Artificial Intelligence

The paper refers to the research direction in which models of social behavior are the methodological basis for the functioning of robot (animat) groups. The purpose of this study is to implement a complex regulatory behavior of animat groups using previously created models and methods. The applicability of this approach is demonstrated by the task of redistributing animats between groups. To accomplish this, the paper proposes to implement a mechanism similar to the phenomenon of slavery that is characteristic of some species of ants. Slavery is a form of social parasitism and can be considered as a method for the redistribution of individuals between families (groups). The paper describes different types of slavery and the behavior of slave owners and slaves among species of ants. The main processes that make up this behavior are: exploring territory, organization of raids, seizure of slaves and their transfer to the slave-maker nests, and slaves adaptation in the new nest. It is proposed that this behavior is based on the “friend-alien” identification and is an evolutionary development of food and territorial behavior. The paper describes previously created methods, models, and mechanisms for implementing similar forms of animats’ behavior: foraging, pack hunting, territory defense, and domination based on aggression. A method for identifying an animat and determining its internal state, which is necessary for organizing the interaction of animats, is proposed. Finally, the paper describes experiments confirming the applicability of the proposed method.

Hierarchical reinforcement learning (HRL) is another step towards the convergence of learning and planning methods. The resulting reusable abstract plans facilitate both the applicability of transfer learning and increasing of resilience in difficult environments with delayed rewards. However, on the way of the practical application of HRL, especially in robotics, there are a number of difficulties, among which the key is a semi-manual task of the creation of the hierarchy of actions, which the agent uses as a pre-trained scheme. In this paper, we present a new approach for simultaneous constructing and applying the hierarchy of actions and sub-goals. In contrast to prior efforts in this direction, the method is based on a united loop of clustering of the environment’s states observed by the agent and allocation of sub-targets by the modified bottleneck method for constructing of abstract machines hierarchy. The general machine is built using the so-called programmable schemes, which are quite universal for the organization of transfer learning for a wide class of tasks. A particular abstract machine is assigned for each set of clustered states. The goal of each machine is to reach one of the found bottleneck states and then get into another cluster. We evaluate our approach using a standard suite of experiments on a challenging planning problem domain and show that our approach facilitates learning without prior knowledge.

The paper describes a hierarchical control architecture for robotic systems with learning that allows combining various goal-directed algorithms. A top-level control algorithm is proposed that switches control between base algorithms: Q-learning, random walk and a rule-based planning. The algorithm is implemented as a software module and is verified by the example of the task of finding a given door in a building of complex planning. The task is considered as a reinforcement learning problem in two distinct cases: with a goal fixed between the episodes and the goal changing from episode to episode. The simulation showed that the proposed method is more stable for different variants of the task than each of the basic ones separately, although it does not give the best result for each individual case.

An application of deep neural networks was studied in the area of stock prices forecasting of pharmacies chain “36 and 6”. The learning sample formation in the time series area was shown and a neural network architecture was proposed. The neural network for exchange trade forecasting using Python’s Keras Library was developed and trained. The basic parameters setting of algorithm have been carried out.

The paper deals with using a Hamming neural network to predict the limiting destruction force under load of a welded connection in shear.

We propose an algorithm for encoding information on dynamic resistance into bipolar signals required for a Hamming neural network tuning and operation.

A computer program that implements a neural network based on the proposed algorithm has been developed. The results of the neural network training and testing are presented. The method proposed in the paper complies with the requirements of ISO 9000:2015 standard for continuous monitoring and documentation of each welded connection. The analysis showed that relative prediction error of the destruction force of a weld does not exceed 10%. Thus, the possibility of using a Hamming neural network to predict bond strength of welded connections based on dynamic resistance of the welding zone has been confirmed. The work was supported by the Russian Foundation of Basic Research (Grant Agreement 15-08-03125 A).

The goal of this paper is to present and compare algorithms for thematic mapping of multispectral satellite images. The paper proposes a nonlinear multi-dimensional filter to combine the results of processing of several multi-temporal multispectral images. Options for implementation of procedures are discussed.

This paper studies the problem of building a modern information society as a task of forming a virtual knowledge space. This task can be implemented on the IT platform of a digital library, which ensures the formation and provision of information resources in various areas to general public users. It shows how electronic copies of objects of a library, archive and museum storage represented in the form of texts, graphic images, and audio/video objects, including 3D-models, can be integrated using digital library facilities. The concept of a hierarchical level of electronic objects is introduced. The definitions of objects of various levels are given and the principles of work with objects of each level are formulated. The hierarchical representation of electronic objects in a digital library environment is proposed. A distinction between thematic, theme-specific, and interdisciplinary collections is shown. The information environment for forming subtheme collections is represented as a set of databases created by information fund holders. This environment is distributed and unified in terms of the software and hardware used, as well as in terms of a set of requirements for digital images being formed. The paper formulates the basic principles of forming collections of various levels in a digital library environment and reflects the developed digital solutions that ensure their implementation. For each hierarchical level, it gives examples of objects and collections provided on the portal of the Scientific Heritage of Russia Digital Library (SHR DL) and formed in compliance with the proposed principles and digital solutions. In particular, the paper describes in detail the thematic collection of publications in mathematics being a database formed through LibMeta DBMS; interdisciplinary collection “Garden of Life” created in collaboration with one of the Russian museums, and some other collections.

