## Sections

### Neuroinformatics - 2014

SESSION 1

Monday, January 27 16:15 – 18:00

Lecture-hall Акт. зал

Chair: Prof. LITINSKII LEONID

Neural network theory

1. MALSAGOV M.YU., KRYZHANOVSKIY V.M.*Scientific Research Institute for System Analysis, Moscow*

**Based on neural networks searching tree with iterative traversal and stop criterion**

The paper offers an algorithm that extends the binary tree search algorithm so that it can deal with distorted input vectors. Perceptrons are the tree nodes. The algorithm features an iterative solution search and stopping criterion.

2. SHATS VLADIMIR

*Independent investigator, St. Petersburg*

**The Index Method of Mashine Learning**

The paper proposes the index method of machine learning, where the principle is implemented multi-level information processing which used by animals. It examines the properties of the indexes attributes, which display the value of the attributes on a finite set of integers. The frequency indices for objects of each class of learning sample it is determined. These frequencies serve to determine the most likely class of test objects. The features of the method in solving individual problems are considered.

3. A.M. SOLOVYEV, M.E. SEMENOV

*Voronezh State University*

**Artificial neural network with hysteresis activation function**

Two-layered artificial neural network with hysteresis activation function was developed. Training algorithm was proposed. Behavior of present neural network was investigated by the example of image’s classification problem.

4. KARLOV I.A., KOSHUR V.D.

1

*National Research University "Higher School of Economics", Moscow*

2

*Siberian Federal University, Krasnoyarsk*

**Neural-Fuzzy control of hybrid model of missing data estimation**

This paper introduce a new and original approach to missing data estimating. The main idea of hybrid method and the realization of some steps of constructing a hybrid model is considered. Special attention is paid to the issue of effective control of hybrid model with usage of adaptive network-based fuzzy inference systems. Conducted a series of experiments that demonstrated the effectiveness of this method on the individual data sets.

5. * VOLOSKOV D.S., MASLENNIKOVA YU.S., BOCHKAREV V.V

*Kazan (Volga region) Federal University*

**Neural networks training algorithm based on the maximum likelihood method for time series prediction**

This paper is devoted to the improvement of neural networks training algorithms for short-term time series prediction. The training algorithm based on the maximum likelihood method is proposed. The algorithm allows improving a forecast precision for time series with non-normal errors distribution. The proposed method was utilized for predicting model time series and dynamics of LAN-network traffic.

6. KISELEV MIKHAIL

*The Chuvash state university named after I. N. Ulyanov*

**HOMOGENOUS CHAOTIC NEURAL NETWORK SERVING AS A CONVERTOR FROM ASYNCHRONOUS TO SYNCHRONOUS SIGNAL CODING**

Nervous system utilizes two main schemes for information coding. The asynchronous scheme assumes that presence and intensity of a stimulus are coded using total intensity of spike stream without exact spike synchronization while highly synchronized sequences of spikes are used in the synchronous scheme. It is demonstrated in the paper that conversion from the former scheme to the latter on can be performed by homogenous chaotic neural network. This effect is observed even without network training and does not require synaptic plasticity.

7. * GORBACHENKO V.I. , ZHUKOV M.V.

*Penza State University*

**Applying a parameters identification method and radial basis functions networks to solve coefficient invers problems of mathematical physics**

The mess-free method is introduced to solve coefficient inverse problems here. It based on both a parameters identification method and radial basis functions networks. The iteration method is used to regularize a solution. A quantity of iterations acts as a regularization factor. It is defended by a residual between the right and the left parts of additional conditions. The approach efficiency was confirmed by example of coefficient inverse problems for elliptic equation.

SESSION 2

Tuesday, January 28 14:00 – 15:30

Lecture-hall 406

Chair: Prof. RED'KO VLADIMIR

Adaptive behavior and evolutionary modelling

8. * IVASCHENKO G.S., KORABLEV N.M.*Kharkov national university of radio electronics, Ukraine*

**The hybrid method of short-term time series forecasting based on the model of clonal selection**

This paper proposes the combined method of short-term time series forecasting using artificial immune systems. A model of the prediction based on the model of clonal selection and the case based reasoning method. Was performed a comparative analysis of the effectiveness of the proposed model and the traditional methods of time series forecasting.

9. A. KABYSH, U. DZIOMIN, V. GOLOVKO

*Brest State Technical University, Belarus*

**HOW TO CONTROL COLLECTIVE BEHAVIOR IN MULTI-AGENT SYSTEM**

We present new algorithm how to build multiagent system with desired collective behavior starting only from individual agents. The algorithm contains two phases. At first phase, «decentralized learning» - each agent learns individual policy separately from other agents. In the second phase, agents coordinate it’s behaviors with each other to create a common multi-agent system with desired behavior. We purpose a coordination method based on virtual leader. Only leader has view of global state of the system and performs the adjustment policies of individual agents to desired global behavior.

10. U. DZIOMIN, A. KABYSH, V. GOLOVKO, R. STETTER

1

*Brest State Technical University, Belarus*

2

*University of Applied Sciences Ravensburg-Weingarten, Germany*

**EFFICIENT CONTROL OF MULTIWHEEL MOBILE ROBOT BASED ON REINFORCEMENT LEARNING**

This paper presents an application of the multiagent reinforcement learning approach for the efficient control of a mobile robot. This approach is based on a multi-agent decomposition applied to multi-wheel control. The robot’s platform is decomposed into driving modules agents. Next we introduce virtual leader, which reinforce control every wheel relative to other wheels. The modified Q-learning used to adjust policy of every module agent. The power reward policy with common error reward is adjusted to produce efficient control.

11. * NEPOMNYASHCHIKH V.A., OSIPOVA Е.А., RED’KO V.G., SHARIPOVA T.I. , BESKHLEBNOVA G.A.

