Hibridni događaj

Program događaja

utorak, 26.5.2026 9:00 - 13:00, Nava 2, Hotel Admiral, Opatija 9:00 - 10:44  Radovi   1. K. Petrinić, T. Špoljarić, B. Peša, G. Malčić (University of Applied Sciences, Zagreb, Croatia) Automation of a Pumpkin Seeds' Drying Process  In this article an automation of a pumpkin seeds' drying machine is examined in two different configuration solutions. The first configuration uses relay devices and requires more interaction with the machine to achieve a successful process. The second configuration utilizes a more complex configuration with programmable logic controller (PLC) to increase the automation level in the process, thus reducing the need for operator intervention. For both versions, a wiring project was designed and component costs were calculated. Additionally, for the version with the PLC, accompanying code was also developed and tested. 2. P. Komljenović, B. Vuletić Komljen, N. Šare, S. Jelčić (Zagreb University of Applied Sciences, Zagreb, Croatia) Design and Development of a Compact Keyboard with 8 Keys and a Rotary Encoder  This paper contains the design and construction of a customized mechanical keyboard. The housing and keys were made according to a custom 3D model. A printed circuit board was designed with mechanical switches, a rotary encoder, hotswap sockets and diodes. Control was achieved with an Arduino microcontroller, and the functionality of the keyboard was achieved by implementing an open source software solution that allows dynamic adaptation of functions. The creation of a fully functional, compact keyboard intended for macro commands, multimedia control and specialized applications was achieved. 3. D. Kokanović, B. Vuletić Komljen, N. Šare, G. Malčić (University of Applied Sciences, Zagreb, Croatia) Simulation of an electric vehicle (EV) charger with dynamic charging power allocation  This paper presents the concept of an electric vehicle (EV) charging station with chargers that enable dynamic power allocation among multiple EVs connected to the system simultaneously. The charging station is designed as a split-type system, where the power cabinet is physically separated from the charging terminals, and each terminal is equipped with a connector for EV connection. The thesis analyzes the logic behind managing power distribution across individual connectors, with a focus on adapting the output power based on each vehicle's battery and its ability to receive power during the charging process. Dynamic power allocation control is implemented through co-simulation between the MATLAB Simulink environment and a simulated PLC system within Siemens PLCSIM Advanced. 4. T. AKKURT EROĞLUER, H. APAYDIN (Ankara University, Ankara, Turkey) Multistep-Ahead Forecasting of Drought Using Neural Prophet and Facebook Prophet Artificial Neural Network models: Beysehir Lake Basin Case  Droughts cause major agricultural, social and economic problems, and, in the worst cases, prolonged droughts can cause catastrophic consequences to the drought-affected region. In this study; in order to determine a potential drought that the Beysehir Lake Basin in Turkey may face in the future, the data obtained from the Seydisehir and Beysehir meteorological observation stations and the Sarkikaraagac, Soguksu-Yesildag, Sugla and Beysehir Lake stations were used. Using the data measured at these stations, data were forecasted for the next 25 years using neural network-based models: The Neural Prophet (NP) and Facebook Prophet (FP) models. Standardized Precipitation Index (SPI), Reconnaissance Drought Index (RDI) and Streamflow Drought Index (SDI) drought indices were calculated with forecasted and historical data. According to the results obtained during the study, drought indices calculated with NP and FP model data showed a similarly increasing drought trend for 3, 6, 9 and 12 months periods at all stations. Frequency rates according to drought classes in drought indices calculated with measured data are very close and consistent with the frequency rates determined using FP and NP model data. The frequency rates of the SPI, RDI and SDI indexes according to drought classes were close to each other. The rate of extreme dry periods was found to be lower in flow-measured stations compared to precipitation stations. 5. J. Franjković, M. Horvat (University of Zagreb, Faculty of Electrical Engineering and Computing, Department of Applied Computi, Zagreb, Croatia), M. Delimar (University of Zagreb, Faculty of Electrical Engineering and Computing, Department of Energy and Powe, Zagreb, Croatia) Identification of Emerging Viral News Topics Using HDBSCAN Outlier Projection  Conventional dynamic topic modeling often disregards low-density regions as stochastic noise. This paper challenges that notion by interpreting HDBSCAN outliers as latent semantic precursors to emerging topic narratives. We propose a predictive projection framework applied to a comprehensive dataset of Croatian news articles organized longitudinally and collected during a specific time period. Using a dual-pipeline BERTopic architecture, we generate word embeddings with paraphrase-multilingual-MiniLM-L12-v2 and apply lemmatization for interpretable c-TF-IDF representations using CLASSLA library. By calibrating density models on a reference corpus and projecting future data into the fixed manifold, we isolate a specific distributional