Tag Archives: neural network

LOW RANK APPROXIMATIONS FOR NEURAL NETWORKS


UDC 004.89
DOI:10.26102/2310-6018/2020.30.3.018

N.V. Shaposhnikova

Today, artificial neural networks (hereinafter ANN) and deep learning have become almost indispensable in applications related to the tasks of machine vision, machine translation, speech to text conversion, text rubrication, video processing, etc. However, despite the presence of a number of classical theorems substantiating the approximating capabilities of neural network structures, the current successes in the field of ANNs in most cases are associated with the heuristic construction of the network architecture applicable only for the specific problem under consideration. On the other hand, deep ANNs have millions of parameters and require powerful computing devices for their functioning, which limits the possibilities of their application, for example, on mobile devices. Significant progress in solving these problems can be obtained using modern powerful algorithms of low-rank approximations for the parameters of the ANN layers, which will both simplify the process of developing a neural network architecture and will lead to significant compression and acceleration of the training of deep ANNs. Considering, for example, the core of the convolutional ANN as a four-dimensional array (tensor), we can construct a low-rank approximation for it with the effective implementation of its convolution with the vector (direct signal propagation in the network when generating the prediction) and differentiation with respect to the parameters (back signal propagation in the network when training). In this paper, we will consider the modern paradigm of machine learning and low-rank tensor approximations, and we will demonstrate the prospects for the tensorization of deep ANNs using a specific model numerical example corresponding to the task of automatic recognition of handwritten digits.

Keywords: machine learning, neural network, deep convolutional network, low rank approximation.

Full text:
Shaposhnikova_3_20_1.pdf

PIECEWISE NEURAL MODEL BASED ON SPLIT SIGNALS FOR BERNOULLI MEMRISTORS


UDC 519.65; 621.3.01
DOI:10.26102/2310-6018/2020.29.2.016

E.B. Solovyeva, H.A. Harchuk

Actuality of the investigation theme is specified by complexity of mathematical modeling of nonlinear dynamic devices, since the analytical solutions of the nonlinear differential equation systems of high size are not always obtained, and numerical solutions are often accompanied by the problem of poor conditionality. In this situation, behavioral modeling is effective, herewith the object of investigation is represented as a “black or gray box”, and its mathematical model is constructed using the sets of the input and output signals. Behavioral modeling is important in conditions of restricted information of new elements and technologies, as well as under the complexity and variety of models built at the component level. The behavioral modeling of memristive devices actively developed using nanotechnology for energy-saving equipment is represented. A method of behavioral modeling of the transfer characteristics of memristive devices by means of piecewise neural models based on split signals is proposed. To reduce the dimension on approximating nonlinear operators and, therefore, to simplify mathematical models, are applied the following: neural networks, the signal splitting method that enables to adapt the model to the type of the input signals, and a piecewise approximation method for operators of nonlinear dynamic systems. On the basis of the proposed method, a piecewise neural model is constructed. This model includes five three-layer neural networks of simple structure (3x2x1, 100 parameters) and provides a significantly higher accuracy of modeling the transfer characteristic of memristors, the current dynamics of which are described by the Bernoulli differential equation, in comparison with the two-layer piecewise neural and piecewise polynomial models. The described results are of practical value for the behavioral modeling of memristors and various memristive devices, as well as of other nonlinear dynamic systems, since they develop a universal approach for approximating nonlinear operators based on neural networks.

Keywords:nonlinear dynamic system, mathematical modeling, nonlinear operator, nonlinear model, approximation, neural network, memristor.

Full text:
SolovyevaHarchuk_2_20_1.pdf

INTELLIGENT ANALYSIS OF VIDEO DATA IN SYSTEM FOR MONITORING COMPLIANCE WITH INDUSTRIAL SAFETY RULES


UDC 004.932
DOI:10.26102/2310-6018/2020.29.2.010

A.M. Vulfin

The use of intelligent cameras and sensors, in combination with the human operator in video analytics systems, from which most of the analytical and visual load has been removed, allows you to increase the efficiency of video surveillance and, as a result, increase the safety and productivity of work in production as a whole. Analysis of the existing data processing methods in the video surveillance systems of industrial facility showed that the use of a non-contact method for analyzing person’s posture and actions in the camera’s field of vision is rare, but it can be critical in certain situations (person in overalls is in the camera’s field of view, but the system is on him does not respond, because he is not in the forbidden zone). The improvement of algorithms for the intellectual analysis of video data in the system for monitoring compliance with industrial safety rules (analysis of the type of dynamics and control “friend or foe”) using neural network processing technologies is considered. Effectiveness evaluation of algorithms for analyzing full-scale video data software implementation showed the correctness of classification in 97% of cases. Effectiveness evaluation of the 5 subjects into two classes of “own” and “alien” classification was carried out by cross-validation and showed an accuracy of 99% on the test sample.

