Tag Archives: machine learning


UDC 004.85:004.056.57

O.N. Vybornova, I.A. Pidchenko

The continuous growth in the number of malicious programs makes the task of their detection urgent: classifying programs into malicious and safe. In this regard, this study is devoted to the development of a malware detection system based on machine learning, namely, training an artificial neural network with a teacher. In the course of the study, we analyzed the structure of Portable Executable files of the Windows operating system, selected characteristics from PE-files to form a training set, and also selected and substantiated the topology (four-level perceptron) and parameters of the antivirus neural network. The Keras library was used to create and train the model. The Ember dataset of safe and malicious software was used to form the training set. We have trained and verified the adequacy of training for the developed malicious code recognition model. The training results of the anti-virus neural network proposed in the study showed a high accuracy of malware detection and the absence of the overtraining effect, which indicates good prospects for using the model. Although the experimental model of a neural network is not able to fully replace the anti-virus scanners, the materials of the article are of practical value for the tasks of classifying programs into malicious and safe.

Keywords: malware, machine learning, anti-virus neural network, neural network training, Keras, Ember, Dropout.

Full text:


UDC 519.683, 519-7

S.L. Sinotova, O.V. Limanovskaya, A.N. Plaksina, V.A. Makutina

Determination of the range of factors affecting the object of research is the most important task of medical research. Its solution is complicated by a large amount of diverse data, including extensive anamnestic information and data from clinical studies, often combined with a limited number of observed patients. This work is devoted to the comparison of the results obtained by various feature selection methods for the search for a set of predictors, on the basis of which a model with the best forecast quality was created, for solving the problem of binary classification of predicting the onset of pregnancy during in vitro fertilization (IVF). The data from the anamnesis of women, presented in binary form, were used as features. The sample consisted of 68 features and 689 objects. The signs were examined for the presence of cross-correlation, after which methods and algorithms were applied to search for a selection of significant factors: nonparametric criteria, interval estimate of the shares, Z-criterion for the difference of two shares, mutual information, RFECV, ADD-DELL, Relief algorithms, algorithms based on the permutation importance (Boruta, Permutation Importance, PIMP), feature selection algorithms using model feature importance (lasso, random forest). To compare the quality of the selected sets of features, various classifiers were built, their metric AUC and the complexity of the model were calculated. All models have high prediction quality (AUC above 95%). The best three of them are based on features selected using nonparametric criteria, model selection (lasso regression), Boruta, Permutation Importance, RFECV and ReliefF algorithms. The optimal set of predictors is a set of 30 binary features obtained by the Boruta algorithm, due to the lower complexity of the model with a relatively high quality (AUC of the model 0.983). Significant signs includes: data about pregnancies in the anamnesis in general, ectopic and regressive pregnancies, independent and term childbirth, abortions up to 12 weeks; hypertension, ischemia, stroke, thrombosis, ulcers, obesity, diabetes mellitus in the immediate family; currently undergoing hormonal treatment not associated with the IVF procedure; allergies; harmful professional factors; normal duration and stability of the menstrual cycle without taking medication; hysteroscopy, laparoscopy and laparotomy; resection of any organ in the genitourinary system; is it the first IVF, the presence of any surgical interventions, diseases of the genitourinary system; the age and BMI of the patient; absence of chronic diseases; the presence of diffuse fibrocystic mastopathy, hypothyroidism.

Keywords: feature selection, binary classification problem, small data analysis, machine learning, assisted reproductive technologies.

Full text:


UDC 004.89

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:


UDC 004.048

A.K. Suleymanov, M.A. Sharipova, O.N. Smetanina, Y.Y. Sazonova, K.V. Mironov

The research works on automatic opinion extraction are still relevant. The article presents a formal description of the term opinion, setting tasks depending on the determined properties of opinion. The problems of solving the tasks of sentiment analysis, approaches to its solution and ready-made software implementations are described. Available corpora of texts in the Bashkir language are presented, and also task statement for sentiment analysis in the Bashkir language. Presented solution, which include an algorithm for tagging the texts, a preprocessing algorithm, a choice of classification features, and classification algorithms. Also, the results of computational experiment, which aimed to define the most effective classifier based on quality metric, are present. The results in this work and the developed software solution based on SVM with stochastic gradient descent, which demonstrated the highest indicators in the criteria of accuracy, completeness, and 𝐹-measure, can be used to sentiment analysis of news sites in the Bashkir language.

Keywords: sentiment analysis, computational linguistics, machine learning, classification features, hybrid intelligent system, support vector machine, random forest.

