Sections

Neuroinformatics - 2018


SESSION 1

Tuesday, October 9                    17:30 – 19:00
Lecture-hall Алексеевский зал

Chair: Prof. TEREKHOV SERGE

Applications of neural networks

1. * GONCHAROV P. V., OSOSKOV G. A., NECHAEVSKIY A. V., UZHINSKIY A. V., NESTSIARENIA I. G.
1Sukhoi State Technical University of Gomel
2Joint Institute for Nuclear Research
Disease detection on the plant leaves by deep learning

Plants disease detection is very popular field of study. Many promising results were already obtained but it is still only few real-life applications that can make farmer’s life easier. The aim of our research is solving the problem of detection and preventing diseases of agricultural crops with the help of deep learning. We collected a special database of the grapes leaves consisting of four set of images, including healthy, Esca, Black rot and Chlorosis diseases. We reached over 90% accuracy using a deep siamese convolutional network. Comparative results of using various models and plants are presented.

2. * IRINA KNYAZEVA (INSTITUTE OF INFORMATION AND COMPUTATIONAL TECHNOLOGIES, ALMATY, KAZAKHSTAN) ALEXANDER EFITOROV (D.V.SKOBELTSYN INSTITUTE OF NUCLEAR PHYSICS, M.V.LOMONOSOV MOSCOW STATE UNIVERSITY) BOY
1The Central Astronomical Observatory of the Russian Academy of Sciences at Pulkovo, Saint-Petersburg
2Lomonosov Moscow State University
3Bekhtereva Human Brain Institute RAS, St.Petersburg
4Skobeltsyn Institute of Nuclear Physics Lomonosov Moscow State University
Single trial EEG classification of tasks with dominance of mental and sensory attention with deep learning approach

In this paper, we present classification algorithms based on single-trial ElectroEncephaloGraphy (EEG) during the performance of tasks with the dominance of mental and sensory attention. Statistical data analysis showed numerous significant differences of EEG wavelet spectra density during this task at the group level. We decided to use wavelet power spectral density (PSD) computed in each channel for single trial as the source of feature extraction for the classification task. To obtain a low-dimensional representation of PSD image convolutional autoencoder (CNN) was trained. With this encoded representation binary classification for each subject with multilayer perceptron (MLP) were performed. The classification error varies depending on the subject with the average true classification rate is 83.4\%, and the standard deviation is 6.6%. So this approach potentially could be used in the tasks where pattern classification is used, such as a clinical decision or in Brain-Computer Interface (BCI) system

3. * OZHEGOV GRIGORY ANDREEVICH, OPRISHKO ALEXANDER VLADIMIROVICH
Bauman Moscow State Technical University
IMAGE RECONSTRUCTION WITH APPLICATION A HYBRID INTELLIGENT SYSTEM

In this article, is considered the solution of the problem of image reconstruction using a hybrid intelligent system. The research work is devoted to the study of machine learning algorithms for the solution of the problem of image reconstruction, in particular for image reconstruction of a CT scan of a damaged skull. The research is based on the method of image reconstruction using the generative adversarial networks, which allows to recognize the context of the image and restore the damaged part of significant dimensions with the necessary accuracy. The method was implemented in python using «tensorflow» - machine learning library. The carried out experimental researches have shown efficiency of the developed method, and also its recommended architecture and a way of preparation of training sample for getting necessary results.

4. * CHERNENKIY MIKHAIL VALERIEVICH, MOGILNIKOV ILIA ANDREEVICH, ILIN VALERY SERGEEVICH, POPOVA MARINA SERGEEVNA
Bauman Moscow State Technical University
Development of an intelligent system for determining the type and route number of a means of ground public transport

The method of solving the problem of determining the type and route number of the means of ground public transport is proposed. Studies have been carried out on the choice of architectures for convolutional neural networks and their parameters for solving the problems of recognizing ground transportation vehicles and digits of the route number in the image. Methods for optimizing the operation of the system for further use in the system of supporting movement in an urban environment for people with visual eye sight are suggested.

5. * O.D.KOTOVA, A.O.EFITOROV, S.A.DOLENKO
1Lomonosov Moscow State University
2Skobeltsyn Institute of Nuclear Physics Lomonosov Moscow State University
Comparative analysis of machine learning methods in solving the inverse problem in spectroscopy

The paper presents the results of a comparative analysis of the use of neural networks (multilayer perceptrons), methods of linear regression and random forest to solve the problem of determining the concentrations of dissolved inorganic salts in multicomponent aqueous solutions by Raman scattering spectra. It is shown that use of neural networks, taking into account the nonlinearity of the problem, allows one to obtain a lower solution error of the obtainedregression model.

6. VALENTIN GANCHENKO, ALEXANDR DOUDKIN
United Institute of Informatics Problems of the National Academy of Sciences of Belarus, Minsk
RECOGNITION OF AGRICULTURAL VEGETATION STATE ON AEROPHOTOGRAPHY DATA BASED ON CONVOLUTIONAL NETWORKS

In this paper we consider task of recognition of agricultural vegetation state according to aerial photography data of different spatial resolution. As a basis for recognition, a classifier is used that allows to classify input image into three classes: "healthy vegetation", "diseased vegetation" and "soil". The proposed classifier is constructed from two convolutional neural networks, allowing to classify image into two classes.

SESSION 2

Wednesday, October 10                    10:30 – 11:15
Lecture-hall Алексеевский зал

Chair: Prof. SHUMSKIY SERGEY

Adaptive behavior and evolutionary modelling

7. SAMER EL-KHATIB, YURI SKOBTSOV , SERGEY RODZIN
1Southern Federal University, Rostov-on-Don
2Saint Petersburg State University of Aerospace Instrumentation
Modified Exponential Particle Swarm Optimization algorithm for medical images segmentation

Modified Exponential Particle Swarm Optimization algorithm is proposed. Done execution time comparison with existing segmentation techniques. Found, that execution time, using proposed method outperforms existing segmentation techniques, including graph-based algorithms. Images from Ossirix image dataset and real patients’ images were used for testing. Developed method was tested using the Ossirix benchmark with magnetic-resonance images (MRI) with various nature and different quality. The results of method’s work and a comparison with competing segmentation methods are presented in the form of a time table of segmentation methods.

8. KOTOV VLADIMIR BORISOVICH
Scientific Research Institute for System Analysis, Moscow
Temporal discounting in the goal-directed behavior model

The model of modification of state value map is considered, supplemented with temporal discounting instruments. It is shown that taking into account the state value discounting results in more “reasonable” goal-directed behavior. Modifiability of the discounting rates leads to differentiation of the short-term goals associated with urgent needs satisfaction and the long-term goals corresponded to strategically comfortable states.

9. DYACHUK P.P., DYACHUK P.P.(JR.), PEREGUDOVA I.P., SHADRIN I.V.
1Krasnoyarsk State Pedagogical University named after V. P. Astafyev
2Siberian Federal University, Krasnoyarsk
MARKOVSKI MODEL OF TRAINING ACTIVE AGENCY IDENTIFICATION OF THE OBJECT, IN CONDITIONS OF INSTITUTIONAL FEEDBACK

We consider the Markov model of object identification as an active agent (AA) in the problem environment, in the process of its interaction with the finite liquidator "Liquidator", which provides control and cancellation of the incorrect actions of AA. The probability distribution of the states of the finite automaton of the "Liquidator" is found. An example of the implementation of the trajectory of total remuneration is given. A diagram of the scattering of target states of AA without aftereffect in the space of total remuneration R and the number of actions k is constructed.