Keywords:video analytics, intelligent analysis, dynamics type recognition, neural network, classifier, pose determination.

Full text:
Vulfin_2_20_1.pdf

STRUCTURIZATION OF ENVIRONMENTAL INFORMATION WITH APPLICATION OF GEOINFORMATION TECHNOLOGIES


UDC 658.512.26
DOI:10.26102/2310-6018/2019.27.4.044

K.A. Fedutinov


The article discusses the development of managerial decisions to improve the environment through the introduction of geographic information technologies, including methods for assessing and predicting the environmental situation based on monitoring approaches. The development of big data processing technologies has identified trends in the widespread implementation of real-time monitoring systems. In this regard, the task of monitoring natural objects is proposed to be solved as the task of determining and controlling the properties and states of a complex object in real time and actively interacting with the environment, as well as developing managerial decisions and recommendations. It is proposed to use the Fuzzy ART neural network as a mathematical apparatus for structuring environmental information, which has proven itself in real-time data processing. To visualize the received information and integrate the results of the network operation of the Fuzzy ART network into a geographic information system, it is proposed to use the Folium Python library, which is intended for graphical display of geographic data and contains all the necessary cartographic information. Using Folium, the results of the structuring of environmental data can be displayed directly on Google maps, which makes it possible to visually determine the boundaries of clusters and possible buffer zones when the map is scaled up.

Keywords: neural network, clustering, machine learning, adaptive resonance theory, Fuzzy ART network, GIS system.

Full text:
Fedutinov_4_19_1.pdf

CONSTRUCTION OF DECISION RULES USING THE ARTMAP NEURAL NETWORK

UDC 004.032.26
DOI:10.26102/2310-6018/2019.26.3.029

I.L. Kashirina, K.A. Fedutinov


This article discusses the ARTMAP neural network architecture, compatible with a symbolic representation based on IF-THEN rules. In particular, the knowledge gained during the training of the ARTMAP network can be transformed into a compact set of decision rules for classifying the source data, which can be analyzed by domain experts, by analogy with interpreted machine learning methods, such as decision trees or linear regression. Similarly, knowledge in the a priori area presented in the form of IF-THEN rules can be transformed into the ARTMAP neural network architecture. The presence of a preliminary set of rules used in the initialization of the network increases the accuracy of classification and the effectiveness of training. The original set of rules can be supplemented using the learning algorithm ARTMAP. Each rule formed in the process of learning a network has a confidence factor that can be interpreted as its importance or usefulness. The architecture, training algorithms and functioning of the ARTMAP network for the extraction of rules are described in terms of the previously proposed generalized model of networks of the ART family proposed by the authors.

Keywords: neural network, machine learning, adaptive resonance theory, ARTMAP, rule extraction.

Full text:
KashirinaFedutinov_3_19_1.pdf

METHOD OF DIFFERENTIAL DIAGNOSTICS OF THE NOSOLOGICAL FORM OF VIRAL HEPATITIS WITH THE APPLICATION OF NEURAL NETWORK OF CASCADE CORRELATION

UDC 004.891.3
DOI:10.26102/2310-6018/2019.26.3.028

A.N. Astafev


An important aspect of determining the nosological form of hepatitis is the combination of input data at the beginning of the study. The use of neural networks in medicine, which have the ability to search for hidden dependencies by learning from the experience of doctors, makes it easier to work in the role of advisor. However, the question of selecting the most effective topology for a specific task remains open. This paper substantiates the need to use neural network algorithms to solve the problem of determining the nosological form of hepatitis. The analysis and selection of input factors characterizing the clinical condition of the patient, and output factors characterizing the specific nosological form of hepatitis, neural network. The algorithm, its use is described, and a cascade neural network is compared with others in the context of the problem under consideration. At the end, a description is made of the established system for determining the nosological form of hepatitis using a cascade correlation neural network, and also describes the clinical efficacy.

Keywords: neural network, viral hepatitis, nosological form of hepatitis, neural network of cascade correlation, classification.

Full text:
Astafev_3_19_1.pdf

NEURAL NETWORK MODELING OF THE INTERACTION OF LABOR MARKET SUBJECTS AND EDUCATIONAL SERVICES

UDC 004.032.26

T.V. Azarnova, I.L. Kashirina, A.N. Schwindt


The economy of modern Russia is characterized by a number of problems: unemployment and the unemployed population, new requirements on the part of employers for vocational education, the discrepancy between the employers ’personnel needs and the professional capabilities of university graduates. All this is a consequence of the disagreement of the most important areas of modern society – the labor market and education. The increasing complexity of tasks requiring solutions in practice leads to an increase in employers’ requirements for the level of training of graduates, which underlies the existing imbalance in the labor market according to qualitative criteria. The article presents the results of neural network modeling of the interaction of subjects of labor markets and educational services. It is shown that to assess the quality of training of specialists according to the criterion of meeting the needs of the regional labor market, indicators of the efficiency of higher education institutions can be used. The rationale for the feasibility of using graduate employment indicators for solving the problems of evaluating and analyzing the effectiveness of higher educational institutions is given. Neural network models of classification, clustering and regression are built for a comprehensive analysis of the relationship between the subjects of the labor market and educational services. Revealed the presence of a strong relationship between the performance indicators of the university and the average salary of graduates in the first year after graduation.