1

*I.D. Papanin Institute for Biology of Inland Waters, Russian Academy of Sciences, Borok, Yaroslavl region*

2

*Scientific Research Institute for System Analysis, Moscow*

**NAVIGATION MODEL ANIMALS IN THE MAZE**

This model is intended for understanding of the searching behavior of animals, in which the animal does not only aim at particular needs, but also aims to the study of the environment. Results of the biological experiment on the navigation fish in the maze are outlined. Computer simulations of the navigation system are described.

12. * MURATOV S.T., LAKHMAN K.V., BURTSEV M.S.

1

*The Moscow Institute of Physics and Technology (State University)*

2

*National Research Centre "Kurchatov Institute", Moscow*

3

*London Institute for Mathematical Sciences, United Kingdom*

**Neuroevolutionary synthesis of a mobile robot's controller in the generating sequences task**

Paper considers the task of automatic formation of the behavioral sequences by mobile robot in the environment with multiple goals. We utilized neurevolutionary algorithm based on neurons' duplication as the method for generation of robot's intelligent controller. We showed that stable behavioral strategies are produced in the course of evolution, that could be implemented even in case of significant changes of the robot's environment.

SESSION 3

Tuesday, January 28 15:30 – 17:00

Lecture-hall 406

Chair: Prof. DUNIN-BARKOWSKI WITALI

Neurobiology

13. I.E. MYSIN*Institute of Theoretical and Experimental Biophysics of RAS, Pushchino*

**Long-term plasticity of glutamatergic synapses**

This article is a review of a problem of synaptic plasticity. Main attention has been concentrated around the plasticity of the glutamatergic synapses, because this is the most common type of the excitatory synapses in the nervous system and plasticity of these synapses is the most better studied. The paper tells about modern view on the mechanism and role of plasticity.

14. V.V. SHAKIROV

*Scientific Research Institute for System Analysis, Moscow*

**THE EMERGING PROSPECTIVE TECHNOLOGIES FOR ARTIFICIAL INTELLIGENCE**

Prospective ways to solve particular problems are considered in brain reverse engineering inspired artificial intelligence. Advantages and disadvantages of the approaches are estimated. The tasks of modalities unification and optimization of existing AI algorithms are considered. The problem of specificity of particular AI algorithms is also treated in connection to the Moore’s law.

15. SMIRNITSKAIA IRINA

*Scientific Research Institute for System Analysis, Moscow*

**The distributed representation of space in superior colliculus, pulvinar nucleus and parietal cortex.**

The distributed representation of visual space in superior colliculus, pulvinar nucleus and parietal cortex is advansed, in which the signals of eye position in orbit is used to bound the separate parts of the whole visual field.

16. * SOLOVYEVA KSENIYA PAVLOVNA

*Scientific Research Institute for System Analysis, Moscow*

**HYPER-RING NEURAL ATTRACTORS**

Bump attractors might represent continuous variables in neural systems. Unlike usually treated associative memorizing networks we consider here me-thods of obtaining preformed networks with inborn connections, which get bump attractors. In particular, we explore networks, which attractors include many times more stable states, than the number of network’s neurons.

17. K.P. SOLOV’EVA, I.M. KARANDASHEV

*Scientific Research Institute for System Analysis, Moscow*

**TOPOLOGICALLY ORDERED POLYATTRACTOR NEURAL NETWORK OF SPIKING NEURONS**

One of the possible mechanisms for quantity representation in central nervous system are attractor neural networks. Continuous attractors of dimension 1, 2 or 3 often are used for modeling of continuous quantities. In the paper, we suggest a hybrid design containing a topologically ordered grid of 0-dimensional attractors using spiking neuron model. Characteristic distinctions of operating modes of such network from networks using 0-dimensional and multi-dimensional attractors are investigated.

18. W.L. DUNIN-BARKOWSKI

*The Moscow Institute of Physics and Technology (State University)*

**ON THE SET OF BRAIN COMPUTING FUNCTIONS**

Elementary brain computational functions, connected to Hopfield waves and neural attractors are described. It is concluded that an execution of computations should be performed by multiplexing them by phases of brain rhythms.

КАЛЕЙДОСКОП ИДЕЙ

Tuesday, January 28 17:00 – 18:00

Lecture-hall 406

Chair: Prof. TIUMENTSEV YURY

*Povolzhskiy State University of Telecommunications and Informatics, Samara*

**Neural network for identification of the user uses specifics of the typing**

In this paper, a method added security and user identification based on neural network technology. Material selection dimension of the input vector, the structure of the network and produced training. We prove the efficiency of the software for a fixed number of computer users.

20. VELTS SERGEY

*Bauman Moscow State Technical University*

**Neural network approach based on hierarchical temporal memory (HTM) for transaction fraud monitoring**

In this article an approach for plastic card operations modeling for detecting fraud actions is proposed. It is based on neural network architecture of hierarchical temporal memory which is able to handle spatial and temporal adjacency in the input data.

21. VEKSHIN N.L.

*Institute of Cell Biophysics, Pushchino, Moscow region*

**On the brief and long memories**

A critical examination of the arguments of the existence of two types of memory - short and long term are carried out. The analysis leads to the conclusion about their identity. All the basic facts can be explained by the model of holographic memory mechanism.

22. Y. NIKONOV

**Kauzogramma as a complex network**

Kauzogramma as a complex network It is assumed that the concept of hidden metric spaces complex networks can be used in in neuroinformatics. For example, for modeling of an assessment of remoteness of events in time. Events – nodes of a complex networkы (kauzogramma). The model uses the modification of the model of "popularity-similarity optimization» (E-PSO).

