Search  English (United States) Hrvatski (Hrvatska)

innovative promotional partnershipFull paper submission deadline: Monday, February 12, 2024

Technical co-sponsorship

Hybrid Event
Event program
Thursday, 5/25/2023 9:00 AM - 12:30 PM,
Camelia 1, Grand hotel Adriatic, Opatija
9:00 AM - 10:45 AMArtificial Intelligence Theory
Chair: Alan Jović 
1.G. Oparin, V. Bogdanova, A. Pashinin (Institute for System Dynamics and Control Theory of SB RAS, Irkutsk, Russian Federation)
Binary Dynamic Models of Structural Synthesis of Programs 
In this paper, we propose a matrix-vector Boolean differential model for constructing plans of computational actions in solving non-procedural problems on a computational model of a modular software system. The conditions of the problem of structural synthesis of a program from pre-implemented modules are formulated as a system of Boolean differential equations, the solutions of which, under given initial conditions (consistent with the non-procedural formulation of the problem), determine the solvability of the problem and give constructive plans for its solution (including parallel ones). The proposed method of structural synthesis is focused on high-dimensional models. It allows highly efficient software implementation due to the high parallelism of performing vector-matrix operations on binary vectors at the level of machine instructions. The developed model is used for planning computations in packages of applied microservices based on the HPCSOMAS-MSC platform.
2.D. Tuličić (College for Information Technologies, Zagreb, Croatia), N. Ivković (Faculty of Organization and Informatics, Varaždin, Croatia)
The Concept of Cognition as Categorization in the Development of New Metaheuristics and Algorithms Inspired by Nature 
The sensorimotor system is a crucial mechanism that enables organisms to categorize information in their environment, a skill necessary for survival. According to cognitive scientist Steven Harnad's "To Cognize is to Categorize: Cognition is Categorization", all living things are essentially sensorimotor systems. This approach is proposed to serve as a novel concept for the development of algorithms inspired by swarm intelligence. Many authors of such algorithms have drawn inspiration from animal species, attempting to emulate their abilities to interact with and navigate their environment. However, these abilities are essentially rooted in the organisms' categorization skills, which are a fundamental characteristic of all sensorimotor systems. Therefore, when designing new algorithms and metaheuristics, it is unnecessary to seek out new species with specific behaviors; instead, the entire sensorimotor system can be observed, and its sensor capabilities can be incorporated even if they do not exist in nature. In order to justify this, abstraction was used as a way to demonstrate the creation of concepts. The paper offers a new perspective on the development of swarm intelligence algorithms and promotes a holistic approach to algorithm design based on Steven Harnad's viewpoint of sensory motor systems.
3.V. Kondratiev (ITMO University, St. Petersburg, Russian Federation), I. Otpuschennikov (Matrosov Institute for System Dynamics and Control Theory of the Siberian Branch of the RAS, Irkutsk, Russian Federation), A. Semenov (ITMO University, St. Petersburg, Russian Federation)
Speeding up the Solving of Logical Equivalence Checking Problems with Disjunctive Diagrams 
In the context of the problem of checking the equivalence of Boolean circuits (LEC), we propose an approach that increases the efficiency of modern SAT solvers on this problem by generating additional constraints of a special kind. To generate such constraints we use a variant of decision diagrams called disjunctive diagrams. In contrast to well-known Binary Decision Diagrams these diagrams can be constructed effectively for an original formula and can also be used to extend the original CNF formula with new clauses which are its logical consequences. In computational experiments, we show that the resulting formulas, encoding difficult LEC variants, extended by the generated constraints are often significantly easier to solve for state-of-the-art SAT solvers compared to the original formulas.
4.S. Kochemazov, V. Kondratiev, I. Gribanova (ISDCT SB RAS, Irkutsk, Russian Federation)
Empirical Analysis of the RC2 MaxSAT Algorithm 
The Boolean satisfiability problem (SAT) and maximum satisfiability problem (MaxSAT) are among the most well-known combinatorial problems in today's computer science. The algorithms for their solving also go hand-in-hand, in that most MaxSAT solvers employ SAT solvers as the so-called oracles. In the present paper we perform a computational study of the RC2 algorithm, which is among the best state-of-the-art algorithms for MaxSAT solving. We view it from the SAT oracle viewpoint and consider how the SAT oracle's runtime is distributed among RC2 procedures and heuristics, how this statistics differs depending on the SAT solver employed as an oracle, devise and test the procedure that can be used to evaluate a standalone SAT solver compatibility with RC2.
5.K. Poje, M. Brčić, M. Kovač, D. Krleža (Faculty of Electrical Engineering and Computing, Zagreb, Croatia)
Challenges in Collective Intelligence: A Survey 
