Software & Systems

Articles of journal № 2 at 2018 year.

Order result by: Public date | Title | Authors |

11. Modern approaches to training intelligent agents in the Atari environment [№2 за 2018 год]
Authors: D.A. Korobov, S.A. Belyaev
Visitors: 8706
The article describes the research and development of machine learning algorithms for the problem of training intelligent agents in the Atari environment. The environment is the game console Atari2600 emulation on OpenAI Gym platform. The main aims of the research are to describe and compare modern learning algorithms with reinforcement in a number of cases, to expose their advantages and drawbacks and to give suggestions in order to increase their effectiveness. The authors describe a mathematical model of the reinforcement learning task in terms of Markov Decision Process and form a system optimality criterion. In addition, they note the Atari environment properties. There is a description of an operation principle of existing solutions, as well as the tools used in them. The paper considers a basic Q-learning approach and its well-known modification that uses a neural network. It also presents the idea of the algorithm based on policy gradient. The paper gives a description of the approach of “actor-critic” algorithms and considers its asynchronous multithreaded version. All the described algorithms are presented in pseudocode. The authors propose a method of increasing the effectiveness of learning using feature extraction. The developed algorithm for feature vector generation is based on image recognition. The paper describes the steps of feature vector selection and its usage during the learning process. The described algorithms are implemented, and the experiment is conducted. A comparative analysis of the results gave conclusions on the efficiency of the algorithms. Finally, the authors propose some ideas for further increasing the speed and efficiency of training of intelligent agents.

12. The algorithm for recognizing situations in a distributed video surveillance system [№2 за 2018 год]
Authors: A.Yu. Kruchinin, D.V. Kolmykov, R.R. Galimov
Visitors: 6123
CCTV systems are the most important means for preventing and timely handling of contingencies, such as crimes, emergency situations. A big number of cameras and a large control area makes it necessary to introduce video analytics to recognize dangerous situations. In this case it is necessary to take into account the data from a number of video cameras both for detecting a motion path of a recognized object and for increasing recognition reliability. The article proposes an algorithm for recognizing emergencies for a distributed video surveillance system based on stochastic grammars. Recognition of a situation occurs at 3 levels: images are recognized at the lower level, events are recognized at the average level, and situations are recognized at the top. To reduce the system response time, it is proposed to use a multi-agent architecture that allows distributing the load between intelligent cameras. Data exchange occurs only between nearby nodes, so network traffic reduces. The use of a large number of cameras involves zones controlled by several nodes. Combination of detection results from several cameras makes it possible to increase the estimate reliability. However, it is required to know the mutual arrangement of chambers and the angles of their turns. The article suggests some methods for automatic calibration of cameras in a distributed video surveillance system, ways of combining images from different cameras, in particular, based on speed rate vectors of objects. Taking into account certain features of a distributed video surveillance system, there is a developed algorithm for recognizing emergencies for an intelligent surveillance camera. Each camera generates probable situations based on previously recognized events. When a threshold value of probabilistic evaluation of the detection result is exceeded, its refinement is carried out in the process of interaction with neighboring nodes.

13. An enterprise management system [№2 за 2018 год]
Author: Yu.M. Lisetskiy
Visitors: 33778
The article considers an enterprise as an open economic and social system that acts in terms of situational approach. This supposes changes of enterprise management forms, methods, systems, style depending on the objective conditions of the environment, as well as accelerated convergence of management models. It is demonstrated that management is a purposeful and continuous process of management subject influence on a controlled object, which is aimed at changing a state according to a defined plan and which is an essential function of any system. There are basic principles (cybernetic laws) of building management systems that include variety, differences of general from particular, external addition, feedback, anti-entropy, as well as cybernetic and functional models of a management system. The paper considers an enterprise management system as a form of real implementation of managerial interconnections and its composition including the subsystem of management methodology, process, structure and technique.

