## Sections

### Neuroinformatics - 2015

POSTER SESSION 1

Monday, January 19 16:30 – 16:45

Lecture-hall Акт. зал

Chair: Prof. USHAKOV VADIM

Neural networks and cognitive sciences

1. I.A. REBRUN, K.V. SIDOROV, S.A. TEREKHIN, N.N. FILATOVA , P.D. SHEMAYEV*Tver State Technical University*

**Biotechnical system for research of cognitive activity in various emotional conditions of the examinee**

The biotechnical system for research of electric activity of a brain when performing a limited set of tasks in conditions and in the absence of external emotional incentives is offered. The technique of research is based on objective confirmation of change of an emotional condition of the examinee by results of monitoring of EEG. A process of creation of linguistic scales for an assessment of the sizes and color of two-dimensional objects of a simple form is considered.

2. KOROLEVA M.E., BAKHCHINA A..V., USHAKOVA I.L., NEKRASOVA M.M., KRUPA V.V., PARIN S.B.

1

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

2

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

3

*Nizhny Novgorod State Medical Academy*

4

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

**Autonomic regulation of some form of social interaction**

The problem of the relationship of human functional state and its social conditions is considered. We consider the specific forms of social interaction: public performance and a student of driving school with instructor during driving instruction. It was shown that the context of public speaking is a stress factor for speakers. We also found a significant positive relationship between the parameters of heart rate of teacher and student at the first practical driving lesson.

3. E.V. KORYAGIN, I.A. MEDYANSKIY, A.E. SHIRKIN

*Immanuel Kant Baltic Federal University, Kaliningrad*

**Development of associative memory model for AR-600 robot for clusterization and generalization tasks**

Several neural network models are considered for associative memory reconstruction perspective. Effectiveness and eligibility of existing methods is analyzed. Modified self-organizing incremental neural network model is presented as a main method for associative memory modeling of any kind input data. Perspectives of further work are described.

SESSION 1

Monday, January 19 16:45 – 18:30

Lecture-hall Акт. зал

Chair: Prof. USHAKOV VADIM

Neural networks and cognitive sciences

4. RADCHENKO G.S., PARIN S.B. POLEVAYA S.A., KORSAKOVA-KREYN M.N., FEDOTCHEV A.I.1

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

2

*Privolzhsky Research Medical University*

3

*Touro College and University System, New York*

4

*Institute of Cell Biophysics, Pushchino, Moscow region*

**Influence of characteristics of tonal modulation of musical fragments on EEG**

We investigate how conditions of modal modulation of musical fragments and distance of modulation influence on EEG characteristics. We found that differences between major and minor mode conditions depend on the distance of modulation. We found significant differences in the total power that depend on the degree of modulation for stimuli with same mode condition major-major and minor-minor.

5. S.V. BOZHOKIN, I.B. SUSLOVA

*Peter the Great St. Petersburg Polytechnic University*

**Non-stationary correlation of EEG bursts ensemble: wavelet analysis**

We modeled EEG activity bursts as a superposition of elementary non-stationary signals. Using the technique of continuous wavelet transform and spectral integrals analysis, we calculated a set of new quantitative parameters characterizing the time variation of spectral properties for each brain activity burst and for the ensemble of bursts. The problem of non-stationary correlation of different EEG channels has been solved. The application of the approach for the analysis of non-stationary EEG during functional tests has been discussed.

6. * A.G. TROFIMOV, I.V. KOLODKIN, V.L. USHAKOV, B.M. VELICHKOVSKI

1

*National Research Nuclear University (MEPhI), Moscow*

2

*National Research Centre "Kurchatov Institute", Moscow*

**Agglomerative method for isolating microstates EEG related to the characteristics of the traveling wave**

We propose a method for the isolation of microstates of the brain according to electroencephalography (EEG) based on the characteristics of traveling waves. To assess the severity of the traveling waves introduced indicators synchrony and coherence. Experimental studies on real EEG data show that the proposed method is highly competitive with traditional methods of segmenting EEG and reflects the new structure of the EEG segmentation.

7. * TROFIMOV A.G., IVANITSKIY I.I., VELICHKOVSKIY B.M.

1

*National Research Nuclear University (MEPhI), Moscow*

2

*National Research Centre "Kurchatov Institute", Moscow*

**Greedy algorithm of composition committee of classifiers EEG signals**

We propose the greedy algorithm of composition committee of classifiers EEG signals, which work on the simplest one-dimensional feature space. It is shown that the accuracy of classification of committee exceeded the accuracy of the best of the weak classifiers more than eight times.

8. * ATANOV M.S., IVANITSKY G.A.