Full text:


UDC 004.048

M.A. Sazonov, S.V. Shekshuev

In article discusses the relevance of solving problems class publication activity analysis for users of social networks. An analysis of existing approaches identifying public opinion about publications in social networks is given, in which the prevalence is substantiated of methods based on the analysis of the texts sentiment. The disadvantages of these methods are given, which reduce the process of assessing public opinion regarding the publication activity of users of social networks efficiency. It is suggested that it is possible to use message metadata without the need a texts sentiment analysis procedure to eliminate this problem. The primary and derived indicators of messages in social networks are determined, obtained from the set of metadata. Approaches to solving the problem of binary classification based on the indicated markers, both based on statistical methods and using machine learning methods, are considered. An assumption is made about the acceptable accuracy of a class of models based on machine learning that provide a solution to the specified problem. A machine learning model based on a random forest is proposed for solving the problem of classifying a positive attitude towards publications in social networks, based on the analysis of primary and derived indicators of messages.

Keywords: social network, data, social networks publications indicators, machine learning, random forest.

Full text:


UDC 004.942

V.P. Okhapkin, E.P. Okhapkina, A.O. Iskhakova, A.Y. Iskhakov

The article discusses the problem of the destructive information influence in social networks revealing. It is noted that the tasks that are associated with the rapid detection of destructive information influence are prerequisites for the development and improvement of methods and means for identifying such influences in social networks. To understand the social dynamics of social networks groups we consider: the communication model proposed by Theodore Newcomb, Kurt Levin’s “planar map”, and Fritz Haider’s theory of cognitive balance. UN documents on the counteraction of the use of the Internet for the extremist purposes and radicalization were analyzed. The role of the cognitive approach to the analysis of social network messages and the main scenarios implemented by influence actors in texts aimed at different audiences are considered. The study presents a systematic approach to the task of designing a multi-agent platform. Special attention is paid to the block of pattern analysis of user’s messages in social networks both from the position of mathematical modeling and social dynamics. The article describes the architecture and methods of the multi-agent system for the destructive information and humanitarian impact detection. The system consists of the administration interface, subsystems for the multi-agent system administration and agents management, clustering agents, network messages analysis and dispersion analysis. The description of the main blocks of agents and subsystems is given.

Keywords:multi-agent technologies, cluster analysis, information security, aggression, radicalization, machine learning, personality, information and psychological impact destructive informational impact, socio-cyberphysical system.

Full text:


UDC 658.512.26

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:


UDC 004.4

A.M. Bershadsky, A.S. Bozhday, Y.I. Evseeva, A.A. Gudkov

The article discusses development and application issues of software self-adaptation method based on machine learning technology. The differences between the Model-Based and Model-Free approaches in reinforcement learning are considered, the choice of the Model-Based approach for creating a software self-adaptation method is substantiated. The definition of an expanded Markov decision-making process that takes into account the role of the situation in the course of program self-adaptation is considered. A mathematical model of the state space of the software system is proposed, based on the hypergraphic formalization of the model of characteristics. Based on the expanded definition of the Markov decision-making process, the proposed model of the state space of the system, and the concept of the Model-Based approach to machine learning with reinforcement, a new method of software self-adaptation was developed that takes into account the effect of the actions performed by the system on the state of the environment. A practical example of using the method is given.

Keywords: multimodal data, repository, research organization, distributed platform, interdisciplinary research.

Full text:


UDC 004.032.26

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:


UDC 612.743, 612.817.2
doi: 10.26102/2310-6018/2019.24.1.017

R.Y. Budko, N.N. Chernov, N.A. Budko, A.Y. Budko

The relevance of this study is due to one of the main problems existing today in the field of building man-machine interfaces – is the creation of an effective management system that interacts directly with the user and external devices replacing functions (prostheses, wheelchairs, etc.). In this regard, this work is devoted to the study of the possibility of using physiological gestures from the daily life of a person to control the prosthesis with the safety of the forearm for at least one third. The leading approach to the study of this problem is the use of methods of statistical processing of experimental data, digital signal processing, machine learning algorithms and pattern recognition. This approach allows a comprehensive study of the electromyogram (EMG) of the forearm when making voluntary movements at different levels of the implementation of the myo-control system. The article presents the results of the EMG study recorded for 11 arbitrary movements from a group of subjects, describes the procedure for pre-processing the EMG and identifying characteristic features for signal recognition, discloses a method for classifying movements using an artificial neural network based on radial basic functions (RBF). Eight of the most suitable for classification movements were identified and ranked according to the classification accuracy: relaxation (like zero movement), hand opening, fist, hand flexion, hand supination, hand extension, hand pronation, pinch. The materials of the article are of practical value for building systems based on the human-machine interface, as well as for classification tasks in electrophysiology applications.

Keywords: :electromyogram, prosthesis, biocontrol, human-machine interface, machine learning, artificial neural networks.

Full text: