATIONS»
Scientific journal
Volume 4, Issue 2, 2010

Content | Bibliography | About  Authors

Abstract and Keywords.

ON TASK FLOW PLANNING IN COMPUTATIONAL RESOURCE SYSTEMS.

  • M.G. Konovalov   IPI RAN, mkonovalov@ipiran.ru

Abstract:  The problem of flow distribution analysis, optimization, and pricing in shared computational resource systems is examined. The appropriate literature review is given. An approach to mathematical models construction is suggested where task flows being described as dynamic balance equations and quality of service relations are used. The subjects in the systems possess own strategies of behavior and purpose individual aims in terms of cost and quality of service. Distributed gradient algorithmis one of possible system member strategies. A numerical example is given and model development and employment for future trends are discussed.

Keywords:  computational resource systems; flow distribution; quality of service; cooperative behavior

NONPARAMETRIC ESTIMATION OF BAYESIAN CLASSIFIER ELEMENTS.

  • M. Krivenko   IPI RAN, mkrivenko@ipiran.ru

Abstract:  The problemof constructing an empirical Bayesian classifier, providing recognition of the text, where some symbols have different picture sizes, is considered. A combined method of constructing an evaluation of Bayesian classifier is proposed. The method includes nonparametric kernel estimation and parametric estimation with the help of the density of normal distribution. This combined assessment allows to deal effectively with the task of handling small amounts of training set. Productivity of the proposed ideas is illustrated by an example of recognizing the real text.

Keywords:  Bayesian classifier; combined multivariate density estimation; adaptive kernel estimation; text recognition

SOLVABILITY PROBLEMS IN THE PROTEIN SECONDARY STRUCTURE RECOGNITION.

  • K. V. Rudakov   Computing Center of RAS;Moscow Institute of Physics and Technology, rudakov@ccas.ru
  • I. Yu. Torshin   Russian Center of the Trace Element Institute for UNESCO, tiy135@yahoo.com

Abstract:  The purpose of the work is to develop a formalism for the application of the algebraic approach to recognition of the protein secondary structure. Paper presents rigorous formal description of the problem and considers its solvability, regularity, and locality. Key terms for the analysis of locality, such as a neighborhood, mask, mask system, monotony, and irreducibility of the mask systems were proposed. An algorithm for constructing nonredundant mask systems was formulated. The formalism has allowed to formulate a correct description of the hypothesis of the local character of the dependence of the secondary structure on the primary and to obtain constructive criteria of the solvability of the problem.

Keywords: algebraic approach; the secondary structure of protein; bioinformatics

ASYMPTOTIC PROPERTIES OF RISK ESTIMATE OF WAVELET-VAGUELETTE COEFFICIENTS THRESHOLDING IN TOMOGRAPHIC RECONSTRUCTION PROBLEM.

  • A. V.Markin   Department ofMathematical Statistics, Faculty of Computational Mathematics and Cybernetics, M. V. Lomonosov Moscow State University, artem.v.markin@mail.ru
  • O. V. Shestakov   Department ofMathematical Statistics, Faculty of Computational Mathematics and Cybernetics, M. V. Lomonosov Moscow State University, oshestakov@cs.msu.su

Abstract:  Tomographic image reconstruction problem using wavelet-vaguelette decomposition is considered. Consistency and asymptotic normality of risk estimate of vaguelette coefficients thresholding are studied.

Keywords:  wavelets; tomography; thresholding; risk estimate; limit distribution

ANALYSIS OF A LINK PROTOCOL WITH A GENERAL CONTENTION WINDOW BACKOFF FUNCTION.

  • A. S. Lukyanenko  Helsinki Institute for Information Technology HIIT, Aalto, Finland, firstname.secondname@hiit.fi
  • E. V.Morozov   Institute of Applied Mathematical Research, Karelian Research Centre RAS; Petrozavodsk State University, emorozov@krc.karelia.ru
  • A. Gurtov   Helsinki Institute for Information Technology HIIT, Aalto, Finland, gurtov@hiit.fi

Abstract:  A set of medium access (backoff) protocols, where the collision resolution window for a station depends on the number of successive collisions, is analyzed. Under mild common assumptions for the network properties, a general backoff protocol is studied. An optimal criterion yields a backoff protocol possessing the minimal service time. An asymptotic analysis is offered for the unboundly growing number of stations. Bounded and unbounded protocols are considered in the analysis. Finally, a continuous time model is introduced as an extension for the slotted model, which allows slots of different sizes.