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

**Optimization of current type of cognitive activity recognition based on EEG data.**

The problem of improving the performance of the BCI, designed for learning the subjects to solve cognitive tasks utilizing biofeedback learning, is considered in the paper. The main point is the application of ICA method to EEG data. Moreover, an attempt was made to interpret the resulting components from the physiological point of view to find the processes, forming the thinking, common for different subjects.

9. * I.V. TAROTIN, G.A. IVANITSKY

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

**A model for cognitive activity monitoring in a space of psychological features**

New method of mental activity real-time visualization in the cognitive space was presented. Its main idea is to quickly identify types of thinking using EEG spectra with the help of artificial neural network (perceptron). Cognitive space construction is based on one of the multidimensional scaling methods (Sammon mapping). The developed technology will be applied in future to study thought processes during complex cognitive activity.

10. V.L. VVEDENSKY

*National Research Centre "Kurchatov Institute", Moscow*

**Averaging of the brain magnetic signals**

We observe in MEG records triangular peaks, which precede for about 700 ms the instant of the self-paced finger movement. Amplitude of the peak varies considerably from trial to trial. This makes possible separation of trials into groups, where the brain behaves differently during planning and execution of the same action. Separation into groups reveals other processes, active in the cortex before voluntary movement.

SESSION 2

Wednesday, January 21 12:00 – 13:00

Lecture-hall 406

Chair: Prof. TIUMENTSEV YURY

Neural network theory

11. * PROSTOV YU.S., TIUMENTSEV YU.V.*Moscow Aviation Institute (National Research University)*

**A study of neural network model composed of hysteresis micro-ensembles**

A model of the neural ensemble consisting of firing-rate neurons with a complex activation function was proposed. It is shown that in such ensemble a new feature emerges, namely existence of three stable states of the model (inactive, half-active and active states) with hysteresis transitions. We demonstrate that the proposed ensemble can be used as a pattern recognition structural element of artificial neural networks. In addition, further usage possibilities are analyzed for the obtained properties of the ensemble.

12. M.S. TARKOV

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

**Mapping of neural network weighting coefficients matrix onto memristor crossbar**

The problems of programming memristor matrix (crossbar) are considered. An estimate for the voltage pulse duration to set the required memristor resistance value, based on the initial resistance value, is evaluated. An algorithm is proposed for mapping neurons layer weighting coefficients matrix onto memristor crossbar with the given memristors conductivity value boundaries. The results can be used both in mathematical modeling, and in the physical realization of neural networks with interneuronal memristor connections.

13. * KUKIN K.A.,SBOEV A.G., SOBINOV A.R.

1

*National Research Centre "Kurchatov Institute", Moscow*

2

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

**Comparison of methods of learning spiking neural networks on base of different neurostimulators**

Investigation of different factor influence. on the learning process were performed by using spiking neural networks (NEST, CSIM, HEM). The next factors were analysed: choice of spike pairing scheme, shapes of postsynaptic currents and the choice of input type signal for learning. Best factors for learning performance were extracted.

POSTER SESSION 2

Wednesday, January 21 14:00 – 14:45

Lecture-hall 405

Chair: Prof. DOLENKO SERGEY

Applications of neural networks

14. SOLOVYEV, A.M.*Voronezh State University*

**Stabilization of flexible inverted pendulum with presence of backlash in the basis by means of artificial neural network with hysteresis properties**

Construction principles of artificial neural networks with hysteresis activation function based on the S-operator are developed. The physical model of the flexible inverted pendulum with presence of backlash in the basis of its mounting is considered. The problem of the pendulum stabilization in the vicinity of the vertical position is solved using two-layer artificial neural network with hysteresis properties.

15. GORBATKOW, S. A., BELOLIPTSEV, I. I., FARKHIEVA, S. A.

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

**Iterative method for constructing of neural network dynamic model for bankruptcies forecasting**

We propose a neural network iterative hybrid method for estimating the probability of developing in time the risk of bankruptcy of enterprises in difficult conditions modeling: the incompleteness of the data and their strong noising. Computational experiments have confirmed the effectiveness of the basic idea of the hybrid neural network method for constructing a dynamic model with continuous time. The method can be used in a wide range of applications: evaluation of the borrower's creditworthiness, assessment of investment objects and others.

16. . IVANOV, E. O., ZAMYATIN, N. V.

*Tomsk Institute of Radio Electronics and Electronic Technics*

**A Hopfield network for the control a group of objects (pumps)**

A statement of multi-objective optimization problem for control of pumps group is given. The problem is reduced to the Hopfield network dynamics. Application of the obtained network to real optimization problem is considered. The results of simulation are analyzed and compared with those of other algorithms.

