ABSTRACT. In this paper, we consider the problem of controlling an intelligent agent that simulates the behavior of an unmanned car when passing an road intersection together with other vehicles. We consider the case of using smart city systems, which allow the agent to get full information about what is happening at the intersection in the form of video frames from surveillance cameras. The paper proposes the implementation of a control system based on a trainable behavior generation module. Agent's model is implemented using reinforcement learning (RL) methods. In our work, we analyze various RL methods (PPO, Rainbow, TD3), and variants of the computer vision subsystem of the agent. Also, we present our results of the best implementation of the agent when driving together with other participants in compliance with traffic rules.

Link to poster

ABSTRACT. The state of a physical system usually described by its position and movement relatively to its surroundings. This representation is used widely in control and reinforcement learning to define the state and compute the cost/reward value. In real-world applications, such representations are available only in structured environments. State representation learning exploits the advances of deep learning to learn useful state representation from raw sensor data. To learn a model for a robotic manipulator using video data we are interested in representation learning from RGB images. We suggest a method for which we require the user to provide us with just a couple of videos demonstrating the task. Our approach uses a sequential contrastive loss to learn latent space mapping, and task-related descriptors in each state. Our framework intended to be used in robotics control scenarios, especially with model-based reinforcement learning algorithms. The resulted representation eliminates the need for engineered reward functions or any explicit access to positioning systems, aiming to improve the applicability of learning to control physical systems. Our framework emphasis reducing the learning time, and to work with low-resource scenarios.

Link to poster

ABSTRACT. Multi-agent path finding with continuous movements and time (denoted MAPF^R) is addressed. The task is to navigate agents that move smoothly between predefined positions to their individual goals so that they do not collide. Recently a novel solving approach for obtaining makespan optimal solutions called SMT-CBS^R based on satisfiability modulo theories (SMT) has been introduces. We extend the approach further towards the sum-of-costs objective which is a more challenging case in the yes/no SMT environment due to more complex calculation of the objective. The new algorithm combines collision resolution known from conflict-based search (CBS) with previous generation of incomplete SAT encodings on top of a novel scheme for selecting decision variables in a potentially uncountable search space. We experimentally compare SMT-CBS^R and previous CCBS algorithm for MAPF^R.

Link to poster

ABSTRACT. The report describes data analysis methods used in DSM systems with non-atomistic strategies. The methods are based on representing of the fact base as similarity spaces and comparing these spaces.

Link to poster

ABSTRACT. The paper proposes an algorithmic structure of inforware for assessing the condition of a technological process for the production of phosphorus from apatite-nepheline ores waste. The structure is based on the use of an en-semble of deep recurrent neural networks for forecasting process parameters with subsequent aggregation of their outputs for clustering, the results of which can be used to analyze repeatability and stability for the process. The results of checking for structure operability on a software model created in Python are presented

ABSTRACT. An approach of automatized object-oriented code refactoring is described that applies LP structures theory to type hierarchy transformations by merging at-tributes that share common subclasses. A distinctive feature of the employed algebraic structures is their ability to model aggregation not as a relation be-tween two independent sets of types and attributes, but as a relation between two specific types. The property enables a more adequate modeling of type hi-erarchies. The described approach is dual to the “Pull Up Field” refactoring method that was considered in the previous works related to the applications of LP structures theory. The paper demonstrates the way the adopted model can be extended to handle a wider range of type hierarchies and incorporate exter-nal constraints on the refactoring process. Also, the paper details the process of model construction and application.

Link to poster

ABSTRACT. В работе рассматривается алгоритм для автоматизированной оценки качества признаков, которые описывают объекты в ДСМ-системе. По результатам оценки можно сократить неинформативные признаки и повысить качество предсказаний ДСМ-системой. Алгоритм используется для оценки качества признаков в ДСМ-системе прогнозирования медицинских данных.

Link to poster

ABSTRACT. The paper describes the capabilities of a hybrid expert system (ES), developed by the authors, for processing streaming information when determining the moment of transition of a technological process to a new path of development and detection of bifurcation sources. The system architecture and implementation of a streaming data model based on the modified Map Reduce technology are considered.

ABSTRACT. This article presents new program tools used for the development of the OntoMathEdu ontology. The ontology includes more than 600 mathematical concepts that are distributed across hierarchies of types, roles, materialized rela-tionships, and a network of points of view. The first direction is related to the task of designing a multilingual ontology, focused on teaching mathematical disciplines in different languages in secondary school. We use the developed program tools for processing mathematical texts to extract ontology-relevant data, in particular, to annotate ontology concepts and extract definitions of ontology concepts from educational mathematical texts.

Link to poster

ABSTRACT. The paper focuses on the estimation of the parameters of intensity of publishing posts in social media with the data on the times of the three last posts. This type of behavior is modeled by the Gamma Poisson model. An estimation procedure for the intensity parameters is based on the copula decomposition of the cumulative probability distribution of the inter-episode intervals.

Link to poster

ABSTRACT. The basis for making a decision on responding to computer incidents is information about characteristics of the identified security breach, the values of which are determined during the diagnosis procedure. To reduce the time spent on determining the values of characteristics and increase the reliability of the diagnosis results, it is proposed to use machine learning methods implemented on the basis of artificial neural networks. The paper considers a combined artificial neural network as the basis of a model for diagnosing computer incidents. In this model, through the use of deep learning, the drawback of the classical multilayer perceptron associated with the need to form a massive base of training examples for the specific structure of the neural network is elimi-nated. At the same time, the addition of new training examples requires proce-dures for optimizing the network structure and relearning. The results of the experiments showed that the proposed model allows one to reduce the duration of training, while not reducing the values of the quality indicators of the functioning of the neural network.

Link to poster

ABSTRACT. Information retrieval (IR) chatbot is a special class of virtual assistants, which are widely used nowadays in customer support services. However, the work of modern IR retrieval systems is limited by simple queries to the database, which does not utilize all the potential of interaction with the user. In this paper we implement an FCA-based approach to deliver the relevant information the user has requested. A developing approach integrates a concept-based model build upon the database and intelligent traversal through it. The proposed algorithm has been implemented as an additional function within the existing IR chatbot. In this paper we also enlighten the perspectives for further development of the proposed system. Formal Concept Analysis (FCA) technique and Pattern Structures as its extension are proposed to process unstructured data (objects with a text description), which has become a common way of presenting various items recently.

Link to poster

ABSTRACT. The article introduces inference methods for Mamdani-type fuzzy systems, which can be implemented with polynomial computational complexity for any t-norms and multiple fuzzy inputs. Center average and center of gravity defuzzification methods were used for case of multiple rules in rule base. Network architectures of systems corresponding to inference methods introduced in the article are provided.

Link to poster

ABSTRACT. The computer-aided engineering of software modules for decision-making intelligent systems requires the development of specialized methods, algorithms and software. The use of a model-driven approach that implements the principles of generative and visual programming, as well as model transformations, is promising. In this paper, we propose an approach to the development of rule-based intelligent system software components in the form of decision-making modules for web applications by specializing and using main principles of a model-driven development. The proposed specialization includes: using a step-by-step development scheme (chain of a model transformation) from information models to source codes and specifications; a method for the automated creation of computation-independent models based on the transformation of spreadsheets; domain-specific tools for formalization, visualization and generation of codes. The developed approach was applied for creating decision-making modules for web-oriented rule-based intelligent systems.

Link to poster

ABSTRACT. The article describes the ontology compilation from the relational database. The process is based on the model called initial. The model is applicable for the design and redesign of existing software products and tuning of the final ontology, including its size reduction. The described model allows obtaining data from a relational schema. The data can be subsequently used for other more flexible formalizations. Moreover, the model can be extended by new parameters which cannot be derived from the schema but be quite useful for design tasks. The paper brings in a short overview of different approaches to ontology compilation based on relational data sources. Additionally, two extra strategies are presented. Finally, the article contains an example of these strategies implementation which is measured by performance tests on two different data sources.

ABSTRACT. The theory of modified descriptive image algebras allows us to formally describe any process of image processing and analysis. Image processing and analysis is formulated as a nonlinear optimization problem with linear or nonlinear constraints. The solution of this problem is equivalent to the parametric synthesis of the mathematical model of the computer vision system. The article formulates the basic theoretical principles of the theory of modified descriptive image algebras. The general formulation of the problem of parametric and structural synthesis of vision systems is described.

