Кафедра інформатики (ІНФ)
Постійний URI для цієї колекції
Перегляд
Перегляд Кафедра інформатики (ІНФ) за темою "artificial intelligence"
Зараз показано 1 - 3 з 3
Результатів на сторінку
Варіанти сортування
Публікація Application of video data classification models using convolutional neural networks(International Journal of Academic and Applied Research, 2023) Tvoroshenko, I.; Pomazan, V.; Gorokhovatskyi, V.; Kobylin, O.This paper explores the classification of video data with convolutional neural networks. It discusses how convolutional neural network architecture can do this task with fastness and maximum efficiency. By analyzing the different convolutional neural network models model was proposed that showed great results in this video classification task. The strengths and weaknesses of the neural network model for video classification have been identified, and the prospects for further work have also been outlined.Публікація Development and research of a method for the combined use of large language models for text generation(2025) Suprun, A.; Tvoroshenko, I.; Gorokhovatskyi, V.; Yakovleva, O.This paper presents a comparative analysis of three advanced large language models (GPT-4o, Claude 4 Opus, and Gemini 2.5 Pro) applied to creative text analysis and generation tasks. The study examines each model’s performance across multiple narrative scenarios of varying complexity, assessing both analytical accuracy and literary expressiveness using integrated linguistic, logical, and stylistic metrics. Experimental evaluation showed that Claude 4 Opus achieved the highest analytical consistency with minimal hallucination rate and strong logical reasoning, while Gemini 2.5 Pro excelled in generation quality, demonstrating superior stylistic coherence, emotional depth, and grammatical precision. GPT-4o, in turn, maintained high contextual completeness but revealed a tendency toward interpretive creativity and higher variance in factual precision. Building on these findings, a new method for combined utilization of LLMs was developed and tested. In this approach, Claude 4 Opus serves as the analytic module, performing structured narrative decomposition and contextual synthesis, while Gemini 2.5 Pro acts as the generative module, transforming the processed analytical output into artistically refined text. Experimental validation demonstrated that the proposed method achieved an average generation quality index that surpassed each model’s individual results in coherence, emotional integrity, and stylistic harmony. These outcomes confirm the effectiveness of inter-model collaboration for enhancing both analytical precision and creative depth in LLM-based literary text generation, offering a promising direction for future hybrid human–AI creative systems.Публікація Development of a hybrid method to enhance context memory for a chatbot application based on large language models(2025) Bohdan, N.; Tvoroshenko, I.; Gorokhovatskyi, V.; Kobylin, O.This paper addresses the critical challenge of maintaining long-term contextual coherence in LLM-based chatbots, particularly in scenarios demanding multi-stage reasoning and complex information recall. The core scientific novelty of this research lies in the development and justification of a Hybrid Methodology for Contextual Memory Enhancement. This methodology synergistically integrates the temporal continuity of recent messages’ context with the precision recall capabilities of RAG and AI analysis agent, effectively mitigating the common issue of contextual dilution over extended dialogues. The research involved the development of chatbot application implementing this method and subsequent rigorous scenario-based testing, which validated the superior performance of the proposed hybrid approach. The results provide definitive recommendations for optimizing LLM memory management, paving the way for more robust and reliable conversational AI systems capable of advanced, multi-turn reasoning and complex task resolution.