Section 4 - Track 3

On this page, talks in the format of posters are posted. They can be viewed during the whole time of the Conference. To send your question to the authors, please follow to Miro and write your question in the comments. You can also use the feedback form on our website. Authors will answer questions using Miro comments or via email. To get manual go to Instructions (eng / rus).

Vadim Pechenin, Michael Bolotov, Nikolay Ruzanov and Ekaterina Pechenina
Creation neural network models for solving the problems of predicting the products geometric accuracy
The article discusses the problems of creating a tool for operational forecasting of quality indicators (assembly parameters) of high-tech products. The basis of forecasting is the creation and use of valid geometric models of parts containing data of their geometric deviations, as well as numerical models of mates of parts. Valid geometric models are created based on coordinate measurements of parts. On the example of an assembly unit consisting of three parts of an aircraft engine turbine rotor, the developed models were tested. To reduce computing resources, the use of a radial basis neural network for calculating assembly parameters is considered. The training and test samples were simulated, the network operation parameter was optimized, and the obtained results were generalized.

Maximilian Khotilin, Alexander Kupriyanov, Natalia Kravtsova and Igor Rytsarev
Сlassification of objects of natural hyperspectral images
This paper describes the process of determining the classification of an object / zone in a hyperspectral image by class: forest, water, earth. Methods and algorithms for finding membership in a particular class are described. Further prospects for the development of the algorithm on big data are described.

Nikita Davydov, Alexander Khramov and Artem Nikonorov
Functional MRI recurrent real-time quality assessment estimation using OpenNFT
The quality assessment of functional magnetic resonance imaging (fMRI) in real-time is important in clinical applications and research. The loss of blood oxygen level-dependent (BOLD) sensitivity due to the susceptibility-induced T2* signal dropouts, head movement, and motion-by-susceptibility interaction during head movement remain the major sources of fMRI artifacts [1]. Furthermore, retained head motion in the pre-processed fMRI data could also predict anthropomorphic, behavioral and clinical factors. Since visual definition of image distortions is complicated, an automatic comprehensive real-time quality assessment approach is needed. The signal-to-noise ratio (SNR) is a fundamental parameter for the complex quality assessment of fMRI. In addition, the contrast to noise ratio (CNR) is a preferential quality measure in the activation paradigms where noise is inflated by the activation. Head movements could be assessed using conventional six head motion parameters and their joint estimates, e.g. framewise displacement (FD). The FD could be used for real-time quality assessment (rtQA) of fMRI in real-time, as well as the recurrent SNR and CNR.

Vadim Moshkin, Ilya Andreev, Vladimir Belov, Dmitry Drozdov and Roman Shakurov
An integrated approach to mapping user profiles on social networks
This paper considers an integrated approach to the search for similar user profiles in social networks. The described method allows you to identify a person’s profile in various social networks based on the analysis of structured, unstructured and graphical profile data.

Vadim Moshkin, Albina Koval and Anton Zarubin
Ontology-based classification model of text resources of an electronic archive
A modern design organization has a significant electronic archive of documents in an unstructured form. Solving the problem of using the experience of previous projects to solve new problems can be based on the use of intelligent methods and algorithms for analyzing text documents of an organization in order to build a classification system for electronic archives. This work presents an ontological model of a text document as an electronic archive resource. The paper also presents an ontologically-oriented сlassification algorithm for technical documents. In conclusion, the results of experiments confirming the effectiveness of models and algorithms in solving the problem of classifying a document archive are presented.

Alexander Yumaganov
Searching for similar code sequences in executable files using siamese neural network
This work is dedicated to solving the problem of finding similar code sequences (functions) in executable files. The description of functions obtained using the proposed method for solving this problem is based on the mutual spatial position of processor instructions and the corresponding operands in the function body. The word embedding model word2vec is used to form an intermediate description of the executable file functions. The final description of the functions is formed using the siamese long short-term memory network (Siamese-LSTM). Then it description directly used to search for similar functions. The results of experimental studies of the developed method are presented in comparison with some previously known methods.

Kseniya Medvedeva and Vladimir Fursov
Application for operational linear-nonlinear correction of mobile images
The paper considers the technology of forming a two-stage linear-nonlinear filter to eliminate various types of distortion in images recorded using mobile devices running Android. The purpose of the development is to provide high quality distortion correction at the minimum computational cost with the possibility of obtaining processing results in real time. The linear filter parameters are tuned due to blind parametric identification of the model with a radially symmetric quadratic-exponential frequency response. Non-linear filter parameters are adjusted based on a visual assessment of the processed images. The algorithm, pseudo-code of the program and the results of experiments are shown, showing the effectiveness of the developed application for eliminating distortions in images.

