## Sections

### Neuroinformatics - 2017

SESSION 1

Monday, October 2 17:00 – 18:30

Lecture-hall Алексеевский зал

Chair: Prof. LITINSKII LEONID

Neural network theory

1. YURY S. PROSTOV, YURY V. TIUMENTSEV*Moscow Aviation Institute (National Research University)*

**Adaptive Gateway Element Based on a Recurrent Neurodynamical Model**

Dynamic model of a recurrent neuron with a sigmoidal activation function is considered. It is shown that with the presence of a modulation parameter its activation characteristic (dependence between input pattern and output signal) varies from a smooth sigmoid-like function to the form of a quasi-rectangular hysteresis loop. We demonstrate how a gateway element can be build using a structure with two recognizing neurons and one output neuron. It is shown how its functional properties change due to changes in the value of the modulation parameter. Such gateway element can take the output value based on a weighted sum of signals from the recognizing neurons. On the other hand it can perform a complex binary-like calculation with the input patterns. We demonstrate that in this case it can be used as a coincidence detector even for disjoint-in-time patterns. Futhermore, under certain extreme conditions it can be triggered even if only the one input pattern was recognized. Also the results of numerical simulations presented and some directions for futher development suggested.

2. MIKHAILYUK T.E., ZHERNAKOV S.V.

*Ufa State Aviation Technical University*

**Implementation of a gate neural network based on combinatorial logic elements**

The disadvantages of the application of existing approaches to the construction of a neural network basis are analyzed. A model of a gate neural network using a mathematical apparatus of Boolean algebra is developed. Formal representations of the gate network are derived. The learning algorithm is proposed. The features of the model are considered, and the conclusion about its possibilities and fields of application is made.

3. NUZHNY ANTON SERGEEVICH

*The Nuclear Safety Institute of the Russian Academy of Sciences*

**Bayesian regularization in the problem of weight coefficients fixing for neural nets and decision trees ensembles**

The supervised-training problem is discussed. Solution is searched in a form of ensemble of weighted weak predictors. Weights are chose with orthogonalized basic functions method. To reduce overfitting training error is minimized with Gaussian stabilized functional and regularization parameter is searched with Bayesian method.

4. I.V. ISAEV, S.A. DOLENKO

*Skobeltsyn Institute of Nuclear Physics Lomonosov Moscow State University*

**ADDING NOISE DURING TRAINING AS A METHOD TO INCREASE RESILIENCE OF NEURAL NETWORK SOLUTION OF INVERSE PROBLEMS: TEST ON THE DATA OF MAGNETOTELLURIC SOUNDING PROBLEM**

In their previous studies, the authors proposed to use the approach associated with adding noise to the training set when training neural networks to solve inverse problems. For a model inverse problem it was shown that this allows increasing the resilience of neural network solution to noise in the input data with different distributions and various intensity of noise. In the present study, the observed effect was tested on the data of the problem of magnetotelluric sounding.

5. V.I. TEREKHOV, I.M. CHERNENKY, S.V. MINAKOVA, YU.E. GAPANYUK

*Bauman Moscow State Technical University*

**The implementation of training method with node splitting to fully connected neural network with two hidden layers**

The article discusses a brief description of the method of training of the neural network with node splitting. Recommendations for network initialization and requirements for training settings are given. The comparison and analysis of results of proposed network training are performed. The results of the network training with node splitting using various initialization settings are discussed. The accuracy on the test samples, the speed of network training and the optimal number of neurons in the hidden layers of the neural network are used as quality metrics.

6. * GLYZIN S. D., KOLESOV A. YU., MARUSHKINA E. A., PREOBRAZHENSKAIA M. M.

*P.G. Demidov Yaroslavl State University*

**Relaxation auto-oscillations in the system of two synaptic-coupled of impulse neurons**

We consider a mathematical model of synaptic interaction between two pulse neuron elements. Each of the neurons is modeled by a singularly-perturbed difference-differential equation. Coupling is assumed to be at the threshold, and time delay is taken into consideration. It is shown that the appearance of solutions containing impulse packets with suitable parameter values in this model can be a consequence of the delay in the chain of connection between the oscillators.

SESSION 2

Tuesday, October 3 14:00 – 15:30

Lecture-hall Алексеевский зал

Chair: Prof. USHAKOV VADIM

Applications of neural networks

7. BONDAREV V.N.*Sevastopol State University*

**Pulse Neuron Learning Rules for Processing of Dynamical Variables Encoded by Pulse Trains**

The supervised temporal learning rules for pulse neural network executing the required linear transformations of dynamical variables represented by pulse trains are proposed. The rules minimize the weighted difference between the actual and required pulse train in general. Rule properties are demonstrated by the examples of mappings of regular and dynamical pulse trains.

8. * POLEVAYA S.A., KARPOVA N.I., TSIRKOVA M.M., SAFONOV A.Y., ANTONETS V.A.

1

*Nizhny Novgorod State Medical Academy*

2

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

3

*Volga District Medical Centre (VDMC) under Federal Medical and Biological Agency (FMBA)*

4

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

**"HANDTRACKER" for dynamic imaging of sensomotor activity in virtual envieroment**

The work presents instrumental technology "HANDTRACKER" for imaging of individual and group interaction. Examples of objective instrumental real time recording of sensorimotor activity of one or several subjects in the management of virtual objects are given. The effectiveness of the diagnosis of CNS impairments, management of the resource of working groups, and development of sensory "human-human" interfaces is shown.