SESSION 3

Wednesday, October 10                    11:15 – 13:00
Lecture-hall Алексеевский зал

Chair: Prof. LEV TSITOLOVSKY

Neurobiology

10. O.E. DICK
Pavlov Institute of Physiology of the Russian Academy of Sciences, St.Petersburg
CHANGES IN DYNAMICAL COMPLEXITY OF TREMOR WITH INCREASING THE MOTOR DISTORTIONS

The task is to test the hypothesis that dynamical complexity of patterns of the human hand tremor accompanied the motor task performance decreases with increasing the disease severity. For solving the task the tremor patterns are examined by the wavelet-transform modulus maxima method and the recur-rence quantification analysis. Our results demonstrate that the dynamical com-plexity of tremor patterns decreases in larger degree for Parkinson’ disease, than for patients with the syndrome of the essential tremor, that is, it declines with increasing the severity degree of motor disorders.

11. ANTON V. CHIZHOV
Ioffe Physical Technical Institute, Russian Academy of Sciences, St Petersburg
Epileptic Seizure Propagation Across Cortical Tissue: Simple Model Based On Potassium Diffusion

Mechanisms of epileptic discharge generation and spread are not well known. Recently proposed simple biophysical model of interictal and ictal discharges is generalized here to the case of spatial propagation. Diffusion of the extracellular potassium concentration is assumed to govern the spatial spread of spiking activity across cortical tissue. Simulations are consistent with experimental registrations of waves in pro-epileptic conditions, propagating at a speed of about 0.5 mm/s.

12. * A.V. BURDAKOV, A.O. UKHAROV, M.P. MYALKIN, V.I. ТEREKHOV
1Bauman Moscow State Technical University
2PH Informatics
FORECASTING OF INFLUENZA-LIKE-ILLNESS INCIDENCE IN AMUR REGION WITH NEURAL NETWORKS

Influenza-like-Illness forecasting strengthens disease control and prevention. The virus, host and host behavior factors influencing outbreaks are thoroughly studied. A range of statistical and machine learning forecasting methods was developed. This paper discusses selection of the main influencing factors, development of the forecast evaluation criteria, and prediction with a Long-Short Term Memory model. It was trained and tested on 2007-18 data set, and compared to ARIMA, LOESS, MVR.

13. IRINA KNYAZEVA, BOYTSOVA YULIA, SERGEY DANKO, NIKOLAY MAKARENKO
1The Central Astronomical Observatory of the Russian Academy of Sciences at Pulkovo, Saint-Petersburg
2Bekhtereva Human Brain Institute RAS, St.Petersburg
Spatial and temporal dynamics of EEG parameters during performance of tasks with dominance of mental and sensory attention

The purpose of this work was to analyze the spatial and temporal dynamics of EEG parameters during performance of tasks with dominance of mental and sensory attention. In addition, an attempt to identify differences in dynamics of EEG parameters in such close mentally oriented tasks as a productive and reproductive imagination was made. EEG wavelet spectra and phase coherence (relationships between EEG-channels) were studied for theta, alpha1, 2 and beta1, 2 frequency ranges of EEG. Analysis of the data showed significant numerous differences of EEG wavelet spectra and phase coherence between tasks with the dominance of mental and sensory attention and also between such close mental states as productive and reproductive imagination. Time dynamics of EEG differences, obtained in different frequency ranges, allows us to trace which processes were consequentially involved in realization of investigated states on different time intervals and provide new information for understanding of brain mechanisms of these states.

14. OLEG KUZNETSOV, NIKOLAY BAZENKOV, BORIS BOLDYSHEV, LIUDMILA ZHILYAKOVA, SERGEY KULIVEC, ILYA CHISTOPOLSKY
1V. A. Trapeznikov Institute of Control Sciences of Russian Academy of Sciences, Moscow
2Koltzov Institute of Developmental Biology of Russian Academy of Sciences
An asynchronous approach to discrete modeling of nonsinaptic interac-tions between neurons

An asynchronous discrete model of nonsynaptic interactions between neurons is proposed. In the model, neurons interact by emitting neuro-transmitters to the shared extracellular space. We introduce dynamics of membrane potentials that comprises two factors: the endogenous rates of change depending on the neurons firing type and the exogenous rate of change, depending on the concentrations of neurotransmitters that the neuron is sensitive to. Differences in endogenous rates of different neurons leads to asynchronous neural interactions and significant variability of phase durations in the activity patterns. The algorithm computing the behavior of the proposed model is provided.

15. BIBIKOV NIKOLAY
N.N. Andreyev Acoustics Institute, Moscow
Interaction of neural elements of the auditory system in the analysis of signals with noise amplitude modulation

The results of the study of correlations between the reactions of single neurons of the auditory center from the middle brain of the tra-vyan frog in the process of coding of tonal signals modulated by repetitive segments of low-frequency noise amplitude modulation are presented. Registration of activity of nearby cells is carried out using the classification of neurons in the form of spike. Along with the study of the activity of each of the selected cells, the interdependence of neural elements registered by one microelectrode is investigated, the task of assessing their mutual correlation is carried out. It is found that in most of the selected pairs of cle-current in the presence of a pronounced correlation determined by the influence of sound, the internal correlation independent of the signal is weakly expressed

POSTER SESSION 1

Wednesday, October 10                    14:00 – 15:30
Lecture-hall Алексеевский зал

Chair: Prof. TEREKHOV SERGE

Applications of neural networks

16. KISELEV M. V.
The Chuvash state university named after I. N. Ulyanov
A GENERAL PURPOSE ALGORITHM FOR CODING/DECODING CONTINUOUS SIGNAL TO SPIKE FORM

An algorithm called Symmetric Integro-Differential Conversion (SIDC) serving to encode continuous real-valued signal to spike form is described. One dynamic numeric signal is converted to 4 spike trains. 2 spiking signal lines serve as a low-pass filter, while 2 other play the role of high-pass filter. This approach allows accurate encoding signals with high variability of spectral properties. A reverse conversion algorithm is proposed which is used to assure that the resulting spike signal preserves information about the original signal to a sufficient extent.

17. EGOR A. DEGILEVICH, ALEKSANDRA M. KOBICHEVA, VALERY A. KOZHIN, ANASTASIA D. SUBBOTA, ILYA U. SURIKOV, DMITRY А. TARKHOV, VALERY A. TERESHIN
Peter the Great St. Petersburg Polytechnic University
Comparative testing of the neural network and semiempirical method on the stabilization problem of inverted pendulum

In this paper we conduct a comparative testing of control methods for dynamic systems on the example of solving the problem of bringing a pendulum into an unstable equilibrium position in a minimum time and in conditions of limited control. We have compared a method based on a neural network approach and a method using author modifications of the algorithm for constructing approximate solutions of a nonlinear system. In the Wolfram Mathematica system we have presented the results of computationa

18. EGORCHEV M.V., TIUMENTSEV YU.V.
Moscow Aviation Institute (National Research University)
Homotopy Continuation Training Method for Semi-Empirical Continuous-Time State-Space Neural Network Models

Recurrent neural networks present a powerful class of nonlinear dynamical system models that hold great potential for real-world applications. However, the traditional gradient-based optimization methods often fail to find a sufficiently good solution unless the initial guess for parameter values lies very close to it. We propose a training method based on the homotopy continuation approach with variable prediction horizon which allows circumventing this issue. Simulation results confirm the efficiency of this training method.