ABSTRACT. The article is devoted to the automatic planning system integrated into the Unity Engine game environment, which controls agent behavior. A graphical notation for domian knowledge definition was developed. The goals reached in this paper will allow for the quick complex behaviour pattern implementation for non-player characters within the game environment.

Link to poster

ABSTRACT. A computationally adequate method is proposed for design of a neural network capable of performing an important group of symbolic operations on a sub-symbolic level without initial learning: extraction of elements of a given structure, conditional branching and construction of a new structure. The neural network primitive infers on distributed representations of symbolic structures and represents a proof of concept for the viability of implementation of symbolic rules in a neural pipeline for various tasks like language analysis or aggregation of linguistic assessments during the decision making process. The proposed method was practically implemented and evaluated within the Keras framework. The network designed was tested for a particular case of transforming active-passive sentences represented in parsed grammatical structures.

Link to poster

ABSTRACT. В работе рассматривается задачный подход к Искусственному Интеллекту, который способен интегрировать различные подходы к ИИ. Описывается методология, теория и возможные приложения задачного подхода. Дается формальное определение понятия задача. Основой задачного подхода является концепция логико-вероятностной семантического моделирования.

Link to poster

ABSTRACT. Social engineering attacks based on the human factor have long been the most frequently used in violation of the information security policies. One of the ways to increase the organization’s level of protection against social engineering attacks is building a social graph of the organization’s employees and its analysis. The nodes of such graph associated with users of the information system, and edge designate the relationships between them. Moreover, this kind of information can be obtained by analyzing social networks. However, often users have accounts in different social networks, and the information presented in them is different. The purpose of this article became to propose approaches to merging probabilistic estimates of the relationship between users, which are linguistic values of linguistic variable "type of relationship". The theoretical significance of the results lies in the proposal of new approaches to the merging of probabilistic estimates of linguistic variables, the practical significance consist in creating the basis for further analysis of the social graph of the organization’s employees, in particular, for detecting the most critical trajectories of attack development or solving backtracking tasks of social engineering attacks, e.i. the investigation of cyber crime committed by using social engineering techniques.

Link to poster

ABSTRACT. This article attempts to give an overview of several algorithms for extracting rules from an artificial neural network. The second goal of this article is to establish fundamental connections between two important areas of artificial intelligence - fuzzy logic and deep learning. Such an approach will allow re-searchers in the field of fuzzy logic to develop ap-plied systems in the field of strong artificial intelligence, which are also of interest to specialists in the field of machine learning.

Link to poster

ABSTRACT. The paper proposes a conceptual model of heterogeneous data based on the concept of a multidimensional formal context. The model is an n-dimensional tensor constructed using n-grams extracted from natural language texts. Various variants of n-grams and their corresponding multidimensional formal contexts are considered. An algorithm for clustering multidimensional contexts is proposed to avoid the appearance of a "bag of words" in clusters when used in question-answering systems. The method was tested on the texts of annotations of scientific articles on biomedical topics.

Link to poster

ABSTRACT. The multidimensional abstract intelligent systems structure is proposed. It reflects knowledge aspects developed within multiple subject areas and deals with simu-lating models for these areas knowledge representation and processing. The four-dimensional intelligent systems architectures is considered. Unified knowledge representation format of abstract semantic hierarchies’ formalisms allows performing formal analysis and homogeneity achieving for intelligent systems’ structural and functional components. Such components adaptation to weakly formalized knowledge models attributes allows creating applied intelligent sys-tems prototypes with further possibility of models development into applied sys-tem specification by homomorphic transformations and extensions as base for complex knowledge synthesis processes. Abstract knowledge processing operations combinations considered the foundation for developing the subject domain goals realization templates. These templates composed as assigned to intelligent systems components diagrams. The cross-component knowledge transfers descriptions allow realize knowledge transformations multidimensional modelling. Abstract knowledge processing flows templates used for modeling the subject ar-eas goals realizations schemes distributed among intelligent systems architecture components. Knowledge processing operations’ formal specifications applied as such modeling parameters. Ontologies used for accumulating the templates realizations diagrams processes formal descriptions as components creating technolo-gy elements. This allows tools creating for knowledge processing templates exhaustive analysis realization and control, based knowledge processing operations descriptions.

Link to poster

ABSTRACT. An algorithm for solving the travelling salesman problem, based on DNA calculations, is described. The ant algorithm to design DNA molecules is proposed. A modification of the original algorithm is given to obtain a more reliable and quick result.

Link to poster

ABSTRACT. In the work the local parametric training of Algebraic Bayesian networks is considered. The theorem about the change of Dirichlet distribution parameters during transition from a priori to a posteriori probability distribution on propositional quantum formulas is formulated and proved. The proof is based on the conjugation property of the multinomial and Dirichlet distributions.

Link to poster

ABSTRACT. The purpose of the article is to study the relationship of personality traits with the content of images posted in social networks. The paper attempts to identify in-formative features and appropriate ways to configure artificial neural networks. The developed technique includes obtaining several color-bright-based and statis-tical characteristics of image collections in the form of histograms and BoW dic-tionaries with further construction of classifiers based on artificial neural net-works to test the hypothesis about the interrelation between the available graphic data and the five-factor personality model of the tested. The questionnaire, which allowed the formation of training and test samples, was carried out by employees of the Psychological Institute of RAE with the “NEO-FFI” test, which included 60 questions. The collections of images used are datasets that published by users of the “VKontakte” social network. The problems of determining personality fac-tors were experimentally solved with using classifying and predictive artificial neural networks. The work confirmed the prevailing opinion that there is no sig-nificant interrelation (correlation) between placed images and “Big Five” personal factors. With the help of published images, the factors “Openness” and “Agreea-bleness” are predicted best, worst of all – “Neuroticism”. The results of forecast-ing personality recognition traits improve as the number of layers of neural net-works grows, up to overtraining moment.

Link to poster

ABSTRACT. Dataset clustering could have more than one “right” result depending on a user intention. For example, texts could be clustered according to their topic, style or author. In case of unsatisfactory results, a data scientist needs to re-construct a feature space in order to change the results. The relation between the feature space and the result are often quite complicated. The latter results in building several clustering models to explore useful relations. Interactive clustering with feedback is aimed to cope with this problem. In this paper an approach to user feedback processing during clustering is presented. The approach is based on end-to-end clustering and uses an autoencoder neural network. This technique allows to adjust iteratively the computing clusters without changing feature space.

Link to poster

ABSTRACT. The paper considers the modeling of the process of monitoring the implementation of the plan for solving the target task posed to the coalition of autonomous robots. To solve the problem of monitoring the implementation of the plan and the consumption of system resources, it is proposed to use a distributed registry with support for smart contracts. The achievement of consensus is based on the mechanism of the Byzantine agreement. Negotiations between robots and the decision to allocate resources are implemented through the implementation of smart contracts on the HyperLedger Fabric platform, which process the stages of the plan for solving the problem and resource requirements from robots, store them and distribute them in a distributed registry. To create a model for the process of solving the problem, a stand based on the Gazebo/ROS environment is used, which provides software and hardware modeling and visualization of objects of the physical world and the interaction of robots in a coalition.

Link to poster

ABSTRACT. The paper presents results of using deep learning methods for automatic recognition of metal nanoparticles on probe microscopy images. As a result, the trained neural network recognized nanoparticles in the test dataset with an accuracy of 0.93 and a recall of 0.78. The accuracy of determining the average particle size is 0.87-0.99. The “ParticlesNN” web service based on a trained neural network has been developed.

Link to poster

ABSTRACT. In the previous works, a probabilistic generalization of the formal concepts analysis was determined. This generalization is induced by the problem of formal concepts determining under noise conditions, when the lattice of formal concepts exponentially grows. In this paper, probabilistic formal concepts with negation are determined, as well as a statistical method for detecting these probabilistic formal concepts. The purpose of this paper is to show that probabilistic formal concepts have a deeper meaning. It is argued that probabilistic formal concepts formalize the “natural” concepts described in cognitive sciences by “causal models”, which are characterized by a highly correlated structure of attributes. The same structure is specific for the "natural" classification of objects of the external world. The definition of "natural" classification given by J. Stuart Mill is fairly accurately formalized by probabilistic formal concepts.

Link to poster

ABSTRACT. The paper presents two forms of compressed table constraint representation: the C- and D-systems. The hybrid methods have been developed to solve non-quantitative constraint satisfaction problems formalized by table structures introduced. To make search more eective under the increase of the solution space, the hybrid methods integrate the original methods, such as non-quantitative constraint propagation and local search, as well as structural constraint graph decomposition methods.

