Research Seminar in Computer Science, Informatics and Mathematics
Istraživački seminar za računarstvo, informatiku i matematiku

Juraj Dobrila University of Pula
Organizers: Tihana Galinac Grbac, Darko Etinger, Neven Grbac

The purpose of this seminar is to foster collaboration and to focus and enhance efforts of the research groups at the Juraj Dobrila University of Pula, Croatia, in computer science, informatics, mathematics and related STEAM fields, through joint seminars with speakers from both academy and industry. It is meant as a platform for discussion and exchange of ideas among visiting scholars, industrial experts, local researchers and students.

Upcoming seminars

  • Comparison of two features (support for multiple inheritance and support for covariance without generic classes) in seven OOP languages
    Zlatko Sirotić, Istra Tech, Pula
    Thursday, June 20, 2024, at 10:30, room: FET-302
    Abstract. The lecture talks about two features of seven OOP languages: C++, Eiffel, Java, C#, Scala, Kotlin, Swift (chronological order, from the oldest to the youngest). All of these OOP languages have static type checking, which was one of the main criteria for their selection. First, very briefly, the origin of these languages is discussed. Then the possibility (or impossibility) of multiple inheritance in these languages is shown. After that, the (im)possibility of covariance without using generic classes is shown (generic classes are the general name for C++ templates, Eiffel generic classes, Java generics...).

Past seminars

  • Evolution of a software system architecture - explained in a model situation
    Csaba Szabó, Technical University of Košice, Slovakia
    Wednesday, May 29, 2024, at 09:00, room: FOOZ-25
    Abstract. The most important decision in the early stage of software development is the design of system architecture. In this talk, we set up a model situation and present such a decision. Then, evolution of the software is simulated. Since in parallel to the evolution of the system and its environment there is also a kind of evolution in the understanding of the underlying domain and problems by the software team, it is obvious that the initial architectural design becomes also a subject of evolution. Kinds and ways of this evolution are the main focus point of this talk. In our model situation, we evolve the system architecture of a web application from a client-server architecture with full single-technology stack to a microfrontends' architecture with distributed heterogenous technology stack. We simulate the evolutionary growth from custom application to software product line. We consider several architectural patterns and (one-way) transformations between them. We discuss the underlying topics to emphasise the importance of a proper architectural envisioning for software projects.

  • Leveraging Large Language Models for automated formative assessment generation in educational settings
    Nikola Tanković, Juraj Dobrila University of Pula
    Thursday, May 9, 2024, at 10:30, room: Rovinjska-RV
    Abstract. The recent Cambrian explosion of Large Language Models (LLMs) offers new possibilities for improving educational methods by generating automated formative assessments. This talk will explore the practical application of state-of-the-art LLMs like OpenAI's GPT models and Anthropic's Claude, paired with Retrieval-Augmented Generation (RAG), to create customized student assessments. Educators can use these models with Python and the Instructor framework to generate tailored quizzes that closely follow intended learning outcomes. The talk will demonstrate a Python-based implementation that uses these technologies, covering the development process and strategies for integrating AI-generated into assessments. We will also discuss implications and lessons learned from applying these tools in both offline and online classroom settings, highlighting potential improvements in student engagement and learning outcomes. The session aims to provide attendees with the technical know-how and practical insights needed to implement these technologies in their educational environments.

  • Collaborative perspectives in mathematical education: Development of the "geometric eye"
    Branka Antunović-Piton, Juraj Dobrila University of Pula
    Thursday, Mar 14, 2024, at 10:30, room: Rovinjska-RV
    Abstract. Click here for an extended abstract in English and Croatian.

  • RELYSOFT Reliable and safe complex software systems: From empirical principles to theoretical models in view of industrial applications
    Tihana Galinac Grbac, Juraj Dobrila University of Pula
    Thursday, Feb 8, 2024, at 10:00, room: FET-403
    Abstract. This is the final presentation of the research project HRZZ-IP-2019-04-4216 RELYSOFT Reliable and Safe Complex Software Systems: From Empirical Principles to Theoretical Models in View of Industrial Applications, 2020-2024, funded by the Croatian Science Foundation. Software is progressively entering every aspect of human life. The number of devices and platforms surrounding and supporting humans in every day activities is continuously growing. These devices and platforms are software-based and interconnected over Internet in the new 5G network, which is virtual, software defined and reconfigurable. Furthermore, there is a continuous growth of the number of software services, variety of software applications, and big data that need software processing. On the other hand, the global software market is highly competitive and dynamic, which introduces challenges for software development companies. In such conditions, current software and system engineering knowledge is inadequate to cope with requirements on reliability and safety system properties. New theories and knowledge are needed. In this project we integrate the software and system engineering research with telecommunication and automotive industry background, and with the aid of mathematical knowledge, to yield new knowledge needed for engineering software system reliability and safety in global competitive market. In this project we focus our efforts in the following three branches of research: 1. Fault behavior in evolving software systems (reliability modeling, fault distribution and fault prediction), 2. Reliability and safety in dynamic environments (Cloud, software defined networks and global market), 3. Software development organization aspects.