19. KOZLOV DMITRY S., TIUMENTSEV YURY V.
Moscow Aviation Institute (National Research University)
Neural Network Based Semi-empirical Models of 3D-Motion of Hypersonic Vehicle

We consider the problem of mathematical modeling and computer simulation of nonlinear controlled dynamical systems represented by differential-algebraic equations of index 1. The problem is proposed to be solved in the framework of a neural network based semi-empirical approach combining theoretical knowledge for the object with training tools of artificial neural network field. Special form neural network based semi-empirical models implementing an implicit scheme of numerical integration inside the activation function are proposed. The training of the semi-empirical model allows elaborating the models of aerodynamic coefficients implemented as a part of it. A semi-empirical model using as theoretical knowledge the equations of the full model of the hypersonic vehicle motion in the specific phase of descent in the upper atmosphere are presented. The results of simulation for the identification task for the aerodynamic lift coefficient implemented as an ANN-module of the semi-empirical model of the hypersonic vehicle motion are presented.

20. IGOR ISAEV, EUGENY OBORNEV, IVAN OBORNEV, MIKHAIL SHIMELEVICH, AND SERGEY DOLENKO
1Skobeltsyn Institute of Nuclear Physics Lomonosov Moscow State University
2Russian State Geological Prospecting University n. a. Sergo Ordzhonikidzе, Moscow
Neural Network Recognition of the Type of Parameterization Scheme for Magnetotelluric Data

The inverse problem of magnetotelluric sounding is a highly non-linear ill-posed inverse problem with high dimension both at the input and at the output. One way to reduce the incorrectness is to narrow the scope of the problem. In our case, this can be implemented in the form of a complex algorithm, which first makes the choice of one of the narrower classes of geological sections and then performs the solution of the regression inverse problem within the selected class. In the present study, we investigate the effectiveness of the implementation of the first phase of this algorithm. The neural network solution of the problem of classification of magnetotelluric sounding data was considered. We estimate the maximum accuracy of classification, perform search for optimal parameters, and test the results for resilience to noise in the data

21. D.A. TARKHOV, M.R. BORTKOVSKAYA, T.T. KAVERZNEVA, D.R. KAPITSIN, I.A. SHISHKINA, D.A. SEMENOVA, P.P. UDALOV AND I.U. ZULKARNAY
1Peter the Great St. Petersburg Polytechnic University
2Bashkir State University, Ufa
SEMIEMPIRICAL MODEL OF THE REAL MEMBRANE BENDING

The solution of the problem of modeling the deflection of a loaded circular mem-brane in the symmetric case is considered. The models that express the depend-ence of the deflection of the membrane from the distance to the center are com-pared. The first is based on the analytical solution of the equations of equilibri-um conditions. The second one was obtained with the help of the original modi-fication of the refined Euler method. The third is constructed in the form of an output of a neural network. The coefficients of the models were chosen from the data obtained experimentally.

22. ENGEL EKATERINA ALEKSANDROVNA , ENGEL NIKITA EVGENEVICH
Katanov Khakass State University, Abakan
TEMPERATURE FORECASTING BASED ON THE MULTI-AGENT ADAPTIVE FUZZY NEURONET

This article presents a multi-agent adaptive fuzzy neuronet for the average monthly ambient temperature forecasting. The agents of the multi-agent adaptive fuzzy neuronet are fulfilled based on two-layered neural network. An automatic generation of the optimal architecture’s parameters of a neuronet is the most complex task. In order to train the effective multi-agent adaptive fuzzy neuronet we modified the Ant Lion Optimizer and combined it with the Levenberg-Marquardt algorithm. The simulation results show that the proposed training algorithm outperforms modified Ant Lion Optimizer and Levenberg-Marquardt algorithms in training the effective multi-agent adaptive fuzzy neuronet for the average monthly ambient temperature forecasting.

23. YURIY S. FEDORENKO, YURIY E. GAPANYUK
Bauman Moscow State Technical University
The Neural Network with Automatic Feature Selection for Solving Problems with Categorical Variables

In data analysis, feature selection depends on the domain area and requires a lot of manual work. In this paper, the regression task with categorical input data is considered. The neural network for solving this task is proposed. Its architecture and details of learning are discussed. Experiments have shown that proposed neural network works better than logistic regression with handcrafted chosen levels.

24. SELIVERSTOVA A.V., PAVLOVA D.A., TONOYAN SLAVIK A., GAPANYUK Y.E.
Bauman Moscow State Technical University
The time series forecasting of the company’s electric power consumption

The purpose of this paper is to choose the appropriate method for time series forecasting of the company’s electric power consumption. The ARIMA, GMDH, LSTM, and seq2seq methods are considered. The MSE, MAE, and MAPE metrics are used for the forecasting quality evaluation. The experi-ments results show that the best method is GMDH. The ARIMA and the LSTM may be considered as second-place methods.

25. VALENTIN MALYKH, VLAD LYALIN
The Moscow Institute of Physics and Technology (State University)
What Did You Say? On Classification of Noisy Texts

Text classification is a fundamental task in natural language processing and huge body of research has been devoted to it. But one important aspect of noise robustness received relatively low attention. In this work we are bridging this gap, introducing results on noise robustness testing of modern classification architectures on English and Russian.

26. GORBATKOV STANISLAV, FARKHIEVA SVETLANA
Financial University under the Government of the Russian Federation, Ufa Branch
AGGREGATION OF VARIABLES IN NEURAL NETWORK MODELS OF BANKRUPTCIES ON THE BASIS OF HARRINGTON'S FUNCTION. PART I.

The original neural network model of bankruptcies of corporations is developed for support of decision-making on restructuring of credit debt in bank technologies of financial management. Algorithms of optimum selection of factors for model to their subsequent compression which provide resistance of model to change of entrance data are offered. The adequacy of neural network model is estimated on real data of construction branch.

27. GORBATKOV STANISLAV, FARKHIEVA SVETLANA
Financial University under the Government of the Russian Federation, Ufa Branch
AGGREGATION OF VARIABLES IN NEURAL NETWORK MODELS OF BANKRUPTCIES ON THE BASIS OF HARRINGTON'S FUNCTION. PART II.

Results of computing experiments on check of efficiency of operation of optimum selection of factors and their subsequent compression by means of the generalized function of desirability of Harrington for the neural network logistic method (NSLM) offered in part I of the report are given. The offered NSLM allows to reduce number of exogenous variables in tens of times at simultaneous regularization of model and increase in her predictive force.

28. BURAKOV M.V.
Saint Petersburg State University of Aerospace Instrumentation
Algorithm for synthesis of an inverse neurocontroller

The problem of constructing a neurocontroller for a dynamic plant is considered. The synthesis algorithm is implemented on the basis of the concept of the inverse model and the given reference process. The inverse model is used to obtain information on the reference control signal necessary for training the neurocontroller. The neural controller is built on the basis of a feedforward neural network with delay lines for input to the output signal. The synthesis procedure involves several consecutive steps. At the first step, the control plant is identified, the neural network traced using the back propagation algorithm is used as the model. The same algorithm is used in the second step to synthesize the inverse model of the control plant. With the help of this model, the third step prepares data for training the neurocontroller in the fourth step of the algorithm. The results of a computational experiment in Simulink MatLab for determining the parameters of a neural controller for a second-order dynamic link, which confirm the practical utility of the proposed approach for a wide range of control plants, are presented.