Link to poster

ABSTRACT. The objective of this work is to study the role of human emotional states, specifically, those related to empathy, during social interactions in a virtual environment. The previously created platform based on the video game "Teleport" was used in this study. The platform allows for anonymous social interaction of actors of various nature, both, humans and automata. The main hypothesis was that the level of empathy is related to the human ability to distinguish artificial intelligent agents (bots) from humans in a social-emotional videogame paradigm based on behavior. Resulting correlational data confirm the hypothesis, revealing the relationship between the number of errors in human identification of bot vs. human behavior and the level of empathy measured in several monitored channels. The outcomes allow us to implicate the developed technique in suggested general tests and metrics for future artificial intelligent agents.

Link to poster

ABSTRACT. Our paper is dedicated to the selection of neuropsychological tests effective for experimental assessment of cognitive reserves, which are increasingly studied as the world population ages. In this, we separately consider the systems of attention, memory and intelligence and use the list of 36 candidate tests composed by domain expert relying on OntoNeuroLOG – Mental State Assessment ontology, which we extend in OWL format. The names of the tests are employed in extracting the publications from PubMed (MEDLINE) bio-medical database with the tools provided by E-utilities (EDirect). To assess the application context, represented as the subject group and the studied pathology, we perform word frequency analysis for the publications’ titles and itemize the most prominent journals. Ultimately we select in total 5 tests that are most popularly used by the research community and whose application context is relevant for the cognitive reserves studies. These neurological instruments include Stroop task, Fluency test, Divided attention tasks, Dual task, and Rey complex figure. We believe that both the mining method that we used and the resulting test battery can be of interest to experimental researchers in Neuropsychology.

Link to poster

ABSTRACT. The paper presents the concept of a software complex for solving interdisciplinary problems based on self-organization features. In particular, the basic principles and stages of self-organization process during solving an interdisciplinary problems of designing complex technical systems, as well as the architecture of proposed software along with some details of implementation are considered. The architecture of the software complex includes subject and problem ontologies, data and knowledge bases, set of "solvers" as well as intelligent scheduler. The intelligent scheduler implements the self-organization algorithm and provides creation the computing environment for solving considered problem using "solvers" as a buildings blocks. The implementation of self-organizing algorithm base on combination of knowledge representation as an ontologies, group decision-making, component and model-oriented approaches. Self-organization features in context of designing complex technical systems task are implemented on the stages of defining the design methodology, determining the source data, solving the problem, and training the system. The intelligent scheduler can analyze the state of current task with set of indicators and manage it through a set of local rules. This paper presents examples of local rules for each stage. The stages specifications in the form of technological diagrams that contains components used ("solvers"), their connections along with the results of their work describe the self-organization process features related to each stage. The software implementation are based on the capabilities of the used platform for creating knowledge-based systems developed by the authors and component approach applied to specialized component development.

Link to poster

ABSTRACT. A new methodological approach to solving the poorly formalized problem of modeling the dynamics of the phytomass of the plant community is presented. At the first stage, the prognostic model NDVI speaker is formed using the technology of artificial neural networks and a fuzzy-possible approach. At the second stage, the obtained values are metrized by switching to the chlorophyll index and the phytomass of the community is determined.

Link to poster

ABSTRACT. The paper describes the principles of construction, experience of application and prospects for the development of the telemedicine system (specialized medical messenger), including a subsystem of intelligent agents, designed to improve the efficiency and convenience of remote interaction between the patient and the doctor.

Link to poster

ABSTRACT. The mathematical model of decision making in complex dynamic systems under conditions of uncertainty is considered. Decision support is presented as support for the decision-maker activities in the processes of constructing and transforming the semiotic model of the decision-making situation, based on the extraction, processing, and structuring of relevant information from the Internet. The model is based on the model of intelligence as a form of organization of a person’s mental experience, which is described using three models related by parameters: syntactic, semantic, and pragmatic. A decision-making support method based on the representation of solutions of the inverse problem in the form of a qualitative ontology of decision classes (a conceptual framework of decision classes) is proposed. The methods for determining and interpreting the names of decision classes of the decisions conceptual frame-work are based on extracting relevant information from the Internet and struc-turing it using lexical-semantic pattern approaches and distributive semantics methods.

Link to poster

ABSTRACT. The paper describes the approaches of risk assessment in the development of an intelligent system for the rehabilitation and development of cognitive abilities of people with disabilities using virtual reality and existing bio-information technologies. The development of an intelligent system for the rehabilitation and development of cognitive abilities is based on the principles and methods of creating the dominance of plasticity in the brain structures and the rehabilitation process setup.

Link to poster

ABSTRACT. the most radical transformation of the system of continuing education, according to the authors of the article, will occur on the basis of the development and implementation of the "digital angel" technology. This technology will mark a new stage in the development and application of artificial intelligence. This direction of digitalization of continuing education in the context of a productive strategy for its modernization can serve as a basis for its preparation for changing the main targets related to the needs of the future labor market, where creative, reflexive and communicative (meta-) human competence will become the most popular.

ABSTRACT. Uncertainties and inconsistencies in knowledge are considered together in the works of A.S. Narignani as a part of nonfactors. However, a unified mathematical approach has not yet been proposed for their modelling and analysis. This paper considers a possibility of using a unified approach for modelling the following types of knowledge with nonfactors: knowledge with uncertainties in the form of variants of values and intervals; knowledge for which nonmonotonic and modal logics are applied; incorrect knowledge that contains paradoxes and contradictions. A generalized approach to modelling and analysis of the listed nonfactors on the basis of n-tuple algebra developed by the authors is proposed.

ABSTRACT. The paper scope is autonomous operational group control over the assembly of a complex product in manufacturing performed by a team of interacting autonomous robots. Its contribution is agent-based model of robots’ individual behavior formalized in terms of finite state machines with internal states and distributed self-organizing algorithm (protocol) of goal-oriented collective behavior of robots’ assembly team operating without any external intervention. An example illustrates the main paper contribution.

Link to poster

ABSTRACT. The paper presents a new way to monitor the level of human cognitive activity by analyzing biomedical signals, which are electroencephalograms (EEG) and electromyograms (EMG). Biomedical signals were recorded and processed using a bioengineering system “EMG/EEG”, which is a hardware and software tool that includes an electroneuromyograph and an electroencephalograph connected to personal computers with appropriate software. The paper shows the points of application of electrodes for recording EMG and EEG and describes a set of testees. During the experiments, the testees performed homogeneous cognitive operations (multiplication tasks). There is a detailed experiment scenario. The task of monitoring cognitive activity has been solved by forming a new system of features that characterize spectral characteristics of biomedical signals. The formed feature space allows analyzing a level of cognitive activity by EEG patterns and tracking human emotional responses by EMG patterns. Emotional stimulation procedures lead to an increasing amplitude spectrum in EMG patterns. For EEG patterns, there is an increase in the total values of spectral characteristics in delta 1 and delta 2 rhythms after stimulation, as well as a decrease in response time and a decrease in the number of mistakes made by testees while performing cognitive tasks. These trends are observed approximately 30 minutes after ending stimulation procedures. In the future, the results will be applied to create a monitoring tools and to correct emotional responses and human cognitive activity of a person.

ABSTRACT. The paper proposes two classification algorithms for solving problems containing only quantitative or only categorical features. Both algorithms proceed from the assumption that the object classes are different probability distributions of individual features, and both are based on the information granulation method, which offers the simplest dependencies for estimating the frequencies of attributes. However, in the algorithm for quantitative features, these estimates depend on the values of four parameters that adjust in each task in the course of cross-validation. In another algorithm (Algorithm 2), these frequencies are calculated based on the distribution of pairwise frequencies of features described by matrices pairwise frequencies of features. The estimated object class corresponds to the maximum probability. This method reflects the sensory process models of animals and is aimed at recognizing an object class by searching for a prototype in information accumulated in the brain. When implementing Algorithm 2 for some objects, the solution will not be determined due to the sparseness of these matrices. For such objects, an analogue of the method k-nearest neighbors is provided, which based on the calculation of the class to which belong majority of the nearest objects by each feature individually. The effectiveness of both algorithms was confirmed on seven databases, on two of them the error was less than 1%.