Collective intelligence and prediction markets have been gaining attention as a powerful tool for decision making and forecasting. Prediction markets are designed to aggregate the opinions and knowledge of a diverse group of participants in order to make more accurate predictions than any single person could make alone. They are based on the principle that the wisdom of the crowd is often superior to the wisdom of any individual. This paper summarizes the potential challenges and failures of prediction markets as a tool for collective intelligence. These include several issues which can lead to poor decision making. Additionally, poor communication and a lack of understanding and agreement among group members can also lead to failure of prediction markets. To improve the effectiveness of prediction markets, it is crucial to be aware of these factors and take steps to address them. This includes fostering open communication, encouraging participation from all group members, and promoting a culture of creativity and innovation.
6.M. Smirnov (ITMO University, Saint Petersburg, Russian Federation), S. Kochemazov (ISDCT SB RAS, Irkutsk, Russian Federation), A. Semenov (ITMO University, Saint Petersburg, Russian Federation)
The Study of the Target Set Selection Problem under Deterministic Linear Threshold Model Using Evolutionary Algorithms 
We carry out the computational study of the well-known Target Set Selection (TSS) problem related to activation dynamics in networks. To the best of our knowledge, there is a limited set of computational algorithms for TSS presented in the literature so far. The main novelty of our research lies in the fact that we apply evolutionary algorithms to solve TSS in networks of large size. We compare the developed algorithms with other known approaches, including greedy algorithms and algorithms based on the local search principles.
7.E. Otović, J. Lerga, D. Kalafatovic, G. Mauša (Center for Artificial Intelligence and Cybersecurity, University of Rijeka, Rijeka, Croatia)
Neuroevolution for the Sustainable Evolution of Neural Networks 
The predictive performance of a neural network depends on its weights and architecture. Optimizers based on gradient descent are most commonly used to optimize the weights, and grid search is utilized to find the most suitable architecture from the list of predefined architectures. On the other hand, neuroevolution offers a solution for the simultaneous growth of neural network architecture and the evolution of its weights. Thus, it is not limited by the user-defined list of possible architectures and can find configurations optimal for a specific task. Both approaches can be effectively parallelized and take advantage of modern multi-process systems. In this research, we compare neuroevolution and backpropagation in terms of the time consumed by the algorithm, the predictive performance of the neural network, and the complexity of the neural network. The total time for each algorithm is measured along with the times for each section of the algorithm and the time spent on synchronization due to the multi-process setting. The neural networks are compared by their predictive performance in terms of Matthews correlation coefficient score and their complexity as the number of nodes and connections. The case study is based on two synthetic and two real-world datasets for classification tasks in a 5-fold cross-validation.
10:45 AM - 11:15 AMBreak 
11:15 AM - 12:30 PMImage and Video Processing and Analysis
Chair: Marina Ivašić-Kos 
1.I. Dorkić, M. Brisinello (TTTech Auto Central and Eastern Europe, Osijek, Croatia), R. Grbić, M. Herceg (Faculty of Electrical Engineering, Computer Science and Information Technology Osijek, Osijek, Croatia)
Influence of Quality of Pixel Level Annotations on Text Detection Performance in Natural Images 
Text detection in natural images is a task that arises in many computer vision applications. State-of-the-art text detection methods are mainly based on deep neural networks designed for instance segmentation task. However, most of the available datasets for text detection do not have fine annotations at the pixel level which are required during supervised learning of such networks. Usually, a whole or reduced text bounding box is used as a segmentation mask. In this paper, a method that generates a synthetic dataset with precise annotations at the pixel level is proposed. The method is based on the available Synthtext script for generating synthetic datasets with text instances. By creating synthetic datasets with precise and imprecise annotations at the pixel level we explore the efficiency of the state-of-the-art text detector TextFuseNet.
2.V. Diklic (dSPACE d.o.o., Đakovo, Croatia), F. Matković (Faculty of Electrical Engineering and Computing, Zagreb, Croatia), K. Pardon (dSPACE d.o.o., Zagreb, Croatia)
Effects of Applying Identified Road Lane Lines on Vehicle Autopilot Model Driving Performance 
Vehicle autonomous driving system is one of the most actively developed system in automotive industry. Producing a reliable autonomous driving system is complex and multistage problem. The road lanes are a traffic sing with high level of semantics that determine how vehicle should move, what actions are allowed to be performed, etc. Comparing different approaches of identifying road lane lines and analysing the implication on vehicle autopilot driving performance is the topic of this manuscript. The lane lines are identified by neural network learned by processing images with previously marked lane lines from CULane dataset. Vehicle autopilot driving performance is analysed using autopilot model running on Euro Track Simulator. Input to neural network model are captured screen frame from simulator, while vehicle controlling is limited to steering wheel angle to simplify problem of autonomous driving and it is implemented through virtual steering wheel. Proposed models are analysed and compared on multiple driving and weather environments including highway, off-road and urban driving during day, night and rain conditions. The lane line recognition (classified by U-Net neural network model from images) is compared with additional control and path planning weather it leads to better autonomous driving performance and smaller driving system model.