Keywords: : labor market, monitoring, evaluation, university efficiency, neural network.

Full text:
AzarnovaSoatori_4_18_1.pdf

PREDICTION OF EYE-DIAGRAM PARAMETERS FROM TRANSIENT AND GAIN-FREQUENCY CHARACTERISTICS USING NEURAL NETWORK

UDC 621.396

A.V. Smirnov


A capability of prediction of the eye-diagram width and height with using artificial neural network (ANN) was investigated. For this purpose, were simulated more than 750 examples of telecommunication channels with different transfer functions. Eye-diagrams were composed for all examples by means of convolution of random pulse sequence and pulse response and parameters of these eye-diagrams were measured. Some ANN was learned. Their input variables were transient characteristic delay time, raise time, magnitude of voltage peak and oscillation duration as well as a gain value at the half of clock rate. For each of predicted parameters distinct ANN was chosen for different ranges of input variables. Root mean square errors of eye-diagram parameters prediction using these ANN were in the range of 2 – 4%. Correlation coefficient of predicted and known values was more then 0,98. Sufficient decreasing of computational time is achieved compare with estimation of the eye width and height using eye-diagram modeling. This method can be used for optimization of communication channel characteristics when eye-diagram parameters are the components of the goal function.

Keywords: eye-diagram, transient characteristic, gain-frequency characteristic, neural network, approximation.

Full text:
Smirnov_3_18_1.pdf

APPLICATION OF FUZZY ARTMAP NETWORK IN INTELLIGENT SYSTEMS OF INVASION DETECTION

UDC 004.032.26

I.L.Kashirina, K.A.Fedutinov


The article deals with the organization of intelligent intrusion detection and detection systems. Research in the field of development of information security tools shows that today the most promising and flexible solutions are based on machine learning methods that can prevent damage from intrusions that were not noticed by standard means of combating computer attacks. In the proposed approach, it is proposed to use a sequential reverse search with a return to select significant features and the Fuzzy neural network ARTMAP to detect and diagnose attacks. Network Fuzzy АRTMAP is able to adapt to the dynamics of computer attacks and allows you to recognize intrusions in the information system in real time, without the need to load datasets in batches. This makes it possible to automate the analysis of safety protocols in a continuous mode. The extensive use of ART family networks in intrusion detection tasks makes it possible to consider the search for approaches that improve their performance. In this paper, the control hyperparameters network Fuzzy ARTMAP proposed to adjust automatically with the use of a genetic algorithm According to the results of the computational experiment, the reduced set of characteristics reduces the computation time by 21%. The accuracy of the classification algorithm was 100% and 99.89% for the detection stage and the diagnostic stage, respectively.

Keywords: neural network, Fuzzy ARTMAP, genetic algorithm, intrusion detection, intelligent information security systems.

Full text:
KashirinaFedutinov_3_18_1.pdf

APPLICATION OF A NEURO NETWORK APPROACH TO LOGICAL DATA PROBLEMS AND BUILDING INTELLIGENT DECISION-MAKING SYSTEMS

UDC 519.71

D.P. Dimitrichenko, R.A. Zhilov


The need to reduce the dimensionality of large data sets while maintaining the logical structure, as well as the detection of hidden patterns and the removal of information noise and redundancy in the description of diagnostic (recognition) objects leads to the need to construct an effective method for classifying objects in weakly formalized areas of knowledge. Logical functions that describe objects using variable-valued predicates allow us to reveal hidden regularities and eliminate redundancy in the description of objects. Ordered by means of variable-valued logical functions, object classes are the basis for the formation of the structure of cognitive maps. The purpose of this study is to create an algorithm for constructing a logical neural network based on the variable-valued logic function and justifying the possibility of applying the results obtained in the construction of cognitive maps. The theoretical possibility and algorithms allowing to make the transition from variable-valued logic functions to cognitive maps using the neural network approach are grounded. The result of this work is the procedure for constructing a cognitive map using logical neural networks built on the basis of variable-valued logical functions. The advantage of the obtained cognitive map is the possibility of functioning within the framework of fuzzy logic.

Keywords: predicate, predicate significance, variable-valued logical function, logical neural network, cognitive map, cluster analysis, neural network.

Full text:
DmitrichenkoZhilov_2_18_1.pdf