ABSTRACT. A model of coexistence of people and artificial creatures (robots) under conditions of robot dominance is considered. The mechanism of robot dominance regulates the distribution of the produced resource between people and robots. Two types of the people’s role in resource manufacturing are examined. The first is the interchangeability of people and robots; the second is the people's indispensability. Different scenarios of civilization evolution including extinction of the human population are described. Conditions that provide humanity with a more or less prosperous future are presented.

ABSTRACT. One of the critical tasks that have to be solved when forming a convolutional neural network (CNN) is the choice of the values of its hyperparameters. Existing attempts to solve this problem are based, as a rule, on one of two approaches. The first of them implements a series of experiments with different values of the hyperparameters of CNN. For each of the obtained sets of hyperparameter values, training is carried out for the corresponding network versions. These experiments are performed until we obtain a CNN with acceptable characteristics. This approach is simple to implement but does not guarantee high performance for CNN. In the second approach, the choice of network hyperparameter values is treated as an optimization problem. With the successful solution of such a problem, it is possible to obtain a CNN with sufficiently high characteristics. However, this task has considerable complexity, and also requires a large consumption of computing resources. This article proposes an alternative approach to solving the problem of choosing the values of hyperparameters for CNN, based on an analysis of the processes taking place in the network. We demonstrate the efficiency of this approach by solving the problem of classifying functional dependencies as an example.

ABSTRACT. The architecture of a hybrid intelligent information system for the analysis of the judicial practice of arbitration courts is discussed. The structure of the subsystems of consciousness and subconsciousness in the architecture of the proposed sys-tem is considered in detail. The text fragments extraction module plays a crucial role in the subconsciousness subsystem of the proposed system. The principles of operation of the text fragment extraction module are examined in detail. The ar-chitecture of a deep neural network, which is the basis of the module, is pro-posed. The aspects of the training of the proposed deep neural network are con-sidered. Variants of text vectorization based on the tf-idf and fasttext approaches are investigated; vectorized texts are input data for the proposed neural network. Experiments were conducted to determine the quality metrics for the proposed vectorization options. The experimental results show that the vectorization option based on tf-idf is superior to the combined vectorization option based on tf-idf and fasttext. The developed text fragments extraction module makes it possible to implement the proposed system successfully.

ABSTRACT. Mirror Neurons in the Interpretation of Action and Intentions

Yuri V. Bushov1,*, Vadim L. Ushakov2, Mihail V. Svetlik1, Sergey I. Kartashov2, Vyacheslav A. Orlov2

1National Research Tomsk State University, Tomsk, Russia 2 National Research Center «Kurchatov Institute», Moscow, Russia * bushov@bio.tsu.ru

Abstract. The aim of the research was the studying the activity of mirror neurons in humans during the observation and reproduction of rhythm. As markers of mirror neuron activity, we used depression of the EEG mu-rhythm in the alpha and beta frequency ranges, cortical interactions at the frequency of this rhythm, as well as the results of fMRI brain mapping. The research involved volunteers men and women aged from 18 to 27 years (University students). Research has shown that monitoring the reproduction of a five-second rhythm is accompanied by activation of not only those areas of the cortex where the «motor» mirror neurons are located, but also other cortex areas, as well as the basal ganglia and cerebellum. This findings suggest that mirror neurons themselves do not provide an understanding of actions and intentions, although they are involved in these processes. It is assumed that these neurons provide interaction between the prefrontal, sensory and motor areas of the cortex, as well as places where motor programs are stored in the brain. The result of the interaction of these structures is an understanding of the actions and intentions of other people.

Keywords: mirror neurons, observation and reproduction of rhythm, interpretation of actions.

1. Introduction

Studying the mirror neurons functions is an relevant scientific and practical problem that is important for human social behaviour understanding. According to the currently popular hypothesis [1], mirror neurons can serve as a neural basis for interpreting actions, imitating learning, and imitating the behaviour of other people. According to researchers [2], this is achieved by copying the observer's brain actions of another person by updating the corresponding motor programs. However, from this point of view, it is not clear why mirror neurons are activated not only when observing, but also when performing and mentally reproducing the same action. The purpose of this research was to study the activity of mirror neurons in humans during the observation and reproduction of rhythm.

2. Materials and Methods

During the preliminary examination, the features of the lateral organization of the brain were studied with the determination of the leading hand (using the questionnaire method) and the speech hemisphere (dichotic test). As a model of cognitive activity, the subjects were offered activities related to the observation and reproduction of a five-second rhythm. As markers of mirror neuron activity, we used EEG mu-rhythm depression in the alpha and beta frequency ranges, cortical interactions at the frequency of this rhythm, and results of fMRI brain mapping. The electroencephalographic study involved volunteers, nearly healthy men (31 people) and women (34 people), and students aged from 18 to 23 years. All subjects gave informed consent to participate in this study. Several series of experiments were carried out. In the first series («monitoring the reproduction of rhythm»), the subject observed the operator's hand, which first memorized the five-second rhythm, then the middle and index fingers of the leading hand reproduced this rhythm, periodically pressing the «Space» key. The rhythm period was set by a visual stimulus (a light square with a side of 2 cm, appearing periodically for 200 MS in the center of the darkened monitor screen). In the second and third series, the subject first memorized a five-second rhythm, then reproduced this rhythm with the fingers of the left hand, then with the right hand. Before and during cognitive activity, EEG was recorded in the frontal, middle, temporal, parietal, and occipital leads according to the «10-20%» system. When processing the obtained data, the maximum values of cross-correlation functions and spectral power estimates were calculated at short (1.5 s) artifact-free segments of EEG recording for 3 s («Background») and 1.5 s («Preparation») before pressing the key and immediately after the specified event («action Execution»). For statistical data processing, we used the "MatLab v6.5" package and the Wilcoxon criterion for related and independent samples. In part of the experiments, fMRI was used to study brain activity during observation and perception of time. These studies involved volunteers - 20 men and 20 women aged from 19 to 27 years (University students). The study included several series of experiments. In the first series, the subject is shown a video clip in which a white square with a side of 2 cm appears periodically in the screen center (with a period of 5 seconds). The subject must remember this rhythm. Then a video clip is shown showing the operator's hand playing a five-second rhythm by pressing the «space» key with the middle and index fingers of the leading hand. After that, they show a video clip with the image of the operator's motionless hand. In the second and third series, the subject reproduces the five-second rhythm by pressing buttons with the left or right hand, depending on the instructions. The subject is then shown a video clip with a picture of the stimulus (a white cross on a dark background in the center of the screen), on which his eyes should be directed during rest. The results of functional MRI were obtained in the complex of NBICS technologies of the Kurchatov Institute using a SIEMENS Magnetom Verio 3 Tesla tomograph. All fMRI data was pre-processed using the SPM8 package. Within each of the paradigms, pairwise comparisons were made based on student statistics and individual and group maps with a significance level of p<0.001 were obtained. All the obtained statistical maps were underlaid on a template T-1 image and anatomically linked the «active» voxels to the CONN Atlas.