3.D. Potoč, D. Petrinović (Faculty of Electrical Engineering and Computing, Zagreb, Croatia)
Estimating a Nonradial Vignetting Shape 
Vignetting is a phenomenon characterized by a decrease in illumination towards the edges of an image. This effect is typically represented by a radially symmetrical model, however, this paper aims to demonstrate a non-radial model of vignetting and estimate its shape. To accomplish this, a synthetic image was created and the angular vignetting shape has been modeled as a sum of harmonics. The magnitudes and amplitudes of these harmonics were obtained and used to construct the desired angular vignetting shape. Once the synthetic image with the modeled vignetting shape and added noise was created, it was used as input into a function for vignetting estimation. Also, the inputs have been a fixed vignetting center and different initial values of harmonics’ magnitudes and phases. With that inputs, despite the level of the noise, we have successfully estimated vignetting function by non-linear optimization. The function has attempted to determine the original harmonics’ used to create the vignetting angular shape. When the vignetting model is calculated, we removed it in order to get a homogeneous image. While it may be difficult to obtain the exact original values of the harmonics’, the shape can be estimated with a high level of accuracy. The paper shows that highly accurate models can be estimated for a lower number of angular harmonics, with a residual gain error standard deviation of less than 0.03%. Even in the presence of 5dB noise in the images, the gain error standard deviation remains below 3%, as long as proper parameter initialization is performed prior to optimization.
4.M. Ferenčak, P. Grd, I. Tomičić (Fakultet organizacije i informatike, Varaždin, Croatia)
The Impact of Image Processing on Perceptual Hash Values 
In the last few decades image processing has become the focus of research in different fields. It can be used to extract information from images, improve their quality and make it easier for computers to understand them. Some of the challenges that pose a great problem for working with images are detecting modified images or detecting similar images. In order to address those challenges, image hashing can be used. Hashing is calculating a digest value from images and perceptual hash algorithms are a type of hash algorithms with the main idea that similar data has similar hash values, which means that the hash values remain approximately the same if the content is not significantly modified. This paper will compare different perceptual hash algorithms, apply them to differently modified images and analyse the impact of those modifications on perceptual hash values provided by different algorithms. The obtained values will be compared and advantages and disadvantages of different algorithms will be discussed.
5.G. Oreški, S. Aničić (Sveučilište Jurja Dobrile u Puli, Pula, Croatia)
Usporedba funkcija gubitka za semantičku segmentaciju objekata u prometu 
Autonomna vožnja postaje sve važniji čimbenik u svakodnevnom životu, čime raste značaj dubokog učenja na kojem se takvi sustavi temelje. Posljednjih godina istraživanja, predložene su mnoge funkcije gubitka za problem semantičke segmentacije, čiji odabir ima direktan utjecaj na sposobnost modela da uči i uspješno generira predikcije. Ovo istraživanje analizira utjecaj različitih funkcija gubitka na performanse semantičke segmentacije na 22 različita objekta prisutna u prometu. Istraživanje je provedeno na skupu podataka prikupljenom na CARLA simulatoru prometa u urbanom okruženju. U radu ćemo obuhvatiti sve najčešće korištene funkcije gubitka kao što su: Binary Cross Entropy, Dice loss, Focal loss, Bounary loss, te pod-varijante, da bi analizirali njihov utjecaj na konvergenciju modela. Cilj rada je donijeti preporuke prikladnosti pojedinih funkcija na problem semantičke segmentacije objekata u prometu.
Thursday, 5/25/2023 3:00 PM - 7:00 PM,
Camelia 1, Grand hotel Adriatic, Opatija
3:00 PM - 5:15 PMNatural Language Processing
Chair: Darko Huljenić 
1.S. Lovrenčić (Fakultet organizacije i informatike Sveučilišta u Zagrebu, Varaždin, Croatia)
The Role of Knowledge Management in Transition to Industry 5.0 
Fifth industrial revolution, or Industry 5.0 is gaining a momentum in last several years, additionally enhanced with many disruptive events in global economy. Its main keywords are sustainable, human-centric and resilient, and it should bring these benefits to industry and society. Knowledge management enables the cycle of knowledge through organization and is a human oriented dicipline that continually embraces new technologies, helping organizations to achieve their goals. This paper explains that sustainability, human-centricity and resilience are naturally supported by knowledge management. Investigation of connection of elements of Industry 5.0 action plan with knowledge management processes and technologies shows that knowledge management has an important role in achieving Industry 5.0, especially through processes of knowledge creation/acquisition and storing and supporting artificial intelligence technologies.
