Кафедра штучного інтелекту (ШІ)
Постійний URI для цієї колекції
Перегляд
Останні подання
Публікація Практикуми для навчання в умовах гібридних загроз(2024) Атаманенко, А.; Балашов, Е.; Головянко, М.; Гришко, С.; Кайкова, О.; Комаров, С.; Рева, Т.; Радіонова, Т.Освітній процес в умовах гібридних загроз стикається з великими методичними викликами, які в контексті зміни навчальної парадигми, що супроводжує розвиток інформаційного суспільства, стають усе більш загрозливими. Тому одне з ключових завдань проєкту WARN (warn-erasmus.eu), окрім формування відповідного навчального контенту, – розробити методичні підходи для повноцінного формування у студентів навичок протидії гібридним загрозам як у професійній діяльності, так і в особистому просторі. Цей посібник містить рекомендації щодо проведення лекційних і практичних занять, тематичних практикумів, майстер-класів, використання технологій мозкового шторму та симуляцій, змагальних ігор та інтелектуальних спарингів. Упродовж 2023–2024 років команда проєкту почала не лише застосовувати ці підходи у власному навчальному процесі, але й проводити практичні семінари, курси та мікрокурси з підвищення кваліфікації для колег-викладачів.Публікація Еволюційні архітектури в штучних нейронних мережах(ХНУРЕ, 2024) Лобанов, А. Д.This thesis delves into the concept of neuroevolution, which combines the merits of these two systems, highlighting how artificial intelligence can harness natural evolution principles to enhance neural network efficiency. Key facets such as the Ant Colony Optimization, a bio-inspired algorithm modelled on ant foraging behavior, are explored in-depth. Challenges like architectural optimization, algorithm limitations, and scalability are analyzed, highlighting potential solutions like surrogate models. As a conclusion, the research underscores the dynamic potential of evolutionary algorithms in powering artificial neural networks, foreseeing continued advancement in this domain.Публікація Дослідження та розробка стратегії оптимізації та підвищення продуктивності веб-додатків на базі NodeJS та React з використанням штучного інтелекту(ХНУРЕ, 2024) Дробицький, Д. С.This study focuses on the development of web applications based on NodeJS and React technologies, which have become integral components of the digital paradigm, with their performance and optimization emerging as strategically important tasks. The integration of Artificial Intelligence (AI) into this ecosystem can interact with existing technologies, contributing to their enhancement and evolution.Публікація Використання low-code платформи у розробці інформаційно-довідникової системи «кінотеатр»(ХНУРЕ, 2024) Пиріг, Н. Я.The rapid development of the modern world, characterised by the widespread use of digital devices, free access of various segments of the population to the Internet and constant technological progress, make the task of data generation and consumption one of the most important tasks of nowadays. One of the social sectors that has been significantly affected by the information explosion is the entertainment industry, especially cinemas. This paper outlined the features of the subject area "Cinema", demonstrated the advantages of using to build an information system using low-code platforms, especially Oracle Apex.Публікація Розробка застосунку психологічної допомоги при повітряній тривозі(ХНУРЕ, 2024) Савєльєва, В. Ю.In today's world people often encounters circumstances that in one way or another affect their stress level. Considering this, the creation of applications for psychological help has become relevant, so the user can turn to these applications to prevent and overcome states of anxiety, panic, etc.Публікація Огляд типів машинного перекладу(ХНУРЕ, 2024) Павленко, О. С.Nowadays, there is a wide variety of information available in many languages. In order for people from different parts of the world to be able to watch and understand any video without any problems, regardless of language, translation is needed. Automatic generation and translation of subtitles became the best solution to this problem. After speech is transcribed into text, it needs to be translated, and translation needs to be both fast and accurate. The purpose of this article is to analyze machine translation methods and determine which one is best suited for real-time subtitle translation.Публікація Research on centrality measures in scale-free networks(ХНУРЕ, 2024) Ivashyn, S.Understanding the behavior of centrality measures in the context of various network models is crucial for grasping the structural properties of complex systems. This study explores the constraints imposed on centrality measures by different network models, revealing the maximum achievable centrality values for individual nodes and for the network as a whole. We adopt a systematic approach, selecting specific network models and centrality metrics, modeling, and assessing centrality through empirical experiments. The study encompasses maximum centrality for nodes, the distribution of node centrality, and network centrality (Freeman's centrality) in various network models and centrality metrics. Additionally, we investigate the possibility of deriving theoretical estimates for these centrality limits. The findings of this study not only contribute to a deeper understanding of network dynamics but also pave the way for further analyses of real networks, allowing for a comparison between model-generated and real networks. This research bridges the gap between theoretical modeling and practical network analysis, offering insights into the fundamental principles governing the structure of complex networks.