9. * A.V. PODOPROSVETOV, V.E. PAVLOVSKY, V.S. SMOLIN

1

*Keldysh Institute of Applied Mathematics, Moscow*

2

*Lomonosov Moscow State University*

**QUASISTATIC MODEL OF THE MANIPULATOR MANGO WITH NEURAL-BASED CONTROL SYSTEM**

The control problem of a two-tier flat-manipulator with pneumatic drive is considered. Such a drive has a complex mathematical model because of significant and poorly predicted elasticities in the contours of the system. These conditions make it difficult to form a manipulator movement and synthesize the control in traditional means. To build the manipulator control based on the experience of interaction with the drives, a neural-like system with training is used.

10. MIGALEV A.S.

*National Research Centre "Kurchatov Institute", Moscow*

**Searching algorithm for sound wave transform in to spike sequence**

Application of pulsed or spiking neural networks in autonomous mobile robots for sound data processing leads to the need of data volume optimization, minimization of required neurons amount by increasing computation efficiency. This paper considers the task of sound signal coding and decoding in pulse or spike sequence and accuracy of data transform. Existing algorithms are investigated and new are proposed.

11. * ZHUKOV R.V., ILIN V.S., TEREKHOV V.I.

*Bauman Moscow State Technical University*

**CONSTRUCTION OF THE RECEPTOR LAYER OF THE SPIKING ARTIFICIAL NEURAL NETWORK FOR SOLVING A PROBLEM OF DANGER DETECTION FOR DISABLED BY SIGHT PEOPLE**

The article describes the methodology of building a receptor layer of the spiking artificial neural network for solving problem of danger detecting on the lower levels to device develop for disabled by sight people. This research demonstrates the correctness of constructed receptor layer and make possible to solve problem of danger detecting on the lower levels that help with designing devices for disabled by sight people.

12. YAKHNO V.G., NUIDEL I.V.,TELNYKH A.A., SHEMAGINA O.V., KOVALCHUK A.V.

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

**Software for studying dynamics of sensory signal transformation in neuron-gliya networks**

It is considered the description of the version of the integrated mathematical model of processing signals in neurons and glial cells and an software that allows to conduct a wide range of research works on modeling the transfer of neuron activity in super-large neuron-glial networks of cortical structures of the brain. Such studies of dynamic regimes of signal patterns are aimed at understanding the fundamental principles of information processing in neural and neuron-glial brain systems.

SESSION 3

Wednesday, October 4 14:00 – 15:30

Lecture-hall Алексеевский зал

Chair: Prof. SHUMSKIY SERGEY

Applications of neural networks

13. * I. S. FOMIN, V.V. MIKHAILOV, A.V. BAKHSHIEV, MERKULYEVA N.S., VESHCHITSKII A.A., MUSIENKO P.E.1

*Russian State Scientific Center for Robotics and Technical Cybernetics, Saint-Petersburg*

2

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

3

*Sanct-Petersburg State University*

**Detection of neurons on images of the histological slices using convolutional neural network**

An automatic analysis of images of the histological slices is one of main steps in process of description of structure of neural network in norm and pathology. Understanding of structure and functions of that networks may help to improve neuro-rehabilitation technologies and to translate experimental data to the clinical practice. Main problem of the automatic analysis is complexity of research object and high variance of its parameters, such as thickness and transparency of slice, intensity and type of histological marker, etc. Variance of parameters make every step of neuron detection very hard and complex task. We represent algorithm of neuron detection on images of spinal cord slices using deep neural network. Networks with different parameters are compared to previous algorithm that based on pixels’ filtration by color.

14. TROFIMOV A.G., VELICHKOVSKIY B.M., SHISHKIN S.L.

1

*National Research Nuclear University (MEPhI), Moscow*

2

*National Research Centre "Kurchatov Institute", Moscow*

**An approach to use convolutional neural network features in eye-brain-computer-interface**

We propose an approach to use the features formed by a convolutional neural network (CNN) trained on big data for classification of electroencephalograms (EEG) in the eye-brain-computer interface (EBCI). Their use together with the EEG amplitude features improved the sensitivity of a linear binary classifier applied to an EEG dataset obtained in an EBCI experiment by more than 30% at a fixed specificity of 90%.

15. ANASTASIYA S. POPOVA, ALEXANDR G. RASSADIN, ALEXANDER A. PONOMARENKO

*Higher School of Economics, Nizhny Novgorod Branch*

**Emotion recognition in sound**

In this paper we consider the automatic emotions recognition problem, especially the case of digital audio signal processing. We consider and verify an straightforvard approach in which the classification of a sound fragment is reduced to the problem of image recognition. The waveform and spectrogram are used as a visual representation of the image. The computational experiment was done based on Radvess open dataset including 8 different emotions: "neutral", "calm", "happy," "sad," "angry," "scared", "disgust", "surprised". Our best accuracy result 71% was produced by combination “melspectrogram + convolution neural network VGG-16”.

16. EGORCHEV M.V., TIUMENTSEV YU.V.

*Moscow Aviation Institute (National Research University)*

**Neural network semi-empirical modeling of the longitudinal motion for maneuverable aircraft and identification of its aerodynamic characteristics**

The simulation problem for longitudinal motion of a maneuverable aircraft is considered including identification of its aerodynamic characteristics, such as the coefficients of aerodynamic axial and normal forces, as well as the pitch moment coefficient. This problem is solved in the class of neural network based semi-empirical dynamic models that combine the possibilities of theoretical and neural network modeling. The efficiency and prospects of such models are confirmed by the simulation results.