29. KKARITSKY D.K., USTYUZHANIN K.YU., SHIRIYAZDANOV R.R., RUDNEV N.A.
Ufa State Petroleum Technological University
Machine learning for system of predictive control of fracture ennoblement plant

The case of creation of digital twin of the fracture ennoblement plant by using resent machine learning techniques is described. The method of creation of plant output prediction system based on integration of classical kinetic models and LSTM networks, of system of express estimation of catalyst activity by means of integration of SGD Regressor algorithm and Random Forest.

30. YAKOVENKO A.A., ANTROPOV A.A.
Peter the Great St. Petersburg Polytechnic University
Dynamic emergent self-organizing map for multidimensional data stream of the auditory nerve response patterns

The paper deals with the problem of multidimensional data analysis in real-time conditions. Based on the subject domain review, an architecture combining emergent self-organizing maps and the UbiSOM dynamic learning algorithm is implemented. Proposed method shows effectiveness in auditory neuronal response analysis, obtained as a result of the simulation, in response to pure tone stimuli.

31. SHEMAGINA O.V., BAKHSHIEV A.V., STANKEVICH L.A., MIKHAILOV V.V., KLIMASHOV V.YU., EMELIANOV A.A.
1Peter the Great St. Petersburg Polytechnic University
2The Institute of Applied Physics of the Russian Academy of Sciences, Nizhny Novgorod
3
4Russian State Scientific Center for Robotics and Technical Cybernetics, Saint-Petersburg
MULTI-USER SYSTEM FOR SIMULATION OF RECONFIGURABLE NEUROMORPHIC NETWORKS FOR INFORMATION PROCESSING

The paper describes a multi-user system for simulating neural networks in which each participant can create models of the study network, create complex hierarchical structures from network components, and create collections of such components that complete a united cloud knowledge base. At the same time, the peculiarity of the system is the ability to model complex irregular neural networks, based on the training rules of which lies not only parametric, but also structural training.

32. GAVRILOV ANDREY V.
Novosibirsk State Technical University
On usage of Neuromorphic Engineering in Intelligent Autonomous Robots

The article considers the opportunities and problems of using neuromorphic engineering in intelligent autonomous robots. The digital and digital-analog kinds of neurochip architecture for usage in robotics are discussed, in particular, three kinds of digital architecture: hard architecture, flexible architecture and growth architecture of neural networks.

33. V.D. KOSHUR, S.V. CHENTSOV
Siberian Federal University, Krasnoyarsk
THE DESIGN OF SMART SYSTEMS FOR ON-LINE DIAGNOSTICS AND CONTROL OF RESPONSIBLE TECHNICAL DEVICES ON THE BASE OF ELECTRONIC CLONES

The models of neural computer electronic clones for diagnostics and control are suggested to inspect nondestructively major elements of technical systems during their operations. The expediency of use of electronic clones is shown.

34. TELNYKH A.A., NUIDEL I.V.
The Institute of Applied Physics of the Russian Academy of Sciences, Nizhny Novgorod
CHARACTER RECOGNITION ON THE EXAMPLE OF THE LICENSE PLATES OF RAILWAY CARS

The problem of recognition of license plates of railway cars is considered: capture of an object from a video stream and character recognition. The attrib-utes of the object are formed, which are nonlocal binary patterns within the area of the formal code description. A weak and strong classifier is being built. The learning process performs minimization of the recognition error on the refer-ence database. In the present work, 11 types of detectors are proposed: the de-tector of the number on the railway car and the detector of all the digits from zero to nine.

POSTER SESSION 2

Wednesday, October 10                    15:30 – 16:45
Lecture-hall Алексеевский зал

Chair: Prof. BONDAREV VLADIMIR NIKOLAEVICH

Neural network theory

35. A.R. GALYAUTDINOVA, J.S. SEDOVA, D.A.TARKHOV, E.A. VARSHAVCHIK, AND A.N. VASILYEV
Peter the Great St. Petersburg Polytechnic University
COMPARATIVE TEST OF EVOLUTIONARY ALGORITHMS TO BUILD AN APPROXIMATE NEURAL NETWORK SOLUTION OF THE MODEL BOUNDARY VALUE PROBLEM

The paper considers the neural network approach to the construction of an approximate solution of the Dirichlet boundary value problem for the Laplace equation in the unit square. We study the work of two algorithms for adding neu-rons to a configurable network in comparison with training a neural network of constant size. The quality of approximate solutions constructed with the help of neural networks for different sets of system parameters (the number of points at which the operator is calculated; the number of test points on one side of the square; the proportion of non-generated points) is estimated.

36. SMOLIN VLADIMIR
Keldysh Institute of Applied Mathematics, Moscow
AGI developers manifest project

Objective obstacles to AGI (artificial general intellectence) are the physical world nonlinearity phenomena and the curse of dimension. The subjective difficulties are the lack of common views on the ways and goals of AGI creating. The complexity implies a collective approach to problem solving. It is necessary not only to identify ways of achieving AGI, but also to coordinate views among developers. It is proposed to discuss and improve the list of fundamental principles for AGI creation.

Neurobiology

37. MARUSHKINA E.A.
P.G. Demidov Yaroslavl State University
Bursting in a system of two coupled pulsed neurons with delay

The system of two singularly perturbed differential-difference equations with delay modeling the synaptic interaction of a pair of pulsed neurons is considered. The connection between elements is selected threshold, taking into account the time delay. In the framework of the problem an algorithm for finding periodic solutions with bursting behaviour is proposed. The appearance of such relaxation oscillations in the system is a consequence of delay in the communication circuit between the oscillators. Of particular importance is the fact that the system can have a large number of coexisting relaxation oscillations.

38. A. L. PROSKURA, T. A. ZAPARA
The Institute of Computational Technologies of SB RAS (ICT SB RAS), Novosibirsk
FUNCTION AND MOLECULAR DESIGN OF THE SYNAPSE

The majority of excitatory synapses in the mammalian forebrain and the hippocampus incorporate dendritic spines. Even though their intricate molecular composition remains obscure in many aspects, available evidence suggests that these structures are highly specialized to support the short- and long-term plasticity crucial for flexible information processing. One concept that is extensively used to describe synaptic function is the Hebb’s postulate. However, the knowledge accumulated throughout following decades advocate for a broader scale in intercellular connection functions. Synaptic activity depends on interactions among sets of proteins (synaptic interactome) that assemble into complex supramolecular machines. Molecular biology, electrophysiology, and live-cell imaging studies have provided tantalizing glimpses into the inner workings of the synapse, yet fundamental questions regarding the functional organization of these “molecular nano-machines” remain to be answered. The presence of accessory receptors for secondary messengers in synapses along with receptors for primary mediator gave us the idea that these molecular constructs could be responsible for initial processing of the incoming signals. The purpose of this initial processing could be determined by analyzing and reconstructing this molecular informational machine that postsynaptic zone essentially is.

39. TEREKHIN S.A., SIDOROV K.V., FILATOVA N.N.
Tver State Technical University
Monitoring of human emotional reactions by using facial electromyography signals and brain electrical activity

This article investigates the possibility of monitoring emotional reactions from person during audiovisual stimulation. Article contains the results of experiments with “EEG/EMG” bioengineering system allowing registration of brain electrical activity signals (EEG) and facial electromyography signals (fEMG). A number of regularities were found, manifested in changes of characteristics of attractors, reconstructed from EEG and fEMG signals during stimulus perception stage and after their removal.