2.B. Penkova, M. Mitreska, K. Ristov, K. Mishev, M. Simjanoska (iReason LLC, Skopje, Macedonia)
Learning Translation Model to Translate Croatian Dialects to Modern Croatian Language 
The task of translating dialects into modern language is a challenging task since it requires enormous parallel data. Such data is hard to find especially when it comes to low-resource languages. Among them is the Croatian language which has very few datasets in the standardized version, let alone enough resources for its dialects. In the Croatian language, there are three main groups depending on the geographical position Shtokavian, Kajkavian, and Chakavian which also include other more specific local dialects. For solving these kinds of problems, unsupervised neural machine translation (UNMT) models are considered a good solution since they can be trained on monolingual data. In this paper, we propose an application of a modified version of the state-of-the-art UNMT model for dialect translation on monolingual data of the standardized Croatian language and one of the dialects. We experimented with several types of cross-lingual embeddings of the input data to determine the best approach that can leverage the similarities and differences between the language and the dialect. All techniques are evaluated on a small parallel dataset using the BLEU metric. Translating these dialects to the modern Croatian language helps in improving communication and access to information for all speakers.
3.A. Poleksić, S. Martinčić-Ipšić (Faculty of Informatics and Digital Technologies, Rijeka, Croatia)
Sentiment of the Tweets on Russo-Ukrainian War: The Social Network Analysis 
This paper presents the analysis of social network posts using standard natural language processing (NLP) methods and undirected graph representations. The Twitter data used in this work is based on two keywords: "Ukraine" and "Russia" in May, October, November and December 2022. After standard pre-processing of the raw data and sentiment classification using "Valence Aware Dictionary and sEntiment Reasoner" (VADER), we proceed to the construction of a weighted graph. We construct the network with hashtags as nodes and hashtag co-occurrences are modelled as edges. This representation enables easier topic extraction and community detection. Combining sentiment classification with simple graph structures gives us a deeper insight into public opinion on the Russian-Ukrainian war.
4.D. Ševerdija, T. Prusina, A. Jovanović, L. Borozan, J. Maltar, D. Matijević (Department of Mathematics, University J. J. Strossmayer of Osijek, Osijek, Croatia)
Compressing Sentence Representation with Maximum Coding Rate Reduction 
In most natural language inference problems, sentence representation is needed for semantic retrieval tasks. In recent years, pre-trained large language models have been quite effective for computing such representations. These models produce high-dimensional sentence embeddings. An evident performance gap between large and small models exists in practice. Hence, due to space and time hardware limitations, there is a need to attain comparable results when using the smaller model, which is usually a distilled version of the large language model. In this paper, we assess the model distillation of the sentence representation model SentenceBERT by augmenting the pre-trained distilled model with a projection layer additionally learned on the Maximum Coding Rate Reduction (MCR2 ) objective, a novel approach developed for general purpose manifold clustering. We demonstrate that the new language model with reduced complexity and sentence embedding size can achieve comparable results on semantic retrieval benchmarks
5.A. Kitanovski, M. Toshevska, G. Mirceva (Faculty of Computer Science and Engineering, Skopje, Macedonia)
DistilBERT and RoBERTa Models for Identification of Fake News 
The proliferation of fake news has become a significant issue in today’s society, affecting the public’s perception of current events and causing harm to individuals and organizations. Therefore, the need for automated systems that can identify and flag fake news is critical. This paper presents a study on the effectiveness of DistilBERT and RoBERTa, two state-of-the-art language models, for detecting fake news. In this study, we trained both models on a dataset of labelled news articles and evaluated them on two different datasets, comparing their performance in terms of accuracy, precision, recall and F1-score. The results of our experiments show that both models perform well in detecting fake news, with RoBERTa model achieving slightly better results in overall. Our study highlights the ability of these models to effectively identify fake news and help combat misinformation.
6.J. Katalinić (Faculty of Humanities and Social Sciences, Zagreb, Croatia)
Analysis of Pro-Russian Tweets during Russian Invasion of Ukraine 
