Search

The rapid development of technological advances, due to which computer algorithms for image analysis compete with professionals in terms of accuracy, but remain unchanged in speed and volume of cases. In recent years, advances in machine learning and the implementation of convolutional neural networks have improved the ability to classify and identify objects. There is strong evidence that such models can match or outperform experts in complex tasks such as imaging and word processing, recognition and classification, decision-making based on abstract representations, and even clinical decision-making. The purpose of this work is to study the existing methods of developing the architecture of the convolutional neural network and design our own model with subsequent implementation for the diagnosis of digital medical (X-ray) images for the presence of lung disease. As a result of the work, the methods of designing the architecture of convolutional neural networks were analyzed and researched, the main advantages and disadvantages of network optimization means were determined, and the software in which the trained model was implemented was developed.

This article analyzes the algorithms of computer vision, the features of the application of augmented reality technology and existing software modules, frameworks and libraries. The result is a computer vision mobile system (application) using augmented reality technology, which allows users (for example, students) to obtain additional virtual information about the research object and to be able to interact with it. The functional model of the system is formed, the process of application development using the Vuforia library is described and the results of the work are given. The result is an Android application, which using augmented reality tools, allows the user to obtain a virtual environment object in a real-world. This computer vision mobile system is intended for educational purposes, in particular for use in schools and universities for more effective interaction between users and educational material.

The number of patients with liver diseases has been continuously increasing because of excessive consumption of alcohol, inhale of harmful gases, intake of contaminated food, pickles, and drugs. Early diagnosis of liver problems will increase patients’ survival rates. Liver disease can be diagnosed by analyzing the levels of enzymes in the blood. Creating automatic classification tools may reduce the burden on doctors. To achieve this numerous classification algorithm (Decision Tree, Random Forest, SVM, Neural Net, Naive Bayes, and others) from different machine learning libraries (Scikit-learn, ML.Net, Keras) are tested against existing liver patients’ dataset, considering appropriate for each algorithm preliminary data processing. These algorithms evaluated based on three criteria: accuracy, sensitivity, specificity.

Over the last decade, there has been a significant growth in web applications for data processing and output, most of them working through the REST API to communicate with third-party applications. Tools for automatically testing web services through their REST API and verifying the reliability and security of these services are still in their infancy. The most sophisticated testing tools currently available for the REST API scan all API traffic in real-time and then analyze, modify, and reproduce it. Many of these tools were born as extensions of more established web site testing and crawling tools. As these REST API testing tools are all recent and not widely used, it is unknown at this time how effective they are in finding errors and how important they are for security. In this paper, using the latest researches in the field, several methods and approaches for fuzzing web interfaces are analyzed. Their comparative analysis of existing techniques allows to see the current state, performance, and appliance to real-world web application and widely used REST API architecture in general.

Research in the field of human machine interaction and machine learning contributed to the revival of the chatbots. They are virtual interlocutors whose logical apparatus is based on artificial intelligence. However, recent reviews show that chatbots are perceived as unwise systems. These results contributed to the rapid introduction of chatbots in social networks. At the same time, the question of choosing the structure of a neural network for learning dialogue systems, the principles and features of human machine interaction remains important. In this paper various architectures of neural networks are being compared, and it's own chatbot using encoder-decoder architecture with attention mechanism is developed. For implementation, the Python programming language is used. TensorFlow framework is used for deep learning. The simulation results confirm the effectiveness of the proposed approach to speech recognition and human machine interaction.

The article is a review of the perspective methods and approaches to the design of fuzzy decision support systems (DSS) with the application of discrete fuzzy inference engine. The authors also developed a two-stage method for fuzzy rule base (RB) correction in the case of changing the structure of the input vector. In addition, the fuzzy DSS with a hierarchical structure for the best selection of the marine delivery company was developed. Simulation results confirm the effectiveness and feasibility of fuzzy DSS structure with variable input coordinates vector, in particular, in marine practice.

The paper analyzes the existing methods and approaches to correcting bases of fuzzy rules of decision support systems (DSS). There was developed and implemented a two-stage method of fuzzy rule base correction for hierarchically-organized DSS with discrete logic output at variable structure of input vector. The appropriate method interactively provides reduction of database rules structure and correction of rules consequents at changing the number of input data, which enables to avoid results deformation in decision making. The feature of this method is the ability to process fuzzy input data with a high level of uncertainty, given in the form of fuzzy sets. Simulation results confirm the effectiveness and feasibility of the two-stage correction method of fuzzy rule bases of hierarchically-organized DSS developed by the authors, including the choice of models of cooperation within the consortium of the “University — IT company”.