pattern for nascent topics. Contrary to the assumption that novel topics appear as orthogonal vectors and cannot be inferred, our results indicate that viral events reported in the news articles manifest within a distinct cosine similarity interval (0.60-0.80) relative to reference outliers. This latent transition interval is significantly distinguishable from uncorrelated noise (similarity < 0.60) and semantic duplicates (similarity > 0.80). These findings demonstrate that viral themes or narratives initially occupy sparse vector space regions before converging into high-density clusters, confirming that outlier projection substantially reduces the search space for event detection. The codebase created for this research is freely available for research purposes at: https://www.kaggle.com/code/jurefranjkovi/emerging-topic-detection. 6. S. Petković, I. Črnjak, S. Zentner Pilinsky (Zagreb University of Applied Sciences, Zagreb, Croatia) Desktop Application for Processing 4G and 5G Mobile Network Radio Parameters  At Zagreb University of Applied Sciences, as a part of the internal project ICT Green Routes for Recreation in Zagreb and student research in the field of mobile communication networks, a desktop application was developed for processing and visualising mobile network radio parameters measured by the university’s Android application for measuring mobile network radio parameters. The application is designed to evaluate mobile network coverage, signal quality, and user-experience. The application processes measured data and stores the results in a local database, where they can be further examined through interactive maps and graphical representations. Each point of measurement is associated with geographic coordinates and a timestamp, allowing the user to observe how key indicators vary along a measurement route. To evaluate the functionality of the application, an analysis was carried out related to the coverage of 4G and 5G mobile network radio parameters on a chosen route. 7. H. Ivančić, M. Šagovac, A. Novalić, F. Blažević, E. Vuran, D. Potoč Starčević, T. Jagušt (University of Zagreb Faculty of Electrical Engineering and Computing, Zagreb, Croatia) Adaptive Task Difficulty Adjustment in Primary Mathematics Education Using Interpretable Machine Learning  This Work-in-Progress paper presents the design and preliminary classroom evaluation of Smart Math, an AI-supported mathematics learning application that adapts task difficulty to individual student performance in lower elementary education. The study was conducted in two classrooms, each from a different public elementary school, with students aged 7–8. The application supports classroom use and enables continuous monitoring of student progress through interpretable machine learning models. Students begin at a predefined difficulty level. After completing an initial set of ten tasks, the system generates adaptive difficulty recommendations based on interaction data, including answer accuracy, response time and hint usage. Two interpretable approaches are explored: logistic regression and Explainable Boosting Machines (EBM). Recommendations are displayed on a teacher dashboard with explanatory notes that clarify the reasoning behind each suggestion. These explanations allow teachers to review and, if necessary, override recommendations, ensuring transparency and pedagogical control. The system is evaluated in a real classroom setting, where student performance and teacher interactions are analysed. The study examines the suitability of interpretable models for adaptive learning and teacher acceptance of AI-supported decision-making, aiming to inform the design of transparent, teacher-centered systems for primary education. 8. A. Dergez, V. Zuppa Bakša, A. Bednjanec (Zagreb University of Applied Sciences, Zagreb, Croatia) Battery Systems of Large Capacity in Power Systems  Electrical power plants, as the fundamental part of the entire power system, must be stable and reliable. Their stability depends on balance between electricity production and consumption. Any imbalance between electricity production and consumption can lead to system instabilities. Battery systems provide a solution to these instabilities, such as reducing peak loads and managing daily load levelling. To effectively support the system, battery systems must have sufficient capacity. This paper compares the main classifications of secondary batteries used in electrical power plants, as well as the challenges they address during peak load events. 10:44 - 10:57  Pauza  10:57 - 12:54  Radovi  9. M. Muđer, A. Bednjanec, V. Zuppa Bakša (Zagreb University of Applied Sciences, Zagreb, Croatia) Quality Management in the Automotive Industry  The paper addresses quality management in the automotive industry as a key factor in ensuring vehicle safety, reliability, and competitiveness. Special emphasis is placed on an integrated approach to quality that encompasses quality control, quality assurance, and continuous quality improvement throughout all stages of the production process, from material procurement to final vehicle inspection. The paper analyzes modern quality management methods, with particular focus on the application of the 5S and Six Sigma methodologies, which enable error reduction, process optimization, and increased productivity. In addition, the role of international standards, such as ISO 9001, ISO 26262, and IATF 16949, in establishing an effective quality management system in the automotive industry is presented. In conclusion, it is emphasized that the combination of a methodological approach, standardized processes, and continuous improvement represents the foundation for long-term sustainability and customer satisfaction in the automotive sector. 