With the Russian invasion of Ukraine on February 24, 2022 a new aspect of the conflict opens up, the fight on social networks. By analyzing the most influential pro-Russian Twitter profiles that cover the daily events and impacts of the Russian invasion of Ukraine as well as outside it, the number of followers, people following as well as tweets, time series comparison of publishing, sentiment through polarity and subjectivity, and the classification of individual tweets as malicious by building an SVM (Support Vector Machine) classifier were collected and analyzed. For the purpose of training the SVM classifier, a sample from dataset of 3 million identified malicious tweets deleted by Twitter after being linked to the Russian IRA agency known for operating malicious user accounts, available through the Kaggle data science community, was used. The results of the work show that pro-Russian Twitter profiles have clear and defined influence operations that oscillate through different time periods as a reaction to the dynamics and development of the conflict, and thus certain events become narrative elements that can influence the emotions of the target audience.
7.I. Hrga (Sveučilište u Rijeci, Rijeka, Croatia)
Augmentacija podataka za klasifikaciju kratkih tekstova 
Augmentacija podataka postala je neizostavni korak u procesu učenja većine sustava baziranih na dubokim neuronskim mrežama. Velike količine podataka potrebne da bi sustav uspješno naučio rješavati zadatke često je teško pribaviti u dovoljnoj količini i kvaliteti. Zbog toga se pribjegava različitim postupcima kojima se primjenom raznovrsnih transformacija može višestruko povećati osnovni skup podataka. U području računalog vida već su uvriježene brojne tehnike koje relativno lako transformiraju sliku bez da se promijeni što ta slika predstavlja. Međutim, u području procesiranja prirodnog jezika takve su transformacije znatno manje zastupljene. Već i promjena samo jednog slova neke riječi može znatno promijeniti smisao rečenice, stoga je potrebno puno više opreza prilikom odabira pogodnih transformacija. U ovom radu analiziramo postupke za augementaciju podataka prikladnih za rad s tekstom. Dajemo pregled postupaka temeljenih na promjenama pojedinih znakova, zamjenama riječi baziranih na WordNet sinonimima te tehnika koje stvaraju reprezentaciju teksta u vektorskom prostoru. Uz prikaz njihovih prednosti i mana, vršimo i praktičnu usporedbu te analiziramo kako se promjena pojedine tehnike odražava na rezultate klasifikacije kratkih tekstova.
8.M. Mitreska, K. Zdravkova (University Ss Cyril and Methodius, Skopje, Macedonia)
Syllable and Morpheme Segmentation of Macedonian Language 
Communication is the key to human development. Approximately 5% of the world’s population experience some form of hearing disability. Modern assistive devices and technologies can improve the communication skills of hearing impaired people by transcribing the speech into text. The creation of such an application depends on the language specific morphosyntacic properties. It usually starts with the syllabification. The research presented in this paper focused on the development of an automatic system for rule-based and sonority-based syllable and morpheme segmentation of Macedonian language, which can be easily incorporated into an efficient speech recognition system. The segmentation rules for breaking the words down into syllables and into morphemes were created according to the new orthography of the Macedonian language. For the sonority-based approach, a novel phonological distance measure was introduced capable of efficient syllable clustering. The implementation of the framework is developed in Python using several data structures for optimized performance and CPU usage. Both segmentation strategies were evaluated using the electronic lexicon consisting of more than one million words. A linguistic expert was also consulted during the entire process. The consistency of the obtained results promises their sustainability for further speech processing applications.
9.M. Pajas, A. Radovan (BISS Ltd., Zagreb, Croatia), I. Ogrizek Biškupić (Visoko učilište Algebra, Zagreb, Croatia)
Multilingual Named Entity Recognition Solution for Optimizing Parcel Delivery in Online Commerce: Identifying Person and Organization Names 
This paper presents a comprehensive solution to enhance parcel delivery in online commerce by implementing multilingual named entity recognition. The solution is designed to accurately identify person and organization names, with a primary emphasis on correctly identifying recipients. The ultimate goal is to use this information to automatically validate recipients and select the most accurate one to improve data accuracy and reliability for parcel delivery. The process begins by collecting a large dataset of online commerce data, including customer search queries, and annotating it with person and organization names. The data is then preprocessed, cleaned to eliminate irrelevant information, and prepared for training a named entity recognition model. Next, the model is trained and evaluated using this data to ensure its ability to identify named entities and extract recipients from queries accurately. The process employs an iterative training process and data generation techniques, while also addressing the issue of noisy data and iterative training introducing unwanted patterns by retraining the model on the subset of the original annotated dataset. Our experiments conclude a consistent increase of F1 score over the baseline and best iteration using this method of training and fine-tuning.
5:15 PM - 5:45 PMBreak 
5:45 PM - 6:45 PMMachine Learning Applications
Chair: Alan Jović 
1.J. Lehtonen, T. Aaltonen (SAMK, Pori, Finland)
Using Generated LIBS Data as a Base for Neural Network Architecture Development 