SESSION 4

Wednesday, January 29 10:00 – 12:00

Lecture-hall 404

Chair: Prof. KAZANOVICH YAKOV

Neurobiology

23. SONKIN K.M., STANKEVICH L.A., NAGORNOVA G.V., HOMENKO Y.G., SHEMYAKINA N.V.1

*Peter the Great St. Petersburg Polytechnic University*

2

*International Scientific Center "Arktika", FEB RAS, Magadan*

3

*Bekhtereva Human Brain Institute RAS, St.Petersburg*

4

*Sechenov Institute of Evolutionary Physiology and Biochemistry of the Russian Academy of Sciences, St.Petersburg*

**Analysis of possibility of EEG pattern of imagined and real one hand fingers movements recognition**

The problem of imaginary and real movements recognition on the EEG data is examined. The study on real and imaginary movements of one hand fingers discrimination is implemented. Spatial localization and duration of the interval for analysis is determined for the EEG signals used for classification. Neural network classifier with the preliminary regression analysis is implemented. The accuracy of the pattern recognition of real and imaginary thumb and forefinger presses is evaluated

24. * S.A. KOZHUKHOV, N.A. LAZAREVA, R.S. IVANOV, A.S. TIKHOMIROV, D.YU. TSUTSKIRIDZE, I.V. BONDAR

*Institute of Higher Nervous Activity and Neurophysiology of RAS, Moscow*

**Analyzing neuronal responses to orienting stimuli in primary visual cortex neurons**

Considered in the report will be responses of V1 neurons to bars of different orientations. All responses are characterized by spike density function and quantitatively estimated using principal component analysis. As a result, novel properties of neuronal responses were discovered.

25. MEDVEDEVA I.V., VISHNEVSKY O.V., SAFRONOVA N.S., KOZHEVNIKOVA O.S., SUSLOV V.V., KULAKOVA E.V., SPITSYNA A.M., AFONNIKOV D.A., KOCHETOV A.V., ORLOV Y.L.

1

*Institute of Cytology and Genetics, Siberian Branch of the Russian Academy of Sciences, Novosibirsk*

2

*Sechenov University*

**Genomic structure and context characteristics of genes with higher expression in brain**

We consider problems of statistical analysis of genes in brain cells based on data from high throughput sequencing and microarrays. Short review of main bioinformatics approaches and existing databases is given. We analyzed genomic orgasnization and context characteristics of genes differentially expressed in brain cells, including number of exons, alternative transcripts and non-coding RNA in gene structure.

26. ARKHIPOV VLADIMIR IVANOVICH, KAPRALOVA MARINA VLADIMIROVNA, GORDON RITA YAKOVLEVNA, PERSHINA EKATERINA VIKTOROVNA

1

*Institute of Theoretical and Experimental Biophysics of RAS, Pushchino*

2

*Institute of Cell Biophysics, Pushchino, Moscow region*

**COGNITIVE FUNCTIONS DURING NEURODEGENERATION, INDUCED BY EXCITOTOXIN**

Wistar rats were trained with food-procuring task before intrahippocampal microinjection of kainic acid in subconvulsive dose. It was shown, that neurodegenerative processes in the hippocampus, developed after kainic acid, deteriorate memory and learning; and disturbances increased along with increase of degeneration processes. Deterioration of memory retrieval were smaller and its recovery was faster if behavioral experiments started after 3 days, rather than 6 days after hippocampal damage. It is assumed that the expression of cognitive defects determined by compensatory mechanisms with one hand, and progressive structural and functional disorders in the hippocampal system, on the other.

27. SERGEY.V.BOZHOKIN, IRINA.B.SUSLOVA

*Peter the Great St. Petersburg Polytechnic University*

**WAVELET ANALYSIS OF THE MATHEMATICAL MODEL OF BRAIN ELECTROENCEPHALOGRAM BURSTS**

The brain activity bursts are simulated as a superposition of elementary non-stationary signals. Continuous wavelet transform (CWT) has been obtained analytically for the model of the EEG signal. The analysis of the spectral integrals in the given frequency range has been carried out. The quantitative parameters that characterize brain activity bursts have been introduced and calculated. The applications of the method to the analysis and classification of transients, describing the properties of the central nervous system, have been discussed.

28. * BAKHCHINA ANASTASYA VLADIMIROVNA, PARIN SERGEY BORISOVICH, POLEVAYA SAPHY ALEKSANDROVNA

1

*Institute of Psychology of Russian Academy of Sciences, Moscow*

2

*N.I. Lobachevsky State University of Nizhni Novgorod*

**Non-liner components of heart rate of drug addicts**

We consider the problem of finding physiological markers of drug addiction, the study of participation of the endogenous opioid system in the regulation of cardiac rhythm man. The data of the results of spectral analysis of nonlinear dynamics of the components of heart rate in drug users in the contexts of lying and making functional tests

29. * IRINA A. ISCHENKO, EUGENIA I. BELOVA, VIACHESLAV A. VASILKOV AND RUBEN A. TIKIDJI-HAMBURYAN

*A.B. Kogan Research Institute for Neurocybernetics Southern Federal University, Rostov-on-Don*

**LOCAL SYNCHRONY FEATURES OF DIFFERENT TYPES NEOCORTICAL NEURONS**

We discuss here the features of different types neocortical neurons involving in local synchrony. First, we provided the criteria for neuron identification during extracellular recording under visual stimulation. The results of these experiments reviled that neurons with «narrow» spikes and bursting activity first involve in synchrony. Then we used computational model to show that neurons, with «narrow» spike and noise current, respond to stimulus with more synchrony than «broad» spike neurons

30. A.V.CHIZHOV, S. RODRIGUES

1

*Ioffe Physical Technical Institute, Russian Academy of Sciences, St Petersburg*

2

*Plymouth University, United Kingdom*

**Simple model relating extracellular potential to membrane currents of neuronal population**

Simple models, linking intracellular parameters of neuronal activity to extracellular potential, are still debated. Basing on a neuron with a cylindrical dendrite and lumped soma, we propose a formula for extracellular potential as a function of synaptic and extrasynaptic membrane currents at soma and dendrite. This formula is applied to the firing-rate type model of two-compartment neuronal population.