10. Z. Bosković Gobo, S. Gaši, M. Todorić, K. Marasovic, A. Mutka (Rochester Institute of Technology Croatia, Zagreb, Croatia) The Making of RITGPT - A University AI Assistant with RAG and APIs  RITGPT is an AI-powered conversational assistant developed to support domain-specific academic and administrative queries within a university environment, using RIT Croatia as a case study. The system is implemented as a web-based application built on a Retrieval-Augmented Generation (RAG) architecture, enabling responses grounded in institutional documents rather than relying solely on a general-purpose language model. In addition to document-based knowledge access, the system supports selected real-time scheduling queries through integration with institutional APIs. The knowledge base can be continuously updated without retraining the underlying language model. To evaluate the current prototype, a structured benchmark of 160 representative queries was used to assess functional correctness, category-level behavior, and response latency under controlled conditions. The results indicate that the proposed hybrid architecture is technically viable as a prototype university assistant, with strongest performance observed for document-based and API-based queries and the main limitations arising in edge-case and multi-intent query handling. 11. V. Valentić, L. Calcich, A. Bendra, A. Višković (RITEH, Rijeka, Croatia) Territorial Voltage Management Using a Technical Virtual Power Plant  This paper explores the use of Technical Virtual Power Plants (TVPPs) for voltage regulation in the power system of the Istrian region. The growing penetration of distributed energy resources, particularly photovoltaic generation and battery energy storage systems, poses significant challenges to grid operation by modifying conventional power flow patterns and increasing sensitivity to disturbances. As traditional voltage control and reactive power management approaches become insufficient under these conditions, more advanced and coordinated control strategies are required. To address this, a comprehensive set of simulations was performed, encompassing a wide range of operating states, from normal system conditions to critical scenarios such as transmission line outages and periods of high network loading. The analysis focuses on the integration of a 100 MW photovoltaic power plant connected at the Funtana substation and a 20 MW battery storage system installed at the Rovinj substation, with particular emphasis on U/Q control capabilities. System performance is evaluated through a comparative assessment of cases with and without TVPP-based coordination. The results indicate that the implementation of TVPPs substantially improves voltage profile robustness, enhances reactive power sharing, and strengthens overall supply reliability. In addition, the study identifies existing grid constraints and outlines technical measures to increase system flexibility and resilience. These findings underline the role of TVPPs as a crucial enabler of modern power systems and a key driver of a secure transition toward renewable-based energy systems. 12. P. Lončar (Faculty of Electrical Engineering and Computing, Zagreb, Croatia), F. Grebenar, A. Jukić, A. Kekelj (Croatian Transmission System Operator Plc., Zagreb, Croatia), D. Sabolić (Faculty of Electrical Engineering and Computing, Croatian Transmission System Operator Plc., Zagreb, Croatia) Impact of Increasing Distributed Generation on the Accuracy of Transmission Congestion Forecasts  The increasing share of renewable energy sources, particularly photovoltaic systems (PVs) connected to the distribution network, poses significant challenges for transmission system operators (TSOs) in managing the power system. Due to the inherent uncertainty and variability of renewable generation, accurate load forecasting has become increasingly difficult, especially at the substation level, which is crucial for effective congestion management. Nodes that traditionally operated as consumers often intermittently shift to net generation, resulting in highly variable and less predictable net load profiles. Under these conditions, prediction methods developed for systems with low renewable penetration can no longer be applied without significant adaptation. Therefore, it is necessary to identify and analyze the limitations of existing forecasting models and to propose improvements or develop more advanced approaches that incorporate a broader set of relevant parameters. By analyzing Day-Ahead Congestion Forecast (DACF) and IntraDay Congestion Forecast (IDCF) outputs for selected substations in the Croatian transmission system with varying shares of distribution-level generation, a clear correlation was identified between the deviation of forecasted and actual load and prevailing weather conditions. Furthermore, the deviation patterns exhibit predictable characteristics, making them suitable for enhancing existing forecasting models. The analysis shows that under stable clear-sky conditions, forecast deviations typically remain below 1 MW, while in cases of high day-to-day solar variability, deviations exceeding 10 MW may occur during central daytime hours. Furthermore, short-term IDCF forecasts (one-hour ahead) reduce forecast errors under variable conditions, although temporal shifts may still be present. 