Active Logic is a conceptual system with reasoning formalism that allows for correlation of their results with specific points in time and that has tolerance to inconsistencies. Currently, tolerance to inconsistencies (paraconsistency) in Active Logic systems is theoretically justified in the works of the authors of this paper and is attributed to the so-called formalisms of stepping theories, which integrate the principles of Active Logic and Logic Programming. More specifically, the argumentation semantics of so-called formalisms of stepping theories with two kinds of negation has been proved to be paraconsistent. This formalism has more expressive power than the other formalisms of stepping theories and to a greater extent satisfies the principles of Logic Programming. This case study proposes the declarative semantics for formalisms of stepping theories and represents its equivalency with respect to the argumentation semantics of this type of formalism. This, in turn, means that the proposed declarative semantics is also paraconsistent, and logical inconsistencies existing in these theories do not result in their destruction.

Issues concerning intelligent data analysis occurring in machine learning are investigated. A scheme for synthesizing correct supervised classification procedures is proposed. These procedures are focused on specifying partial order relations on sets of feature values; they are based on a generalization of the classical concepts of logical classification. It is shown that learning a correct logical classifier requires solution of an intractable discrete problem to be solved. This is the dualization problem over products of partially ordered sets. The matrix formulation of this problem is given. The effectiveness of the proposed approach for solution of the supervised classification problem is illustrated on model and real-life data.

A lot of projects on ontologies focus on describing some aspect of reality: objects, relations, states of affairs, events, and processes in the world. Another approach is using ontologies for problem-solving. In this paper we discuss an approach for designing NLP tasks based on a multilevel system of ontological models. We developed a system of ontological models which is used for ontology-driven computational processing of unstructured texts. The components of the system are the ontology of task designing, the ontology of applied models, and the domain ontology. We discuss the general schema of designing solutions of applied tasks and some applications.

The paper presents a text mining approach to identifying technological trajectories. The main problem addressed is the selection of documents related to a particular technology. These documents are needed to identify a trajectory of the technology. Two different methods were compared (based on word2vec and lexical-morphological and syntactic search). The aim of developed approach is to retrieve more information about a given technology and about technologies that could affect its development. We present the results of experiments on a dataset containing over 4.4 million of documents as a part of USPTO patent database. Self-driving car technology was chosen as an example. The result of the research shows that the developed methods are useful for automated information retrieval as the first stage of the analysis and identification of technological trajectories.

The paper presents the results of a preliminary analysis of the argumentation indicators observed in the corpus of popular science texts in Russian. Main pragmatic aspects of the argumentation signaled by discursive indicators are outlined. The classification of indicators takes into account pragmatic meaning and the type of language means used. Special attention is paid to insufficiently studied indicator constructions and classes of their core content words. We consider constructions with verbs and nouns of mental state, speech, inference, and mental impact. The process of creating a lexicon of argumentation indicators is described. Indicators are presented in the form of lexical units and lexical-grammatical patterns, which are automatically generated from annotated text fragments and can be manually corrected by the expert. The pattern description language allows to represent grammatical and semantic constraints, nested constructs, alternatives, and discontinuity. The lexicon of indicators will be used for automatic annotation of argument indicators in unannotated text, as well as for experiments in argument mining.

According to the CHAOS report from Standish Group during 1992–2017, the degree of success of projects in the development of software intensive systems (Software Intensive Systems, SIS) has changed insignificantly, remaining at the level of 50% inconsistency with the initial requirements (finance, time and functionality) for medium-sized projects. The annual financial losses in the world due to the total failures are of the order of hundreds of billion dollars. The majority of information about software projects has textual representation. Analysis of this information is vital for project status understanding, revealing problems on the early stage. Nowadays the majority of tasks in NLP field are solved by means of neural network language models. These models already have shown state-of-the-art results in classification, translation, named entity recognition, and so on. Pre-trained models are accessible in the internet, but the real life problem domain could differ from the origin domain where the network was learned. In this paper an approach to vocabulary expansion for neural network language model by means of hierarchical clustering is presented. This technique allows one to adopt pre-trained language model to a different domain.