Laser-Induced Breakdown Spectroscopy (LIBS) is a rapid atomic spectroscopy technique used to measure the concentration of elements in samples. This research paper uses supervised deep learning with a LIBS database provided by National Institute of Standards and Technology (NIST) to find out what kind of neural network architecture can best predict element concentrations from metal alloy samples. Accuracy of predictions was used to evaluate the network architectures. Because collecting and storing real samples of metal alloys with a wide range of element concentrations requires a lot of resources, network’s ability to predict outside of training set’s range of concentrations was also tested. The goal of the research was to find out if NIST database together with machine learning can be used to reduce the amount of real alloy samples needed for LIBS calibration.
2.S. Goluža, T. Bauman, T. Kovačević, Z. Kostanjčar (Faculty of Electrical Engineering and Computing, Zagreb, Croatia)
Imitation Learning for Financial Applications 
Algorithms for solving decision-making problems under uncertainty are often employed in complex and unstructured environments such as financial markets. Reinforcement learning (RL), a mathematical framework for sequential decision-making, is utilized in many financial decision-making problems, such as portfolio optimization, optimal execution, and market making. RL methods require manual engineering by providing the reward function, which the agent uses as a feedback mechanism. In such noisy environments, the financial decision-making problem is often easier to approach by demonstrating the desired behavior rather than manually engineering it. Imitation learning (IL) algorithms extract knowledge from expert's demonstrations by directly replicating the desired behavior or by learning the expert's reward function. This paper aims to introduce the background and recent advances in the field, present the differences between IL and more familiar frameworks like supervised learning and reinforcement learning, and provide a survey of the current state-of-the-art applications of IL in finance.
3.N. Rodin, D. Pinčić, K. Lenac, D. Sušanj (Faculty of Engineering, Rijeka, Croatia)
The Comparison of Different Feature Extraction Methods in Musical Instrument Classification 
In this paper, we analyze four different methods for audio feature extraction and compare their efficiency in the context of musical instrument classification. We study spectrograms, Mel spectrograms, Linear-Frequency Cepstral Coefficients (LFCCs) and Mel-Frequency Cepstral Coefficients (MFCCs) in combination with three different Deep Learning architectures: VGG16, ResNet34 and a custom CNN. We investigate the behavior of our models in two different classification scenarios to determine a possible correlation between the number of classes and the efficiency of each method. For this purpose, we took samples from the London Philharmonic Orchestra dataset and ran the experiment for three and fifteen classes of musical instruments belonging to three different instrument families: Woodwinds, Strings and Brass.
Friday, 5/26/2023 9:00 AM - 10:30 AM,
Camelia 1, Grand hotel Adriatic, Opatija
9:00 AM - 10:30 AMOther Topics
Chair: Marko Horvat 
1.K. Šiber Makar (IN2 d.o.o., Zagreb, Croatia)
Driven by Artificial Intelligence (AI) – Improving Operational Efficiency and Competitiveness in Business 
Artificial intelligence has already become a necessary factor of advanced digitalization of companies and is one of the main prerequisites for global competitiveness and highly efficient operations. With cloud computing wider usage, decreased costs of AI technology and openAI, advanced AI technology and algorithms have become easily accessible to everyday users and even small to mid-size companies, which has widely extended its application. Based on the relevant research studies, this paper analyzes areas where AI can provide added value with its application in different industries like healthcare, retail, telecommunications, manufacturing, etc. Hence, an overview on different areas of AI implementation potential, like marketing and sales, product and service development, finance, supply chain management, customer experience, maintenance, quality control, services’ and products’ personalization, etc. and its benefits and challenges will be analyzed in detail. Main focus of this paper is on the organizations that have already started digitalizing their business from end to end and have exact benefits from applying AI by increasing revenues and decreasing costs.
2.M. Horvat (Faculty of Electrical Engineering and Computing, Zagreb, Croatia), P. Jerčić (Graz University of Technology, Institute of Interactive Systems and Data Science, Graz, Austria)
A Survey on Usage of Multimedia Databases for Emotion Elicitation: A Quantitative Report on How Content Diversity Can Improve Performance 
Affective picture databases provide a standardized set of images to elicit controlled and consistent emotional responses in research participants. They are a valuable tool for studying various emotion-related phenomena across several research domains. These domains include emotion perception, emotion regulation, and the neural basis of emotion. However, affective picture databases have diverse schemas, structures, and content that make them difficult to use. Searching and retrieving optimal pictures relevant to affective stimulation may be challenging and time-consuming. In this context, we surveyed domain experts about their practices and experiences working with affective multimedia databases such as IAPS, NAPS, OASIS, GAPED, and others. The survey identified a need for novel data observatory software. This finding motivates the authors' intention to develop and validate such software platform that relies on AI. Such a platform would describe better, retrieve, and integrate various semi-structured affective multimedia datasets. The results prominently indicate the overwhelming dissatisfaction regarding stimuli content diversity and cultural bias, specifically regarding emotional and semantic context. This survey follows up on a similar survey conducted ten years ago and explores the differences in researchers' opinions and experiences during that time. The complete aggregated results are publicly available at