13. T. Špoljarić, E. Čižmak, J. Kordek, K. Martinčić (University of Applied Sciences, Zagreb, Croatia) Development and implementation of PWM control device for laboratory observation of control possibilities in automation education  In this article a practical educational role of PWM control in automation laboratory is developed and described. For this purpose an adequate electronic board device (ECB) is designed and implemented. ECB design takes into account the requests for three different frequency modes and the possibility to control the output voltage on different load types. These types include active power load such as light bulb and reactive power loads such as: series connected RL load and DC motor. Possibilities of PWM control are observed through voltage and current waveform for simpler loads such as light bulb and RL load. For more complex load (DC motor) PWM control possibility is discussed through observation results of waveforms, vibration detection and speed measurements. These results are used in formulation of students' laboratory assignment with goal in understanding the design, results and capabilities of PWM control type in automation. 14. M. Milenković, L. Pacek, E. Rožić (University of Zagreb Faculty of Transport and Traffic Sciences, Zagreb, Croatia) Automated Traffic Surveillance and Oversight: Legal Limits of Personal Data Processing by Public Authorities in the Republic of Croatia  The deployment of video surveillance, automatic number plate recognition (ANPR), and electronic toll collection systems enables public authorities to systematically collect and process road users’ personal data. However, citizens are not given a genuine choice or explicit consent under the GDPR, while the purposes, responsible authorities, retention periods, and security safeguards (including authentication and anonymization/pseudonymization) often remain unclear. This raises significant concerns regarding lawfulness, proportionality, and purpose limitation. This paper analyzes the legal limits of personal data processing in automated traffic control and surveillance systems in the Republic of Croatia. It examines the application of the GDPR, the Croatian Act on its Implementation, the EU Charter of Fundamental Rights, and relevant EU instruments such as the ITS Directive, Directive 2016/680, and the AI Act, particularly in relation to AI-based and biometric technologies. The paper identifies regulatory gaps and emphasizes the need for clear legal bases, transparency, accountability, and proportionate use of technology to ensure effective protection of fundamental rights under EU and national law. 15. I. Veličković, A. Drobnjak (Faculty of Electrical Engineering and Computing, Zagreb, Croatia), L. Jakić (LAFCO TECH j.d.o.o., Zagreb, Croatia), T. Rajnović, I. Botički (Faculty of Electrical Engineering and Computing, Zagreb, Croatia) Mixed Reality – An Exploration of Practical Capabilities and Constraints  Mixed reality (MR) has gained popularity in recent years due to its ability to combine the real world with virtual content. Unlike virtual reality (VR), which places the user in a fully virtual environment, MR displays the real world while integrating virtual elements that the user can interact with. This paper investigates the practical capabilities and limitations of mixed reality technologies in application development. Unity was used as the primary development environment, while Meta SDK packages provided support for passthrough, spatial tracking, and user interaction. The work includes prototype implementations for manipulating virtual objects in three-dimensional space and for detecting QR codes within mixed reality environments. The main limitations observed were related to tracking stability, lighting conditions affecting QR detection, and performance constraints of the mixed reality platform. Overall, the paper presents an initial exploration of MR development, outlining both its potential and its current practical constraints. 16. F. Buhiniček, A. Drobnjak, I. Botički (Faculty of Electrical Engineering and Computing, Zagreb, Croatia) Design and Evaluation of a Local LLM-Based Architecture for Intelligent Educational Applications  The growing availability of locally deployable large language models presents a transformative opportunity for developing intelligent educational applications while avoiding the privacy, cost, and latency concerns associated with cloud-based services. This study examines the use of locally executed language models to automatically generate educational questions and contextual hints. A systematic evaluation of multiple LLMs was conducted, comparing response latency, output quality, and adherence to predefined generation rules. Based on the evaluation results, suitable models were selected a experimental web-based application designed for interactive question generation and adaptive hint delivery was proposed. Experimental results demonstrate that model initialization time has a significant impact on the first inference request, while subsequent interactions exhibit stable and efficient performance. Furthermore, the study shows that smaller language models can be effectively utilized for auxiliary tasks such as hint generation, maintaining satisfactory output quality while improving overall system efficiency. These findings indicate that locally deployed LLMs provide a viable, scalable, and secure foundation for next-generation intelligent educational systems. 