Публікація Моніторинг фізичних рухів(ХНУРЕ, 2024) Гаденко, В. Ю.Nowadays, where health and physical well-being are becoming increasingly important, regular exercise is considered a key element in maintaining overall well-being. However, improper exercise techniques can lead to a decrease in their effectiveness and, even worse, a risk of injury. This is especially true for beginners, who often face difficulties in mastering the correct techniques without the direct supervision of a trainer or specialist. The use of video monitoring can help many people monitor their activity and exercise technique, provide useful feedback, and ultimately increase physical activity levels and overall public health.Публікація Дослідження сучасних методів аугментації текстових даних(ХНУРЕ, 2024) Абросімов, Є. О.The goal of this work is to explore methods of text data augmentation, which involves creating new synthetic data similar to real ones, for machine learning tasks where available data is limited. Generative data augmentation is used to combat overfitting, but it has found limited application in Natural Language Processing. Simple augmentation methods like random insertions, replacements, and shuffling are too limited in their effectiveness. Substituting n-grams with synonyms is another method that can be used for data augmentation, as well as the application of intelligent models like Back translation and Style augmentation. The use of generative models such as C-BERT is a popular solution for the augmentation task. Prompt engineering is also becoming increasingly popular for creating queries that prompt the model to provide optimal responses.Публікація Дослідження комбінування навчання з підкріпленням та мовних агентів для реалізації діалогових агентів(ХНУРЕ, 2024) Бовдуй, Р. В.The chatbot market has been growing rapidly lately. They allow you to optimize business processes and meet user needs. With the advent of large language models based on transformers, interest in the end-to-end approach to building such systems has increased. However, when building task-oriented dialog systems based on transformers, there is no certainty that they will pursue the intended goal, and here reinforcement learning comes to the rescue. The paper discusses the combination of these approaches in order to improve the characteristics of task-oriented dialogue systems which are needed for business.Публікація Нечітка сегментація з використанням глибоких нейронних мереж(ХНУРЕ, 2024) Шатило, І. Ю.This work delves into the usage of deep neural networks for fuzzy image segmentation to improve data analysis in various fields. Recognizing the limitations of conventional semantic segmentation methods, this research advocates for a nuanced approach that combines fuzzy logic principles with the computational power of deep neural networks. This hybrid methodology is designed to enhance the accuracy and adaptability of segmentation processes, making it particularly relevant for applications requiring precise image interpretation. Additionally, the proposed approach could significantly impact the future of image segmentation techniques.Публікація Аналіз інструментів для створення NLP чат-бота(ХНУРЕ, 2024) Селін, Я. Ю.Examination of different methodologies employed in the development of NLP chatbots is presented in this article. The study compares two primary strategies for chatbot development: constructing from the scratch and utilizing pre-existing no-code platforms. Various criteria, including flexibility, deployment complexity, and customization options, are used to evaluate these approaches. The paper also includes examples of well-known NLP libraries like NLTK and spaCy, as well as off-the-shelf platforms such as Dialogflow and Microsoft Bot Framework. It is emphasized that the selection of an approach depends on factors such as technical expertise, customization and integration requirements, and the specific functional demands of the chatbot.Публікація Аналіз та розробка алгоритмів сегментації клітин на мікропрепаратах(ХНУРЕ, 2024) Яковенко, А. П.This research explores the development of algorithms for cell segmentation in microscopic slides, addressing the challenges posed by the high variability of cell structures and noise in microimages. A comprehensive review of existing segmentation methods, including the influence of different neural network architectures, forms the foundation for the proposed deep learning-based approach. The algorithms, adapted to the specific features of microscopic preparations, will be able to show promising experiments on real cell samples, showcasing their potential impact on advancing medical diagnostics and contributing to the automation of research laboratories.