Hybrid Event

Event program

Thursday, 5/26/2022 9:00 AM - 1:00 PM, Camelia 1, Grand hotel Adriatic, Opatija 9:00 AM - 10:45 AM Artificial Intelligence Theory Chair: Darko Huljenić  1. A. Krajna, M. Kovač, M. Brčić, A. Šarčević (University of Zagreb, Faculty of Electrical Engineering and Computing, Zagreb, Croatia) Explainable Artificial Intelligence: An updated perspective  Artificial intelligence has received a noticeable momentum during the last decade due to the rapid rise in computing power and data storage capabilities. Enormous advances are evident in image recognition, speech analysis, strategic planning, etc. Despite the unprecedented advancements, the key obstacle to greater use of these intelligent systems is their lack of transparency and understandability. Since artificial intelligence already has a significant impact on various sectors, the problems and development of explainable artificial intelligence is gaining importance. Explainability is important for: model result justification, to root out errors in model functioning, to reach some innovations, or to improve the model itself. There are many factors that an AI model can consider to make its predictions. This research offers an overview of methods for generating explanations that can be used throughout the AI model development process in general, starting with the pre-modeling explainability stage until the point when the model is already generated. This paper's focus is on recent explainability methods for supervised learning, which includes today's very relevant graph neural networks. 2. J. Juros (Student at University of Zagreb, Faculty of Electrical Engineering and Computing, Zagreb, Croatia), M. Brcic (University of Zagreb, Faculty of Electrical Engineering and Computing, Zagreb, Croatia), M. Koncic (KONZUM plus d.o.o., Zagreb, Croatia), M. Kovac (University of Zagreb, Faculty of Electrical Engineering and Computing, Zagreb, Croatia) Exact Solving Scheduling Problems Accelerated by Graph Neural Networks  Scheduling is a family of combinatorial problems where we need to find optimal time arrangements for activities. Scheduling problems in applications are usually notoriously hard to solve exactly. Existing procedures, based on mathematical and constraint programming, usually make manually-tuned heuristic choices. Manual tuning can be eliminated by using machine learning, while simultaneously finding better heuristics. In this paper, we apply a graph convolutional neural network model on the problem of learning branching decisions in a general branch&bound solver. We test the augmented solver on job-shop scheduling problems and specific delivery scheduling problems in the supply chain of a local retailer. We get promising results and point to possible improvements. 3. G. Oparin, V. Bogdanova, A. Pashinin (Institute for System Dynamics and Control Theory of SB RAS, Irkutsk, Russian Federation) Implicit Boolean Network for Planning Actions  It is the usual to classify the automated searching of a plan as an artificial intelligence problem. In planning, the environment is defined as a set of both states and transitions between them. Planning consists in searching for a finite transitions sequence (actions) that transfers the initial state to one of the states, satisfying the objective. Intelligent planning is based on mathematical logic. The reasoning required for forming the plan is reduced to logic inference. As a logical model of the planning environment, we propose to use implicit Boolean networks – binary dynamic systems, the transition function of which is defined by the implicit Boolean equation relative to the current state variables vector and some variables vectors following the current state. In the framework of this model, the forming of a plan is reduced to qualitative analysis of dynamic properties of the implicit binary dynamic system. In particular, these are reachability or cyclicity properties over a given discrete time interval. At the final stage of this method, the verification of the Boolean formula satisfiability is performed. This formula includes the equation of the dynamics of the implicit Boolean network and the specification of the dynamical property being verified. The considered approach is demonstrated on the problem of planning a river crossing and constructing a Hamiltonian cycle in a graph. A specialized parallel algorithm for solving deeply structured Boolean satisfiability problems is proposed, which provides scalability with an increase in their dimension. 