14. An automation technology and means for regional security control process simulation [№2 за 2018 год]
Authors: A.V. Masloboev, V.A. Putilov
Visitors: 8189
The general line of research is related to development of information technologies and computer simulation tools for information and analytical support of regional socio-economic systems security management. The research is carried out within implementation of the development strategy of the Russian Federation Arctic zone and national security efforts until 2020 in the Murmansk region. It is focused on engineering high-end information infrastructure for regional development security control problem solving. The paper is focused on development of toolkit to solve the problem of decision-making information support in the field of regional security management. For that purpose a software training simulation system for network-centric control information support of regional security has been developed. The paper considers system architecture and development features of the software package. The package is a multi-agent modeling environment and provides agent-based model automated synthesis and analysis of multi-agent models of network virtual managerial structures for regional security support in crisis situations in socio-economic sphere. System simulation and software toolkit allows formation, analysis and extension of a spectrum of alternative modeling scenario for regional crisis situations. That provides efficient managerial decision making through information support quality and validity enhancement. The program suite consists of autonomous software agents with framework-integrated simulation toolkit and auxiliary software. The system core and components form network-centric multilayer virtual space as an integration framework for problem-oriented coalition-based multi-agent systems for management information support of each domain of regional security. Application of the proposed software system makes it possible to configure modeling environment fast according to specific features of a control problem and provides high variability of computing experiment implementation.

15. Extracting a data schema from the SPARQL endpoint [№2 за 2018 год]
Authors: А.О. Mochalov, Muromtsev D.I.
Visitors: 5618
The paper is devoted to the problem of extracting a data schema from the SPARQL endpoint. The data schema is necessary for writing queries and searching information, as well as for optimizing query execution. The paper considers the existing methods for extracting schemes, notes minuses and pluses of each method. It defines the notion of a data schema to develop own method. A data schema is a dictionary of all triplets, where the predicates are all fixed predicates from the RDFS namespace, as well as triplets that logically follow from the dataset according to RDFS semantics, except for triplets that are elements of well-known dictionaries (RDF, RDFS, Void, OWL, XML Schema, XSD). The elements from the SPARQL endpoint schema will be retrieved using RDFS rules. The work does not use all RDFS semantics rules as not all rules return the elements of the schema according to the definition of the data scheme adopted in this paper. The paper describes the developed method in detail and presents its architecture. It uses the business rules management system Drools to work with inference rules. The paper presents the advantages and disadvantages of the developed method that showed the expected test results. It it noted that the number of classes and properties can be different for the schemes developed and obtained by the developed method. This is only due to the fact that the developed method uses RDF rules.

16. The system for parallelizing the load on computer resources [№2 за 2018 год]
Authors: E.V. Palchevsky, A.R. Khalikov
Visitors: 7987
The article is devoted to parallelization of the load on computer physical resources using an algorithm based on Markov chains. The IDLP hardware module (I distribute the load processes) is designed to parallelize computation processes to physical resources of a computing cluster. The paper shows load dependence of physical resources on the tasks being run on a computing cluster. It also substantiates the expediency of using the IDLP hardware-software module. The created hardware and software module allows efficiently handling transfers of computing processes and prevents system congestion. One of the module elements is a user-friendly interface that allows remote managing IDLP from both a smartphone and a personal computer. The load parallelization system consists of three stages. The first one is algorithm development, the second one is technical implementation, the third one is hardware-software module testing. At the first stage, the authors present the functional of the developed hardware and software module, which includes start, stop, restart, set the kernel load value. The paper presents the scheme of the developed hardware and software module. At the second stage, a source code fragment (in Python) is responsible for uniform distribution (parallelization) of computational processes on physical and logical computer cores, followed by decreasing load on a central processor. At the third stage, the authors test the hardware and software module. The testing confirmed the decrease in the workload of physical server computing resources, and, as a result, it led to increased productivity. The developed solution not only raises productivity by factor of 1.72 (on the equipment used; the load may differ in other cases), but also allows parallel running of complex and resource-intensive computing processes without disrupting computer performance.

17. Software tools of a wavelet/fractal correlation method for detecting space debris objects [№2 за 2018 год]
Authors: Palyukh B.V., I.I. Zykov
Visitors: 5826
The problem of space exploration efforts depends on orbiters, which in turn require their safety precautions from the effects of space debris. Each satellite, space probe or manned mission might be a potential source of space debris. The paper presents a method of detecting various objects of space debris. Images of cosmic space from an optoelectronic device placed on a spacecraft are transformed into halftone images. They are further processed by wavelet transforms. The next stage includes binarization and filling operations for the obtained results. At the end of this stage, there is a formation of the minimum rectangular areas covering the supposed objects of space debris. As a result, these areas have formed statistics in the form of the maximum eigenvalues of autocorrelation matrices and fractal dimensions for making a decision on detection space debris according to the Neyman-Pearson criterion. The obtained results show that space debris objects can be successfully detected.