3.N. Boudjani, V. Colas (SogetiLabs, Paris, France), C. Joubert (Capgemini Engineering, Paris, France), D. Ben Amor (SogetiLabs, Paris, France)
AI Chatbot for Job Interview 
In this paper, we propose an interactive AI chatbot in French language which able to ask questions to a potential candidate, to detect incomplete answers and to ask additional questions in order to obtain a complete answer to a given question. Additionally, the proposed chatbot allows an interactive environment where candidates can also ask questions during the interview. We also propose a system of interview based on an exhaustive behavioral diagram. All the chatbot’s functionalities have been validated by experimentation. The results show that our chatbot is complete in terms of the questions asked and information to be collected during the interview. Moreover, experimentation has shown that the diagram covers all cases of scenarios between an interviewer and a candidate during a job interview.
4.T. Aaltonen (SAMK, Pori, Finland)
Consumer Class Side Scanning Sonar Dataset for Human Detection 
One of the leading problems for using modern deep neural networks with sonars, is the lack of datasets, even more rare are datasets that are collected with consumer class sonars. This paper introduces a novel side scanning sonar dataset for humans under water. Data is collected with consumer class Garmin 8400 XSV sonar with GT54UHD-TM transducer. Dataset is collected in shallow coastal water of the Baltic Sea, near Rauma Finland. Dataset contains 331 images of human, and 364 images with other objects like tires and rocks. Dataset contains cropped images from objects, and full resolution pictures. Data is collected from two different locations, with different sonar settings. All images are from two rescue divers in the bottom of the sea. This Paper also introduces standard data split for collected dataset for training, validation, and test data for benchmarking different models with dataset, and the data collection system based on ROS.
5.A. Đukić, R. Bjelošević, M. Stojčić (Faculty of Transport and Traffic Engineering, Doboj, Bosnia and Herzegovina), M. Banjanin (Faculty of Philosophy, Pale, Bosnia and Herzegovina)
Network Model of Multiagent Communication of Traffic Inspection for Supervision and Control of Passenger Transportation in Road and City Traffic 
As a rule, the supervision and control of the transportation of persons in different types of traffic is the responsibility of the Traffic Inspection of the state inspectorates of individual countries. This paper investigates the intelligent supervision and control of the transportation of persons in road and city traffic managed by the Traffic Inspection in the geo-space of the Republic of Srpska (RS). The aim of the research is to create a multiagent communication (MAC) network model of an ensemble of intelligent interactive teams of human agents (HA), software agents (SA) and cyber-physical agents (CPA) located in six distributed nodes of a star network topology. All nodes of this topology represent Intelligent communication agents, and a special team of intelligent management agents for Inspection Affairs is located in the sixth node of the MAC network model. In the proposed model, the communication platform for MAC is supported by process-adaptable applications of machine learning methods, and the ability of agents to learn from experience and predict based on digital data processing. The main goal of MAC model is optimization of solutions for timely planning, reliable functioning, continuous monitoring, preventive supervision and effective management of Traffic Inspection tasks in uncertain situations contexts.
6.S. Glumčević, Z. Mašetić (International Burch University, Sarajevo, Bosnia and Herzegovina), B. Viteškić (The Arctic University of Norway, Tromso, Norway)
Closed-loop Artificial Pancreas Development: A Review  
Abstract: Diabetes is a widespread disease, suffered by millions, including children. Treatment of diabetes type 1 and sometimes even of type 2, entails multiple blood sugar checks and insulin injections per day, and can thus be extremely exhausting, especially for very young children. Open-loop system of insulin delivery, insulin pumps, used today commercially require human interaction which can lead to low blood sugar control due to human mistakes. Fully automated closed loop systems of artificial pancreas, as one-hormone as well as dual-hormone systems, are being developed. This paper is the literature survey of the latest research on the automated closed-loop artificial pancreas. Objective of this paper is to explore development of devices and techniques to facilitate the daily life of diabetic patients with emphasis to the latest research on the topic. From so called pens, to open loop systems of insulin pumps, closed-loop systems with user’s interaction - hybrid closed-loop, to the of latest fully automatised closed-loop – artificial pancreas. In total 300 articles are reviewed from which 150 articles are retained for the literature survey and 50 are analysed in this literature review.