4. J. Maltar, D. Matijević (Department of Mathematics, J. J. Strossmayer University of Osijek , Osijek, Croatia) Optimization Techniques for Image Representation in Visual Place Recognition  In visual place recognition we aim to match a given query image from a query database with the most appropriate reference image from a reference database. One of the main issues is how to represent a place. Although an ordinary RGB representation can represent a place, various, either handcrafted or learned representations such as deep convolutional neural networks achieve better quantitative results. By using optimization techniques, both convex and non-convex, we can adapt a place representation such that it fits into the problem of visual place recognition. Therefore, in this paper we examine numerous optimization techniques and incorporate them in the context of our problem. Quantitatively, in terms of the area under a curve (AUC) measure, conducted experiments show how such optimized representation outperforms unoptimized one. 5. V. Kondratiev (ITMO University, Saint Petersburg, Russian Federation) Using Disjunctive Diagrams for Preprocessing of Conjunctive Normal Forms  In this article, in the context of the Boolean satisfiability problem, the problem of speeding up the SAT solvers on specific input formulas is considered. The speedup is achieved using the CNF preprocessing algorithm which is based on the use of decision diagrams of a special kind, called Disjunctive Diagrams. These diagrams allow one to naturally test some sets of partial variable assignments and add more stringent constraints to the original formula. Several families of SAT instances are considered and used to compare the solving time before and after preprocessing, including a selection of tests from the SAT Competition 2021 and some tests related to the problem of checking the equivalence of Boolean circuits. Computational experiments have shown that for hard instances in more than half of the cases the proposed preprocessing algorithm can speed up the solving time of considered CNFs. 6. A. Gribl Koščević, D. Petrinović (University of Zagreb, Faculty of Electrical Engineering and Computing, Zagreb, Croatia) Maximizing Accuracy of 2D Gaussian Profile Estimation Using Differential Entropy  The goal of this paper is to find an optimal width of circular region-of-interest (ROI) for the precise estimation of 2D Gaussian profile parameters in the presence of additive noise. The radius of circular ROI for the rotationally symmetrical profile can be represented as a product of the profile's STD and the factor of Mahalanobis distance k. The centre of ROI coincides with the centre of the profile being estimated. It was shown that in the case of a random sampling within such circular ROI, the estimation accuracy of the least-squares method is highly affected by the chosen factor k for the constant number of random input samples and given SNR. The differences in estimation accuracy are the results of variations of profile data informativity for different ROI widths. If sample positions are random variables uniformly distributed within the circular ROI, it was derived that the 2D Gaussian profile values as a function of random variables follow the log-uniform distribution. Therefore, in the paper we derive the differential entropy of log-uniform distribution which is maximized with respect to the factor of Mahalanobis distance k, thus yielding the optimal ROI width. The theoretical results are verified using Monte-Carlo simulation and we show that the loss of estimation accuracy for other non-optimal widths is proportional to the reduction of the profile’s differential entropy. Such a solution is valid under a fixed number of samples as an estimation constraint. However, for the case of sample density constraint, the solution is different, as we will demonstrate in the paper. 7. T. Bronzin, B. Prole, A. Stipić (CITUS d.o.o., Zagreb, Croatia), K. Pap (University of Zagreb, Faculty of Graphic Arts , Zagreb, Croatia) The Proposed Method of Measuring How Mixed Reality Can Affect the Enhancement of the User Experience  The rapid development of technology and the acceptance of augmented reality (AR) and mixed reality (MR) opens a new chapter in human behavior. AR & MR is changing how we walk, interact with other people, and live in physical and digital worlds. The line between the two worlds is becoming increasingly blurred. With a wide range of possibilities, the potential of AR & MR to create new value for a person lies in its ability to create an enhanced personalized user experience in different aspects of human life. This paper proposes the methodology of measuring the extent of mixed reality influence on the user experience enhancement for the museum visitors. It measures the visitor’s emotion type and level of emotional intensity during the artwork observation. Microsoft Kinect is used as a measurement device to determine emotion type and level of emotional intensity. Microsoft HoloLens will guide a visitor through the correct sequence of steps and ensure that all parts of artworks are visited in the correct order. The high precision of measurements obtained by Kinect and HoloLens make all measurements highly objective. 