31. N.G. BIBIKOV, I.N.PIGAREV, S.V. NIZAMOV

1

*N.N. Andreyev Acoustics Institute, Moscow*

2

*Institute for Information Transmission Problems (Kharkevich Institute), Moscow*

**The randomness of neuronal background activity at the periphery of the frog’s auditory system and in the cortex of sleeping cat**

Two functions that characterized the randomness of stochastic point time processes (the dependence of the Fano factor and Allan from the interval of analysis) was used to describe the statistical characteristics of the background firing of neurons in two different areas of the brain in different vertebrates (cochlear nucleus of the frogs and cerebral cortex of the sleeping cat). The growth of these functions was observed in majority of cells. It was more pronounced at the periphery of the auditory system. Refractoriness also was more clearly manifested on the periphery.

POSTER SESSION 1

Wednesday, January 29 12:00 – 13:00

Lecture-hall 404

Chair: Prof. KAGANOV YURI

Neural network theory

32. BELOV D.E., BOGOMOLOV Y.V.*P.G. Demidov Yaroslavl State University*

**Multilayer perceprton based on ΣΠ-neuron with alternative synapses**

In clause the model of multilayer network based on the model of sigma-pi neuron is investigated. The modification of sigma-pi neural network with alternative synapses is proposed and the learning algorithm based on backpropagation method is also described.

33. MILOVANOV A.V.

*Voronezh State University*

**Investigation of neural network model based on the System Hodgkin's - Huxley**

The results of mathematical modeling of the modified model of Hodgkin - Huxley, different from the classic by the presence of two parameters: s, u, are considered. We study the time dependence of the membrane potential of nerve cells in the squid variations in the values of parameters s, u. Calculated the boundary of the stability of limit cycles and the boundary of the stability of a permanent solution to systems that do not match. The conditions for the emergence of a stable limit cycle are considered.

34. ANUFRIENKO SERGEY EVGENIEVICH

*P.G. Demidov Yaroslavl State University*

**THE MODIFIED MODEL OF A SALTATORY CONDUCTION OF EXCITATION**

In this article, a model of conduction of impulses along a nerve fibers covered with myelin is considered. The modified model of the neuron-detector is used to describe the Ranvier captures. This model takes into account several types of gates that regulate the currents of sodium and potassium. The model of saltatory conduction contains the differential equations with delay and ordinary differential equations of. All results are obtained analytically.

35. SHIBZUKHOV Z.M.

*Moscow State Pedagogical University*

**Pointwise and aggregational correct operations on algorithms**

A class of recognition algorithms which are correct with respect to given aggregation functional of algorithm’s quality. A class of aggregationally correct operations on such algorithms, which transforms any finite tuple of aggregationally correct algorithms to new aggregationally correct algorithm, is discussed. By this way one can extend classes of basic recognition algorithms to extended classes, which are preserve correctness property in aggregate manner.

36. NAKONECNAYA SVETLANA VYACHESLAVOVNA, TIMCHENKO LEONID IVANOVICH, YAROVYY ANDRIY ANATOLIYOVYCH, KOKRYATSKAYA NATALIYA IVANOVNA

*State University for Transport Economy and Technologies, Kyiv, Ukraine*

**PRINCIPLES OF ORGANIZATION OF MULTILEVEL PARALLEL-HIERARCHICAL NETWORKS**

The principles of constructions of multilevel parallel-hierarchical network are examined. Unlike neural networks in multilevel parallel-hierarchical networks is present when processing in its branches computational algorithm that considerably increases their functionality. Suggested the models of parallel processing which represent a pyramidal process of converting numeric fields.

Neurobiology

37. V.A. DEMAREVA, S.A. POLEVAYA*N.I. Lobachevsky State University of Nizhni Novgorod*

**The dynamics of the role of interhemispheric asymmetry during English language acquisition**

The work is devoted to the search for psychophysiological markers of linguistic competence and individual optimal states for doing linguistic tasks in English by students of the 3rd and 9-10th forms. The study revealed the relationship between the lability, irritability and hemispheric stability and success of solutions of language tasks. The level of English language proficiency was significantly higher in students who had left hemisphere dominance.

38. POKROVSKY

*Sanct-Petersburg State University*

**COMPONENTS OF THE CAUSED POTENTIAL OF CORTEX OF THE BRAIN**

The problem of allocation of components of a various origin from records of the caused potential of bark of a brain is considered. The caused potentials usually are registered at many levels of bark at the same time; dependence of signals on time at different levels of depth of bark rather difficult. The work purpose – full division of the caused potential (CP) into components.

39. KRYLOV A.K.

*Institute of Psychology of Russian Academy of Sciences, Moscow*

**Neural activity analysis by the symbolic dynamics method**

A method of description of the interspike intervals is proposed. The idea behind the method is based on the analogy with heartbeat intervals analysis by symbolic dynamics. The quantitative characteristics of the neural activity are presented according to the method.

Adaptive behavior and evolutionary modelling

40. RED'KO V.G., SOKHOVA Z.B.*Scientific Research Institute for System Analysis, Moscow*

**Agent-based model of a transparent economic system**

An agent-based model of a transparent economic system is designed. A community of interacting investors and manufacturers is analyzed. Functioning of the model is demonstrated. First results of computer simulation are represented. It is shown that the proposed scheme of interaction between investors and manufacturers is effective.