  • Application of artificial intelligence in medicine: Trends and new perspectives
    Ivan Lorencin, Istrian University of Applied Sciences
    Thursday, May 4, 2023, at 10:00, room: FET-302
    Abstract. Click here for an extended abstract in English and Croatian.

  • Decentralized training of machine learning models
    Romeo Šajina, Juraj Dobrila University of Pula
    Thursday, Mar 23, 2023, at 10:00, room: FET-302
    Abstract. In the field of sports, computer vision can be used to detect and track players or the ball, detect player actions, detect objective score change, player pose estimation, etc. In this work, we will focus on action evaluation, i.e. comparison of poses while the player is executing certain action (e.g. handball shot) against the template sequence performed by a professional. By comparing player poses to the template poses we can provide them with the information of the needed corrections to improve their action execution. This application can be especially useful to beginners in the sport or in rehabilitation exercises purposes. Comparison of poses is the final step in action evaluation which requires that we have collected players' poses in 3D space. Previous research on action evaluation mostly used markers and sensors to capture players' poses that provide accurate poses, but require a controlled environment that is often expensive and time-consuming. For that reason, a system based on a monocular camera would be more applicable in real world scenarios. In that environment, obtaining accurate players' poses becomes a challenging task. To tackle this task we use a 2D and 2D-to-3D pose-estimation method to produce poses in 3D space. Given that the distance of the player from the camera can vary, we need to standardize pose sequences before comparison. The final step before the comparison is sequence alignment, a task where the two sequences are aligned so that actions are performed at the same time in both sequences. Once the sequences are aligned we can inspect each frame and visualize the difference between poses. Further analysis can be made on aligned sequences, such as a comparison of execution speeds for a particular action primitive, action style recognition, etc.

  • Box dimension of stuffed sets
    Siniša Miličić, Juraj Dobrila University of Pula
    Thursday, Mar 2, 2023, at 10:00, room: FET-302
    Abstract. Box dimension (box-counting dimension, Minkowski-Bouligand dimension) is one of the tools for the study of complex geometric properties of sets in metric spaces, with consequences in different parts of mathematics. The collections of sets of special interest are those for which non-trivial phenomena are detected by the box dimension, but not by the other fractal dimensions that have nice theoretical properties (Hausdorff, affine, inductive, Lebesgue dimension...). Stuffed sets form such a collecction, and the goal of the seminar is, besides an introduction to the fractal analysis, to exhibit previous and current results in the direction of a more general algorithm for computing these dimensions.

  • A survey of end-to-end congestion mechanisms in the field of IoT
    Dalibor Fonović, Juraj Dobrila University of Pula
    Thursday, Feb 16, 2023, at 10:00, room: R-1.14
    Abstract. The Internet of Things (IoT) is a global network of devices capable of communicating and exchanging data with other devices and systems mainly via the Internet using specific communication protocols. The number of such devices that can be connected to the Internet due to progress in various technological areas (such as communication technologies, microelectronic circuits, sensors, embedded systems, smartphones...) is constantly growing. This leads to network congestion due to the large amount of data exchanged between devices. In addition, IoT devices are resource-constrained, further exacerbating network congestion. Network congestion leads to additional communication delay, low network bandwidth and waste of computer resources. This is one of the important aspects when some IoT applications exchange critical information (e.g., monitoring the patient's health condition in the application of IoT in smart healthcare). As the number of IoT applications increases, so does the need to modify or introduce new protocols to deal with the problems of adapting to network conditions. In the IoT layered architecture, the transport layer plays an important role in managing the end-to-end connection to the services of the upper application IoT layer. An important function of the transmission layer of the IoT layer is congestion control. This talk presents an overview of related research on the congestion control mechanism of the IoT architecture. transport layer, advantages, disadvantages, and current problems of the transport layer in IoT applications.