Keywords:  data communications; performance analysis; backoff protocol; stability; medium access control

DEVELOPMENT OF PARALLEL HEURISTIC ALGORITHMS OF WEIGHTS COEFFICIENTS SELECTION FOR ARTIFICIAL NEURAL NETWORK.

  • O. V. Kryuchin   G.R. Derzhavin Tambov State University, kryuchov@gmail.com

Abstract:  A gradients alogirithm of artificial neural network and heuristic algoritms QuickProp and RProp which are based on it are described. Possible applications of cluster systems have been considered.

Keywords:  artificial neural network; heuristic algorithms of teaching; cluster systesm

APPLICATION OF THE COORDINATE METHOD OF COMMUTATED NEURAL NETWORK FRAGMENTATION FOR TRAFFIC REDUCTION.

  • S. Y. Stepanov   Moscow State Technological Institute STANKIN, cympak shade@rambler.ru

Abstract:  The problem of increasing traffic in scaling commutated neural network, amethod and algorithmof its solution are outlined. An example of the developed algorithm is also presented.

Keywords:  commutated neural network; scaling; traffic

ON ASYMPTOTIC BEHAVIOR OF THE POWERS OF THE TESTS FOR THE CASE OF LAPLACE DISTRIBUTION.

  • V. E.Bening   Faculty of ComputationalMathematics and Cybernetics, M.V. Lomonosov Moscow State University, bening@yandex.ru
  • R.A.Korolev   Faculty of ComputationalMathematics and Cybernetics, M.V. Lomonosov Moscow State University, stochastique@gmail.com

Abstract:  A formula for the limit of the normalized difference between the power of the asymptotically most powerful test and the power of the asymptotically optimal test for the case of Laplace distribution was proved. Due to the nonregularity of the Laplace distribution, the logarithm of the likelihood ratio admits nonregular stochastic expansion, and an analog of Cram‚er condition is not valid for the sign statistic which is the basis of the asymptotically optimal test. Then direct use of theorem 3.2.1 from [1] or theorem 2.1 from [2] is difficult, and in the present paper, their proofs for the case of Laplace distribution are revisited.

Keywords:  power function; conditional probability measure; conditional moment; Laplace distribution

AN IMPROVEMENT OF THE KATZ–BERRY–ESSEEN INEQUALITY.

  • M. E. Grigorieva   Department of Mathematical Statistics, Faculty of Computational Mathematics and Cybernetics, M.V. Lomonosov Moscow State University, maria-grigorieva@yandex.su
  • I.G. Shevtsova   Department of Mathematical Statistics, Faculty of Computational Mathematics and Cybernetics, M.V. Lomonosov Moscow State University, ishevtsova@cs.msu.su

Abstract:  The upper estimates of the absolute constant in the Katz–Berry–Esseen inequality for sums of independent identically distributed random variables with finite absolute moments of order between 2 and 3 are sharpened and an alternative inequality with sharpened structure and evaluated constants is proposed.

Keywords: central limit theorem; Katz–Berry–Esseen inequality; Lyapounov fraction

LINGUISTIC FILTERS IN STATISTICAL MACHINE TRANSLATION MODELS.

  • E. B. Kozerenko   IPI RAN, kozerenko@mail.ru

Abstract:  The paper focuses on the problems of linguistic filters development for statistical machine translation and advancement of parallel texts alignment methods for enhancing precision and adequacy of translations. Statistical and heuristic models of alignment and translation are considered. The solutions proposed are based on the hybrid grammar formalism comprising the linguistic rules and probability characteristics of language structures.

Keywords:  statistical models; machine translation; parallel texts; alignment; linguistic filters