17. Ž. Trzun, S. Morić (Zagreb University of Applied Sciences, Zagreb, Croatia) SUSTAINABILITY OF MOBILE CORE ARCHITECTURE DEVELOPMENT  With the growing popularity of smart mobile devices, the introduction of 4G and 5G mobile networks, the increasing demand for digital content, the number of network servers in the core network has also increased. The signaling of network functions, data and IP packet control, and the entire process of connecting the end user with another network participant or the internet is becoming increasingly demanding, leading to greater complexity in the mobile network infrastructure. The research results confirm the theoretical assumption that by developing and designing a new type of infrastructure of network functions, compared to the previous one, significant energy savings and reduction of total operating expenses for mobile operators are achieved. utorak, 26.5.2026 15:00 - 18:00, Nava 2, Hotel Admiral, Opatija 15:00 - 16:18  Radovi  1. J. Simon, G. Sirovatka (TVZ, Zagreb, Croatia), I. Branica (Locus d.o.o., Zagreb, Croatia), M. Žagar (TVZ, Zagreb, Croatia) Identification of optimal intrusion path via graph-based cost modeling and Dijkstra’s algorithm  This work addresses the critical challenge of identifying the optimal intrusion path. This path is the most efficient trajectory an attacker traverses from initial entry to a sensitive information, and has the least resistance. At the core of this methodology is a graph-based cost model where network nodes represent system states or vulnerabilities and edges represent the computational cost required for exploitation. Mathematical determination of the path of least resistance is done by applying Dijkstra’s algorithm. Main contribution of this work is made through the development of a Python simulation prototype. Python was selected as it has an extensive library ecosystem, specifically leveraging NetworkX for complex graph manipulations and Matplotlib for the real-time visualization of the pathfinding process. The implementation utilizes a custom class structure to represent network topologies. This allows the tool to simulate how an adversary might move through a network when faced with varying defensive hurdles. The simulation demonstrates the algorithm’s high efficiency, proving that Dijkstra’s approach remains computationally viable even for large network complexity. Beyond theoretical validation, the tool serves a dual purpose: as an educational platform for cybersecurity students and as a decision-support system. By visualizing attach paths security architects can prioritize patching efforts on the specific nodes, thus significantly hardening system resilience. 2. V. Bilogrević, A. Bednjanec, A. Kiričenko (Zagreb University of Applied Sciences, Zagreb, Croatia) An ISO/IEC 27001-Oriented Approach to Quality Management in the Computing Sub-Sector  Quality management in the computing sub-sector is a strategically important component of modern organisations that increasingly depend on information systems and digital infrastructure. The quality of computing systems is manifested through their functional reliability, information security, performance, scalability, and compliance with relevant standards and regulatory requirements. In this context, information security represents a critical element of overall system quality, as security incidents and system failures may cause significant operational, financial, and reputational damage. This paper examines key aspects of quality management within the computing sub-sector, including functional system management, protection of sensitive data, performance optimisation, and regulatory compliance. Special emphasis is placed on the role of IT professionals in infrastructure design and implementation, configuration and change management, continuous system monitoring, security incident response, and business continuity and disaster recovery planning. The internation standard ISO/IEC 27001 is identified as a fundamental framework for establishing and maintaining an Information Security Management System (isms). Its structured approach enables systematic risk management, protection of information assets, and continuous improvement of security controls, contributing to higher reliability and overall quality of computing systems. 