Early detection of mental disorders risk is an important task for modern society. A large set of clinical works showed that five-factor personality traits model (Big Five) can predict mental disorders. In this paper, we consider the problem of automatic detection of personality traits from user profiles of Russian social network VKontakte. We describe the preparation of user profiles dataset, propose several features sets and evaluate machine learning methods for predicting personality traits. The results of experiments show that different features set demonstrate promising results on the task of a personality prediction.

Rhetorical structure theory defines the relations between predicates and larger discourse units, but it does not consider the extralinguistic nature of text-writing at all. However, the text-writing process is totally related to the particular targeted activity. This paper presents a new approach that does not model a text as a result of a researcher’s cognitive activity embodied in it, but it models cognitive activity reflected in the scientific text. We also propose and evaluate a framework for detection of text fragments, which is related to cognitive operations in scientific texts. The obtained results confirm the usefulness of the suggested set of cognitive operations for the analysis of scientific texts. Moreover, these results justify the applicability of the proposed framework to cognitive operation extraction from scientific texts in Russian.

The paper considers a novel intelligent approach to the design of fuzzy systems based on Multi-Objective Evolutionary Fuzzy Systems (MOEFSs) theory. The presented approach is based on the principle of Pareto optimality using evidence combination schemes of Dempster-Shafer. The evidence combination scheme is used in evolutionary operators of search algorithms during implementation of fitness assignment and solution selection. The paper proposes new representation forms for integral and vector criteria reflecting not only accuracy and complexity of multi-objective fuzzy systems, but also their interpretability characterizing readability of fuzzy-rule base and semantic consistency. The main advantage of the considered MOEFSs is that they satisfy to many criteria simultaneously, which include interpretability properties of fuzzy systems, such as compact description, readability, semantic consistency and description completeness. The novel technique of solution selection and combination based on fusion of fitness estimations from several individuals using Dempster-Shafer theory is proposed. Here, Dempster-Shafer theory allows to select those solutions from Pareto-optimal ones, which are most satisfactory in multi-objective design terms. Solution selection and combination based on probability theory of evidence combination increase objectivity of the best solution selection in evolutionary algorithms. The novel techniques of fitness ranging in evolutionary algorithms and expert preferences integration into MOEFSs design based on Dempster-Shafer modified network models are proposed. The comparison results of MOEFSs design using several evolutionary algorithms are shown in the example of railway task decision. These results prove that the proposed evolutionary design provides a better compromise between accuracy and interpretability in comparison with conventional algorithms.

The paper is devoted to the application of the method of randomized general indices in assessing potential damage to a company if confidential information is leaked due to malefactor’s social engineering attack. This assessment is used in a comparative analysis of the effectiveness of various measures aimed at increasing the level of user protection from the malefactor’s social engineering attacks.

The paper presents a solution to the multiclass classification problem based on the Convolutional Fuzzy Neural Networks. The proposed model includes a fuzzy self-organization layer for data clustering (in addition to convolutional, pooling and fully-connected layers). The model combines the power of convolutional neural networks and fuzzy logic, it is capable of handling uncertainty and imprecision in the input pattern representation. The Convolutional Fuzzy Neural Networks could provide better accuracy in multiclass classification tasks when classified objects are often characterized by uncertainty and inaccuracy in their representation.

The task of model learning arises in algebraic Bayesian networks as one of the probabilistic graphical models. Several approaches to machine learning of algebraic Bayesian networks are known. This research is dedicated to the algorithm of machine learning of algebraic Bayesian network represented by a knowledge pattern on missing data. Besides this algorithm, some examples of machine learning on artificial and real data from social media are considered.

The paper presents the inference methods for fuzzy systems of Mamdani type, which, for many fuzzy inputs and for any t-norms, are implemented with polynomial computational complexity. Center of sums and the discrete version of the center of gravity methods are applied for the rule base at the defuzzification stage. The network structures of the systems corresponding to these methods are given.

The paper is intended to discover a chaotic behavior in collective of a number of interacting dynamic systems with the goal oriented learning. Previously it was shown that these semi-independent systems are able to achieve their individual goals iff the vector of these goals belongs to a certain area \( \Uplambda^{(n)} \). We call this behavior pattern as collective automata behavior that maybe successful or not. Past research showed a similar property in collective of a pair of finite state automata that also have a similar restriction for successful behavior.

Computer experiments performed in this paper for case \( n = 2 \) have demonstrated that some chaos is present both outside of the area \( \Uplambda^{(n)} \), and within it. Analogous chaotic phenomena were not mentioned before in collective behavior of intelligent agents. However, these phenomena are undoubtedly important for practical applications where goals for intelligent dynamic systems either established from outside, or chosen by systems independently. They may produce some undesirable, even harmful global effects in implemented technical devices.