17. VASILYEV A.N., TARKHOV D.A., TERESHIN V.A., BARMINOVA M.S., GALYAUTDINOVA A.R.

*Peter the Great St. Petersburg Polytechnic University*

**SEMI-EMPIRICAL NEURAL NETWORK MODEL OF REAL THREAD SAGGING**

An approach to the construction of multilayer neural network approximations for solutions of ordinary differential equations was successfully used by the authors earlier. In accordance with the proposed concept of building models of complex objects or processes, this method is used by the authors to build a neural network model of a freely sagging real thread. A rough initial model is refined on the basis of experimental data.

18. V.E. GAI, P.A. RODIONOV, M.O. DERBASOV, D.A. LYAKHMANOV, A.A. KOSHURINA

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

**TEXTURE RECOGNITION FROM POSITIONS OF THE THEORY OF ACTIVE PERCEPTION**

Recognition of textures is one of the topical tasks of computer vision. The key step in solving this problem is the formation of feature description of the texture image. A new approach to the formation of texture features based on the theory of active perception is proposed. The results of a computational experiment based on the Brodatz-32 database are presented, and the accuracy of the classification is demonstrated. The application of the proposed feature systems for recognition of snow and land textures in the solution of the problem of auto piloting in complex natural and climatic conditions is considered.

POSTER SESSION 1

Wednesday, October 4 14:00 – 15:30

Lecture-hall Петровский зал

Chair: Prof. MALSAGOV MAGOMED

Neural network theory

19. M. E. MAZUROV*Plekhanov Russian University of Economics*

**MATHEMATICAL MODEL OF A NEURON AXIOMATIC TYPE, EFFECTIVE IN APPLICATIONS**

The most common axiomatic models of neurons. This description of the mathematical model of a neuron axiomatic type, effective in applications. Given the application considered axiomatic model for the study of various types of neural activity: pulse generation, utility generation, synchronization of neuronal ensembles, mutual synchronization of neuronal ensembles to form a single rhythm.

20. SHIBZUKHOV Z.M., KAZAKOV M.A., DIMITRICHENKO D.P.

1

*Center «Antistikhia», EMERCOM of Russia*

2

*Institute of Applied Mathematics and Automation*

**Empirical Risk Minimization And Averaging Aggregation Functions**

The paper offers an extended version of the principle of minimizing empirical risk for solving the problems of machine learning. It is based on the application of averaging aggregating functions to calculate risk based on the values of losses instead of the arithmetic mean. A gradient scheme is proposed for solving the problem of minimizing broadened empirical risk. Illustrative examples of constructing a robust procedure for estimating parameters in a linear re-greasing problem and classifying them on the basis of using the averaging function approximating the median or quantile are given.

Applications of neural networks

21. NAILYA T. GABDRAKHMANOVA*Peoples’ Friendship University of Russia, Moscow*

**Construction of a neural network model of network traffic on the basis of multifractal and topological analysis of the complexity of its time series**

Abstract. The dynamics of changes in traffic intensity of the network is analyzed based on the measurement data at the entrance to the boundary switch. We use a tool based on the methods of the geometry of random fields and computational topology. We evaluate topological invariants - Betti numbers for the time sequence of traffic intensity of the network. Preliminary results indicate the existence of possible topological precursors before changing traffic intensity. The main goal is to build an adequate forecast model of network traffic. Research is focused on ensuring efficient transmission of packets in the network.We define concept of integral persistent diagram whichinvolve geometrical characteristics of excursion sets and prove stabilityof such diagrams.

22. FOMIN I.S., GROMOSHINSKII D.A., BAKHSHIEV A.V.

*Russian State Scientific Center for Robotics and Technical Cybernetics, Saint-Petersburg*

**Object detection on images in docking tasks using deep neural networks**

In process of docking of automated apparatus there is a problem of determining of them relative position. This problem can be solved by algorithms for relative position calculation, based on information about positions of static visual landmarks in frames formed by camera installed in apparatus and them relative positions in real world. For example, study of detection of visual landmarks in space docking videos using deep convolutionam neural network based object detection system will be discussed.

23. ALEKSANDR BAKHSHIEV, LEV STANKEVICH

1

*Russian State Scientific Center for Robotics and Technical Cybernetics, Saint-Petersburg*

2

**Prospects for the development of neuromorphic systems**

The article is devoted to the analysis of neural networks from the positions of the neuromorphic approach. The analysis allows to conclude that modern artificial neural networks can effectively solve particular problems, for which it is permissible to fix the topology of the network or its small changes. In the nervous system, as a prototype, the functional element - the neuron - is a fundamentally complex object, which allows implementing a change in topology through the structural adaptation of the dendritic tree of a single neuron. Promising direction of development of neuromorphic systems based on deep spike neural networks in which structural adaptation can be realized is determined.

24. V.N. SHATS

*Independent investigator, St. Petersburg*

**The Classification of Objects Based on a Model of Perceptionv/N/**

This paper proposes a model of perception that allows animals to classify objects in the environment. We consider the sequence of transformations of external information in four blocks of the model that simulate information interactions in animal sensory systems. The principles of information processing are established and lead to a simple and universal algorithm for solving the problem of classifying of objects with high accuracy.