3. Results and Discussion

The conducted research allowed us to detect statistically significant changes in the spectral characteristics of the EEG mu-rhythm in the middle regions of the cortex at different stages of the performed activity in men and women. It turned out that the nature of these changes depends on the frequency of this rhythm, gender, lateral organization of the brain, the type and stage of the activity performed. In particular, in the series «monitoring the reproduction of rhythm» on stage «Performing actions» in comparison with men, there is a statistically significant decrease in spectral power of EEG leads C3 at a frequency of 9 Hz to 10.1% (p<0.005), and at a frequency of 10 Hz at 9.4% (p<0.05). At the same time, at the «Preparation» stage, there were no statistically significant changes in the spectral power of the EEG at mu-rhythm frequencies in comparison with the background. The dependence of changes in the spectral characteristics of the mu-rhythm on its frequency, type and stage of the performed activity was also found in women. In particular, when reproducing a five-second rhythm with the right hand at the stage Of «filling in the action» in comparison with the background, they have a statistically significant increase in the spectral power of the EEG in the C4 lead at a frequency of 8 Hz (p<0.01), and at a frequency of 9 Hz in the same lead - a decrease of 10.8% (p<0.02). Probably, the decrease in the spectral power of the mu-rhythm detected at certain frequencies at the stages of «Preparation» and «action Execution» reflects the activation of «motor» mirror neurons [3]. Analysis of cortical interactions at the mu-rhythm frequency between the middle and other cortical regions during the observation and reproduction of the five-second rhythm allowed us to detect in men and women the dependence of these interactions on the type and stage of the performed activity. Comparison of the series with the right and left hands rhythm reproduction allowed us to detect interhemispheric differences in the levels of cortical connections. In particular, when the reproduction of a rhythm with the left hand observed a statistically significant increased right-brain connections, and, if reproduction rhythm in the right hand – strengthening ties left hemisphere. A clear dependence of cortical interactions on the type and stage of the activity performed was also found in women. In particular, when reproducing the rhythm with the left hand at the «Preparation» stage, they have a statistically significant (p= 0.01÷ 0.02) increase in the levels of cortical connections between the middle and temporal, middle and occipital EEG leads compared to the background. So, if in the background the correlation coefficient between leads C3 and T5 is r=0.634, then at the «Preparation» stage it increases statistically significantly (p=0.02) to r=0.649. If in the background the correlation coefficient between leads C3 and O1 is r=0.598, then at the «Preparation» stage it increases statistically significantly (p=0.01) to r=0.617. Along with this, there are clear gender differences in the dynamics of cortical connections in the observation and reproduction of the rhythm. Thus, if when observing the reproduction of the rhythm at the stage «performing an action» in men, a statistically significant (p<0.05) increase in cortical connections between the middle and parietal (CZ and P4), middle and temporal (CZ and T6, C4 and T6) leads was found in comparison with the background. Such changes in women were not detected. The comparison of the fMRI brain mapping results while monitoring the operator's hand, which plays a five-second rhythm, with the rest (viewing the video image of a stationary on the operator's hands) in men and women made it possible to detect activation not only in regions of «motor» mirror neurons [3], but also in other ones of the cortex and, in particular, right and left supramarginal gyrus, right and left angular gyrus, right and left lateral occipital cortex, right temporal courts, right and left frontal pole as well as the basal ganglia and some cerebellar areas. Gender differences were shown, in particular, in the fact that men have more activated angular gyrus on the left, while women have more activated angular gyrus on the right. At the same time, the area of the right and left middle temporal gyrus is more active in women. Probably, these gender differences are related to the peculiarities of the lateral organization of the brain in men and women. The comparison of the results of a fMRI during reproduction five-second rhythm with the rest state allowed to detect men and women are partially activated the same regions of the cortex like when observing the reproduction of rhythm and also precuneus, the occipital pole and some other regions of the cortex. However, when reproducing the rhythm, the activity of these brain regions is much more pronounced. So, if in men, when reproducing the rhythm in the area of the right and left lateral occipital cortex, the number of activated voxels is 878 and 1007 respectively, then when observing the reproduction of the rhythm - 28 and 60 voxels respectively. Gender differences were shown in the fact that women's brain regions involved in providing the proposed activity are activated more strongly than in men. So, if the number of activated voxels in the area of the right precentral gyrus in men is 179 voxels, then in women - 509 voxels. Thus, studies have shown that the observation and reproduction of the rhythm are accompanied by depression of the mu-rhythm at certain frequencies and, most often, an increase in cortical connections at the mu-rhythm frequency between the middle and other cortical areas. It turned out that the nature of these changes depends on the frequency of mu-rhythm, the type and stage of activity performed, gender and lateral organization of the brain. It was found that the observation of rhythm reproduction is accompanied by activation of the right and left lateral occipital cortex, right fusiform gyrus, right and left middle temporal gyrus, right and left precentral gyrus, right and left temporal pad, right and left supracranial gyrus, right and left angular gyrus, right and left frontal pole as well as basal ganglia and some areas of the cerebellum. These structures probably form a functional system that provides an understanding of actions and intentions. It is important to note that in addition to the temporal regions of the brain and the prefrontal cortex, where the «motor» mirror neurons are located [3], this system includes other areas of the cortex, as well as the cerebellum and basal ganglia, which are considered the storage place for motor programs. These results and some literature data [4] suggest that mirror neurons themselves do not provide an understanding of actions and intentions, although they are involved in these processes. Analysis of cortical connections at the mu-rhythm frequency suggests that these neurons provide interaction between the sensory, motor and prefrontal cortex zones, as well as places where motor programs are stored in the brain. The result of the interaction of these structures is probably an understanding of the actions and intentions of other people.

Conclusion

Thus, the research shows that mirror neurons themselves do not provide interpretation of actions and intentions, although they are involved in these processes. The results suggest that these neurons provide interaction between the prefrontal, sensory, and motor cortical regions, as well as storage locations for motor programs in the brain. The result of the interaction of these structures, apparently, is an understanding of the actions and intentions of other people. The study was funded by RFBR according to the research project № 18-001-00001. References

1. J. R. Skoyles, Gesture Language Origins and Right Handedness. Psychology. 2000. 11, pp. 24 -29. 2. G. Rizzolatti, C. Sinigaglia, F. Anderson, Mirrors in the Brain: How Our Minds Share Actions, Emotions, and Experience // Oxford University Press; 1 edition, 2008, P. 242. 3. Kosonogov V., Why the mirror neurons cannot support action understanding. Neurophysiology. 2012. 44 (6), pp. 499-502. 4. M. B. Schippers, A. Roebroeck, R. Renken, L. Nanetti, C. Keysers, Mapping the information flow from one brain to another during gestural communication. Proceedings of the National Academy of Sciences. 2010. 107 (20), pp. 9388-9393.

ABSTRACT. Variable resistors have proven to be potential elements of massively parallel computation systems. Non-resistor components (transistors, diodes, capacitors) are indispensible is such systems. However it might be desirable to integrate only resistor elements into large systems at the first stage of integration. The question arises of the adequate data representation in this sort of systems. In this connection a system totally consisted of variable resistors located at junctions of conductors is considered. The use of individual resistors as data medium is shown to be impracticable because of nonlocality of the data recording process. The use of group properties of a conductance matrix looks more promising. It is convenient to take a representative graph of the conductance matrix as data-encoding entities. Being the result of the thinning of the matrix by zeroing small conductivities, this sort of graph is noise resistant and can determine the effective two-terminal conductivity to good accuracy. The dynamics of the representative graph is defined by the interaction between initial conditions, the signature graph and input signals.

ABSTRACT. The solution of boundary value problems describing piecewise homogeneous media on radial basis functions networks is considered. An algorithm is proposed based on the solution of individual problems for each region with differ-ent environmental properties associated with the conjugation conditions. This removes the restrictions on the radial basis functions used. An algorithm for training the network by the method proposed by the authors is used. The results of solving the model problem showed the effectiveness of the proposed algorithm.

ABSTRACT. This paper presents the solar plant intelligent control system which includes: 5 convolutional blocks and then a two-layered recurrent network, fuzzy units, and two-layered recurrent networks. The experimental results demonstrate that the proposed intelligent control system based on solar plant images and data of basic sensors under uniform and non-uniform solar irradiance provides the best control speed and performance, as compared to a common solar plant control system.

ABSTRACT. The probabilistic Hopfield model known also as the Boltzman machine is a basic example in the zoo of artificial neural networks. Initially, it was designed as a model of associative memory, but played a fundamental role in understanding the statistical nature of the realm of neural networks. The close relation between the Boltzman machine and the Ising model was a challenging observation in [1]. In this note we go further, we establish another type of structural similarity between these models sharing the methods of the Bethe ansatz family of integrable statistical mechanics. We examine the asymmetric model on the triangular lattice with arbitrary weights. We show that the probability of passing a trajectory in time dynamics obeys the Gibbs distribution with a partition function of the Ising model on the cubic lattice with additional weights on diagonals.

ABSTRACT. The problem of the implementation of fully hardware neural networks based on programmable logic circuits is considered. A method for minimizing the hardware costs of artificial neural networks is proposed. The model of the learning logic gate network is given. For programmable logic gate array a model of a learning logic gate is proposed. On the base of the methods of algebra of logic the decomposition of a two-layer gate neural network with a trained hidden layer is proposed. Models of learning logic gates based on conjunction and disjunction functions are being developed. The intellectual properties of such models are shown. A method of mapping to the basis of learning logic gate networks is proposed. The results of discrete optimization of network parameters are presented. It is shown that the genetic algorithm has the lowest training error. A conclusion is drawn on the applicability of the obtained models when constructing optimal combinatorial circuits for neural network processing.

ABSTRACT. The problem of constructing a criterion, that allows to evaluate the complexity of the problem to solve by evaluating the inner layer of the variational autoencoder, is considered. A variational autoencoder was modeled for image classification problems of several levels of complexity. A complexity estimate, based on the study of the inner layer of the autoencoder in the form of measuring distances between distributions was formed. Calculated complexity estimates allow you to rank classification tasks according to the expert rating.