41. V.B. LEBEDEV

*Southern Federal University, Rostov-on-Don*

**INTEGRATION OF MODELS OF ADAPTIVE BEHAVIOUR OF THE BEER COLONY AND EVOLUTIONARY ADAPTATIONS**

The swarm algorithm paradigm on the basis of integration of models of adaptive behaviour of a beer colony and evolutionary adaptation is offered. Integration of models is reduced in creation of the hybrid agent of adaptive behaviour of a beer colony serially carrying out function and genetic algorithm. The integration essence consists that in the course of performance of search procedure alternation of separate procedures of beer and genetic algorithms is made, and agents exchange functions

42. B. K. LEBEDEV, O.B. LEBEDEV

*Southern Federal University, Rostov-on-Don*

**THE DECISION OF TRANSPORT PROBLEMS WITH TIME RESTRICTION ON THE BASIS OF MODELS OF THE ANT COLONY ADAPTIVE BEHAVIOUR**

The modified paradigm of an ant colony for the decision of a transport problem with restriction on time is considered. On the basis of the analysis of ant colony self-organizing model, methods and search engines of decisions are developed. Formation of routes is carried out on the basis of the parallel-serial approach. Distinctive feature of the presented algorithm is that the ant colony is broken into groups and formation of the concrete decision of a problem is carried out by group of ants. In comparison with existing algorithms improvement of results is reached.

Neural networks and cognitive sciences

43. BAZYAN ARA SAHAKI*Institute of Higher Nervous Activity and Neurophysiology of RAS, Moscow*

**THE NEURAL NETWORK MECHANISMS REALIZATION OF VOLUNTARY AND MOTIVATED BEHAVIOR, THE ROLE OF MIRROR NEURONS**

The neural network mechanisms are analyzed for the realization of voluntary and motivated behavior. Describes the role of mirror neurons in the realization of behavior. The mirror neurons are activated both when a particular action, and in monitoring the realization of this action. It is assumed that mirror neurons are involved in the processes of imitation, empathy and understanding of another's consciousness. That is, play a key role in leraining and social orientation.

44. V.V. LAVROV, A.V. RUDINSKY

1

2

*Center of System Consulting And Education "Synergia", St.Petersburg*

**STRATEGIES OF MANAGEMENT OF PROCESSING BY FRAGMENTARY SENSORY INFORMATION**

Information handling in brain is provided by manipulation of information fragments to use them as sign of the image and as arguments of decision mak-ing. We have revealed 3 strategies of management of information processing. Choice of strategy is conditioned by degree of the cut-in of emotional structures of the brain. Communicative system, providing issue of information between neurons in process decision making, is considered. We call attention on subservience of brain cortex to subcortex management.

SESSION 5

Wednesday, January 29 14:00 – 16:00

Lecture-hall 404

Chair: Prof. EZHOV ALEXANDER

Applications of neural networks

45. * KALINOWSKI ILYA ANDREEVICH, SPITSYN VLADIMIR GRIGORIEVICH*National Research Tomsk Polytechnic University*

**Face detection on video stream with the use of convolutional neural network**

There have been shown an implementation of the convolutional neural network for detection of faces in the images and video streams. Have been given the results of training and testing of the constructed neural network. The quality of the trained convolutional neural network with the algorithm of Viola - Jones to search for faces has been compared.

46. * MURAVYOV A.S., BELOUSOV A.A.

*National Research Tomsk Polytechnic University*

**Neural network for natural image quality evaluation based on the NSS model**

In this paper we present a neural network-based natural image quality evaluator. Statistical features obtained from the NSS model describing a reference image set are used as inputs. Training utilizes an evolutionary algorithm, which maximizes the correlation between neural network output value and user opinion scores. It is shown that this method can in some cases perform as well as some full-reference quality metrics while avoiding their inherent drawbacks.

47. * YURI N. LAVRENKOV, LUDMILA G. KOMARTSOVA

*Bauman Moscow State Technical University Kaluga Branch*

**Neural network system of generating random numbers to ensure secure transmission of digital information**

We propose a neural network algorithm to generate random numbers based on the use of multiple Hopfield neural networks. The description of the source of entropy, which is the basis of a stochastic process of recharge two tanks. Developed neural network logic required to break the dependency of random sequences. Substantiates the effectiveness of such systems is to organize the secure transfer of information using the scrambling.

48. * V.R. SHIROKY, I.N. MYAGKOVA, I.G. PERSIANTSEV

*Skobeltsyn Institute of Nuclear Physics Lomonosov Moscow State University*

**NEURAL NETWORK PREDICTION OF RELATIVISTIC ELECTRON FLUX IN THE OUTER RADIATION BELT OF THE EARTH**

The difficulty of prediction of the time series of relativistic electron flux in the outer radiation belt of the Earth is conditioned by the complexity and non-linearity of the magnetosphere of the Earth as a dynamic system, and by the features of the satellite data used. This study is devoted to elaboration of methods of data preparation and of neural networks design for implementa-tion of the prediction and for comparison of indicators of the quality of forecasts, with a horizon of one to four hours.

49. * EGORCHEV M.V., TIUMENTSEV YU.V.

*Moscow Aviation Institute (National Research University)*

**Training of neural network based semi-empirical models for controlled aircraft motion**

A simulation approach is discussed for controlled aircraft motion under multiple and diverse uncertainties including knowledge imperfection concern-ing simulated plant and its environment exposure. The main goal of the paper is an advance on semi-empirical dynamical models combining theoretical knowledge for the plant with training tools of artificial neural network field. A generation of training sets needed for these models is studied.

50. D.V. GRACHIKOV

*Voronezh State University*

**Applying model of biological neural network with hysteresis nature for image segmentation**

A model of biological neural network, based on the model of the biological neuron SA Kashchenko - VV Mayorov and memory model of hysteretic nature by AN Radchenko is investigated in the paper. It is shown that the use of biologically-based model allows to solve problems of computer vision. In particular, the algorithms of image segmentation using biological neural network model of hysteresis nature is proposed in the paper.

51. * S.A.DOLENKO, S.A.BURIKOV, K.A.GUSHCHIN, T.A.DOLENKO

1

*Skobeltsyn Institute of Nuclear Physics Lomonosov Moscow State University*

2

*Lomonosov Moscow State University*

**Application of Kohonen neural networks for analysis of composition of multi-component solution**

This paper presents the results of application of clusterization with Kohonen neural networks for analysis of an array of Raman spectra of multi-component solutions of inorganic salts, with the purpose of determination of the types of salts present in the solution. It is demonstrated that the approach is efficient in the whole, although the method of its application requires revision. The directions of future work are formulated.