17. GABDRAKHMANOVA, N. T.

*Peoples’ Friendship University of Russia, Moscow*

**Neural network models for management tasks on the main oil pipeline**

Basic principles and results of constructing identification model of electricity consumption for oil transit are set out. Constructed neural network model is tested on one of the linear sections of the main oil pipeline. The model can be used for planning of electricity consumption for oil transit. Several separate tasks that can be solved by proposed neural network model, are considered.

18. SHEPELEV, I. E., NADTOKA, I. I., VYALKOVA, S. A., GUBSKY, S. O.

1

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

2

*Platov South-Russian State Polytechnic University (Novocherkassk Polytechnic Institute)*

**Optimal metaparameters identification for neural network short- term forecasting of electricity consumption of a large city**

We consider the problem of optimal metaparameters identification for neural network, which is designed for short-term forecasting of electricity consumption by the example of Moscow in the spring and summer season. These are the following metaparameters: length of the delay line at the input of the neural network, the size of its hidden layer, the depth of the training sample, the radius of the training surroundings, set of significant inputs and neural network regularization coefficient. Construction of the forecast is based on a multi-layer perceptron.

19. ISAEV, I. V., DOLENKO, S. A., OBORNEV, I. E., OBORNEV, E. A., SHIMELEVICH, M. I.

1

*Skobeltsyn Institute of Nuclear Physics Lomonosov Moscow State University*

2

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

**Improving the accuracy of neural network solution of multi-parameter inverse problems by group determination of parameters: verification on model data**

When solving multi-parameter inverse problems (IP), the problem is usually solved separately for each parameter. In previous studies it was shown that aggregation of parameters into groups with simultaneous values determination of all the parameters of the group in some cases allows improving the accuracy of the solution for solving multi-parameter IP of electrical prospecting. In this study, the observed effect was verified on model data defined explicitly.

20. BURAKOV, M. V.

*Saint Petersburg State University of Aerospace Instrumentation*

**Design of neuro-emulator for nonlinear dynamic plant**

In this paper, we consider the problem of synthesis a neuro-emulator for a class of nonlinear dynamical plants which can be described by the Hammerstein model with unknown parameters. Learning of neuro-emulator used genetic algorithm. The quality of the neuro-emulator is checked by modeling in MatLab Simulink package.

21. ABATUROV, V. S., DOROGOV, A. YU.

*Saint Petersburg Electrotechnical University "LETI"*

**The service-oriented analytical platform infrastructure for embedded intelligent subsystems**

The service-oriented analytical platform infrastructure for embedded intelligent subsystem based on DBMS PostgreSQL is considered. The architecture of analytical platform that satisfies requirements of SQL/MM and PMML is proposed. The unified management interface for analytical platform is described. The diagram of main knowledge extraction phases (training, testing and application phases) is shown. The logical data model of analytical platform infrastructure is shown. The advantages of analytical platform are presented.

22. TARKOV, M. S., OSIPOV, M. I.

1

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

2

*Novosibirsk State University*

**Object tracking by Bayesian network**

We propose an algorithm for tracking objects in a video stream. The algorithm is based on the use of hierarchical Bayesian network. A feature of the proposed algorithm is the use of multidimensional scaling which significantly shortens the network learning time. Algorithm is robust to the temporary tracked object disappearance. It can keep track of multiple objects on a complex background and can be well parallelized.

23. OREKHOVA, E. E., ABRAMOV, A.A., ANDREEV, V. V, ANDREEVA, O.V.

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

**The creation of an information system based on artificial neural networks to determine the endurance limit of metals under the influence of external factors**

This article is devoted to the problem of determining the parameters of the fatigue of metals. As is known, basically the parameters of the fatigue of metals are determined experimentally. But experiments require long time and essential expenses. In addition, the existing experimental database does not cover all materials and their operating conditions in industry. Therefore, attempts have been made to create a rapid method of determining the parameters of the fatigue of metals, using artificial neural networks.

SESSION 3

Wednesday, January 21 14:45 – 17:00

Lecture-hall 406

Chair: Prof. DOLENKO SERGEY

Applications of neural networks

24. KOZLOV D.S., TIUMENTSEV YU.V.*Moscow Aviation Institute (National Research University)*

**Neural network based semi-empirical models for dynamical systems described by differential-algebraic equations**

A simulation problem is discussed for nonlinear controlled dynamical systems described by differential-algebraic equations. It is proposed to seek a solution of the problem within the semi-empirical modeling approach combining theoretical knowledge for the plant with training tools of artificial neural network field. The simulation results are presented for a semi-empirical model generated in respect to reentry hypersonic vehicle.

25. * EGORCHEV M.V.