10:45 AM - 11:15 AM Break  11:15 AM - 1:00 PM Image and Video Processing and Analysis Chair: Marina Ivašić-Kos  1. B. Liberatori, C. Mami, G. Santacatterina, M. Zullich, F. Pellegrino (University of Trieste, Italy, Italy) YOLO-based Face Mask Detection on Low-End Devices Using Pruning and Quantization  Deploying deep learning (DL) based object detection (OD) models in low-end devices, such as single board computers, may lead to poor performance in terms of frames-per-second (FPS). Pruning and quantization are well-known compression techniques that can potentially lead to a reduction of the computational burden of a DL model, with a possible decrease of performance in terms of detection accuracy. Motivated by the widespread introduction of face mask mandates by many institutions during the Covid-19 pandemic, we aim at training and compressing an OD model based on YOLOv4 to recognize the presence of face masks, to be deployed on a Raspberry Pi 4. We investigate the capability of different kinds of pruning and quantization techniques of increasing the FPS with respect to the uncompressed model, while retaining the detection accuracy. We quantitatively assess the pruned and quantized models in terms of Mean Average Precision (mAP) and FPS, and show that with proper pruning and quantization, the FPS can be doubled with a moderate loss in mAP. The results provide guidelines for compression of other OD models based on YOLO. 2. A. Kralevska, R. Trajanov, S. GIevska (Ss. Cyril and Methodius University, Faculty of Computer Science and Engineering, Skopje, Macedonia) Real-time Macedonian Sign Language Recognition System by using Transfer Learning  The objective for developing a real-time sign language recognition system is twofold: improving interpersonal communication and supporting inclusive human-computer interaction with hearing-impaired population using a particular sign language. This study describes the design and implementation of a system for real-time Macedonian Sign Language recognition in images and videos. A robust and lightweight model was proposed based on transfer learning of suitable pretrained architectures, namely, Single Shot Detector (SSD) MobileNetV2 and SSD MobileNetV2 FPNLite. The proposed models were fine-tuned and extensively evaluated in a number of diverse scenarios to account for the inherent difficulties in recognizing particular letters. In the absence of publicly available dataset, we have created a dataset consisting of two-handed images of 28 out of 31 letters of the Macedonian alphabet; the three letters expressed by dynamic gestures were excluded from the study. The results point out to a state-of-the-art prediction accuracy on the classification task of Macedonian sign language alphabet. 3. A. Fahmy, M. Tayel (Faculty of Engineering Alexandria, University, Alexandria, Egypt) Advanced Medical Images Recognition and Diagnosis of Respiratory System Viruses  Detection of respiratory viruses (COVID-19) is a perplexing task which regularly requires taking a quick look at clinical images of patients on and on. Hence, there is a need to propose a model to predict the respiratory viruses (COVID-19) cases at the earliest possible to control the spread of disease. Deep Learning makes it possible to find out that viral pneumonia is more or less same as COVID-19. In this research paper, the deep learning technique has been applied to clinical images of different types of respiratory viruses (COVID-19). This shows the knowledge gained by model trained for detecting viral pneumonia can be transferred for identifying COVID-19. This makes the extraordinary work easier by using existing model for determining COVID-19. It is difficult to detect the abnormal features from images due to the noise impedance from lesions and tissues. For this reason, Mel Frequency Ceptral Coefficient (MFCC) feature extraction is consummated which focus only on the area of interest to detect COVID-19 out of CT image. MFCC is a very common and efficient technique for signal processing. In this research, a MFCC – CNN learning model to accelerate the prediction process is proposed that assist the medical professionals. MFCC is used for extracting the image's features concerning existence of COVID-19 or not. Classification is performed by using convolutional neural network. This makes the time-consuming process easier and faster with more accurate results for radiologists and this reduces the spread of virus and save lives. Experimental results shows that a CNN using CT image converted to Mel-frequency cepstral coefficient spectrogram images as input can achieve high accuracy results; with classification of validation data scoring an accuracy of 98.5% correct classification of COVID and NOT_COVID labeled images. Hence, can be used practically for detection of COVID-19 from CT images. The work here provides a proof of concept that high accuracy can be achieved with a moderate dataset, which can have a significant impact in this area. The