25. YAKOVENKO A.A., MALYKHINA G.F.

*Peter the Great St. Petersburg Polytechnic University*

**Adaptive pattern classification for neural responses of the auditory model**

In this paper the problem of interpretation of neural responses is considered as adaptive pattern classification task. Speech signals processing and analysis approach using auditory model and twofold neural network classifier is proposed. In proposed architecture the model of auditory periphery used as a preprocessor for acoustic stimuli. To discriminate the patterns obtained, self-organizing maps and radial basis functions are used.

26. V.D. KOSHUR, P.I. ROZHKOV

*Siberian Federal University, Krasnoyarsk*

**Multi-agent image recognition system**

In this paper we consider the development image recognition neural network system with convolutional neural network and multi-agent architecture.

27. A.V. KOLESNIKOV, S.V. LISTOPAD, V.I. DANISHEVSKIY

1

*Institute for Problems of Informatics RAS, Kaliningrad Branch*

2

*Immanuel Kant Baltic Federal University, Kaliningrad*

**IMAGE RECOGNITION SUBSYSTEM OF THE HYBRID INTELLIGENT SYSTEM WITH HETEROGENEOUS VISUAL AND MODEL FIELDS**

To ensure relevance to the teams of experts, the development of hybrid intelligent systems with heterogeneous visual and model fields to find solutions in visual and symbolic languages, which is associated with the recognition of graphic images is urgently needed. The subsystem of perception of visual information of such systems is realized by convolutional neural networks.

28. BURAKOV M.V.

*Saint Petersburg State University of Aerospace Instrumentation*

**APPLICATION OF THE HOPFILD NEURAL NETWORK FOR IDENTIFICATION OF THE DYNAMIC PLANT**

The problem of identifying the parameters of dynamic plants using the Hopfield neural network is considered. The structure of the identifier and the calculation formulas for determining its parameters are described. The results of the computational experiment in Simulink MatLab for determining the parameters of the DC motor model are presented, which confirm the practical usefulness of the proposed approach.

29. BRYNZA A.A., KORLYAKOVA M.O.

*Bauman Moscow State Technical University Kaluga Branch*

**Using of fractal curves for description of images to solve the classification problem in neural networks**

Considered the problem of reducing the time spent on training and improving the quality of the classification of a neural network for recognizing visual images with a small dimension of the training sample. Offered the Variants of the bypass of image fragments when forming the description of examples. In the course of the experiments, was adopted a model for traversing images by fractal curves of different orders. The results of the analysis of real objects are given and the estimation error is made.

Neurobiology

30. BAKHCHINA ANASTASIIA VLADIMIROVNA*Institute of Psychology of Russian Academy of Sciences, Moscow*

**COMPLEXITY OF HEART RATE DURING MORE AND LESS DIFFERENTIATED BEHAVIORS**

Autonomic nervous system is the main way for the brain-body coordination, of which mode can be evaluated by dynamics of heart rate variability (HRV). HRV analysis is used for evaluation of different psychological states, which can be considered as characteristics of behavior that formed at different stages of ontogeny. We investigated whether HRV differs between the early-formed behavior and the later-formed behavior.

31. KRYLOV A.K.

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

**A neuron activates to recieve metabolites**

The neuronal activity is considered as a tool for the neuron to receive metabolites from glial cells. The phenomenon of overcompensatory energy activation is modeled. The received metabolites are spent by the neuron for a biosynthesis. The approach compared with modeling a neuron as an input-output device.

32. G.S. VORONKOV

*Lomonosov Moscow State University*

**SIGNIFICANCE ANALYSIS OF THE "SPECIFIC VISION DEFICIT" PHENOMENON**

The author suggests his understanding of the nature of the objective existence of the “specific vision deficit” phenomenon (SVDph), that is the SVD-spot created on the retina. The author offers two possible explanations for the properties ”duality” of the subjective manifestations of SVDph – the image of SVD-spot generated by the Central visual structures. The author considered the SVDph importance in three aspects – the neuro- and psycho-physiological, and medical. The author likewise assumes that it is possible to use a hardware.recorder for fixation of the SVD-spot created on the retina.

33. BOZHOKIN S.V., SUSLOVA I.B., TARAKANOV D.E.

*Peter the Great St. Petersburg Polytechnic University*

**WAVELET CORRELATION OF BURST ACTIVITY IN NEURONS GENERATING SPIKES**

Mathematical model of neurons generating bursts comprising several spikes is developed. Time-frequency properties of the bursts are studied analytically by using continuous wavelet transform (CWT). To reveal the synchronization in the activity of neurons, we introduce wavelet correlation function, which is calculated analytically for the case of two neurons generating the given number of spikes.

Adaptive behavior and evolutionary modelling

34. KOTOV VLADIMIR BORISOVICH*Scientific Research Institute for System Analysis, Moscow*

**On Generating Goals**

Based on simple models of value map modification, the opportunities of generating goals for reasonable behaviour are considered. It is demonstrated that suitable choices of trajectories in the state space enable us to generate new goals which are different from the primary goals.

SESSION 4

Wednesday, October 4 16:00 – 18:30

Lecture-hall Алексеевский зал

Chair: Prof. EZHOV ALEXANDER

Neural networks and cognitive sciences

35. MEILIKHOV E.Z., FARZETDINOVA R.M.*National Research Centre "Kurchatov Institute", Moscow*

**IMAGE CONTRAST – COLOR OR LUMINANCE ?**

In the framework of a simple analytical model, we quantitatively validate the statement that the "color world" is amenable to much more accurate and faster segmentation than the "gray world". That results in significant facilitating conditions required for originating indispensable pop-out effect, and, probably, forms the basis of various cognitive phenomena connected with the color vision.