*Moscow Aviation Institute (National Research University)*

**Training of neural network based semi-empirical models for spatial 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. Training of the dynamical neural network model for multi-step ahead prediction is performed in a sequential fashion.

26. * A.O.EFITOROV, S.A. BURIKOV, T.A. DOLENKO

*Lomonosov Moscow State University*

**Comparison of quality of solving inverse problem of multi-component solutions spectroscopy by neural networks and by method of projections to latent structures**

This study provides comparative analysis of application of neural networks and of method of projections to latent structures for simultaneous determination of types and concentrations of inorganic salts dissolved in multicomponent water solutions by Raman spectra. It is shown that the method of projection to latent structures has several advantages, such as the quality of the solution and the time of creation of a regression model, for solving problems with moderate nonlinearity

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

*Lomonosov Moscow State University*

**Data dimensionality reduction and clustering quality estimaton in problems of composition analysis of multi-component solutions**

The paper presents the results of searching optimal combination of data dimensionality reduction method with clustering algorithm for analysis of Raman spectra of multi-component solutions of inorganic salts. The most informative clustering quality measure is presented. It is demonstrated that use of specialized algorithms together with dimensionality reduction techniques leads to improvement in quality and stability of the result of solution of clustering problem.

28. * V.R. SHIROKY

*Skobeltsyn Institute of Nuclear Physics Lomonosov Moscow State University*

**Comparison of neural network models for prediction of geomagnetic Dst index on different datasets, and comparison of methods for evaluating model quality**

Complexity of neural network prediction of the state of the Earth’s magnetosphere is determined inter alia by the small fraction of samples obtained during geomagnetic disturbances in the total number of samples. Consequences of this fact are high integral statistics of predicting models, trivial as well as neural network based. This study is devoted to the comparison of models trained on different data sets, and to the comparison of their quality estimation indexes.

29. * V.A. SVETLOV, I.G. PERSIANTSEV, YU.S. SHUGAY

1

*Lomonosov Moscow State University*

2

*Skobeltsyn Institute of Nuclear Physics Lomonosov Moscow State University*

**Testing of new software implementation of algorithm for adaptive construction of hierarchical neural network classifiers**

The article presents development of the algorithm for adaptive construction of hierarchical neural network classifiers based on automatic modification of the desired output of perceptrons with small number of neurons in the single hidden layer. The conducted testing of the new program implementation of this approach demonstrated that the considered algorithm was more computationally efficient and provided higher quality of solution of classification problems in comparison with standard multi-layer perceptron.

30. * SENYUKOVA O.V., ZOBNIN D.S., PETRAIKIN A.V.

1

*Lomonosov Moscow State University*

2

*Pirogov Russian National Research Medical University (RNRMU), Moscow*

**Brain MRI registration algorithm based on key points matching**

This work is devoted to the problem of registration (matching) of human brain MRI. In order to register two input brain MR images of different subjects, one should find such a transformation of the first image that makes it the most similar to the second image. Anatomical features should be taken into account. The paper describes the registration algorithm based on key points matching using spline transform. The proposed algorithm was implemented and tested on real data.

31. B.V. KRYZHANOVSKIY, L.B.LITINSKIY

*Scientific Research Institute for System Analysis, Moscow*

**General method for calculation of partition function**

We develop a new method for calculation of partition sum that plays a central role when analyzing complex physical systems by means of statistical physics methods. Our method is based on decomposition of all the set of states into non-overlapping classes. Then we approximate the distribution of energies of the states from each class with the aid of the Gaussian density. We succeeded in exact calculation of the mean energy and the variance of the distribution of energies of the states from each class. The obtained general expressions were tested by the example of the Ising model on the hypercube.

POSTER SESSION 3

Thursday, January 22 14:00 – 14:45

Lecture-hall 405

Chair: Prof. TEREKHOV SERGE

Applications of neural networks

32. IVANOV N.A., VULFIN A.M.*Ufa State Aviation Technical University*

**A neural network algorithm for constructing the language model in statistical machine translator**

The goal of this research is to construct a model of natural language for statistical machine translator. The initial data for the model is redundant information extracted from the corpus of natural language texts. The algorithm for the language model is based on neural network.

33. BONDAREV V.N.

*Sevastopol State University*

**Colored Gaussian signal extraction based on cascade neural network**

Artificial neural network for consecutive colored Gaussian signal extraction from its mixture with other signals or noise is considered. It is proposed a cascade neural network learning rule based on criteria of minimum of mean square predictive error, which allows to make simpler the on-line realization of the network. Examples of extracted signals and values of performance indexes are demonstrated.