40. VORONKOV G.S.
Lomonosov Moscow State University
WHY SHADOWS, WHICH CREATE THE "SPECIFIC VISION DEFICIT" SPOT, DO NOT MANIFEST THEMSELVES UNDER NORMAL CONDITIONS OF VIEW: AN ANALOGY WITH THE "BLIND SPOT"

A manifestation of the "specific vision deficit" (SVD) phenomenon is the visible structured spot (SVDs-image) at the position of a bright point light source, placed on a shaded background. SVDs-image is created due to the shadows presence in the projection of a bright source on the retina; shadows (Sh) created due to heterogeneities (H) in the eye transparent structures. However, in normal circumstances we do not see any evidence of the HSh, even if the visual space is a bright white screen. Here is an attempt to analyse this observation.

41. SMIRNITSKAYA IRINA
Scientific Research Institute for System Analysis, Moscow
The role of hypothalamic suprachiasmatic nucleus in contagious itch in the mice

The structure of the system of contagious itch is disscussed. It is postulated, that contagious itch is not a goal directed behavior. That is a reason for it not to use a mirror neurons system

Adaptive behavior and evolutionary modelling

42. SOKHOVA Z. B.
Scientific Research Institute for System Analysis, Moscow
RESEARCH OF PROCESSES OF SELF-ORGANIZATION IN EVOLUTIONARY MODEL OF TRANSPARENT ECONOMY

The article is devoted to the evolutionary model of a multi-agent economic system consisting of investors and producers. Investors and producers interact in a dynamic "transparent" external environment. A specific mechanism for allocating the capital of each investor is proposed with the help of iterative estimates, which allow the investor to take into account the decisions of other investors. The influence of evolutionary processes and learning on the behavior of the model is analyzed.

43. B.K. LEBEDEV, O.B. LEBEDEV, V.B. LEBEDEV
Southern Federal University, Rostov-on-Don
Mechanisms of the swarm intellect in searching in the affine space of optimal tree-like representations of the decision

The paper describes a modified paradigm of particle swarms that provides, unlike the canonical method, the possibility of using positions with integer parameter values in affine space. Mechanisms for moving particles in affine space to reduce the weight of affine bonds are considered. The developed position structures (chromosomes) are focused on the integration of swarm intelligence and genetic evolution.

Neural networks and cognitive sciences

44. MALAFEEV S.I., MALAFEEVA A.A.
1
2Vladimir State University named after Alexander and Nikolat Stoletovs
Algorithms of Distribution of General Loads under the Joint Work of Aggregates

The technique and algorithms for searching for the optimal distribution of the total load between parallel units operating under discrete control of their modes are considered. The description of the method of direct search of variants, the use of the genetic algorithm, the search algorithm by moving along the extremal is given. A variant of the solution of the problem is proposed with variable productivity of the units. On the basis of the considered algorithms, the software for the automated dispatch control system of the heat and power complex was developed. The use of optimal algorithms for gas boiler houses provides gas savings of up to 5% compared to traditional dispatch control.

45. VADIM L. USHAKOV, VYACHESLAV A. ORLOV, SERGEY I. KARTASHOV, DENIS G. MALAKHOV, ANASTASIA N. KOROSTELEVA, LYUDMILA I. SKITEVA, LYUDMILA YA. ZAIDELMAN, ANNA A. ZININA, VERA I. ZABOTKINA, BORIS M. VELICH
National Research Centre "Kurchatov Institute", Moscow
Contrasting Human Brain Responses to Literature Descriptions of Nature and to Technical Instructions

Semantic brain mapping of the large continuous segments of Russian-language texts

46. V.A. ORLOV, V.L. USHAKOV, S.I. KARTASHOV, D.G. MALAKHOV, A.N. KOROSTELEVA, L.I. SKITEVA, AND A.V. SAMSONOVICH
National Research Centre "Kurchatov Institute", Moscow
Functional neural networks in behavioral motivations

Functional magnetic resonance imaging (fMRI) is an effective non-invasive tool for exploration and analysis of brain functions. Here functional neural networks involved in behavioral motivations are studied using fMRI. It was found that behavioral conditions producing different motivations for action can be associated with different patterns of functional network activity. At the same time, connection can be made to dynamics of socio-emotional cognition, decision making and action control, described by the Virtual Actor model based on the eBICA cognitive architecture. These preliminary observations encourage further fMRI-based study of human social-emotional cognition. The impact is expected on the emergent technology of humanlike collaborative robots (cobots) and creative cognitive assistants.

47. A.А. FETISOVA, A.V. VARTANOV
1
2Lomonosov Moscow State University
Semantic Space and Homonymous Words

Significant differences were found in the organization of electrobiological responses of the brain to the words-homonyms, presented in different meanings by priming a certain context in the electrophysiological experiment on a sample of 14 people As a result of comparison of series before and after negative reinforcement by the method of Semantic Radical of the word entering the semantic field of one of the two studied meanings of the word-homonym, the fact of indirect long-term influence of the emotional component is revealed. This is manifested not only in the change in the meaning of the homonym word (presented in the relevant context) in the late latency (300-500 MS) in the frontal leads, but also in the reconfiguration of its meaning in a non-relevant context (in the earlier – 200 MS – latencies in the central leads). Thus, the results objectively indicate the possibility of two types of reconfiguration of the semantic map (space) 1) operational reconfiguration to the current context defined in this experiment by the corresponding prime, and 2) long-term reconfiguration determined by the control action

48. FILATOVA N.N., SIDOROV K.V., SHEMAEV P.D., BODRINA N.I., REBRUN I.A.
Tver State Technical University
The bioengineering system for monitoring and cognitive activity management through emotional stimulation

The paper considers principles of construction of a two-channel bioengineering system providing a solution for two classes of tasks: monitoring of a user’s state characteristics, and cognitive activity management through brain emotional stimulation using natural sensory systems for conduction and perception of information. This article cites the testing results of laboratory version of the bioengineering system.

49. CHUGROVA M.E. BAHCHINA A.V. PARIN S.B. POLEVAYA S. A.
1N.I. Lobachevsky State University of Nizhni Novgorod
2Institute of Psychology of Russian Academy of Sciences, Moscow
3Nizhny Novgorod State Medical Academy
Autonomic regulation of students ' behavior in the process of monologue and dialogue with a public

We study the functional state of students in a context of communication of the educational process. The method of our research is a dynamic analysis of heart rate variability. We found that the presence of the public affects the functional state at the beginning of the monologue. Evaluation changes the functional state in the time of completion of the monologue. We observed the maximum level of strain of regulatory systems of organism in the middle of communication in dialogue with the public.

50. V.B. KOTOV
Scientific Research Institute for System Analysis, Moscow
Natural mathematics in the artificial brain

Natural mathematical capabilities of the human brain and prospective imbedding them in the artificial brain are topics of the paper. Steps to build the human-like apparatus for processing natural numbers and logical quantities are discussed. The principal component of this apparatus is the advanced prefrontal cortex, specifically its dorsal-lateral section.

SESSION 4

Wednesday, October 10                    17:15 – 19:00
Lecture-hall Алексеевский зал

Chair: Prof. DOLENKO SERGEY

Applications of neural networks

51. * SOROKIN DMITRY, NUZHNY ANTON
1The Moscow Institute of Physics and Technology (State University)
2The Nuclear Safety Institute of the Russian Academy of Sciences
Development of a neural network system for analysis of specialized documentation

For the analysis of information on the objects of radioactive waste disposal, it is necessary to solve the problems of thematic modeling and text classification. The growth of documentation leads to the need to automate the data processing. The analysis includes search for documents, topics, retrieval of attributes, relationships and facts. The automation of the task requires mathematical methods of information retrieval and text mining.