36. KHARYBINA ZOIA SERGEEVNA

*Southern Federal University, Rostov-on-Don*

**Environmental borders effect on spatial representation provided by place cells in the model of spatial navigation neurodynamics based on even cyclic inhibitory networks**

Study focuses on modelling of spatial behavior neurodynamics near environmental borders and far from them. Sequential place field formation algorithm based on interaction of self-movement information and information about boundaries was proposed. It is assumed that information about boundaries predominates at the environment periphery whereas self-motion information guides spatial navigation in the center of environment.

37. S.A. POLEVAIA, Т.V. CHERNIGOVSKAYA, I.S. PARINA, A.A. KONINA, S.A. ALEXEEVA, V.A. DEMAREVA, M.A. KNABENGOF, S.B. PARIN

1

*Nizhny Novgorod State Medical Academy*

2

*Sanct-Petersburg State University*

3

4

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

**DYNAMICS OF HEART RATE IN INTERACTION OF INFORMATION IMAGES IN THE PROCESS OF SIMULTANEOUS TRANSLATION AND IN STROOP TEST**

With the help of technology of event-related telemetry (ERT) of the heart rhythm, the features of vegetative provision of record-breaking energy efficiency and stressogenic activity - simultaneous interpretation (SI) were investigated. By comparing the processes of interaction of information images with SI and solving the bilingual test of Stroop - a laboratory model of the conflict of information images of a different nature - possible physiological mechanisms for solving cognitive tasks of increased complexity are analyzed.

38. V.L. VVEDENSKY

*National Research Centre "Kurchatov Institute", Moscow*

**SYNCHRONY OF CORTICAL ALPHA AND BETA OSCILLATIONS**

We analyzed behavior of cortical alpha and beta oscillations during preparation of self-paced finger movement. Field patterns during the rhythmic events were complex implying several simultaneously active sources. For high amplitude events sharp synchrony between oscillations in cortical sites separated by a few centimeters was observed

39. I.S. KNYAZEVA, JU. A. BOYTSOVA, N.G. MAKARENKO, S.G. DANKO , A.O. EFITOROV , D.A. ZELENKINA

1

*The Central Astronomical Observatory of the Russian Academy of Sciences at Pulkovo, Saint-Petersburg*

2

*Bekhtereva Human Brain Institute RAS, St.Petersburg*

3

*Lomonosov Moscow State University*

4

*Sanct-Petersburg State University*

**Application of wavelet phase coherence analysis for investigation of states with different level of mental and sensory attention**

The work is devoted to the application of wavelet phase coherence for analyzing brain activity in states with different degrees of mental and sensory attention. In the analysis of electroencephalographic correlates of mental states, the focus is on the analysis of the spectral power of a quasi-stationary EEG or task-related power of time-frequency EEG spectra. A much smaller amount of work has been devoted to the analysis of inter-channel coherence. As a result of the present work, it is shown that the analysis of the wavelet phase coherence provides additional information on the organization of brain activity, which provides states of mental attention.

40. FILATOVA N.N., SIDOROV K.V., SHEMAEV P.D.

*Tver State Technical University*

**Monitoring of the attractors characteristics for assesment of the human emotional state**

An approach to monitoring emotional reactions using estimates of the characteristics by EEG signals. The article describes methods and results of experiments illustrating changes in the properties of attractors during the exposure on the human by the packs of acoustic stimuli with the identical emotional interpretation. Based on fuzzy estimation of the attractors density increment sign reconstructed for several consecutive monitoring windows the trend direction of participant’s emotional state was determined.

41. SAMSONOVICH A.V., EIDLIN A.A.

*National Research Nuclear University (MEPhI), Moscow*

**Building weak semantic maps of word senses**

The problem of building weak semantic maps of word senses from synonym-antonym dictionaries is considered. The approach suggested addresses the problem of polysemy, which is one word having multiple meanings. Using Microsoft Word thesaurus, semantic maps of word senses have been built both for English and Russian language. The semantic maps of senses are compared to corresponding semantic maps of words.

42. SILKIS ISABELLA

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

**NEURAL NETWORK PROVIDING PROCESSING OF PHYSICAL PROPERTIES AND SUBSECOND TIME PARAMETERS OF SENSORY STIMULI**

A mechanism for processing the physical and sub-second temporal parameters of sensory stimuli in neuronal circuits cortex-basal ganglia (subthalamic nucleus-pedunculopontine nucleus)-thalamus-cortex is proposed. It follows from this mechanism that any factor enhancing (weakening) the neuronal representation of physical properties of the stimulus in the neocortex decreases (increases) the duration of activity circulation in these circuits, increasing (decreasing) the clock rate of processing, which will lead to overestimation (underestimation) of perceived time.

43. OLEG KUZNETSOV, LIUDMILA ZHILYAKOVA, VARVARA DYAKONOVA, DMITRI SAKHAROV, NIKOLAY BAZENKOV, SERGEY KULIVEC

1

*V. A. Trapeznikov Institute of Control Sciences of Russian Academy of Sciences, Moscow*

2

*Koltzov Institute of Developmental Biology of Russian Academy of Sciences*

**Discrete Modeling of Multi-Transmitter Interactions in Simple Nervous Systems**

We propose a novel discrete model of multi-transmitter interactions in simple nervous systems. The model emphasizes the role of nonsynaptic interactions, the diversity of neuronal phenotypes and endogenous activity in nervous systems. Neurons in the model release different neurotransmitters into the shared extracellular space (ECS) so each neuron with the appropriate set of receptors can receive signals from other neurons. We consider neurons of different types represented as finite-state machines functioning in discrete time.