34. BEKIREV A.S., KLIMOV V.V., KUZIN M.V., SHCHUKIN B.A.

1

*National Research Nuclear University (MEPhI), Moscow*

2

*BPC Banking Technologies, Moscow*

**Credit card fraud detection using neural network committee and clustering**

The task of credit card fraud detection using account information is considered. We apply two approaches to neural networks interaction: neural network committee and approach based on pre-clustering. Finally, these two methods are compared.

35. ANIKIN V.I., KARMANOVA A.A.

*Volga Region State University of Service, Togliatti, Samara region*

**Modeling and study of cellular Kohonen’s neural network in spreadsheets**

The work of Cellular Kohonen’s Neural Network in Microsoft Excel was emulated. The physical features and high temporal efficiency of multidimensional data clustering and classification by the cellular network were studied. An original method for the initial deployment of a self-organizing maps, and useful application of the edge effect and multilinked self-organizing maps to solve the problem of "dead" neurons and reliable identification of cluster boundaries were proposed.

36. S.A.DOLENKO, I.N.MYAGKOVA, I.G.PERSIANTSEV

*Skobeltsyn Institute of Nuclear Physics Lomonosov Moscow State University*

**Neural network segmentation of multi-dimensional time series as an instrument for study of dynamics of the Earth's magnetosphere**

This paper presents preliminary analysis of the results of multi-dimensional time series segmentation by means of Kohonen neural networks. The time series includes parameters of solar wind, interplanetary magnetic field, geomagnetic indexes (Dst-variation) and the flux of relativistic electrons of the outer Earth’s radiation belt with energy > 2 MeV. It is expected that this method of study will allow one to advance in understanding of the dynamics of the Earth’s magnetoshere, to improve quality and to increase the prediction horizon for the studied quantities.

37. PROTASOVA M.A.

*Saint Petersburg Electrotechnical University "LETI"*

**Neural network qualifier of anomaly telecommunication**

Application of artificial neural networks to solve the problem of recognition and classification of anomalies in the telecommunication-network is considered. The implementation of the neural network classifier using the R programming language is considered. The feature space is selected, and the training sample of neural network is collected. The implementation of the telecommunications network classifier anomalies is presented.

38. SHATS V.N.

*Independent investigator, St. Petersburg*

**On new computing technology in machine learning**

Living organisms have the elements that provide multi-layered and multiple processing incentives information on a single rule. The task of supervised learning is solved on the basis of the model of this technology. Data matrix is considered as one of the images of the sample, a countable set of other images are indexes matrix, which are found by the quantization features. Indexes approximately describe the feature and allow us to find the frequency of occurrence of combinations of indices corresponding to objects of a certain class.

39. KOMARTSOVA L.G., LAVRENKOV Y.N.

*Bauman Moscow State Technical University Kaluga Branch*

**Use of neural networks for analyzing characteristics of the elements of the telecommunication network**

This article provides the description of the settings tuning algorithm for neural network built on the basis of sigma-pi neurons. The core of the learning algorithm is a combination of random search with the heuristic algorithm. An integrated approach to training neural networks allows one to perform training within the time range required to adjust the neural network to solve the problem. The possibility of using sigma-pi network to estimate the parameters of data transmission channel is considered.

40. BONDAREV V.N.

*Sevastopol State University*

**Application of digital model of pulse neuron for the adaptive signal filtration**

The model of multi input pulse neural element focused on the solution of digital signal processing problems is considered. Discrete expression for calculation of the state of pulse neural element and the learning rule based on criterion of a mean square error minimum of useful signal extraction from the mix with stationary additive noise is proposed.

41. V.E.PAVLOVSKY, A.V.SAVITSKY

1

*Keldysh Institute of Applied Mathematics, Moscow*

2

*Lomonosov Moscow State University*

**The neural network controller for quadrocopter**

In this work the theoretical-mechanical model of the multirotor robot – a quadrocopter taking into account the main aerodynamic effects is constructed. On the basis of results of numerical modeling of flight of the device the neural network regulator for basic trajectories is constructed. Operation of the regulator depending on height sensor error is studied.

SESSION 4

Thursday, January 22 15:00 – 15:30

Lecture-hall 406

Chair: Prof. BURTSEV MIKHAIL

Adaptive behavior and evolutionary modelling

42. KOSHUR V.D.*Siberian Federal University, Krasnoyarsk*

**Amplification of swarm intelligence in method of global optimization by using fuzzy neurel network control for searching process**

The updating of the Particles Swarm Optimization method with the increased adaptive properties is submitted. The introduction in algorithm of Global Optimization the Fussy Neuron Network for control of a choice of making movements of agents-particles and amplification of Swarm Intelligence is used. The results of computing experiments of search of a global minimum for multimodal test functions with two, fifty and hundred variable are given.