52. * MISHULINA O.A., EIDLINA M.A.
National Research Nuclear University (MEPhI), Moscow
CATEGORICAL DATA: AN APPROACH TO VISUALIZATION FOR CLUSTER ANALYSIS

The problem of studying the cluster structure of a set of objects with qualitative (categorical) features is considered. We propose an approach to visualization of source data and categorical data groups in a form that is convenient for human analysis and decision-making. The technique is demonstrated on a model example.

53. * KULAEV MAKSIM ALEKSANDROVICH

Application of attention-based recurrent neural networks for inverse machine transliteration of Cyrillic names

A problem of restoring the original spelling of Cyrillic names (inverse transliteration from the Latin spelling) is considered. A method of inverse transliteration using attention-based recurrent neural networks is proposed, implemented and studied. In addition, the subtask is the preparation of a sample, which makes it possible to train this network.

54. * TROFIMOV A.G., KUZNETSOVA K.E., KORSHIKOVA A.A.
1National Research Nuclear University (MEPhI), Moscow
2INTERAUTOMATIKA
Abnormal Operation Detection in Heat Power Plant Using Ensemble of Binary Classifiers

The problem of abnormal operation detection is considered for prediction of malfunctions in power plant’s equipment. Abnormal operation detection method based on multivariate state estimation technique (MSET) along with machine learning algorithms is proposed. The obtained results demonstrate that the developed model can be used to detect and predict operation anomalies in power plant equipment.

55. * GUSEVA A.I., MALYKHINA G.F., NEVELSKIY A.S.
Peter the Great St. Petersburg Polytechnic University
MULTICENSORY INTELEGENT EARLY WARNING SYSTEM FOR FIRE DETECTION ON SHIPS

The purpose of the study is to develop a multi-sensor intelligent early warning system for fire detection on the ship. An evolutionary algorithm for the optimal placement of sensors controlling such fire factors as temperature, concentration of carbon dioxide, carbon monoxide and the presence of smoke is pro-posed. It is proposed to use the evolutionary algorithm to determine the optimal coordinates of the sensors of the multi-sensor fire system, depending on the position of the source of ignition. Based on the results of modeling, the possibility of determining the type of source of burning was substantiated. Determination of the type of ignition source allows you to respond quickly and correctly to the onset of fire extinguishing. This can reduce the damage from it.

56. GAI V. E., PRESNYAKOV I. A.
Nizhny Novgorod State Technical University named after R.E. Alekseev
A METHOD OF FORMING A DEPTH MAP BASED ON THE STEREO IMAGES

The problem of depth map formation based on stereo image pairs is considered. The depth information recovery result can be used to capture the reference points of objects in film production when creating special effects, as well as in computer vision systems used on vehicles to warn the driver of a possible collision.

57. * A.A.BRYNZA,M.O.KORLYAKOVA
Bauman Moscow State Technical University Kaluga Branch
Estimation of the complexity of neural network convolutional classifier

The problem of constructing a criterion for determining the complexity of a classifier based on the architecture of a convolutional neural network is considered. An analysis was made of a family of architectures distinguished by the number of layers, the size of convolution kernels, and the dimensionality of the analyzed fragments. The estimation of the classification accuracy of the studied structures of the trained networks was performed.

SESSION 5

Thursday, October 11                    14:00 – 15:45
Lecture-hall Алексеевский зал

Chair: Prof. LITINSKII LEONID

Neural network theory

58. MIKHAIL S. TARKOV
Rzhanov Institute of Semiconductor Physics, Siberian Branch of Russian Academy of Sciences, Novosibirsk
Hopfield Associative Memory with Quantized Weights

The use of binary and multilevel memristors in the hardware neural networks implementation necessitates their weight coefficients quantization. In this paper we investigate the Hopfield network weights quantization influence on its information capacity and resistance to input data distortions. It is shown that, for a weight level number of the order of tens, the quantized weights Hopfield-Hebb network capacitance approximates its continuous weights version capacity. For a Hopfield projection network, similar result can be achieved only for a weight levels number of the order of hundreds. Experiments have shown that: 1) binary memristors should be used in Hopfield-Hebb networks, reduced by zeroing all weights in a given row which moduli are strictly less than the maximum weight in the row; 2) in the Hopfield projection networks with quantized weights, multilevel memristors with a weight levels number significantly more than two should be used, with a specific levels number depending on the stored reference vectors dimension, their particular set and the permissible input data noise level.

59. YURY S. PROSTOV, YURY V. TIUMENTSEV
Moscow Aviation Institute (National Research University)
Match-Mismatch Detection Neural Circuit Based on Multistable Neurons

An ability to detect matches and mismatches between the real world and its representation are both crucial for intelligent autonomous systems to work efficiently. Here we recall our previously proposed model of the neuron which activation characteristic (dependence between an input pattern and output signal) depends on the value of the modulation parameter and varies from a smooth sigmoid-like function to the form of a quasi-rectangular hysteresis loop. Then we propose the neural circuit based on this model of multistable hysteresis neuron. Such a neural circuit can compare expectations represented by downward signals and reality represented by upward signals. In case of matching between them the neural circuit transfers both up and downward signals. In another case, internal inhibitory neurons are activated, and transferring becomes blocked until the expected conditions are met. Besides, the changes in the modulation parameter allow fine-tuning the behavior of this neural circuit. In the end, the results of the numerical simulation are presented.

60. L. LITINSKII, I.KAGANOWA
Scientific Research Institute for System Analysis, Moscow
Spectral density of 1D Ising Model in n-vicinity method

For the 1D-Ising system we obtain an exact combinatorial expression for the spectral density in the n-vicinity of the ground state. We compare the obtained expression with the normal approximation of the spectral density that is usually used in the framework of the n-vicinity method. We discuss the reasons why the n-vicinity method does not work in the case of the 1D-Ising system.

61. ALTAISKY M.V., ZOLNIKOVA N.N., KAPUTKINA N.E., KRYLOV V.A.
1Space Research Institute
2
3Joint Institute for Nuclear Research
Quantum Neural Networks: Prospects and Implementation

Quantum neural networks (QNN) generalize artificial neural networks for the case when each neuron can be in quantum superposition of states. Due to quantum parallelism such networks can solve exponentially hard optimization problems in polynomial time. The paper presents modern trends in QNN developments based on magnetic and optical technologies, along with results of authors on the construction of QNNs based on quantum dots

62. V.A. DEMIN, D.V. NEKHAEV
National Research Centre "Kurchatov Institute", Moscow
Spiking Neural Networks learning based on a neuron activity maximizing principle

We propose a simple idea of bioinspired local dynamic rules for changing the synaptic weights of a spiking neural network (SNN) based on LIF-neurons that can be used with or without a teacher. We assume that each neuron in the biological neural network tends to maximize its activity, and we transfer this principle to the SNN training algorithm. The spiking neural network with trained forward, reciprocal and inhibitionconnections was tested on MNIST dataset; recognition accuracy 95% was obtained.

63. SMOLIN VLADIMIR, KOVALENKO EUGENE
Keldysh Institute of Applied Mathematics, Moscow
FROM DROPOUT TO DROPBACK

After the multilayered neural networks adaptation with back propagation error method, a significant number of neurons of each layer can be turned off (dropout) from calculations without degrading the efficiency of the task. The causes of the occurrence of useless neurons and the inverse dropback problem of neurons - candidates for ejection into the number of effective neurons are considered. Properties that lead to the necessity of dropout and ways to make them more effective are considered

64. DOROGOV A. YU.
Saint Petersburg Electrotechnical University "LETI"
REGULAR TWO-DIMENSIONAL TUNABLE TRANSFORMATIONS WITH ADDITIONAL PLANES

The algorithm of construction and training of two-dimensional regular tunable transformations is considered. The learning algorithm has an analytical representation, is absolutely stable and converges in a finite number of steps. Additional planes significantly expand the information capacity of the tunable transformation. The control of planes in the learning mode is realized by numerical codes with a regular structure.