44. KAZANOVICH Y.B., BURYLKO O., BORISYUK R.M.

1

*Institute of Mathematical Problems of Biology RAS*

2

3

*Plymouth University, United Kingdom*

**Winner-take all in a network of phase oscillators**

A network with a central element built from generalized phase oscillators is considered. We show that this network is appropriate for implementation of the winner-take-all procedure in terms of the competitions between peripheral oscillators for the synchronization with the central oscillator. Computer simulations illustrate how the system works under different values of parameters. An application of the system as an oscillatory model of visual search is described.

SESSION 5

Thursday, October 5 14:00 – 15:30

Lecture-hall Алексеевский зал

Chair: Prof. TARKOV MIKHAIL

Applications of neural networks

45. ENGEL EKATERINA ALEKSANDROVNA*Katanov Khakass State University, Abakan*

**DUMP TRUCK FAULT’S SHORT-TERM FORECASTING BASED ON THE MULTI-AGENT ADAPTIVE FUZZY NEURONET**

Dump truck fault’s short-term forecasting is the important step for solving real-time fleet dispatching tasks and to provide reliable, efficient and safe open-pit mining. This paper presents a multi-agent adaptive fuzzy neuronet for dump truck fault's short-term forecasts. The agents of the multi-agent adaptive fuzzy neuronet are fulfilled based on recurrent networks. An automatic determination of the optimal architecture’s parameters of a neuronet is the most critical task. In order to train the effective multi-agent adaptive fuzzy neuronet we use algorithm, in which the multi-dimensional Particle Swarm Optimization is combined with the Levenberg-Marquardt algorithm. The simulation results show that proposed training algorithm outperforms multi-dimensional Particle Swarm Optimization and Levenberg-Marquardt algorithm in training the effective multi-agent adaptive fuzzy neuronet for dump truck fault's short-term forecasts.

46. LE BA CHUNG, Y.A. HOLOPOV

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

**Information environment for neural-network adaptive control system**

The proposed model is a coherent information environment for neural-network control system, with simultaneous record mode of the object state parameters. This mode is important for the control systems objects with unidentified degrees of freedom – typical field of applications of neural systems.

47. MALAFEEV SERGEY SERGEEVICH, MALAFEEV SERGEY IVANOVICH, TIKHONOV YURIY VASILIEVICH

**INTELLIGENT DIAGNOSTICS OF MECHATRONIC SYSTEM COMPONENTS OF CAREER EXCAVATORS IN OPERATION**

The article provides the results of application of artificial neural networks for diagnosis of the condition of electrical mining machinery as well as the description of data collection and processing of intelligent system structure and a condition of components of mechatronic systems analysis algorithms using neural networks. Information is provided on practical implementation of algorithms in information and diagnostic systems of career excavators developed by Joint Power Co. Ltd, Moscow.

48. GLUSHCHENKO A.I.

**Method of Calculation of Upper Bound of Learning Rate for Neural Tuner to Control DC Drive**

A two-loop cascade control system of a DC drive is considered in this research. The task is to keep transients quality in both speed and armature current loops. It is solved by a usage of P/PI-controller parameters neural tuner, which operates in real time and does not require a plant model. The tuner is trained online during its functioning in order to follow the plant parameters change, but usage of too high values of a learning rate may result in instability of the control system. So, the upper bound of the learning rate value calculation method is proposed. It is based on Lyapunov’s second method application to estimate the system sustainability. It is applied to implement adaptive control of a mathematical model of a two-high rolling mill. Obtained results show that the proposed method is reliable.

49. MARUSHKO Y., DOUDKIN A.

*United Institute of Informatics Problems of the National Academy of Sciences of Belarus, Minsk*

**Long-term forecast of the spacecraft power supply system parameters**

The task of forecasting multidimensional time series of telemetry information of a spacecraft is considered. The approach for long-term forecasting of telemetry parameters of spacecraft power supply system using ensembles of neural networks is given.

50. LITVIN A.A., ZHARIKOV O.G., KOVALEV V.A.

1

*Immanuel Kant Baltic Federal University, Kaliningrad*

2

3

*United Institute of Informatics Problems of the National Academy of Sciences of Belarus, Minsk*

**SYSTEM FOR PREDICTION AND DIAGNOSTICS OF INFECTIOUS PANCREONECROSIS ON THE BASIS OF ARTIFICIAL NEURAL NETWORKS**

The results of the development and study of the effectiveness of the system for the prediction and diagnosis of infected pancreoche necrosis, created on the basis of the use of artificial neural networks, are considered. The created decision support system showed good prognostic and diagnostic characteristics. The use of the developed system in clinical practice made it possible to improve the prognosis and diagnosis of severe acute pancreatitis infected with pancreatic necrosis.

SESSION 6

Thursday, October 5 16:00 – 18:30

Lecture-hall Алексеевский зал

Chair: Prof. SMIRNITSKAYA IRINA

Neurobiology

51. BIBIKOV NIKOLAY GRIGOR'YEVICH, MAKUSHEVICH IL'YA VIKTOROVICH*N.N. Andreyev Acoustics Institute, Moscow*

**Comparison of some fractal anslysis methods for studying the spontaneous activity in medullar auditory units**

For the analysis of spontaneous background impulse activity in the auditory neurons the Hurst index was used. It was compared with the Fano and Allan indices. A comparative analysis of these indices by Kendall's rank correlation method made it possible to formulate assumptions about the possibility and efficiency of using Hurst index to analyze the sequence of pulsed discharges of a neuron. In most cells, chaotic changes in impulse density were observed, which is indicative of the trend behavior of neuron’s firing. Anti-trend behavior was not observed.