43. MISHULINA O.A., SUKONKIN I.N.

*National Research Nuclear University (MEPhI), Moscow*

**Evolutionary algorithm for data clustering based on statistical criterion of clusters standard volume**

A new statistical criterion for data clustering is proposed. It can be used for data samples with special properties: different geometrical dimensions of the clusters in the feature space; significant difference between the numbers of sample points contained in the clusters; possible incompleteness of the features. Criterion, called SV-criterion allows for a priori unknown number of clusters in the sample data to obtain a stable estimate of their actual number. A genetic algorithm using the proposed criterion is developed. Model examples illustrate practical possibilities of the algorithm.

SESSION 5

Thursday, January 22 15:30 – 17:00

Lecture-hall 406

Chair: Prof. BURTSEV MIKHAIL

Applications of neural networks

44. MALSAGOV M.YU., KRYZHANOVSKIY V.M., ZELAVSKAYA I.S.*Scientific Research Institute for System Analysis, Moscow*

**Scalar neural network tree for nearest neighbor search problem in high-dimensional binary space**

Nearest neighbor search problem in high-dimensional binary space is considered. To solve this problem neural network tree is proposed and studied. Perceptrons are nodes of tree. Upper bound of algorithm error probability is obtained. Neural network tree is compared with exhaustive search. Theoretical estimations of algorithm computational complexity are obtained.

45. KOVALCHUK A.V. , BAHCHINA A.V., POLEVAYA S.A.

1

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

2

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

3

*Privolzhsky Research Medical University*

**Spectral analysis of cardiointervalography of uneven time series for estimation of human activity vegetative support**

Basic methods of rhythmogram spectrum estimation use well developed regular time series theory. There are different methods of reduction to a regular time grid, such as smoothing, approximation values, etc. Although the quality of these methods is sufficient when using a quasi-uniform rhythmograms but in the case of dynamic changes of heart rate caused by the normal functioning in the social environment may give incorrect results. The paper presents a method of spectral estimation of rhythmogram time series that permits a further spectral analysis with sufficient details.

46. IVASHINA E.A., KORLYAKOVA M.O., PROKOPOV E.YU.

*Bauman Moscow State Technical University Kaluga Branch*

**Generation of the neural networks association for the stereo reconstruction for on board technical vision systems**

Proposed solution of for the stereo reconstruction of coordinates of scene objects is based on the analysis of a stereo pair of projections. The process of stereo reconstruction was modeled in virtual scene with projections errors estimated from real cameras calibration. The results of the synthesis of the solver based on boosting and inhomogeneity error analysis of the neural network was presents.

47. ENGEL E.A.

*Katanov Khakass State University, Abakan*

**Energy saving technology for electrotechnical system on the basis of the adaptive neurocontroller**

Within the Smart Grid concept the energy saving technology of automated system of the electricity’s technical accounting on the basе of the adaptive neural controller organizing the functional interaction of the account and the mode identification of system’s energy consumption is developed. This adaptive neural controller on the basе of the identification of electro technical system’s status and dual neural nets minimizing and forecasting power consumption’s creates the effective control signal under random perturbations.

48. SAVCHENKO A.V., MILOV V.R.

1

*Higher School of Economics, Nizhny Novgorod Branch*

2

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

**An approach to sequential hierarchical image recognition**

The automatic image recognition problem is considered. A hierarchical approach is proposed in which the analysis on the next, more detailed level of object's description is performed only if the decision on the current level is unreliable. The practical examples of the proposed approach in face recognition problem are presented.

POSTER SESSION 4

Friday, January 23 12:00 – 12:30

Lecture-hall 406

Chair: Prof. KAZANOVICH YAKOV

Neurobiology

49. V.B. KOTOV*Scientific Research Institute for System Analysis, Moscow*

**Using two-input neurons for inner image sequence generator**

We consider the operation of a recurrent neural network (RNN) generates inner image sequences. Neurons with two summing inputs are suggested to improve the network performance. It is demonstrated that an RNN based on such neurons has increased capacity as compared with one-input neural networks and enables memorization and recall of the sequences with nearly identical or even identical images.

50. VORONKOV G.S.

*Lomonosov Moscow State University*

**Little-known mysterious phenomenon in vision: description and attempt analysis**

The paper describes a phenomenon of visual perception: in a small peek-a-boo hole there exists a web-like patch that contains inclusions of two main types, static and dynamic. Besides a description of the patch form and manifestation conditions, the paper presents a number of disclosed enigmatic patch characteristics. Based on the performed analysis, it is concluded that the patch image has a composite nature. Suggestions are made to explain some characteristics of the patch. Evidence from the wave optics, vision neurophysiology, and ophthalmology is attracted for reasoning.