3. M. Šoljić, P. Lončar, L. Malarić, F. Čopčić, D. Brozović (Faculty of Electrical Engineering and Computing, Zagreb, Croatia) Enhancing Power System Reliability Through N-1 Contingency Analysis, EMS, and Load Forecasting  This paper examines key operational tools used by Transmission System Operators to maintain power system reliability under increasingly complex and variable operating conditions. First, the role of N-1 contingency analysis is discussed as a foundational method for ensuring system security, while highlighting its limitations in power systems with high penetration of renewable energy sources. Second, the paper reviews the function of Energy Management Systems (EMS) in supporting real-time monitoring, control, and decision-making, emphasizing their role in coordinating generation, transmission, and system security constraints. The importance of load forecasting is then addressed, focusing on its impact on operational planning, congestion management, and risk mitigation. By considering these elements jointly, the paper highlights how traditional security criteria, advanced EMS functionalities, and forecasting techniques must increasingly be viewed as interdependent components of modern power system operation. The analysis underscores their continued relevance for enhancing reliability and resilience in future power systems. 4. I. Vrhovec, A. Domitrović, K. Krajček Nikolić, D. Gerhardinger (Faculty of Transport and Traffic Sciences, Zagreb, Croatia) Mission-Level Endurance and Payload Benefits of a Series-Hybrid Multirotor UAV  Hybrid propulsion combines electric actuation with the specific energy of liquid fuels, enabling longer multirotor missions and higher payload capability than battery-only platforms. This paper presents a MATLAB/UAV Toolbox simulation framework for a series-hybrid multirotor including: (i) a mission-driven electrical load model based on a reduced-order trajectory-to-power mapping, (ii) battery SOC dynamics with discharge energy-throughput accounting, (iii) ICE–generator fuel consumption using a constant SFC model, and (iv) a transparent rule-based Energy Management System (EMS) that allocates generator and battery power using fixed SOC thresholds and fuel safeguards. Comparative simulations on a reference 3D trajectory demonstrate that hybrid propulsion achieves 2.3× longer range than an all-electric configuration of equal base mass (15 kg), despite the hybrid platform using 6.5× smaller battery (400 Wh vs 2600 Wh). The hybrid UAV also outperforms a lighter 5 kg all-electric platform (4.2× range improvement), indicating that fuel-based energy storage enables both longer endurance and higher payload capability. 5. A. Dundov, M. Šimatović, D. Beronić, B. Mihaljević (RIT Croatia, Zagreb, Croatia), A. Radovan (Algebra Bernays University, Zagreb, Croatia) Evaluating Java Virtual Threads for Concurrent Spring Boot and Helidon Services  Virtual Threads were introduced to address the scalability limitations of traditional platform threads in large-scale, highly concurrent, server-side Java applications. By reducing the overhead of thread creation, scheduling, and management, Virtual Threads enable an efficient thread-per-request programming model that was previously impractical at scale. Contemporary Java frameworks, including Spring Boot and Helidon, support this modern model of concurrency. However, the combined influence of framework architecture and thread type on application behavior under high concurrency remains insufficiently explored. Our research presents an evaluation of Java Virtual Threads in Spring Boot and Helidon frameworks, compared against platform threads within the same frameworks. An HTTP-based service benchmark is implemented across all framework-thread combinations, focusing on framework and threading models as the factors of interest. Each configuration is subjected to progressively increasing client concurrency, collecting typical performance-relevant metrics, such as throughput, latency, and resource utilization. The results demonstrate that the Virtual Threads influence application behavior under load and affect scalability and resource efficiency, revealing framework-specific performance patterns in high-concurrency Java services. These findings provide a clearer understanding of Virtual Thread behavior in Java frameworks and offer insights for designing robust, high-throughput Java services and making informed concurrency and framework selection decisions. 6. M. Šimatović, D. Beronić, B. Mihaljević, A. Dundov (RIT Croatia, Zagreb, Croatia), A. Radovan (Algebra Bernays University, Zagreb, Croatia) Memory Constraints and Scalability of Concurrent Java Applications with Virtual Threads  Modern Java applications are increasingly deployed in virtualized environments such as containers and cloud platforms, where computational resources, including system memory, are explicitly provisioned, constrained, and invoiced. The introduction of Java Virtual Threads enables significantly higher levels of concurrency compared to traditional platform threads, making memory-related scalability constraints more pronounced. Compared to traditional platform threads, which are commonly constrained by CPU availability, Virtual-Thread-based execution shifts a greater share of scalability pressure toward memory usage, making memory-related constraints and Java Virtual Machine (JVM) heap configuration particularly important. This work explores the scalability of concurrent Java applications implemented with Virtual Threads under fixed JVM heap-size constraints. An experimental approach is employed in which concurrency is gradually increased across multiple heap configurations while evaluating performance and stability. The analysis considers throughput, latency, memory utilization, CPU utilization, and failure behavior to characterize scalability limits. Selected experiments are conducted with varying system memory capacities to assess how behavior changes beyond the configured heap. By examining the interaction between heap configuration, system memory constraints, and high-concurrency VirtualThread-based execution, this paper provides practical insights into memory-related scalability behavior, supporting more informed memory provisioning and configuration decisions for concurrent Java applications. 