52. * FILATOVA N.N., SIDOROV K.V., KHANEEV D.M.

*Tver State Technical University*

**Use of neurolike hierarchical structure for the classification of emotions' sign**

Examining the use of a hierarchical neurolike structure for the classification of electroencephalogram recordings, representing the variation in the sign of the emotional state of the human during the presentation of incentives of vari-ous emotional color. Describes a method of creating a multi-modal base and system of characteristics for describing of electroencephalograms and speech signals, also considered features of the construction algorithm and the results of the neurolike classifier work with a of fuzzy descriptions of electroencephalograms.

SESSION 6

Wednesday, January 29 16:00 – 17:00

Lecture-hall 404

Chair: Prof. YAKHNO VLADIMIR

Neural networks and cognitive sciences

53. VITALY M. VERKHLYUTOV, PAVEL A. SOKOLOV, VADIM L. USHAKOV1

*Institute of Higher Nervous Activity and Neurophysiology of RAS, Moscow*

2

**ACTIVITY LARGE SCALE CORTICAL NETWORKS WHEN COMPLEX VISUAL STIMULI VIEWING AND IMAGING**

Identified seven large-scale cortical networks when healthy subjects were viewed and imagined complex visual stimuli. Networks were stable in space and it does not depend on the experimental paradigms. Comparison of the identified networks and resting state networks showed that the task's networks are the recombination of the latter.

54. O.D. CHERNAVSKAYA, D.S. CHERNAVSKII, V.P. KARP, A.P. NIKITIN, D.S. SHCHEPETOV

1

*P.N. Lebedev Physical Institute of the Russian Academy of Sciences, Moscow*

2

*MIREA - Russian Technological University*

3

*Prokhorov General Physics Institute of the Russian Academy of Sciences, Moscow*

**THE VERSION OF THE THINKING SYSTEM ARCHITECTURE BASED ON THE DYNAMICAL THEORY OF INFORMATION: SOLVING THE CONCRETE PROBLEMS**

The process of solving the problems of recognition, classification and prognosis in the artificial thinking system, which is built on the basic principals of the dynamical theory of information and the concept of the dynamical formal neu-ron. It is shown that these problems should be solved by means of active participation of both subsystems which could be conventionally related to the left and right cerebral hemispheres. The effects are discussed that could be interpreted as a “sense of humor” of the artificial system.

55. D.S. CHERNAVSKII, V.P. KARP, A.P. NIKITIN, D.S. SHCHEPETOV, O.D. CHERNAVSKAYA

1

*P.N. Lebedev Physical Institute of the Russian Academy of Sciences, Moscow*

2

*MIREA - Russian Technological University*

3

*Prokhorov General Physics Institute of the Russian Academy of Sciences, Moscow*

**NATURAL-CONSTRUCTIVE APPROACH TO THE THINKING SIMULATION: ANALIZING THE DYNAMICS OF THE SYM-BOL FORMATION PROCESS**

The mathematical model of the process of symbol formation is considered within the frame of the natural-constructive approach to the thinking simulation. This model is based on the concept of the dynamical formal neuron. The dynamics of the competitive interaction in the localization processor is analyzed, with the “semi-exited” neuron states being shown to play an important role. A specific version of the dynamics of learning the inter-plate connections is proposed and shown to provide the self-organizing character of the whole process.

56. * V.A. OSINOV, D.G. SHAPOSHNIKOV

1

*A.B. Kogan Research Institute for Neurocybernetics Southern Federal University, Rostov-on-Don*

2

*Scientific Research Center of Neurotechnologies, Southern Federal University, Rostov-on-Don*

**Assessment of the factors influencing gaze shift at image viewing**

The problem of algorithms development of an assessment of visual attention distribution is considered. In this paper, we propose a method based on the formation of viewing trajectories by model. The model includes the foveal input window and determining factors module and their combinations, which determines the next point of "fixation". The assessment of these factors and optimization of model parameters was carried out.

POSTER SESSION 2

Thursday, January 30 14:00 – 14:45

Lecture-hall холл Б-100

Chair: Prof. KAGANOV YURI

Applications of neural networks

57. ILYA POVIDALO, ALEKSEY AVERKIN*Dubna International University*

**Biomorphic neural modular structures for dynamic object identification**

In this article a number of neural structures, based on Kohonen sef-organizing maps, which can be successfully used for dynamic object identification are described. Also modular neural networks, their architecture and possible learning algorithms are described. Some modifications of modular neural networks applicable for dynamic object identification are given along with some examples of application of these networks.

58. RUMOVSKAYA S.B., KOLESNIKOV A.V.

*Institute for Problems of Informatics RAS, Kaliningrad Branch*

**Virtual diagnostics of the arterial hypertension**

We represent the approach to the modelling of the medical concilium for arterial hypertension diagnostics within synergetic paradigm of artificial intelligence. The system domain analysis of the AH diagnosing task and construction of the method for its solving were performed. The developed models are designed in accordance with the recommendations of Society of cardiology of Russian Federation.

59. ENGEL EKATERINA ALEKSANDROVNA

*Katanov Khakass State University, Abakan*

**INTELLIGENT CONTROL SYSTEM OF ROBOT’S MOVEMENT**

The paper describes the control system of multilegged robot’s movement in condition of uncertainty, the efficiency of which is shown by the example of Matlab/Simulink model of the three-legged robot’s movement. The application of these intelligent technologies significantly increases the effectiveness of the control process and control system’s robustness to different environmental condition.