51. LAVROV V.V., RUDINSKY A.V.

1

*Pavlov Institute of Physiology of the Russian Academy of Sciences, St.Petersburg*

2

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

**Interneuronal communication system: A contextual field in information coding**

The discovered conformity between superslow waves of biopotentials (caused by regulatory processes) and frequency-temporal parameters of multicellular impulses of cortical neurons contradicts to the opinion that there exist specific forms of information coding by impulse streams. A hypothesis is put forward that the exchange of information in a nervous network is implemented by a communication system that is formed due to the conformity context of the source and receiver signals, that is under the control of the contextual field.

52. SMIRNITSKAYA I.A.

*Scientific Research Institute for System Analysis, Moscow*

**About one widely used neural circuit motif**

A widely used symmetric neural circuit structure is described. Most appropriate functions of such a circuit are the locomotion control and the control of turns. Examples of this circuit functioning in vertebrates and invertebrates are described. Different types of sensory input transformation for the control of motion direction are discussed. Assumptions about the application of such a circuit for behavioral choice are made.

53. POKROVSKY A.N.

*Sanct-Petersburg State University*

**Presinaptic components of evoked potentials of the brain cortex**

The problem of separation of presinaptic components in the records of evoked potentials of the brain cortex is considered. Usually the evoked potentials are registered simultaneously at many cortical levels. The dependence of the signals on time at different levels of the cortex is rather complex. The work is aimed to realize the full separation of the evoked potential components.

POSTER SESSION 5

Friday, January 23 12:30 – 12:45

Lecture-hall 406

Chair: Prof. LITINSKII LEONID

Neural network theory

54. GOLOSHCHAPOV VLADISLAV*Scientific and cultural center SETI, Moscow*

**Ensembles of synapses as candidates for the role of structures that encode semantic properties in artificial neural networks**

This research addresses the issue of searching ANN structural elements which disentangle semantic properties of the input signal and directly encode network responses at the output. It is shown that ensembles of synapses in ANN could be such structural elements for some types of networks. The research considers the possible ways of encoding semantic properties in ensembles of synapses. By the example of a trained ANN, the author proves the existence of such ensembles and shows a significant difference between the role of synapses placed both within and outside of the ensembles.

55. SHIBZUKHOV Z.M., CHEREDNIKOV D. Y.

1

*Center «Antistikhia», EMERCOM of Russia*

2

*Moscow State Pedagogical University*

**About models of artificial neurons of aggregational type**

A new class of models of artificial neurons is described in this work. These models are based on some assumptions: 1) contributions of synapses are summing with the help of certain aggregation operation; 2) contribution of complex synapse or synaptic cluster is computed with the help of another aggregation operation on the set of simple synapses. These models includes a big part of known functional models of neurons.

56. KISELEV M. V.

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

**Creation of spiking neural networks with desired properties using empirical models**

Relationships between neuron properties, structural characteristics of spiking neural network and its behavior cannot be obtained in explicit analytical form even for simplest neuron models and homogenous networks. It is proposed to use for this purpose empirical models produced by multiple adaptive regression splines and other data mining methods. This approach is illustrated for neural network consisting of LIF neurons with homeostatic properties.

POSTER SESSION 6

Friday, January 23 12:45 – 13:00

Lecture-hall 406

Chair: Prof. LITINSKII LEONID

Adaptive behavior and evolutionary modelling

57. ANFILETS S.V., SHUTS V.N.*Brest State Technical University, Belarus*

**The use of artificial immune system to optimize the management of traffic lights cycle**

The focus of this article is the problem of adaptive control at traffic light object. Control method is based on calculating the length of the cycle phases. Used method for the calculation phases based on the predicted values of traffic at the intersection and algorithm of artificial immune systems.

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

1

*Scientific Research Institute for System Analysis, Moscow*

2

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

**Modeling of predictions by fishes studying labyrinths**

The initial stage of modeling of cognitive abilities of the fishes studying labyrinths is presented. The model of the generation of fish’s predictions is constructed. According the model, fish prediction assurance increases. The results of computer simulations agree qualitatively with experimental biological data.

59. ANOKHIN MIKHAIL

*Institute of the Service Sector and Entrepreneurship of DSTU, Shakhty*

**Development of neuron-like control system for dummy agent in a dynamic environment**

The article discusses the process of building and architectural features of the neural network control system of the animat body placed in dynamic environment. Animat body is a complex dynamic object with many degrees of freedom. Considered control system solves two problems: links body components together into a single system and provides basic body reactions to external dynamic influences.