Basic information:

Darko Huljenić (Croatia), Alan Jović (Croatia)

Steering Committee:

Andrea Budin (Croatia), Bojan Cukic (United States), Marko Đurasević (Croatia), Marina Ivašić-Kos (Croatia), Domagoj Jakobović (Croatia), Ruizhe Ma (United States), Neeta Nain (India), Stjepan Picek (Netherlands), Slobodan Ribarić (Croatia), Vitomir Štruc (Slovenia)

Program Committee:

Mario Brčić (Croatia), Karla Brkić (Croatia), Marko Čupić (Croatia), Marko Đurasević (Croatia), Marko Horvat (Croatia), Ivo Ipšić (Croatia), Domagoj Jakobović (Croatia)

Registration / Fees:

Price in EUR
Up to 8 May 2023
From 9 May 2023
Members of MIPRO and IEEE 230 260
Students (undergraduate and graduate), primary and secondary school teachers 120 140
Others 250 280

The discount doesn't apply to PhD students.


Darko Huljenić

Ericsson Nikola Tesla Inc.
Krapinska 45
HR-10000 Zagreb, Croatia

Phone:+385 1 365 4734

Alan Jović

Faculty of Electrical Engineering and Computing
Unska 3
HR-10000 Zagreb, Croatia

Phone: +385 1 612 9548

The best papers will get a special award.
Accepted papers will be published in the ISSN registered conference proceedings. Presented papers in English will be submitted for inclusion in the IEEE Xplore Digital Library.
There is a possibility that the selected scientific papers with some further modification and refinement are being published in the following journals: Journal of Computing and Information Technology (CIT)MDPI Applied ScienceMDPI Information JournalFrontiers and EAI Endorsed Transaction on Scalable Information Systems.


Opatija is the leading seaside resort of the Eastern Adriatic and one of the most famous tourist destinations on the Mediterranean. With its aristocratic architecture and style, Opatija has been attracting artists, kings, politicians, scientists, sportsmen, as well as business people, bankers and managers for more than 170 years.

The tourist offer in Opatija includes a vast number of hotels, excellent restaurants, entertainment venues, art festivals, superb modern and classical music concerts, beaches and swimming pools – this city satisfies all wishes and demands.

Opatija, the Queen of the Adriatic, is also one of the most prominent congress cities in the Mediterranean, particularly important for its ICT conventions, one of which is MIPRO, which has been held in Opatija since 1979, and has attracted more than a thousand participants from over forty countries. These conventions promote Opatija as one of the most desirable technological, business, educational and scientific centers in South-eastern Europe and the European Union in general.

For more details, please visit and

News about event
Currently there are no news
Patrons - random
Sveučilište u RijeciFER ZagrebPomorski fakultet RijekaTehnički fakultet RijekaFOI Varaždin