60. ANDREEV O.A., TROFIMOV A.T.

*Scientific-Research Institute «Atoll», Dubna, Moscow Region*

**Interpretation of the weights of the neuron**

The paper addresses the issue of interpretation of the operators of artificial neural and bayessian networks by comparing functioning processes of the networks, used as classifiers. Real hydroacoustic sound locator data are used to initiate functioning processes of the networks. Weights, biases and transfer functions of neurons are treated as the operators of the networks. The issue of interpretation is related to possibility of applying polygaussian probabilistic models for synthesis of the operators of artificial neural networks without train process.

61. A.A. TELNYKH, N.S. BELLYUSTIN, O.V. SHEMAGINA, A.V. KOVALCHUK, I.V. NUIDEL

1

*The Institute of Applied Physics of the Russian Academy of Sciences, Nizhny Novgorod*

2

*Scientific Research Institute of Radio-Physics, Nizhny Novgorod*

**Artificial Intellectual Systems for Scene Analysis: Learning and Tuning Issue**

In the present paper we discuss principles of good quality detector forming which allow for two dimension images frrs dependence on an parameter which the parallelization degree of system determines. As an application example the classifier which peoples face in profile position determines is considered.

62. POTANINA M.V., POTANIN N.I., DAVYDOV O.D., BABUSHKIN V.N., PUHOV V.A.

1

*Ural Federal University named after the first President of Russia B.N. Yeltsin, Ekaterinburg*

2

*Ural Scientific Research Institute of Traumatology and Orthopaedics named after V.D. Chaklin, Ekaterinburg*

**USE OF THE NEURAL NETWORK IN FUNCTIONAL DIAGNOSTICS OF DISEASES OF COXOFEMORAL AND KNEE JOINTS AND THE ASSESSMENT OF QUALITY OF SURGICAL TREATMENT**

In this work it is shown that in stabilografichesky space of tests of Romberg with the closed and open eyes neural networks diagnosing an illness of coxofemoral and knee joints and treatments estimating quality in dynamics are under construction. The minimum perseptronny network (by quantity of neurons) which in minimum informative spaces qualitatively diagnoses an illness ρ = 0,97 is found. The pathology index which allows to estimate quality of treatment is entered. The estimated scale is developed for definition of degree of expressiveness of functional violations of the musculoskeletal device at patients with pathology of coxofemoral and knee joints with use of mathematical processing (neural networks) stabilometrichesky indicators.

63. MOKROV A. M., KAGANOV YU. T.

*Bauman Moscow State Technical University*

**Computer system for recognition of show emotion by facial expression of person on the base neural networks**

In this article have been described methods and approaches for designing program system of video monitoring and recognition of person’s face. This program system is based on using algorithms for pick out specific personality traits and identification human. Interaction algorithms of recognition was investigated in detail and merits and demerits were detected in the context of program system. The Cohonen’s maps and Grosberg’s stars have been used as artificial neural nets algorithms. These algorithms are very efficient in comparison with traditional neural nets as perceptrons.

64. M.V. BURAKOV

*Saint Petersburg State University of Aerospace Instrumentation*

**NEURAL NETWORK BASED PID CONTROLLERS**

This paper provides a new style of nonlinear PID controller that is based on artificial neural network (ANN). The ANN is used to realize PID formula and the genetic algorithm is used to search parameter of the nonlinear activation function. The performance of the proposed method is compared with the conventional PID methods for a vehicle active suspension system using MAT-LAB/Simulink software package.

65. BABICHENKO A.V., KARPENKO A.P., KILCHIK A.V., TROFIMOV A.G.

1

*Ramenskoye Instrument-Building Design Bureau, Moscow Region*

2

*Bauman Moscow State Technical University*

3

*National Research Nuclear University (MEPhI), Moscow*

**Neural network prediction model of aircraft fuel consumption**

To improve the effectiveness of engineering and navigational calculations on the aircraft there is important task to use the on-board adaptive mathematical model of the aircraft. It will improve the accuracy and efficiency of on-board solutions of the aircraft engineering and navigational direct and inverse problems. One of the main problems is the problem of fuel consumption prediction. The work is devoted to the evaluation of the effectiveness of neural network prediction model of the aircraft fuel consumption.

66. OREHOVA EKATERINA ANDREEV VYACHESLAV

*Nizhny Novgorod State Technical University named after R.E. Alekseev*

**Defining the thermophysical properties of water and steam with the use of artificial neural network**

In this paper the attempt of application of artificial neural networks to the solution of the problem of determining the thermophysical properties of water and steam at saturation line. As part of this work have been created artificial neural networks to measure the temperature, enthalpy, entropy and specific volume of water and steam at saturation line on the known pressure and the received results are considered.

67. ILYA KALINOWSKI, ALEXANDER SHILIN, VICTOR BOUKREEV

*National Research Tomsk Polytechnic University*

**Using of neural network adaptive critics in the parameter identification problem of mathematical models**

This paper considers the solution of recovery from experimental data values of time-varying parameters of the mathematical model of the hot water heat exchanger of a residential building using neural network adaptive critic.

68. SERGEY ZHERNAKOV, ARTUR GILMANSHIN

*Ufa State Aviation Technical University*

**Application of neural-network and neurаl-fuzzy algorithms of gas-turbine engine control and diagnostics in airborne requirements**

The application of intelligent algorithms to solve problems of control and diagnostics of aircraft such as the implementation of gas-turbine engine (GTE) model and diagnosis of failures of sensors is considered, an approaches to the construction of a purely neural network and neuro-fuzzy models are shown with their advantages and disadvantages.

69. KOZLOVA A.A., KISLYAEV A.S.

1

*Samara State Aerospace University*

2

*Research institute of heat power instrument making*

**Prediction of atrial fibrillation episodes with the help of MLP- and RBF- neural networks**

The paper is devoted to the prediction of actual episodes of atrial fibrillations. MLP- and RBF- neural networks are used for the purpose. The values of intervals of electrocardiograms are used as the input vector of training and testing samples.