SESSION 6

Friday, January 23 14:00 – 15:30

Lecture-hall 406

Chair: Prof. KAZANOVICH YAKOV

Neurobiology

60. V.V. KOZUNOV, A.E. OSSADTCHI1

*Neurocognitive Research Center (MEG Center), Moscow*

2

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

**GALA: Group Analysis Leads to Accuracy, a novel approach for solving the inverse problem in exploratory analysis of group MEG recordings**

Although MEG/EEG signals are highly variable between subjects, they allow characterizing systematic changes of cortical activity in both space and time. Traditionally, to discover sources of such activity a two-step procedure is used. The first step is a transition from sensor to source space by the means of solving an ill-posed inverse problem for each subject individually. The second step is mapping of cortical regions consistently active across subjects. In practice, however, the first step often leads to a set of active cortical regions whose location and activation timecourses display a great amount of interindividual variability hindering the subsequent group analysis. Here we propose Group Analysis Leads to Accuracy (GALA) - a solution that combines the two basic steps into one. In contrast to many other methods GALA takes advantage of individual variations of cortical geometry and sensor locations. Our method exploits the ensuing variability in electromagnetic forward model as a source of additional information. We assume that for different subjects functionally identical cortical regions are located in close proximity and partially overlap and their time courses are correlated (but not identical). This relaxed similarity constraint on the inverse solution can be expressed within a solid probabilistic framework, allowing the development of an algorithm to solve the inverse problem jointly for all subjects.

61. RATUSHNYAK A.S., PROSKURA A.L., ZAPARA T.A.

*The Institute of Computational Technologies of SB RAS (ICT SB RAS), Novosibirsk*

**Analysis and possible directions of the reengineering of self-organizing neural systems based on convergence technologies**

Analysis of mechanisms of neuron’s intracellular information networks allow to reveal the basics of associative and predictive properties of living systems at the molecular level. On the basis of the convergence of technologies it is likely to make it possible to develop molecular neural information nuclei capable as their living prototypes to self-organization in complex with the functional architecture like a brain and the ability to perform fairly complex cognitive functions. Analysis of mechanisms of neuron’s intracellular information networks allow to reveal the basics of associative and predictive properties of living systems at the molecular level. On the basis of the convergence of technologies it is likely to make it possible to develop molecular neural information nuclei capable as their living prototypes of self-organization in complex with the functional architecture like a brain and the ability to perform fairly complex cognitive functions.

62. CHIZHOV, A.V.

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

**Analysis of neuronal sensitivity to input signals with the help of neuronal ensemble model**

Analysis of neuronal sensitivity to input signals is performed by using a realistic model of the statistical ensemble of Hodgkin-Huxley-type neurons, formulated in terms of the refractory density. The amplitude of current step stimuli and the noise amplitude are varied. The model reproduces known experimental data of many-trial patch-clamp recordings of spiking activity in single neurons, which reveal rapid reaction of the population firing rate to the stimulus changes.

63. BIBIKOV N.G., NIZAMOV S.V.

*N.N. Andreyev Acoustics Institute, Moscow*

**Some temporary features, those evokes responses of neurons in the auditory system**

In the vertebrate brain there are several consecutive nuclei those proceed the analysis of the received sound. This analysis based mainly on the identification of temporal characteristics of the signal in different frequency channels. Due to the nonlinearity of neuronal processing, correlation approach is insufficient for understanding these processes. Some methods to assess the contribution in the neuronal response the instantaneous amplitude and the rate of amplitude changes. The example of their application is presented.

64. DICK O.E., GLAZOV A.L.

1

*Pavlov Institute of Physiology of the Russian Academy of Sciences, St.Petersburg*

2

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

**Multifractal analysis of tremor of the human hand under the motor dysfunction**

Multifractal properties of involuntary shaking (tremor) arising during the performance of the motor task (sustaining effort of fingers under isometric conditions) have been examined. The wavelet and multifractal parameters have been gained. These parameters allow us to find correlations between clinic symptoms of tremor before the antiparkinsonian treatment, their decrease after drug administration and changes of the parameters. The suggested analytic approach for study of integrative activity of the central nervous system during realization of the motor task enables to estimate quantitatively the degree of deviation of the motor function from the healthy one.

65. L.N. PODLADCHIKOVA, T.I. KOLTUNOVA, D.G. SHAPOSHNIKOV

1

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

2

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

**The peculiarities of the viewing trajectory local elements: object-return gaze fixations**

The results of experimental data analysis about the features of the return fixations during the complex images and textures viewing are presented. It is shown that the duration of fixations and saccadic amplitude in the vicinity of the return fixations differ from the average values of these parameters. The possibility of using the results obtained to develop a model that provides a realistic simulation of the viewing trajectories in specific psychophysical tests is considered.