16:18 - 16:31  Pauza  16:31 - 17:49  Radovi  7. I. Ahmad (Leonardo da Vinci Telematic University, Chieti, Italy), A. Amelio (University of Chieti-Pescara, Pescara, Italy), M. Aracne (Leonardo da Vinci Telematic University, Chieti, Italy), L. Caroprese, E. Gill, C. Morbidoni (University of Chieti-Pescara, Pescara, Italy) Technologies, Applications, and Challenges of Large Language Models  The rise of Large Language Models (LLMs) marks one of the most significant breakthroughs in artificial intelligence, with the potential to transform knowledge intensive sectors. Among these, the legal field and public administration stand out as particularly suitable environments for the adoption of such technologies, given their highly normative, document based, and procedural nature. Nevertheless, integrating LLMs into these domains raises legal, ethical, and operational issues that call for careful and systematic examination. This paper offers an overview of the key features that define LLMs and examines the main methodologies used to specialize and ground them in factual information, with particular attention to techniques such as Retrieval Augmented Generation and fine-tuning. Building on this foundation, the contribution aims to explore LLMs on two levels: first, by analyzing the legal implications associated with their use, focusing on topics such as liability, algorithmic transparency, personal data protection, bias mitigation, and the reliability of generated outputs; and second, by surveying the principal applications of LLMs in legal and administrative contexts, with the goal of providing an up to date and critically informed overview of the opportunities and challenges linked to their adoption. 8. A. Radovan, A. Klasić, D. Bele (Algebra Bernays University, Zagreb, Croatia), B. Mihaljević (RIT Croatia, Zagreb, Croatia) Proposal for an AI-Driven Learning Platform with Gamification for K-12 Students  The rapid evolution of generative artificial intelligence has introduced transformative possibilities for personalized education, particularly in addressing the diverse pacing needs of primary and secondary school students. However, the manual creation of high-quality, individualized practice material remains a significant burden for educators and learners alike. This paper presents a specialized web application designed to bridge this gap by automating the generation of tailored educational content. By leveraging OpenAI’s Large Language Models (LLMs) integrated via Azure cloud services, the platform transforms static uploaded materials into dynamic practice sets, including contextual references and mock examinations. The system architecture employs a Vue.js frontend for a responsive user experience and a robust C# backend to manage data processing and secure authentication. To sustain student engagement, the application integrates gamification mechanics—such as peer challenges and merit-based rewards—converting traditional study routines into an interactive, competitive environment. Beyond the functional features, this study details critical architectural decisions regarding cost optimization in AI token usage and strategies for system scalability. Preliminary user testing indicates that the fusion of automated content generation with gamified engagement significantly improves knowledge retention and collaborative learning, offering a scalable model for AI-driven pedagogical tools. 9. T. Saneva, M. Stojcheva (Faculty of Computer Science and Engineering, Ss. Cyril and Methodius University, Skopje, Macedonia), J. Prodanova, A. Dedinec (Macedonian Academy of Sciences and Arts, Skopje, Macedonia), A. Dedinec (Faculty of Computer Science and Engineering, Ss. Cyril and Methodius University, Skopje, Macedonia) Analyzing Media Narratives on Renewable Energy in Romania Using Sentiment Analysis and Topic Modeling  This study analyzes Romanian news media coverage of renewable energy sources (RES) using sentiment analysis and topic modeling. The dataset includes 3,329 news articles published between January 2016 and August 2025. Results reveal predominantly neutral coverage, with moderate increases in positive sentiment linked to major policy decisions. BERTopic analysis shows thematic evolution from regulatory framing, through crisis-driven energy security dynamics, toward implementation and system integration narratives. 