70. A.YU. DOROGOV V.C. ABATUROV O.V. ZABRODIN I.V. RAKOV

*Saint Petersburg Electrotechnical University "LETI"*

**IMPLEMENTATION OF NEURAL NETWORK CLASSIFIER IN SQL/MM CONTEXT**

The task of implementation of neural network classifier in SQL/MM context is coincided. Analyses of applicability of SQL/MM standard for implementation of neural network technology is carried out. Unitized interface for control of neural network classifier is described. Scheme of script generation of base phases of data mining is shown. Variant of implementation of analytic subsystem infrastructure for embedding applications is suggested.

71. S.A. GORBATKOV, D.V. POLUPANOV, I.I. BELOLIPTSEV

1

*Financial University under the Government of the Russian Federation, Ufa Branch*

2

*Bashkir State University, Ufa*

**THE COMPARISON OF HYBRID NEURAL NETWORK METHODS OF THE SYNTHESIS OF THE OPTIMAL PLAN OF FIELD TAX MONITORING**

Various approaches, probability, and fractal, to the synthesis of the optimal plan selection of taxpayers for field tax audits based on neural network models are investigated. Testing tool are wide series of computational experiments and comparisons with the natural experiments on the field tax monitoring.

72. D.V. POLUPANOV, N.A. KHAYRULLINA, I.V. ASCHEULOV

*Bashkir State University, Ufa*

**RESTRUCTURING OF THE MARKET OF SHOPPING CENTRES IN UFA CITY ON THE BASIS OF KOKHONEN'S SELF-ORGANIZING MAPS BASED ON A BAYESIAN APPROACH**

Investigated the problem of structuring the shopping center market. The key factors structuring, highlighting the characteristics of shopping centers, which have the greatest impact on their popularity and attendance. For the performance neural network model is proposed segmentation based on Kohonen self-organizing maps. As a result of modeling a robust estimate, allowing to give practical advice to make decisions on the future development of the sector and increase the transparency of the market.

SESSION 7

Friday, January 31 14:00 – 16:00

Lecture-hall 403

Chair: Prof. VVEDENSKY VIKTOR

Applications of neural networks

73. S.A. DOLENKO, I.V. ISAEV, E.A. OBORNEV, I.G. PERSIANTSEV, M.I. SHIMELEVICH1

*Skobeltsyn Institute of Nuclear Physics Lomonosov Moscow State University*

2

*Russian State Geological Prospecting University n. a. Sergo Ordzhonikidzе, Moscow*

**INVESTIGATION OF EFFICIENCY OF JOINT APPLICATION FOR METHODS OF GROUP DETERMINATION AND STEPWISE DETERMINATION OF PARAMETERS IN NEURAL NETWORK SOLUTION OF ELECTRICAL PROSPECTING INVERSE RPOBLEM**

In the electrical prospecting inverse problem (IP), the sought-for distribution of electrical conductivity (EC) in Earth stratum is described by dividing the studied section into blocks arranged in layers, with determination of EC of each block. This IP is usually solved separately for each block – the so-called autonomous determination. In the preceding studies it has been shown that use of methods of group determination or stepwise determination of parameters allows increasing the precision of the solution in a number of cases. In this study, joint application of these methods is investigated.

74. GORBATKOV S. A., BELOLIPTSEV I. I., FARHIEVA S. A.

*Financial University under the Government of the Russian Federation, Ufa Branch*

**Generalized neural network models of enterprise risk bancruptcy**

To support the development of anti-crisis management decisions to improve the financial condition of the company a neural network dynamic logit-model is developed. The model takes into account the factors that characterize the stage of crisis. Developed a method for optimal selection of key economic indicators. Theoretical proposals approved on real data.

75. AKHMETSHINA L.G., YEGOROV A.A.

*Oles Honchar Dnipropetrovsk National University, Ukraine*

**Clustering validity improvement based on using neuro-fuzzy technology**

This work describes the clustering method sFCM, which based on using fuzzy approach and Kohonen neural network inside the algorithm for the pur-pose of automatic detecting the number of the clusters during data processing. Experimental results of using sFCM algorithm for the medical low-contrast images processing are shown.

76. A.B. BAKHSHIEV, L.A. STANKEVICH

*Peter the Great St. Petersburg Polytechnic University*

**APPLICATION OF PULSE NEURAL NETWORK MODEL WITH STRUCTURAL ADAPTATION AT PROBLEMS OF MOVEMENT CONTROL**

In this research a possibility of approach implementation for control of robotic systems based on data on architecture of biological neural structures for move-ment control is considered. Example of creation of neural network regulator in view of simple neural model for spinal level of control by muscles compressing is considered. Algorithm for structural tuning of the regulator neural network for problem of object position stabilization is presented.

77. S.A. BUTENKOV

*Supercomputers and Neurocomputers Research Center, Taganrog*

**Self-Organized Neural Networks Based on Grassmann Neurons**

In this paper the new approach to the single-layered self-organized neu-ral networks for the clustering and classification is presented. The intro-duced approach is based on the very general technique for multi-dimensional data granulation. Be means of new neuron model there are the sufficient restrictions for the neural network amount and complexity are realized. A few practical problems, don’t resolved for the similar SOM neu-ral networks has been solved

78. M.S. TARKOV, S.V. DUBYNIN

1

*Rzhanov Institute of Semiconductor Physics, Siberian Branch of Russian Academy of Sciences, Novosibirsk*

2

*Novosibirsk State University*

**REAL-TIME OBJECT TRACKING BY CUDA-ACCELERATED NEURAL NETWORK**

An algorithm for real-time object tracking is proposed. The algorithm is based on the application of neural network implemented on the graphics card (GPU). The research and optimization of the algorithm parameters are realized. Tracking process is accelerated by 10 times, and the training process is accelerated by 2-fold compared with the sequential algorithm. We calculate the maximum resolution of the frame to track the real-time and optimal sampling frames of video for training the neural network.