ABSTRACT. The article proposes an approach to automated metagraph construction based on the Hopfield neural network. The metagraph model is a kind of complex network model with emergence. Thus, constructing a metagraph model based on a com-pound domain is a complex task that requires automation. A brief description of the Hopfield neural network is given. The algorithm for automated metagraph generation based on Hopfield neural network is proposed. The vertices and metavertices of metagraph are represented as vertices-nodes, which are uniquely defined by a set of features. A feature may be an attribute belonging to vertex (metavertex), the nested vertex (metavertex), or the nested edge. The Hopfield neural network is trained on the metagraph data, and then a new vertex-node is passed to the network for recognition. As a result of recognition, two cases can occur. In the first case, the Hopfield neural network returns a spurious pattern. Therefore, the added vertex-node is not close to any of the existing ones; there-fore, it should be added to the metagraph as an independent vertex-node. In the second case, the vertex-node that the Hopfield network returns is close to the new vertex-node, and it is correct to combine both vertices-nodes into one new metavertex. The results of experiments show that the proposed approach can be used for metagraph construction.

ABSTRACT. The purpose of working out of model of scenario forecasting is a substantiation and forecasting of actions for maintenance of additions of values of functional (a rating of university) and target indicators (indicators) of activity of university in the international institutional rating QS to the demanded values necessary for entering of university in TOP -500 of high schools in 2025. For achievement of the set goal of the research application of methods of solution of weakly structured tasks on the basis of development of the model of scenario forecast (planning) with the use of cognitive maps, which allowed to define probable (possible) tendencies of development of events on alternative variants and possible consequences of accepted decisions with the purpose of a choice of the most preferable alternative, is grounded. The proposed approach, which ensures the guaranteed achievement of the set goal, allows to find the most acceptable scenario of planning the increment of the functional and indicator values to the target values under the set constraints due to the impulse effects on the latent factors affecting the indicators. In the course of the research the following tasks have been solved: a cognitive model of scenario forecasting of measures to achieve the required values of the university performance targets in the international institutional ranking QS has been developed, on the basis of the developed model the calculation of the most preferable variant of the set of required intensity of influence on the control variables (factors - causes) at the given increment of the target factor value has been performed. The obtained results allowed to form a scenario plan of the required step-by-step increase in indicator values considering the latent factors affecting them in the interval between 2020 and 2025.

ABSTRACT. The paper presents the results of an analysis of the residents’ reaction to the implementation of road construction projects based on digital content to prevent urban conflicts. The material for the study was the data of social networks, microblogs, blogs, instant messengers, videos, forums, reviews on the construction of road transport facilities in Moscow, namely the imple-mentation of transport interchange hub projects (Michurinsky, Khovrino, Varshavskaya hubs). The study used the transdisciplinary approach. To iden-tify digital conflicts and interpret the content, neural network text analysis, content analysis and sentiment analysis were used. The topic structure of the database and the semantic network were identified and analyzed; an associative search was performed, and the social stress index was calculated. The study made it possible to identify conflictogenic digital zones and main causes of conflicts, to determine the level of social stress, to predict the growth directions of identified conflicts and to develop recom-mendations to neutralize the situation.

ABSTRACT. The problem of optimal reception of useful signals by adaptive antenna arrays (AAA) in the conditions of a complex signal-to-noise environment is considered. The procedures for compiling a training sample, optimizing the size of hidden layers and the number of training epochs for a feedforward neural network for solving problems of suppressing wideband interference are described. A numerical simulation was performed in the Matlab environment, the results of which show that the neural network algorithm of amplitude-phase control in AAA provides an average depth of interference suppression in the frequency band of up to 10%, comparable in efficiency with the traditional adaptation algorithm by the direct matrix inversion method using the maximum signal-to-interference-plus-noise ratio criterion (-19,2 dB in the neural network algorithm versus -18,6 dB in the traditional one). The possibility of reducing the volume of the neural network with a significant reduction in computational costs in adaptation tasks is also shown. The use of neural networks as an adaptive processor can significantly expand the class of tasks with the help of AAA and provide a high degree of versatility in the use of the antenna for various scenarios of signal-noise situations. The application of the considered algorithms is possible in communication systems, radar, radio navigation and other areas where it is required to improve the quality of received signals.

ABSTRACT. The advantages that the use of neural networks in automatic control systems of dynamic objects can bring are considered. A technique for hierarchical construction of adaptive control laws is proposed and the application results on a simple robotic device - a spherical direct control drive (SDCD), which is part of a vision system with structured illumination, are presented.

ABSTRACT. In this work, we analyzed the signal-to-noise characteristic at the output of an adaptive antenna array for various control modes — traditional, phase, discrete phase, and with a relationship of amplitude and phase for a discrete phase. The choice of phase control was determined from the considerations of maintaining the useful signal power and the simplicity of designing the corresponding adaptive antenna arrays. Because the phase control problem does not have an exact analytical solution, it was decided to use a neural network as a phase control unit of an adaptive antenna array. Because purely phase control, from theoretical considerations, should show the worst results in maximizing the signal-to-noise parameter, it was decided to carry out numerical simulation for an antenna array with discrete phases, where the amplitude was chosen based on traditional control, which is predicted to greatly improve the signal-to-noise parameter compared to with conventional discrete control adaptive antenna array. It was shown that the use of the last three control modes shows results that are close to the results of the general traditional amplitude-phase control.

ABSTRACT. This article discusses neural network approaches that can find application in the tasks of tracking moving objects in video. A segmentation algorithm was proposed, as well as a classical correlation algorithm with constant further training of the reference image

ABSTRACT. The article discusses the development of a search system for similar images. For the analysis of images, the collaboration of methods for extracting information and color components, based on which each image acquires a numerical hash, is proposed. Images are compared solely on the basis of the calculated hash. For comparison, a fully connected neural network is used.

ABSTRACT. This work studies the interaction between learning and evolution at minimizing of the spin-glass energy. Our computer model considers the population of agents. The genotypes and phenotypes of agents are coded by spin glasses. The energy of a spin glass is minimized by means of both learning and evolution. The phenotype of an agent is adjusted by learning during the agent life. The fitness of the agent is determined by the final agent pheno-type after learning. The genotype of the agent-parent is inherited by the agent-child at the beginning of the generation; at this transmission, the genotype mutates. In the considered spin-glass case, there are many local maxima of agent fitness. The properties of the considered interaction are analyzed.

ABSTRACT. Text summarization is a natural language processing (NLP) technique that is used for extracting key ideas from a given document. Advanced summarization methods should be able to comprehend high-level semantic in the text. In this paper, prevalent language models (LM) and frameworks for text summarization based on them are reviewed. It is shown that the best quality summaries are achieved with the use of recent LMs. Furthermore, some of the considered frameworks are suitable for multiple NLP tasks and are able to learn in unsupervised way. Since the ability to learn and communicate in natural language is an important characteristic of artificial general intelligence (AGI), these researches are highly relevant to the development of AGI

ABSTRACT. Although bilinguals’ word processing in L2 is one of the mainstreams of psycholinguistic studies, the actual research on how they process function words (FW) with graded differences in L2 proficiency is still lagging behind plenty of studies available in second language acquisition literature. We used eye movement measures of FW reading in L2 to investigate whether the degree of L2 proficiency modulates L2 FW processing. The results showed that proficient L2 bilinguals process FW faster compared to basic L2 bilinguals. Taken together, our findings are consistent with implicit learning statements, the Weaker Links hypothesis, and the bilingual activation model plus. Thus, the level of L2 proficiency can be predicted based on L2 FW processing features.

ABSTRACT. The article considers one of the aspects of the situational awareness problem for control systems of unmanned vehicles. We interpret this problem as getting information about the current situation in which, for example, an unmanned aerial vehicle (UAV) is operating. This information is required as source data for decision-making in the UAV behavior control process. One possible component of situational awareness is information about objects in the space surrounding the UAV. At the same time, it is important to know along which trajectories these objects move. Also, we need to predict the motion of the observed objects. We consider this task in the article as a formation example for one of the elements of situational awareness. To solve this problem, we prepare a data set using the FlightGear flight simulator. We extract from this set the training, validation, and test sets required to obtain a neural network that predicts the trajectory of the object being tracked. Then, based on the collected data that characterize the behavior of the desired object, we design a neural network model based on recurrent neural networks to solve the problem of predicting the trajectory of a dynamic object.