Viktoriia Evdokimova, Maksim Petrov, Marina Klyueva, Andrey Alekseev, Sergei Bibikov, Roman Skidanov and Artem Nikonorov
Study of GAN-based image reconstruction for diffractive optical systems
This paper studies the causes of artifacts that appear in images after color correction by a generative adversarial network. Input images are captured by diffractive optics. This paper shows overexposing of the original image and markers in training images affect the occurrence of the artifacts. The paper proposes an architecture of a generative adversarial network with a scaling layer. This architecture allows producing reconstructed images with scale factor 2.

Albert Gareev, Evgeniy Minaev, Dmitriy Stadnik, Vladimir Protsenko, Ilia Popelniuk, Ashat Gimadiev and Artem Nikonorov
Investigation of the effectiveness of neural network algorithms for the faults detection in hydraulic systems
This article investigates the effectiveness of neural network algorithms for detecting faults in hydraulic systems. Mathematical modeling of a typical unit of a hydraulic system was carried out, according to the results of which initial data were generated for training neural network models. The effectiveness of classifiers was tested both on model data and as a result of bench tests of a real hydraulic system. The best detection result was 98% of correctly recognized system states.

Alexey Borisov and Evgeny Myasnikov
Dimensionality reduction using the GPU-accelerated gradient descent
In this paper we discuss possible realizations of GPU-targeted gradient descent algorithm used for dimensionality reduction. 4 realizations of gradient descent algorithm were created using HIP - a new framework for GPGPU programming. We got 6 times perfomance improvement over multithreaded CPU version using AMD Radeon RX Vega 56.

Vladimir Mokshin and Dinar Yakupov
Graphs decomposition using modified spectral clustering method
Among a large number of tasks on graphs, studies related to the placement of objects with the aim of increasing the information content of complex multi-parameter systems find wide practical application (for example, in transport and computer networks, piping systems, in image processing). Despite years of research, accurate and efficient algorithms cannot be found for placement problems. It is proposed to consider the solution of the allocation problem in the context of decomposition of the initial network into k regions, in each of which a vertex with some centrality property is searched. This article provides an analysis of sources for solving the problem of placement in graphs, as well as methods of decomposition of graph structures. Following the main provisions of the theory of spectral clustering, the disadvantages of the splitting applied criteria Rcut and Ncut are indicated. It is shown that the application of the distance minimization criterion Dcut proposed in this paper allows to obtain high results in the decomposition of the graph. The obtained results are based on the examples of searching for sensor placement vertices in the known ZJ and D-Тown networks of the EPANET hydraulic modeling system.

Pavel Ostapenko, Kamila Sultantemirova and Oleg Saprykin
Adaptive traffic light control based on machine learning
This article discusses the main causes of traffic congestion on the city roads. Particular attention is paid to modern methods of adaptive traffic control as a means to reduce the waiting time at a signaled crossing. The article outlines modern approaches based on the use of artificial intelligence methods, as well as the known issues of these methods. The authors propose a traffic light optimization method that is based on a modified Q-learning machine learning algorithm. The method was tested on a simulation model of the Aurora-Partizanskaya intersection in the city of Samara.

Anastasia Plisko, Pavel Serafimovich, Artem Nikonorov and Yuri Koush
Detection of step-like head displacements in fMRI head motion data based on machine learning
Accurate artifacts detection in functional magnetic resonance imaging (fMRI) data is important in clinical applications and research. Subjects head movement remains the major source of fMRI artifacts, and retained head motion in the pre-processed fMRI data could also predict anthropomorphic, behavioral and clinical factors. However, an accurate characterization of subject head motion artifacts is lacking. We searched for step-like displacements of the subject’s head in fMRI head motion data using k-nearest neighborhood classification and support vector machine (SVM). Head motion data of six subjects were defined using conventional six head motion parameters as produced by fMRI realignment procedure. We created semi-automatic markup tool for preparing training head motion data-set for classification. The semi-automatic markup was done using the sliding-window statistical anomaly detection and manual mark up of a few characteristic step-like artifacts. Marked up training data set was used to train the k-nearest neighborhood and SVM classifiers and to classify the step-like head displacements. Our approach based on k-nearest neighborhood classification showed the high accuracy in detection of step-like head displacements, which could be used for an accurate detection of specific fMRI artifacts associated with head motions.

Albert Gareev, Dmitry Stadnik, Artem Nikonorov and Asgat Gimadiev
Datasets gathering for hydraulic systems technical diagnosis using machine learning methods
Modern systems for diagnosing and monitoring complex technical systems are based on machine learning methods, which require the accumulation and processing of large amounts of information. The compilation of data based on mathematical or simulation modeling is preferable in comparison with the experiment, since it not only spends less time, but also allows you to simulate malfunctions that are difficult or impossible to implement at the test rig. However, this necessitates the development of a model of the diagnostic object adequate to real processes, as well as the processing and systematization of the information received. In this work, based on the modeling of dynamic processes in the hydraulic system in good and faulty states in the SimulationX software package training sets are compiled that are used in machine learning to diagnose system failures. The transients of the main parameters of the system are calculated that are adequate to experimental data, an algorithm for their processing and the compilation of training sets is developed. The published material may be useful for specialists developing methods for monitoring and diagnosing hydraulic systems of energy and technological complexes based on machine learning methods.

