Category: Research

Mr. Igor Ćulafić successfully defended his master’s thesis titled “Cross-lingual Transfer Learning in Large Language Models: Scaling Laws and Parameter-Efficient Fine-Tuning for Multilingual Applications.” His research provides a comprehensive study of cross-lingual transfer for the Montenegrin language, combining a custom V-shaped semi-automated book scanner, a YOLOv11 + Tesseract OCR pipeline, and the creation of 46,661 parallel paragraph pairs. Using LoRA fine-tuning on Qwen2.5-7B and Qwen3-30B—executed on the Leonardo EuroHPC supercomputer—the work demonstrates parameter-efficient adaptation (only 1.05% trainable parameters) and offers insights into model behavior in cultural understanding, script mixing, and analytical reasoning. This research was supported by NCC Montenegro team and made use of the HPC cluster and EuroHPC JU computational resources.

ABSTRACT – This thesis presents a comprehensive study of Cross-lingual transfer learning in Large Language Models with a focus on parameter-efficient fine-tuning for the Montenegrinlanguage. The research integrates the development of a custom semi-automated book scanner with V-shaped design and a computer vision pipeline using YOLO v11 models and Tesseract OCR to digitize 5000 on Montenegrin and 40000 on English language, from public domain books, resulting in 46661 parallel paragraph pairs. Implementation of LoRA fine-tuning on Qwen2.5-7B and Qwen3-30B models was conducted on Leonardo HPC supercomputer, achieving memory efficiency with only 1.05% trainable parameters. Comparative analysis through a structured benchmark of ten progressively complex questions reveals limited but positive effects of fine-tuning, where larger models show better performance in cultural understanding and analytical tasks, while systematic analysis identifies specific problems such as script mixing and cultural inaccuracies that require specialized approaches.

Ms. Ivana Lalatović successfully defended her master’s thesis titled “Application of Explainable Artificial Intelligence in Medicine” at the Faculty of Information Systems and Technologies, University of Donja Gorica.

The defence took place in October 2025, and the thesis explored how modern XAI techniques—such as SHAP and LIME—can improve transparency and trust in AI models used for analysing the performance and reliability of medical respirators. The development, training, and testing of the machine learning and XAI workflows were supported by the high-performance computing (HPC) resources provided through the EuroCC initiative in Montenegro, enabling scalable data processing, faster experimentation, and reproducible analysis required for medical AI applications. Her work demonstrates how HPC-enabled explainability can strengthen the safety, reliability, and ethical use of AI in healthcare environments, contributing to the growing ecosystem of advanced AI research supported by NCC Montenegro.

ABSTRACT – The need for explainable intelligent systems is growing along with the increase in artificial intelligence products used in everyday life. Explainable artificial intelligence (XAI) has experienced significant growth in the last few years. The reason for this is the wide application of machine learning, as well as deep learning techniques, which have led to the development of highly accurate models. However, they lack explainability and interpretability. This study explores the application of XAI methods in medical applications, with a particular focus on interpreting model decisions. SHAP and LIME methods were applied to interpret the model’s predictions, enabling the identification of key features that have the greatest influence on the model’s decisions. The results of this research confirm the importance of explainable artificial intelligence in critical domains such as medicine, where trust in AI systems must be based on understanding and verifiability of their decisions.

Mr. Elvis Taruh successfully defended his master’s thesis titled “Development of Edge/AI Applications with HPC” at the Faculty of Information Systems and Technologies, University of Donja Gorica.

ABSTRACT – The efficiency of training artificial intelligence (AI) models has become a crucial factor in modern research, especially when dealing with complex systems that require substanial computational power. This study explores how the application of high-performance computing (HPC) and Edge devices can optimize the AI model training process, reducing processing time and improving efficiency. Through an experimental approach, AI model training was analyzed across three different platforms. Local computer, Google Colab and the HPC cluster at the University of Donja Gorica. As a practical example, livestock detection was used. By comparing the training time, memory consumption, and model accuracy, the research demonstrates that HPC clusters significantly accelerate the training process compared to traditional methods, while Edge devices enable faster real-time data analysis.

Mr. Arnad Lekić successfully defended his master’s thesis titled “Development of an AI Tutor Using Large Language Models and HPC” at the Faculty of Information Systems and Technologies, University of Donja Gorica.

ABSTRACT – This thesis explores the development of a personalized AI tutor using large language models (LLMs), with a specific focus on the LLaMA architecture and the application of High-Performance Computing (HPC) resources. The research involves the acquisition, setup, and evaluation of an open-source LLaMA model, with the goal of building a system capable of automated test grading. Special emphasis is placed on the training efficiency and feasibility of running the model locally using the available computing nodes, compared to cloud-based solutions like Google Colab. Beyond the technical implementation, the study also addresses the ethical challenges of using generative AI in education. Through experimental analysis, the research demonstrates that open models can be effectively adapted for educational purposes, with the potential to expand to grading diverse exam formats and generating educational content. The work provides directions for future development of systems leveraging advanced multimodal models for more complex tasks.