ABSTRACT. Despite the efforts of many scientists and philosophers in understanding the nature of consciousness, there is no significant progress in the field. The way out of the impasse is offered by cognitive neuroscience based on the concept of neural correlates of consciousness (NCC). According to the NCC-concept, the studies should determine the minimum sufficient number of neurons or the minimum activity of the nervous system that necessarily accompany this or that conscious experience. The article argues that from the study of consciousness phenomena in all their diversity it can be more effective to study the key property of consciousness - reflection sense. Experiments show that behavior that is impossible without reflection is demonstrated by organisms with a fairly simple (less than 1 million neurons) brain - bees, bumblebees, ants. This indicates that neural correlates of reflection (NCR) can be studied with fairly simple artificial neural networks. In the paper the possibility of detecting NCR in a simple recurrent neural network trained to pass the delayed matching to sample (DMTS) test, which unambiguously indicates the formation of the internal representation of the external world and the use of this representation for decision making is investigated Computational experiments have shown that the identification of portrait neural correlates of reflection is not possible even for the same neural network. This result raises the question of the possibility of revealing in the brain of a human and higher animals neural correlates of consciousness associated with a certain content.

ABSTRACT. A neural network algorithm has been developed to identify in the statistical properties of coherent biological signals present in a mixture with other signals and interference. The algorithm includes real-time determination of the coherence function of signals between fluctuations in systemic blood pressure and blood flow velocities in the left and right middle cerebral arteries and the phase shift function between these signals in the Meyer wavelength range. To reduce the influence of noise, it is proposed to use the technique of a sliding frame, divided into windows. The coherence and phase shift functions obtained in the windows are averaged within the frame boundaries. As a result, smoothed functions can be obtained in the time-frequency domain. To detect infractions of the autoregulation process, it is proposed to use trained neural feedforward network, which generalizing property can be improved as new experimental data are obtained while maintaining a balance between individual and general characteristics of patients

ABSTRACT. The study describes approaches to the description of the functional states of the human cardiovascular system using dynamic patterns of amplitude-phase coupling of the electrocardiographic signal with the purpose of highlighting informative features and patterns of regulation and control mechanisms for analyzing the diagnostics and rehabilitation.

ABSTRACT. The paper considers the task of reconstructing protein-protein interactions in the interactome of the excitatory synapses of the hippocampus under the control of the hormone leptin. A number of effects of this hormone are realized through the accumulation of triphosphoinositides on the membrane, but the molecular mechanisms of these processes and their involvement in the regulation of synaptic plasticity in the CA1 field of the hippocampus are not fully understood.

ABSTRACT. The paper investigates the architecture of deep neural networks for recognizing human emotions from voice. Convolutional neural networks and recurrent neural networks with LSTM memory cells were used as the models of deep neural networks. Computer experiments on the use of constructed neural networks for recognizing emotions in human speech, contained in the RAVDESS database, were conducted. The results obtained showed a high efficiency of the chosen approach, and the accuracy estimates for individual emotions were 92%.

ABSTRACT. The paper considers the problem of forecasting hourly market electricity prices using the artificial neural networks. A Python application has been developed to perform the electricity prices forecasting and analysis, using NeuroLab and Keras libraries. The paper presents the results of numerical experiments on electricity prices forecasting using the developed application.

ABSTRACT. В связи с появлением новых устройств и датчиков для снятия сигна-лов флоуметрии и развитием алгоритмов машинного обучения, появи-лась возможность использовать эти сигналы для диагностики заболева-ний сердечно-сосудистой системы (ССС). Недавние исследования по-казали потенциальную клиническую пользу флоуметрии для диагно-стики и наблюдения определенных заболеваний сердца. В этой статье будет проведено сравнение различных алгоритмов машинного обучения и оценка их эффективности для решения задачи диагностики заболеваний сердечно-сосудистой системы. В результате описанных в статье исследований показана эффективность сверточных нейросетей в задаче диагностики заболеваний сердца по данным до-плеровской флоуметрии.

ABSTRACT. A neural network complex consisting of interacting dynamic recurrent cells with developed feedback systems is being designed to build an adaptive wireless data communication system. The structure of the digital neural network cell has been developed to generate signals of synchronization; it contains reconfigurable memory areas to adjust the synchronization system of the communication network in the learning process of the main neural network expert. Based on the analysis of the internal structures and the method for interaction of neural network blocks, the algorithm for learning and parameter setting has been developed and tested.

ABSTRACT. The accuracy of the Coriolis vibratory gyroscope is determined by the technological factors of manufacturing its resonator and the heterogeneity of its physical characteristics. The elimination of heterogeneities primarily requires the accurate determination of their parameters (amplitudes, locations). In this work, the problem of identifying the parameters of the Coriolis vibratory gyroscope, using time series data for the main and quadrature signals obtained from information pickup sensors is considered. As a dynamic model of the gyroscope, a system of two second-order ordinary differential equations with coefficients associated with the parameters of the device is used. The algorithm for identifying unknown parameters is based on the use of an autoregressive model of the time series and a single-layer neural network that implements forecasting. The method gives us the possibility to get an explicit interpretation of the weight coefficients of the forecast model of the two-dimensional time series.

ABSTRACT. The paper researches the problem of detecting young people’s shift to destructive formats of communicative behavior in the social networks. The problem of detecting and preventing such de-structive behavior can be solved in case its specific features are elicited. The authors propose an inte-grated approach based on using neural networks analysis and application of the following set of techniques: (a) for detecting the features of destructive formats of communicative behavior within users’ profiles in social network and for probabilistic forecasting correlation between users’ profiles peculiarities and the results of Ego-structure test (by G. Ammon); (b) for detecting the shift to destructive formats of communicative behavior by application of original seven-level evaluation scale; (c) for preventing destructive communicative behavior, providing proper psychological and educational tools and guidance. The paper details the first technique. Within the scope of the first task and based on the application of a multilayer neural network, there has been elaborated an algorithm of probabilistic forecasting the results of the Ego-structure test. The latter technique has been introduced and the first results of its implementation investigated.

ABSTRACT. The development of an intelligent video analytics system with modules for detecting, recognizing and tracking objects has led to an understanding of the process of forming detectors of objects of various types in terms of information processing by living systems. The system of semantic image analysis using cascades of strong classifiers based on nonlocal binary templates is presented as a biomorphic model of structural and functional modules, for example, columns of the cortex, for information processing in living systems. The proposed multilayer neural network architecture can be classified as "growing neural networks". This work was supported by the Ministry of Education and Science of the Russian Federation, project No. 14.Y26.31.0022.

ABSTRACT. The structural and functional organization of inter-neural communication channels is considered in the light of morphological and neurophysiological knowledge about non-specific brain formations that control and consolidate the brain. Attention is drawn to the typology of inter-neural connections within neural networks. The heterogeneity of neural modules containing representatives of the Central regulatory apparatus explains the systemic consolidation of the cortical and subcortical parts of the brain.

ABSTRACT. The class of multilayer modular neural networks with a self-similar structure is considered. The paper introduces the concept of morphogenesis and network regularity. Conditions for the morphogenesis of a multilayer network are determined, a theorem on the morphology of a loosely coupled multilayer network is proved, and invariants of system graphs of regular self-similar networks are obtained. A rule for constructing a graph of self-similar multilayer networks is proposed. It is noted that self-similar networks describe the structure of fast Fourier transform algorithms.

ABSTRACT. The neuron connection architecture in the visual system is proposed (for computer modeling), which provides the representation stability on the constant screen of the visual space picture perceived by the eye retina when the gaze is shifted (in the fixed head conditions). The described basic neuron model (architecture) is then discussed in terms of the possibility of reducing its redundancy and simplifying it.

ABSTRACT. Transcranial magnetic stimulation (TMS) is a powerful diagnostic and therapeutic tool that relies on changes in the neuronal activity following application of electromagnetic pulses. Since TMS-induced changes in the neuronal activity are rather focal, to get the desired effect it is necessary to select the optimal position of the stimulation coil. TMS navigation systems allow tracking the position of a simulating coil with a high precision. However, they do not suggest the optimal coil position for a given set of target fibers in a particular individual. Finding the optimal coil position by probing experimentally multiple stimulations sites is a time-consuming and laborious procedure. Existing theoretical models of neuronal excitation by the electromagnetic field suggest that TMS-induced effects should depend on the effective electrical field, i.e. projection of the induced electrical field to the local orientation of the stimulated axons. In this work we propose a stochastic region contraction approach to optimize the TMS coil position by maximizing the TMS-induced effect for a target group of the nervous fibers reconstructed from diffusion MRI data. The found stimulation site showed two-fold higher effect than the best position from an actual experimental TMS mapping session.