Alexandr Astafiev, Anton Demidov and Denis Privezentsev
Analysis of the Applicability of the Bundle Method for the Construction of Multi-Code Labelings
The article presents the results of studies on the applicability of the method of monitoring the integrity of the method of bundles for the organization of multi-code labeling of products in the enterprise. The definition of multi-code marking is given, the purpose of its application is described. A comparative analysis of the integrity control methods has been carried out and their quality indicators are given in the field of organizing the control of product movement. An algorithm for generating multi-code labeling based on integrity control methods, an algorithm for determining falsification and recognition errors are proposed. Examples of generated identifiers are given. The results of the research were expressed in the form of a web application integrated into the online store of the Global33 Group of companies (Vladimir Region, Murom, Kovrov), which is confirmed by the implementation certificate. The purpose of the implemented system is to organize the automatic traceability of the rolls of air-bubble film along the conveyor line.

Vladimir Fursov, Pavel Kuznetsov, Anton Kotov and Boris Martemyanov
The method of generalized functions in the problem of conformed estimation of the dynamic characteristics of video sequences
The task of the image motion analysis has been over forty years old, but its full solution, has not yet been received. This is because images usually contain noise, differ in brightness, spectral range and orientation scale, etc. A scene image motion analysis task is formulated as a task of constructing an optical flow created by a sequence of images. Optical flow is a vector field of image point speeds. Optical flow can be restored by matching, a sequence of image frames. This paper considers the approach to solving the problem of determining parameters of optical flow, based on the ideas of the quality criterion functionalization and construction of matched parameters estimates.

Artem Kabanov, Yelizaveta Morkhova and Natalya Kabanova
PATHFINDER toolkit for analysis of ion migration pathways in solids
We have developed a set of shell-scripts for an efficient generation and analysis of ion conductivity pathways in solids. This set of shell-scripts is called "PATHFINDER" and can be freely download at PATHFINDER toolkit allows to identify ion migration pathways and generate corresponding structural files with pathways. PATHFINDER may help with transition state finding and subsequent precise DFT modeling (within NEB approach). We have shown a few examples how to apply scripts for searching of multivalent ion conductivity in magnesium-, calcium-, and strontium–oxygen-containing compounds.

Ekaterina Popova and Vladimir Spitsyn
Text classification based on the use of convolutional neural networks
The article is devoted to neural network text classification algorithms. This paper presents the main components of the text classification system, as well as the same problems associated with the use of the architecture of convolutional neural networks. The algorithm for obtaining vector representations of the dictionary is described.

Oleg Surnin, Pavel Sitnikov, Anastasia Khorina, Anton Ivaschenko, Anastasia Stolbova and Nataly Yu Ilyasova
Data exchange platform for digital economy
This paper describes an experience of the Data exchange software platform practical use supporting the modern trends of Digital Economy. The platform was initially designed for the suppliers and customers of data sources providing the up-to-date technologies of Big Data processing as an online service. The platform is also open for software developers to upload new algorithms and technologies in order to help them to find new areas of application. There is presented architecture and its software implementation for an intermediary online platform capable of collecting, processing and analysis of various datasets. Modern companies being the members of digital economy can use this platform to process their data and produce business analytics. They can become both suppliers and providers of data, as well as develop and upload new customized algorithms. First results were achieved in the area of retail and social media analysis.

Roman Lobov and Ilia Lobov
Application of the method of automatic decision-making for the construction of the control algorithm for multi-drive systems.
The possibility of using the method of automatic decision-making for the construction of intelligent algorithms for automatic control of multi-drive systems is considered. A modernized method of sequential narrowing of vector estimates is used. The goal of the upgrade is to reduce the amount of computing. The algorithm allowing to implement the method of automatic decision-making in the control system is given.

Artem Nikonorov, Dmitriy Stadnik, Albert Gareev, Pavel Greshniakov and Asgat Gimadiev
Experimental study of neural networks-based fault detection methods efficiency for electro-hydromechanical systems
A key role in ensuring the reliability of hydraulic systems is operational reliability, which is achieved by using modern diagnostic methods. Existing methods for monitoring the technical condition of the hydraulic system during their operation in most cases are based on achieving the required indicators for the output parameters, such as pressure, flow rate of the working fluid, the speed of shifting of the actuators, leakage and other parameters calculated by the duration of the transient processes. This approach allows you to diagnose the overall hydraulic systems and its components in detail. This paper describes an array of data generation and preparation and real-time diagnosis on the operation for a working and faulty hydraulic system in order to develop a tool for operational and proactive maintenance of industrial plants and technological complexes in accordance with their technical condition. Experimental test bench is used for data preparation and online tests of machine learning, neural network fault detection and prediction methods.

Back to Top