10. K. Stefanova, M. Stojcheva, A. Dedinec (Faculty of Computer Science and Engineering, Ss. Cyril and Methodius University, Skopje, Macedonia), J. Prodanova (Macedonian Academy of Sciences and Arts, Skopje, Macedonia) Comparing waste management X posts from Spain and Macedonia using GPT and BERT models  Waste management is one of the key topics for environmental protection, and creating effective waste management strategies relies on positive experiences and public opinions. In this study, sentiment analysis done by comparing BERT and GPT models is used to categorize a large amount of waste management related X posts in Macedonia and Spain. An additional classification of the posts is performed according to Paul Ekman's general emotions with added neutral emotion. Results from the conducted analysis show that 71.9 percent of the emotions expressed in Macedonian posts are classified as negative, while only 8,9 percent of Spanish posts are classified the same. Furthermore, emotions expressed in Macedonian posts are mostly related to anger and disgust, and Spanish posts express happiness and mostly a neutral emotion. This study also presents the various topics achieved from topic modeling done by utilizing gpt-4o-mini and BERTopic. In general, the research presented in this paper shows a huge difference between how residents perceive waste management and which topics were of great importance in 2025 in Spain and Macedonia. Results outlined in this article together with additional analysis of public opinion on waste management can be used to evaluate public awareness and the importance of good environmental practices. 11. D. Krstev, B. Jancheska, P. Kalinovski, M. Toshevska, S. Gievska (FCSE, Skopje, Macedonia) Comparison of Knowledge Transfer Methods for Bias Neutralization in Text  Automated text debiasing with large language models (LLMs) aims to reduce biased language while preserving semantic content. This paper presents a comparative study of multiple transfer learning paradigms for text neutralization using the Wikipedia Neutrality Corpus. We evaluate reinforcement learning approaches, Direct Policy Optimization (DPO) and Group Relative Policy Optimization (GRPO), with rewards provided by an LLM-as-a-Judge, alongside Chain-of-Thought distillation, a supervised fine-tuned (SFT) model, and a pretrained baseline. Results show that GRPO achieves the strongest bias neutralization, followed by SFT, while DPO and CoT distillation fail to meaningfully improve neutrality despite strong lexical and semantic scores. Moreover, traditional metrics exhibit weak correlation with neutrality performance, emphasizing the importance of task-specific evaluation and neutrality-aligned training signals for debiasing tasks. 12. F. Alaber (Fakultet elektrotehnike i računarstva, Zagreb, Croatia), B. Novoselnik (Fakultet elektrotehnike i računastva, Zagreb, Croatia) An Integrated IoT Weather Station for Continuous Environmental Sensing and Analytics  This paper presents the development and improvement of a low-cost mobile weather station designed for scalable deployment across the urban area. The primary objective of the project is to create an affordable system that enables increased spatial resolution in urban meteorological measurements, addressing the limitations of existing weather prediction models based on 4 × 4 km spatial grids and a small number of reference meteorological stations. The proposed system represents a prototype of a larger network of stations capable of measuring key environmental parameters, including air temperature, relative humidity, atmospheric pressure, and precipitation. All sensors are integrated into a single compact hardware unit. Compared to related low-cost weather station designs, the main contribution of this work lies in the integration of a greater number of meteorological sensors within a single deployable package, enabling more comprehensive local measurements without reliance on expensive professional equipment. The paper describes the system architecture, sensor integration, and data acquisition process, and discusses the potential benefits of dense station deployment for improved urban weather monitoring. Future work will focus on enhancing system autonomy and expanding the sensing capabilities to support long-term, large-scale deployment throughout the Zagreb metropolitan area.

Osnovni podaci:

Pavle Ergović (Croatia), Dubravko Sabolić (Croatia), Filip Srdić (Croatia)
 

Prijava/Kotizacija:

Studentski popust se ne odnosi na studente doktorskog studija.

OBAVIJEST AUTORIMA: Uvjet za objavu rada je plaćanje najmanje jedne kotizacije po radu. Autorima 2 ili više radova, ukupna se kotizacija umanjuje za 10%.

Pavle Ergović
KSET
Fakultet elektrotehnike i računarstva
Unska 3
10000 Zagreb, Hrvatska

GSM: +385 99 417 3482
E-mail: pavleergovic@gmail.com

Najbolji radovi bit će nagrađeni.
Prihvaćeni radovi bit će objavljeni u zborniku radova s ISSN brojem. Radovi na engleskom jeziku prezentirani na skupu bit će poslani za uključenje u digitalnu bazu IEEE Xplore.

Opatija je vodeće je ljetovalište na istočnoj strani Jadrana i jedno od najpoznatijih na Mediteranu. Ovaj grad aristokratske arhitekture i stila već više od 180 godina privlači svjetski poznate umjetnike, političare, kraljeve, znanstvenike, sportaše, ali i poslovne ljude, bankare, menadžere i sve kojima Opatija nudi svoje brojne sadržaje. 

Opatija svojim gostima nudi brojne komforne hotele, odlične restorane, zabavne sadržaje, umjetničke festivale, vrhunske koncerte ozbiljne i zabavne glazbe, uređene plaže i brojne bazene i sve što je potrebno za ugodan boravak gostiju različitih afiniteta. 

U novije doba Opatija je jedan od najpoznatijih kongresnih gradova na Mediteranu, posebno prepoznatljiva po međunarodnim ICT skupovima MIPRO koji se u njoj održavaju od 1979. godine i koji redovito okupljaju preko tisuću sudionika iz četrdesetak zemalja. Ovi skupovi Opatiju promoviraju u nezaobilazan tehnološki, poslovni, obrazovni i znanstveni centar jugoistočne Europe i Europske unije općenito.

Detaljnije informacije se mogu potražiti na www.opatija.hr i www.visitopatija.com.