SESSION 6

Thursday, October 11                    15:45 – 16:45
Lecture-hall Алексеевский зал

Chair: Prof. CHIZHOV ANTON

Neurobiology

65. BOZHOKIN S.V., SUSLOVA I.B., TARAKANOV D.E.
Peter the Great St. Petersburg Polytechnic University
CALCULATION OF EEG WAVE PROPAGATION VELOCITIES ALONG THE SURFACE OF BRAIN CORTEX

Brain electroencephalogram is represented as a set of neural electrical activity bursts in different spectral ranges. Time-frequency properties of the bursts are studied by the method of continuous wavelet transform and spectral integrals. The velocities of delta, theta, alfa, and beta waves of neural activity along the surface of brain cortex are calculated.

66. TSITOLOVSKY
The Moscow Institute of Physics and Technology (State University)
MODELING CONSCIOUSNESS: GENERATION OF GOALS AND HOMEOSTASES

The properties of conscious actions are difficult to reconcile with the fundamental laws of nature. There are different goals of behavior, the main of which is the achievement of the preservation of the life. This is function of homeostasis. Damage to neurons provokes the turning on homeostasis and causes an elementary negative sensation. The article argues that the tension of homeostatic compensation can lead to the generation of targeted actions. The question arises whether it is possible to organize consciousness in an artificial system? We answer this question positively and describe the way to solve the problem.

67. SERGEI ALEXANDROVICH KOZHUKHOV
Institute of Higher Nervous Activity and Neurophysiology of RAS, Moscow
A model of temporary changes of optimal orientation in primary visual cortex neurons

as was shown in experiments a peak orientation in primaru visual cortex neurons systematically changes during the response time course. However, such orientation shifts don't reproduce by majority of modern biophysical models. In recent research we improved a ring rate model of orientation hypercolumn and condider asymmetries in spatial distribution of horizontal connections in it. We will show that consideration of such asymmetries is important for reproduction of these dynamic changes

68. SMIRNITSKAYA IRINA
Scientific Research Institute for System Analysis, Moscow
The structure of hypothalamus and it's role in the control of behavior

The structure and function of hypothalamus is discussed. The behavior types under it's control are arrenged in the order of their appearance in the course of evolution. It should emphasize the participation of hypothalamus in the control of those types of behaviors that effect the internal milieu. The role of lateral hypothalamus and orexin in behavior arousal is described.

SESSION 7

Thursday, October 11                    17:15 – 19:00
Lecture-hall Алексеевский зал

Chair: Prof. KAGANOV YURI

Applications of neural networks

69. IGOR ISAEV, SERGEY BURIKOV, TATIANA DOLENKO, KIRILL LAPTINSKIY, AND SERGEY DOLENKO
1Skobeltsyn Institute of Nuclear Physics Lomonosov Moscow State University
2Lomonosov Moscow State University
Artificial Neural Networks for Diagnostics of Water-Ethanol Solutions by Raman Spectra

The present paper is devoted to elaboration of a method of diagnostics of alcoholic beverages using artificial neural networks: the inverse problem of spectroscopy – determination of concentrations of ethanol, methanol, fusel oil, ethyl acetate in water-ethanol solutions – was solved using Raman spectra. The following accuracies of concentration determination were obtained: 0.25% vol. for ethanol, 0.19% vol. for fusel oil, 0.35% vol. for methanol, and 0.29% vol. for ethyl acetate. The obtained results demonstrate the prospects of using Raman spectroscopy in combination with modern data processing methods (artificial neural networks) for elaboration of an express non-contact method of detection of harmful and dangerous impurities in alcoholic beverages, as well as for the detection of counterfeit and low-quality beverages.

70. I.S. FOMIN, S.R. ORLOVA, D.A. GROMOSHINSKII, A.V. BAKHSHIEV
1Russian State Scientific Center for Robotics and Technical Cybernetics, Saint-Petersburg
2Peter the Great St. Petersburg Polytechnic University
Object Detection in Docking Control Tasks Using YOLO and MobileNet

A problem of visual landmarks detection on images received during process of docking of space ship to the International Space Station discussed in this paper. All images received in very hard conditions and from analog cameras and have very low quality. Demonstrated that object detection system YOLO v2 with MobileNet as object detector can be trained on prepared dataset of images, received from video camera during space docking. This network com-bination approaches mAP up to 85,1%.

71. SHAPOSHNIKOV D.G., PODLADCHIKOVA L.N.

Detection of initial moment of head motion by neural network modules

The results of application of the neural network modules to detect the head motion parameters are presented. Each module has a very simple structure and consists of a pair of excitatory and inhibitory neurons that have common center, different sizes of their receptive fields and time delay. During computer simulation was shown that the initial front of membrane potential quick changes corresponding to quick motion of head was equal 12 ms.

72. BONDAREV V.N.
Sevastopol State University
Linear Filtering Based on a Pulsed Neuron Model with an Orthogonal Filter Bank

This paper deals with the model and learning rules of a pulsed neuron, which provide the linear filtering of signals represented by pulse trains. To reduce the number of training parameters of a pulsed neuron we propose using a bank of orthogonal filters as a model of synaptic connections. For this model of pulsed neuron, we derive a supervised learning rule in a general form that can include various orthogonal basis functions. The rules minimize the mean square error between the desired and the actual output signal of a linear filter realized on the base of the pulsed neuron model. We derive two special learning rules: with set of exponential complex orthogonal functions and set of block-pulse orthogonal functions. For both set of these functions, we demonstrate rule’s properties by computer simulation of linear filters that implement high-pass filtering and double integration of the input signal transformed to pulse train. We show the impulse and frequency responses of the filters as well as the dependencies of the normalized mean square error on the number of training iterations.

73. ANTON I. GLUSHCHENKO, VLADISLAV A. PETROV

On Comparative Evaluation of Effectiveness of Neural Network and Fuzzy Logic Based Adjusters of Speed Controller for Rolling Mill Drive

The article deals with a problem of speed control of a DC electric drive of a re-verse rolling mill under the conditions of its mechanics parameters drift and influ-ence of disturbances. The analysis of existing methods to solve it is made. As a result, two intelligent methods are chosen – the neural network (proposed by the author) and the fuzzy logic based tuners of linear controllers, the efficiency of which is to be compared. The neural tuner consists of two neural networks calculating the controller parameters of the electric drive, and a rule base that determines at what moments and speed to train these networks. A general description of the fuzzy tuner is also provided. Experimental studies are made using a model of the electric drive of the rolling mill under the above mentioned conditions. The obtained results show that the neural tuner, contrary to the fuzzy one, keeps the speed overshoot within the required limits and also reduces the time of disturbance rejection by 30%.