  • Statistical analysis of the tail behaviour of dependent sequences
    Darko Brborović, Juraj Dobrila University of Pula and University of Zagreb
    Thursday, Feb 2, 2023, at 10:00, room: R-1.14
    Abstract. In this lecture we present some results regarding the tail behaviour of bivariate random sequences, including the stationary M-dependent sequences. Statistical analysis of the asymptotic tail behaviour of random sequences is often done within the theory of point processes. However, if a larger number of tail events is available it is possible to obtain convergence results towards normal distribution, which is the case for the results that will be presented. Specifically, we present and analyse the permutation test of tail dependence for independent (or tail independent) and identically distributed bivariate random vectors.

  • Decentralized training of machine learning models
    Robert Šajina, Juraj Dobrila University of Pula
    Thursday, Jan 12, 2023, at 10:00, room: R-1.14
    Abstract. Federated Learning is a new paradigm that enables the training orchestration of machine learning models using decentralized data. Decentralized data is private user data that must not leave the edge device (e.g. mobile phone) and can not be used for standard centralized learning where all data is available in one central location. Federated Learning uses a central server that trains a global model by iteratively selecting subsets of decentralized agents that train the model on local data. Then these changes are aggregated into a global model. However, the central server can be a communication bottleneck and can "discover" private data through specific mechanisms. By eliminating the central server, decentralized devices are forced to collaborate in a peer-to-peer network. One of the significant problems in peer-to-peer networks is synchronization, especially when training on heterogeneous data. Due to the heterogeneity of data from decentralized devices, a global solution (model) is often not optimal. New personalization techniques have emerged that, through certain methods, adapt the model for a specific decentralized device (user). Our developed personalization technique uses Batch Normalization layers in deep learning models and the early stopping method to personalize the model for each device. We evaluated this approach on data collected at the user level, specifically on user comments from the Reddit and StackOverflow websites. Our approach achieved a relatively increased accuracy of 26% for Reddit and 20% for the StackOverflow dataset.

  • Bringing social robots to human public environments
    Dražen Brščić, Kyoto University, Japan
    Friday, Dec 23, 2022, at 10:00, room: R-1.2
    Abstract. The talk will discuss the challenges of introducing socially interactive robots into human everyday environments. The presenter will talk about some of the past works that he and his colleagues have been doing, focusing on the observing and modeling of human behavior and the social interaction between robots and humans in public spaces.

  • Langlands program as the Grand Unified Theory of mathematics
    Neven Grbac, Juraj Dobrila University of Pula
    Thursday, Dec 22, 2022, at 10:00, room: R-1.14
    Abstract. Number theory exists almost since the dawn of human conciousness. Modern number theory is the deepest area of mathematics, which serves as the meeting ground and the place of interplay of all three traditionally main parts of mathematics: algebra, mathematical analysis and geometry. Its most prominent problems, such as the Fermat Last Theorem, which is proved by now, or the Riemann Hypothesis, the Birch and Swinnerton-Dyer Conjecture, the Ramanujan and Selberg Conjecture, the abc Conjecture, which are still open, are the source of inspiration and motivation of generations of mathematicians over the centuries, they motivate and guide the development of entire mathematics, and attract the attention of wider public audience. The importance of number theory is testified by the list of laureates of the Fields Medal and Abel Prize, and the fact that three out of seven Millenium Problems of the Clay Institute, whose solution comes with a million dollar prize, are related to number theory. Nowadays, with the rise of computers and communication technologies, number theory has found important applications in criptography, coding theory and internet security. The Langlands program is one of the central research fields in number theory, but also the mathematics as a whole. It is the complex system of conjectures and expected correcpondences which predict a deep functorial (algebraic) interconnection between number theory (Galois representations), mathematical analysis (automorphic and modular forms) and geometry (elliptic curves, Shimura varieties, cohomology).

  • Scientific directions for (non)traditional software engineering researcher(s)
    Vaidas Giedrimas, Panevezys University of Applied Sciences and Vilnius University Siauliai Academy, Lithuania
    Friday, Oct 7, 2022, at 12:30, room: Inf-44
    Abstract. The talk proposes several scientific directions for current and future research in the field of software engineering. In particular, the scientific areas of automated development of distributed systems, advanced methods for software engineering and scientific software development are discussed.