The paper presents an approach and a methodology that were used to collect and store training samples. Preliminary processing in the decision support system for diagnosing complex vision pathologies is described. A training sample consists of various records of biopotentials obtained from a special medical device.

The article describes an example of data consolidation between two relational databases (RDB). The proposed approach involves using of ontological engineering methods for extracting ontologies from RDB data models. The merging of the resulting ontologies is used to organize the information interaction between the RDB. The difference between the traditional and the proposed data consolidation algorithms is shown, their advantages and disadvantages are considered.

The paper describes the technology of forming a list of personalized preventive recommendations. The technology consists of the following main components: human health state, data acquisition module, database, knowledge base, and solver with output explanation. In the version presented, this technology allows one to assess the risks of stroke, myocardial infarction and depression, contains more than two hundred risk factors for these diseases and more than twenty preventive recommendations. Training for this version was based on automated analysis of a large number of publications and expert knowledge.

The scientific and technological experience of the development and use of tutoring integrated expert systems and the creation of a single ontological space of knowledge and skills for the automated construction of competence-oriented models of specialists in the field of methods and technologies of artificial intelligence in the direction of training “Software engineering” obtained at the Department of Cybernetics of NRNU MEPhI is analyzed.

We consider trajectory-tracking problem for an unmanned aerial vehicle (UAV) based on optimal choice of knots of the interpolating spline. As examples, we use typical second-order curves: ellipses, parabolas, hyperbolas, obtained by cutting a cone with planes. The rules are proposed for rational placement of a given number of knots for curves given in parametric form. The use of spline interpolation methods opens the way to developing mathematical tools for tracking complex trajectories, storing geometrical information in a compact form and reproducing trajectories with a predetermined accuracy on a general basis. The research is focused on parametric cubic Hermite spline and Bezier curves, which are characterized by simplicity of computational implementation. We have conducted experimental studies to search for the optimal allocation of knots. The problem of moving along the route represented by a parabola has been investigated under wind loads taking into account the mathematical model of the aerial vehicle. We consider an approach to dynamic motion planning based on strategies and rules that imitate actions of a pilot when rapid actions are needed.

Analysis and visualization of large volumes of semi-structured information (Big Data) in decision-making support is an important and urgent problem of the digital economy. This article is devoted to solving this problem in the field of digital marketing, e.g. distributing outlets and service centers in the city. We propose a technology of adaptive formation of spatial segments of an urbanized territory based on the analysis of supply and demand areas and their visualization on an electronic map. The proposed approach to matching supply and demand includes 3 stages: semantic-statistical analysis, which allows building dependencies between objects generating demand, automated search for a balance between supply and demand, and visualization of solution options. An original concept of data organization using multiple layer including digital map, semantic web (knowledge base) and overlay network was developed on the basis of the introduced spatial clustering model. The proposed technology, being implemented by an intelligent software solution of a situational center for automated decision-making support, can be used to solve problems of optimization of networks of medical institutions, retail and cultural centers, and social services. Some examples given in this paper illustrate possible benefits of its practical use.

The paper presents an approach to modeling and analyzing the argumentation found in popular science literature. Argumentation is modeled using an argumentation ontology based on the AIF base ontology expanded by the means for modeling the target audience and allowing for a more detailed description of the arguments’ content. In terms of this ontology, the authors give the descriptions of argumentation schemes, arguments’ structure and elements, as well as of the network of arguments and their constituent parts extracted from the texts under study. To analyze argumentation, a software system is being developed. It provides tools for modeling and identifying the structure of argumentation on a corpus of texts relating to popular science discourse. In addition, the system can examine the extracted argumentation with the purpose of revealing and analyzing argumentative strategies and rhetorical methods used in scientific and popular science texts. The paper describes the specific features of the proposed argumentation ontology and presents the architecture and functionality of the software system designed for argumentation analysis.

The paper describes an approach to providing intelligent support for solving compute-intensive problems on supercomputers, based on the ontologies and decision rules helping the users choose a computational method suited best to their task and supercomputer architecture. The authors focus on the concept and following components of intelligent support: the ontology of the problem area “Solving Compute-intensive Problems of Mathematical Physics on Supercomputers”, an information-analytical Internet resource based on this ontology, providing the users access to information necessary for solving compute-intensive problems on supercomputers and an expert system helping users develop a parallel code based on ready-made software components. The paper describes in detail the conceptual scheme of intelligent support and the ontologies developed. To avoid possible errors in designing an ontology, in model means for knowledge representation absent in the ontology description language, to systematize and facilitate populating an ontology with concept instances, we have developed and applied a number of structural and content patterns.