74. YAKOVENKO A.A., SIDORENKO E.V., MALYKHINA G.F.
Peter the Great St. Petersburg Polytechnic University
Semi-Supervised Vowel Classification with Additive Noise by the Response Patterns of Auditory Nerve Model through the Label Propagation in Emergent SOM

The paper proposes an approach to stationary patterns of auditory neuronal activity analysis from the point of semi-supervised learning in self-organizing maps (SOM). The suggested approach has allowed to classify and identify complex auditory stimuli, such as speech vowels, given limited prior information about the data. A computational model of the auditory periphery has been used to obtain auditory nerve responses. Label propagation through Delaunay triangulation proximity graph, derived by SOM algorithm, is implemented to classify unlabeled units. In order to avoid the dead unit problem in emergent SOM and to improve method effectiveness, an adaptive conscience mechanism has been realized. The study has considered the influence of AWGN on the robustness of auditory stimuli identification under various SNRs. The representation of acoustic signals in the form of neuronal activity of the auditory nerve has proven more robust compared to that in the form of the most common acoustic features, such as MFCC and PLP. The approach has produced high accuracy classification, both in case of similar sounds and with high SNR.

75. DMITRIY MURODYANTS, OLEG LITVINOV
Bauman Moscow State Technical University
Investigation of spectral characteristics of interference suppression in adaptive antenna arrays on neural network control

In this paper, we analyzed various broadband noise suppression characteristics in adaptive antenna arrays (AAA) with a neural network control algorithm for phase shifters in comparison with the known result of suppressing noise interference data in AAA with the traditional phase shifter and attenuator control algorithm. The success of the neural network in the task of phase control of AAA, as well as its advantage in relation to the power loss of the useful signal, is shown.

SESSION 8

Friday, October 12                    14:00 – 15:00
Lecture-hall Алексеевский зал

Chair: Prof. USHAKOV VADIM

Neural networks and cognitive sciences

76. V.L. USHAKOV, V.A. ORLOV, D.G. MALAKHOV, S.I. KARTASHOV, A.V. MASLENNIKOVA, A.YU. ARKHIPOV, V.B. STRELEZ, M. ARSALIDOU, A.V. VARTANOV, G.P. KOSTYUK, N.V. ZAKHAROVA
1National Research Centre "Kurchatov Institute", Moscow
2Lomonosov Moscow State University
FMRI and tractographic studies of cognitive systems in the human brain at the norm and the paranoid schizophrenia

This study is aimed at a systematic study of the work of neural networks of the human brain and their architecture in norm and in schizophrenia. To obtain the neurophysiological data, a unique complex of experimental equipment for world-class neurocognitive studies was used. The data obtained showed a significant decrease in the structural connectivity relationships for the rich club coefficient for a group of schizophrenic patients compared with the norm. Perception of emotionally negative visual and audio stimuli related to delusions in patients with schizophrenia does not lead to a significant decrease in BOLD signal as compared with the norm in Calcarine_L, Cerebelum_4_5_R, ParaHippocampal_LR, Precuneus_L, Temporal_Sup_R areas. The differences found in the structural and functional patterns of cognitive-affective disorders can serve as prognostic biomarkers in patients with schizophrenia and will make a significant contribution to the development of high-tech diagnostics in the early stages of mental illness

77. KRYLOV A.K.
Institute of Psychology of Russian Academy of Sciences, Moscow
A model of neuronal activity in the search behavior and thinking

A model of search behavior and thinking has been proposed. The model based on systemic psychophysiology. The model results are compared with psychological and behavioral data.

78. MIGALEV ALEXANDER SERGEEVICH
National Research Centre "Kurchatov Institute", Moscow
Application of custom synaptic plasticity model on spiking neural networks

The computations in spiking neural networks are reviewed from a position of construction required sub-threshold function of a membrane potential or defining filter parameters. The synaptic plasticity is introduced as an instrument of forming this parameters. In this paper the custom synaptic plasticity model is proposed.

79. MANSUROVA (JACHMONINA) JU.O., POLEVAIA S.A., VETJUGOV V.V., FEDOTCHEV A.I., PARIN S.B.
1N.I. Lobachevsky State University of Nizhni Novgorod
2Nizhny Novgorod State Medical Academy
3Institute of Cell Biophysics, Pushchino, Moscow region
FEATURES OF COGNITIVE FUNCTIONS AND THEIR AUTONOMIC REGULATION AT DISRUPTION OF ENDOGENOUS OPIOID SYSTEM

Features of cognitive functions and their vegetative support in people with disabilities of the endogenous opioid system are considered. The study was carried out using information technology event-related telemetry of the heart rhythm, providing objective data on the functional state of the subject.

SESSION 9

Friday, October 12                    15:30 – 16:30
Lecture-hall Алексеевский зал

Chair: Prof. USHAKOV VADIM

Neural networks and cognitive sciences

80. TIMOFEI I. VOZNENKO, ALEXEI V. SAMSONOVICH, ALEXANDER A. GRIDNEV, AND ALIONA I. PETROVA
National Research Nuclear University (MEPhI), Moscow
The principle of implementing an assistant composer

The task of developing an assistant composer is a very interesting task, as the composer's work is complex and creative. Lately there has been done a lot of research, devoted both to systems that help composers to write melodies and systems that generate melodies. In the case of melody generating, the system o ers a composition based on some rules embedded in it according to the musical theory, or obtained by analyzing a large number of compositions with the help of neural networks, while the assistant's task is to other its ideas that can help composer with creating the work of music. In this article the basic interaction between the composer and the assistant, who helps with writing the melody, is considered. The melody itself can be conditionally divided into two com- ponents: the main melody (the main theme, the main idea of the work), and its accompaniment (the decoration of the main melody). Accordingly, the composer's assistant can help with composing both the main melody and its accompaniment. In this article we consider possible problems that the composer may encounter with while composing melodies,as well as possible ways of implementing an assistant which will be able to solve them.

81. * TEREKHOV V.I., BUKLIN S.V., CHERNENKIY I.M., YAKUBOV A.R.
Bauman Moscow State Technical University
COGNITIVE VISIBILITY IN THE TASKS OF SUPPORT FOR THE ADOPTION OF ADMINISTRATIVE DECISIONS

As a promising method in support of management decisions, the method of dynamic metaanamorphosis is considered, which allows to activate mechanisms of visual-figurative thinking of the person making a decision in a complex situation. The essence of the method, algorithm and examples of constructed metaana-morphoses are presented in the article with the use of an integral indicator combining the indicators of different physical nature. Directions for further research have been determined.

82. VVEDENSKY V.L., GURTOVOY K.G.
National Research Centre "Kurchatov Institute", Moscow
The Start and the End of the Word Perception

The effect of distortions of the speech sound on the perception of words is considered. A number of characteristic phenomena have been discovered that can become the basis of individualized experiments without the use of averaging and accumulation procedures.

83. MEILIKHOV E.Z, FARZETDINOVA R.M.
National Research Centre "Kurchatov Institute", Moscow
On the memory consolidation in the brain neuronal net

There are two types of memory - short-term and long-term ones. First, the former arises and then the latter one (in the course of the so called consolidation process). Own neuronal networks (engrams) in the brain correspond to each of those memories, and our goal is to understand what is the difference between those networks from viewpoint of their structural properties. It is not about the special biochemical structure of some neurons or synapses arising under the memory consolidation, but about some total topological properties of those brain networks which are associated with the stored pattern. In other words, could the topological reconstruction of the neuronal network promote the memory consolidation and transfer it into the long-term form? The model consideration of that phenomena shows that such a process is quite possible. For that to happen, two conditions have to be met: i) the neuronal net should be, initially, the scale-free one, and ii) the memory consolidation should proceed via the building of long-range links that arise at this stage, for instance, by means of new axon-neuron synaptic contacts.



Российская академия наук Министерство образования и науки Российской Федерации Госкорпорация «Росатом» НИЯУ МИФИ НИИСИ РАН МАИ ТРИНИТИ