52. SMIRNOVA ELENA YURIEVNA, ZEFIROV ARTEM VIKTOROVICH, AMAKHIN DMITRY VALERIEVICH, CHIZHOV ANTON VADIMOVICH

1

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

2

*Peter the Great St. Petersburg Polytechnic University*

3

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

**Effect of persistent sodium current on neuronal activity**

In epilepsy, the number of persistent sodium (NaP) channels increases. To study effects of NaP current on neuronal excitability we applied the dynamic-clamp technique. We have revealed that NaP current decreases the rheobase, promotes the depolarization block (DB) and changes the membrane potential profile between spikes. Applying bifurcation analysis to Hodgkin-Huxley-like model of a neuron, we have found that NaP current shifts saddle-node and Hopf bifurcations, which correspond to the rheobase and DB, in agreement with experiments. By shifting DB, NaP current can make an antiepileptic effect via excitatory neurons.

53. A. L. PROSKURA, A. S. RATUSHNYAK, S.O. VECHKAPOVA, T. A. ZAPARA

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

**SYNAPSE AS A MULTI-COMPONENT AND MULTI-LEVEL INFORMATION SYSTEM**

Synapse, as is known, consists of presynaptic and postsynaptic parts. In a number of brain cells, in particular the cortex and the hippocampus, a small protrusion - a dendritic spinе- appears as a postsynaptic part. The dendritic spine is partly an isolated structure, but it is associated functionally with other spines of dendritic shaft as well as through the vesicular system with the soma of neuron. The main task of the spine interactome is not only to receive a signal from presynaptic cell and to react to it by opening/closing ion channels, thus ensuring its transmission to the axon. The interactome of spine is primarily a detector of environmental signals and through the remodeling of the system of its macro-complexes it recognizes and remembers the pattern of the signal.

54. SAMARIN A.I., PODLADCHIKOVA L. N., PETRUSHAN M.V., SHAPOSHNIKOV D.G.

1

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

2

**Possible strategies of choosing the areas on the vision field periphery for gaze fixation**

The results of psychophysical tests during viewing of 2D images and navigation in virtual 3D environment have been considered. It was shown that both probability of high-amplitude saccades (more 3º) and fixation duration (for saccades more 8º) are decreased in the first case. Poly-modal distribution of saccade amplitudes was revealed during navigation in 3D environment. Differences between viewing scan paths in conditions of active and passive navigation were shown. Strategies of choosing the areas on the vision field periphery for gaze fixation have been discussed.

55. I.A.SMIRNITSKAYA

*Scientific Research Institute for System Analysis, Moscow*

**Hierarchical levels for the locomotion control from spinal cord up to diencephalic locomotor region**

The locomotion control levels are described, - from central pattern generator in spinal cord, to reticulo-spinal, mesencephalic and diencephalic locomotor regions. The role of lateral hypothalamus in locomotion and arousal is discussed. The appearance of each level is interpreted in relation to suitable behavior task.

56. I.A.SMIRNITSKAYA

*Scientific Research Institute for System Analysis, Moscow*

**Pavlovian-to-Instrumental Transfer and the control of the behavioral choice**

The review of the Pavlovian-to-Instrumental Transfer (PIT) research and the participation of nucleus accumbens shell and core, amygdala basolateral and central nuclei in the General Versus Specific Transfer. Habit formation is described and the role of dorsomedial and dorsolateral striatum.

57. CHERNOIVANOV V.I., SUDAKOV S.K., TOLOKONNIKOV G.K.

1

2

*Institute of Normal Physiology P.K. Anohin*

**CATEGORIES THEORY OF SYSTEMS, FUNCTIONAL SYSTEMS AND BIOMACHSYSTEMS**

A categorical approach to the theory of systems, functional systems used in the theory of biomachsystems is discussed. The tasks of integrating biomachsystems into the theory of functional systems are formulated on the basis of the categorical approach on the basis of which the theory of biomass systems develops. Precise mathematical results of models concerning the duality of polygraphs that form neurons and their connections are given.

58. BAZYAN ARA

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

**HEBBIAN MODEL AND MODERN CONCEPTS OF MOLECULAR AND CELLULAR MECHANISMS OF MEMORY**

An algorithm for the plasticity of glutamatergic synapses is described, an experimental analogue of the Hebbian model. The molecular - cellular algorithm of allosteric plasticity of GABA receptors is analyzed. The hypothesis of actualization of neural networks, which is realized at the molecular level by means of mechanisms of internalization and recycling of a specific cluster of the GABA(A) receptor is discussed. It is assumed that this is the process of transferring memory from the storage stage to the stage of working or operating memory, and vice versa.

59. DICK O. E.

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

**About the possibility of nociceptive signal nodulation**

Using the method of bifurcation analysis of the nociceptive neuron model we examine the possibility to regulate nociceptive signals in dorsal root ganglion of rat. It has been shown that two types of ectopic bursting can emerge in the model. Suppression of bursting discharge is connected with changes parameters of the NaV1.8 sodium channels due to the action of 5-hydroxy-γ-pyrone-2-carboxilic acid.