Публікація Виявлення військових цілей за допомогою методів комп’ютерного зору(ХНУРЕ, 2024) Любименко, Р. С.This paper is devoted to the detection of military targets using computer vision methods. Two-stage and one-stage object detection algorithms are considered, along with a more detailed description of the most popular algorithms. The problems related to the detection of military objects are investigated, and possible solutions are proposed.Публікація Нечіткий нелійний беггінг на основі адаптивної метамоделі в задачах прогнозувань(ХНУРЕ, 2024) Іванов, Є. О.Prediction is a fundamental task in artificial intelligence, applied across various domains from finance to marketing and industry. Traditional linear models often fall short in capturing the complexity of data relationships, necessitating the enhancement of predictive model accuracy and reliability. Nonlinear bagging, based on an adaptive meta-model, has emerged as an effective approach for processing large datasets. This method involves creating an ensemble of models with diverse parameters, ensuring both prediction quality and stability in the face of anomalies. However, avoiding overfitting is crucial, requiring the selection of appropriate optimization strategies, notably utilizing an adaptive meta-model. Further development of this method entails exploring various adaptation and optimization strategies for hyperparameters. Overall, the proposed fuzzy nonlinear online bagging procedure synthesizes the ensemble's computational intelligence within the framework of online data processing, offering advantages in handling both sequential and non-stationary data.Публікація Реконфігурація структури зв’язків мереж на основі ігрової моделі(ХНУРЕ, 2024) Гриньов, С. А.; Шергін, В. В.The problem of reconfiguring connections between nodes of scale-free networks is considered. Existing evolutionary models of networks have built-in mechanisms for network growth through the addition of new nodes with edges (links) incident to them, but they do not have a mechanism for redirecting or breaking existing links, so the structure of links remains static. We proposes to use a game model for network rewiring. Nodes are considered as playersplayers – intelligent agents interacting with each other according to the model of the bimatrix game "battle of sexes". Links that are unfavorable for the player are broken and redirected to other nodes. As a result, the network is split into two weakly connected clusters.Публікація Адаптивний подвійний нео-нечіткий нейрон для підвищення ефективності розпізнавання образів(ХНУРЕ, 2024) Плєтньов, В. В.This article explores a combined approach to training the double neo-fuzzy neuron, based on principles of supervised, unsupervised, and lazy learning. This approach aims to optimize the configuration of synaptic weights and the formation of membership functions in real-time mode. Importantly, the computational simplicity and minimal reliance on training data make this method versatile and applicable across various scenarios. It holds promise for developing a double neo-fuzzy system capable of effectively adapting to non-stationary data, even with limited training data.Публікація Методи багаторівневого аналізу приватності бізнес-процесів(ХНУРЕ, 2024) Єрохін, М. А.Proper business process management is a pivotal area for improving the organizational efficacy. Rare research concerns the comprehensive analysis of both BPMN model and the model execution. Moreover, the modern privacy challenges and data flows are scarcely investigated within existing works. This report provides an outlook of the privacy analysis methods in PE-BPMN models, including SQL workflows in collaborative processes. The privacy policies are integrated into the PE-BPMN models and the unwanted data disclosures are identified using novel Pleak tools, advancing the awareness of privacy issues.Публікація Використання автокодувальників для знешумлення зображень(ХНУРЕ, 2024) Марченко, М. Є.In this paper we consider the problem of denoising images by using autoencoders, specifically, the term “image noise”, the place of the denoising problem in intelligent image processing and the general structure and idea of the autoencoder network for solving the image denoising problem.Публікація Кластеризація та дискретизація геоданих(ХНУРЕ, 2024) Котелевець, К. А.This work is devoted to geodata clustering using discretization and its practical relevance in today's geographic information environment. The importance of geodata clustering for identifying patterns and trends in geographic data is analyzed, and the problems associated with non-discrete geodata, such as large data volume and the presence of noise, are highlighted. It is experimentally shown on real GPS data using clustering methods K-Means, DBSCAN and OPTICS that the use of discretization techniques can effectively address these problems by reducing data volume, removing noise and improving the quality of analysis. The advantages of using geodata discretization for clustering are highlighted and the importance of further research and development of these methods to expand their applications in various fields is emphasized.