18. Creating panoramic aerial images from quadcopter [№2 за 2018 год]
Authors: Pakhirka A.I., Zotin A.G., V.V. Buryachenko
Visitors: 3964
The paper considers the approach to forming panoramic images representing highly detailed images of some area. To obtain high-quality panoramic images, there is a need in specialized equipment and, if possible, the maximum survey altitude. These factors can be excluded if stitching the images from cameras of inexpensive unmanned aerial vehicles, i.e. quadcopters. One of the approaches to forming panoramic images is using a set of images or video sequence data obtained when surveying by a quadcopter or other unmanned aerial vehicle. Images are stitched by the methods of matching point features. However, feature points detection algorithms should ensure invariance to any transformations of an image in order to analyze data and generate panoramic images. The research includes applying such algorithms such as FAST, FAST-ER and SURF due to the fact that they are currently the most common solutions for such task. The algorithms for determining correspondences of found feature points are RANSAC and MLESAC. The paper proposes the algorithm of creating a panoramic aerial image from a set of successive landscape images from an unmanned aerial vehicle based on matching key features. The authors also use panoramic image global adjustment through affine processing. A special attention is paid to stitching images using multi-band blending techniques. This provides high-quality visualization in the stitching places of a panoramic aerial image.

19. Continuous execution of system dynamics models on input data stream [№2 за 2018 год]
Authors: I.А. Perl, М.М. Petrova, А.А. Mulyukin, О.V. Kalenova
Visitors: 6301
This article describes a new approach to system dynamics models execution. In most cases when model execution is involved, the input data is a set of static and known data. It is expected that on the model output modeler will get a set of other system or event characteristics computed by the model based on the input parameters. This approach is still used widely in different industries, but it is not the only scenario. With growing popularity of some concepts such as the Internet of Things, the demand on modeling-based solutions, which take continuous data streams as input, has grown significantly. In comparison with stand-alone client-side modeling systems, cloud-based solutions, such as sdCloud, became a reasonable answer to the industry request. Such systems can provide an ability of continuous execution of system dynamics models. In other words, these systems are ready to accept an incoming data stream and perform model execution that will result in streaming modeling results back to the end-user. Running system dynamics models in parallel with the describing process allows performing predictive modeling of the system status in the future. It also allows finding additional hidden external impacts to the model. For example, such approach can be a base for predictive maintenance of complicated technical systems, because it allows computing nearest maintenance time more efficiently.

20. On the issue of assessing stability of functioning of communication network elements [№2 за 2018 год]
Author: Popkov G.V.
Visitors: 5623
The article considers the issues of assessing stability of power grid functioning to destructive destabilizing influences (DDI). The proposed method represents an attacked telecommunication network as dynamic hyper network that allow objective assessing an attacked network state in the context of stability. The author introduces a concept of DDI and the related concept of a DDI channel that affects NE elements of a communication network at different levels of hyper network models. Simplicity and convenience of representing an attacked network in the form of dynamic hyper networks makes it possible to extend the class of problems related to determining stability of communication networks to external destructive influences, in particular, to find correlation links between private intruder models, attack models and vulnerability models of a power grid at the investigated level. Such approach allows designing stable communication networks taking into account dynamically changing external factors connected with threats directed to a communication network structure. It also allows effective detecting and blocking threats related to external information influences while ensuring accessibility, integrity, confidentiality of user information. Due to the proposed approaches, it seems appropriate to create knowledge ontologies based on a network response to DDI at network monitoring points, which in turn will effectively design and install information protection tools in real network structures. The author proposes a private mathematical model of external destabilizing influence, which is based on the probability theory. It makes it possible to trace dynamic changes in a network structure and determine quantitative estimates of QoS applications.

◄ ← Preview | 1 | 2 | 3 | Next → ►