60. A.S. RATUSHNYAK, T.A. ZAPARA, A.L. PROSKURA, E.D. SOROKOUMOV

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

**ABOUT THE APPROACH TO SIMULATION OF NEURONS AS SYSTEMS OF MULTILEVELS, AUTONOMOUS, ASSOCIATIVELY TRAINED, PROGNOSTIC, SUPRAMOLECULAR AGENTS**

On the base of the theoretical-experimental analysis, assumptions and ap-proaches to the modeling of biological information systems are formed. Such systems can be constructed from models of simple basic negentropic elements. Multilevel structures from such autonomous, associatively trained, prognostic agents can be created on the basis of knowledge of the molecular organization of nerve cells and the evolution of biological systems.

SESSION 7

Friday, October 6 14:45 – 15:15

Lecture-hall Алексеевский зал

Chair: Prof. VVEDENSKY VIKTOR

Adaptive behavior and evolutionary modelling

61. VLADIMIR G. RED’KO, ZAREMA B. SOKHOVA*Scientific Research Institute for System Analysis, Moscow*

**Processes of Self-Organization in the Community of Investors and Producers**

The paper analyzes the processes of self-organization in the economic system, that consists of investors and producers. There is intensive information exchange between investors and producers in the considered community. The model that describes the economic processes has been developed. The model proposes a specific mechanism of distribution of investors capital between producers. The model considers the interaction mechanism between investors and producers in a decentralized economic system. The main element of the interaction is the iterative process. In this process, each investor takes into account the contributions of other investors into producers. The model is investigated by means of the computer simulation, which demonstrates the effectiveness of the considered mechanism.

62. YURY SKOBTSOV, OLGA CHENGAR.

1

*Saint Petersburg State University of Aerospace Instrumentation*

2

*Sevastopol State University*

**Experimental study of directed ant algorithm for optimization of the production schedule**

The problem of optimization of the production schedule on the basis of a directed ant algorithm with adaptively variable weights is considered. To evaluate the quality of potential solutions, an object-oriented model of the technological complex of machining is used. The rules are determined for the transition and calculation of the artificial pheromone concentration in the proposed ant algorithms . The influence is investigated on the optimization efficiency of the main parameters of the ant algorithm relative to various criteria.

SESSION 8

Friday, October 6 15:15 – 16:30

Lecture-hall Алексеевский зал

Chair: Prof. KARANDASHEV YAKOV

Neural network theory

63. TARASOV D.S., IZOTOVA E.D.1

*Meanotek, Kazan*

2

*Kazan (Volga region) Federal University*

**COMMON SENSE KNOWLEDGE IN LARGE SCALE NEURAL CONVERSATIONAL MODELS**

It was recently shown, that neural language models, trained on large scale conversational corpus such as OpenSubtitles have recently demonstrated ability to simulate conversation and answer questions, that require common-sense knowledge, suggesting the possibility that such networks actually learn a way to represent and use common-sense knowledge, extracted from dialog corpus. If this is really true, the possibility exists of using large scale conversational models for use in information retrieval (IR) tasks, including question answering, document retrieval and other problems that require measuring of semantic similarity. In this work we analyze behavior of a number of neural network architectures, trained on Russian conversations corpus, containing 20 million dialog turns. We found that small to medium neural networks do not really learn any noticeable common-sense knowledge, operating pure on the level of syntactic features, while large very deep networks shows do posses some common-sense knowledge.

64. T.V. LAZOVSKAYA, D.A. TARKHOV, A.N. VASILYEV

1

*Computing Center of Far Eastern Branch RAS*

2

*Peter the Great St. Petersburg Polytechnic University*

**MULTI-LAYER SOLUTION OF HEAT EQUATION**

One approach to the construction of multilayer neural network approximate solutions for evolutionary partial differential equations is considered. Earlier, a similar approach has been successfully used by the authors in the case of ordinary differential equations. Computational experiments are performed on the test problem for the one-dimensional (in terms of spatial variables) heat equation.

65. YU.S. FEDORENKO, YU.E. GAPANYUK, S.V. MINAKOVA

*Bauman Moscow State Technical University*

**The analysis of regularization strategies in deep neural networks**

The problem of overfitting in deep neural networks is formulated. The ways to solve this problem are discussed. The regularization strategies in deep neural networks are considered. The new approaches such as early stopping and dropout are described. Comparison of different regularization techniques are conducted on MNIST and CoverType datasets. Regularization representation using metagraph approach is discussed. Results of experiments are analyzed.

66. KARANDASHEV I.M., KRYZHANOVSKII B.V., MALSAGOV M.YU.

*Scientific Research Institute for System Analysis, Moscow*

**ANALYTICAL EXPRESSIONS FOR THE TWO-DIMENSIONAL IZING MODEL FOR FINITE SIZES**

The dependence of the thermodynamic characteristics of the two-dimensional Ising model on the number of spins N was studied by numerical methods. A generalization of the Onsager solution to the case of a lattice of finite dimensions was obtained and analytical expressions for the free energy and its derivatives, which describe the experimental results well, were obtained. It is shown that with increasing N the heat capacity at the critical point increases logarithmically. The limitations on the accuracy of determining the critical temperature due to the finite size of the system are indicated.

67. MIKHAIL S. TARKOV

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

**HOPFIELD NETWORK SYNAPSES REDUCTION**

Based on the analogy with oscillator networks, the influence of the Hopfield network synapses number reduction on the network auto-associative behavior is investigated. It is shown that the exclusion of synapses with weights whose absolute values are strictly less than the maximum for a given neuron significantly improves the network performance. In this case, the percentage of the input vector distorted elements filtered by the network increases with its size increase.