ABSTRACT. Robotization of crystal sorting processes in the mining industry is an urgent challenge, for the solution of which it is necessary to develop new methods and algorithms. Promising data sources are cameras that allow one to get crystal images in several projections. In this research, algorithms for automatic crystal sorting based on processing of crystal images obtained in seven projections are developed. Sorting is based on evaluating the transparency and color of the crystals. At the morphological processing stage, a way to obtain a crystal template characterizing its shape is proposed. Application of the template allowed to obtain histograms of crystal colors. The principal components of the color histogram are predictors of the quality of the crystal, which are fed to the input of a multilayer perceptron, which makes a decision on a separate projection of the crystal. A special neural network summarizes the solutions obtained for each projection. This architecture reduces the size of the neural network.

ABSTRACT. Several of the presented works performed by the authors have shown effectiveness of application of neural networks. The article describes a case study of optimal of using neural networks to select the optimal mode of the fractionation process. Creating a digital double of the chemical-technological process is described. The work on modeling several catalytic oil refining processes is presented. It should be noted that the shown effectiveness of machine learning for solving many problems pushes large industrial enterprises to accumulate information about their activities and treats it as a valuable resource while simultaneously increasing their experience in use.

ABSTRACT. Currently, ontologies have become the main means for formalizing and systematizing knowledge in scientific subject domains, therefore, there exist an urgent need for methods and software tools that make it possible to involve specialists in specific subject areas in the process of constructing ontologies. The paper presents an approach to the development of an information-analytical Internet resource that supports the use of ontological design patterns that can effectively solve this problem.

ABSTRACT. The paper examines a new approach to the creation of knowledge bases from empirical research datasets. Several dozens of metadata standards for empirical research data exist, making it difficult for users to choose and apply such standards. Thus, the integration of datasets from empirical studies, even similar ones, is difficult. To resolve this problem, an ontology-based methodology for building knowledge bases for empirical research data (ERD) has been developed. The methodology is based on a unique ontology system that describes the empirical research data (ONTO-ERD) and methods and algorithms of its enrichment and population.

ABSTRACT. To provide security for modern computer systems (i.e. identify threats and employ countermeasures) threat modelling is used on early stages of life cycle (requirements, design). Security patterns can be applied as security design decisions. However there are some challenges, related to management of security patterns, in particular, lack of methods to identify the necessity of security patterns and weak integration with security risk-based models. To overcome these restrictions we have developed an ontological format (schema), which allows a) creation of security pattern catalogs, and b) definition of context labels to map patterns with design decisions and security problems. We have proposed a usage model of security pattern catalogs. The usage model enables creation of domain-specific threat models, used for ontology-driven threat modelling. Also, OWL ontology and a free toolset (Java, OWL API) have been developed to manage security pattern catalogs and motivate development of high-level software tools for maintenance of security pattern catalogs.

ABSTRACT. This article proposes an approach to a comprehensive assessment of expert knowledge that utilizes a model based on scales for various characteristics of diseases. Suggested approach has an option to factor in the fuzziness of the clinical picture of diseases and various combinations of signs in particular cases. Based on the hypotheses, differential diagnostic series and comparison of reference models with personal models of new cases are formed, that allows to rate the degree of similarity and identify the disease. The study was carried out on the mucopolysaccharidoses as an example, which belong to the class of inherited orphan diseases of lysosomal nature.

ABSTRACT. В статье представлено извлечение и структуризация медицинских знаний в области риска развития артериальной гипертонии. Процесс извлечения знаний осуществлялся в два этапа. Знания, извлекаемые вначале из научной литературы, согласовывались с экспертом-кардиологом. Структуризация осуществлялась на основе метода построения концептуальных карт, которые содержали факторы риска развития артериальной гипертонии, а также рекомендации по изменению образа жизни, являющиеся протекторами.

ABSTRACT. The experience in the development and evolutionary development of technology of knowledge acquisition from various sources on the basis of the original combined method of knowledge acquisition, which is an important part of the problem-oriented methodology for building integrated expert systems for static and dynamic problem domains, are analyzed. Particular emphasis is placed on experimental software modeling of the processes of temporal knowledge acquiring from experts, NL texts and temporal databases and analysis of the results (using the example of medical diagnostics).

ABSTRACT. A method of intelligent information search and contextual information provision in distributed data warehouses is proposed, which allows increasing the efficiency and quality of providing information needs for intellectual preparation and decision support. The method is based on the proposed compositional ontological model that provides an interoperable representation of knowledge about the tasks (processes) of the subject area, taking into account user profiles, in combination with functionally oriented information resources formed on the basis of generalization and semantic integration of structured, poorly structured and unstructured data from heterogeneous sources.

ABSTRACT. An extensive and diverse dataset is a key requirement for successful training of a deep model. Compared to on-site data collection, 3D modelling allows to generate large datasets faster and cheaper. Still, the diversity and the perceptual realism of synthetic images remain in the realm of the experience of a 3D modeller. Moreover, hard sample mining with 3D modelling poses an open question: which synthetic images are challenging for an object detection model? We present an Adversarial 3D Modelling framework for training an object detection model against a reinforcement learning-based adversarial controller. The controller alters the 3D simulator parameters to generate complex synthetic images. The aim of the controller is to minimize the score of the object detection model during the training time. We hypothesize, that such objective of the controller allows to maximize the score of the detection model during inference on real-world data. We evaluate our approach by training a YOLOv3 object detection model using our adversarial framework. A comparison with a similar model trained on random synthetic and real images proves that our framework allows to achieve better performance than using random real of synthetic data.

ABSTRACT. The second order effects in the resistive switching of the memristors based on poly-p-xylylene (PPX) and PPX with embedded silver nanoparticles were studied. The decrease of the switching time caused by second order effects under preliminary memristor heating with electric pulses was observed. The obtained results indicate the possibility of the detected effects usage for the creation of the neuromorphic computing systems.

ABSTRACT. In this paper, we present a method for quantization of neural net-work weights by means of maximizing correlations between the initial and quantized weights taking into account the weight density distribution in each layer. We perform the quantization after the neural network training without a subsequent post-training and compare our algorithm with linear and expo-nential quantization discussed in paper [8]. We tested the algorithm using the ImageNet dataset for the neural networks VGG-16, ResNet-50, Xception [2]. In the case of the ResNet-50 and Xception neural networks, the top-1 accu-racy drop was 6-8% when we reserved 5-bit for storing each weight of one layer; for 6-bit, it was 1.5-3% and less than 1.5% when we reserved 7-bit. When we used 5-bit per weight of one layer, for the VGG-16 neural network the top-1 accuracy drop was around 1.5% and it was less 1% when the num-ber of bits per weight was larger.

ABSTRACT. In previous papers, we showed that our n-vicinity method allowed us to calculate with high accuracy critical values of the inverse temperature of Ising models with short-range interaction. We generalized this method to the case of long-range interaction in spin systems. The comparison of our results for the planar and cubic Ising models showed a good agreement with computer simulations.

ABSTRACT. The purpose of this work was to study the structure and operation of the human visual system, as well as its mathematical description and computer modeling. The work included the following stages: analysis of the structure and functioning of the visual system from a biological point of view; con-struction of a mathematical model of the retina; development and research of a computer structural and functional model of the retina, models of eye muscles; research of some reflex movements of the eye on a model that combines the retina and control of the direction of the eye. The general computer model reproduces such reflex eye movements as involuntary sac-cades and eye movements that follow a moving object.

ABSTRACT. The dynamic of ring of m unidirectionally coupled neural oscillators with synaptic coupling are considered. The system of relay differential-difference equations was chosen as the mathematical model of this ring. The unidirectional synaptic coupling was modeled based on the idea of fast threshold modulation (FTM). We show that the system has at least m stable cycles, which are discrete traveling waves. Moreover, the solution corresponding to each of the oscillators is a periodic function with a predetermined number of bursts per period. Also in case of even m it is proved existence and stability of impulse-refractoctive periodic mode. This is a such solution that oscillators with odd indexes have high periodic bursts and with even indexes have a refractory behaviour.