obtained accuracy is better than that obtained of Logistic Regression, Random forest, SVM, CNN (applied to Covid images), CNN (applied to MFCC of Covid coughs). The obtained results are highly encouraging and provide further opportunities for research by the academic community on this important topic. Keywords Biomedical imaging, COVID-19, Computed tomography, feature extraction, MFCC, image classification, CNN 4. D. Potoč, D. Petrinović (Faculty of Electrical Engineering and Computing, Zagreb, Croatia) Creating a Synthetic Image for Evaluation of Vignetting Modeling and Estimation  Vignetting is an effect caused by camera settings or lens limitations. It is also known as the ”light fall-of” where the intensity of light decreases from the middle to the edges. Vignetting has become popular in various scientific papers where attempts are made to remove vignetting by various methods. The idea of this article would be to design a synthetic image of the vignetting effect that would mimic the starry night sky. It would be made by adding random white pixels to a previously set positive constant value modeling the sky glow, also adding Gaussian or Poison noise after applying vignetting effect itself with an arbitrarily chosen center point. Parameters of the vignetting module can be set manually defining the ground truth module, thus modeling the actual multispectral lens response functions. Such an image could be used as ground truth for vignetting modeling and estimation functions to evaluate different methods and their accuracy as a function of signal to noise (SNR) ratio and noise type. The image is tested on two vignetting correction functions, both functions show great results, only the first has larger residual gain variations, especially toward the frame corners. 5. K. Brkić, T. Hrkać, Z. Kalafatić (University of Zagreb, Faculty of Electrical Engineering and Computing, Zagreb, Croatia) A Privacy Preservation Pipeline for Personally Identifiable Data in Images Using Convolutional and Transformer Architectures  Image and video data of people, shared voluntarily and involuntarily, is ubiquitous. There is an increased need for techniques that enable privacy protection via removal of personally identifiable information in such data, spurred by regulatory interest and increased social awareness of privacy implications. In this paper, we introduce a privacy preservation pipeline that enables de-identifying personal data in images and videos via replacement image synthesis while retaining data utility. We utilize the recently proposed convolutional VQGANs combined with autoregressive transformers to synthesize realistic and fully de-identified images of people that are then blended with the original scene. Experimental results show that the method provides very strong de-identification while retaining high realism of the scene. 6. D. Zoraja, T. Petković, T. Pribanić (UniZG FER, Zagreb, Croatia), J. Forest Collado (University of Girona, Girona, Spain) Projector Calibration in a Two-Layer Flat Refractive Geometry for Underwater Imaging  In underwater imaging we observe objects in the water through a flat glass interface which defines a two-layer flat refractive geometry. Using structured light for underwater 3D imaging requires that both camera and projector are calibrated, which is a difficult task. In this paper we discuss how to calibrate a projector for underwater imaging under a two-layer flat refractive geometry. We propose to model the projector as an inverse camera which enables the use of known procedures for camera calibration in flat refractive geometry. We discuss how to efficiently collect the calibration data using only a planar calibration board and an uncalibrated camera. Performed experiments show that the results of projector calibration are comparable to the results of camera calibration making it highly applicable in practice. 7. Z. Kalafatić, T. Hrkać, K. Brkić (University of Zagreb, Faculty of Electrical Engineering and Computing, Zagreb, Croatia) Multiple Object Tracking for Football Game Analysis  We consider the problem of tracking football players in video captured from one side of the football field. The task is challenging due to frequent occlusions of players, varying size of players' projections, changing illumination and similar appearance of players of the same team. We approach the problem of multiple object tracking using the tracking-by-detection paradigm, where the object detector is applied to individual video frames and the tracker tries to associate the detector responses and form the trajectories by using some motion model. We provide experiments using a classic object detector based on background modeling as well as a deep learning based detector. Thursday, 5/26/2022 3:00 PM - 6:15 PM, Camelia 1, Grand hotel Adriatic, Opatija 3:00 PM - 3:30 PM Invited Lecture Chair: Alan Jović  Željka Motika (Central State Office for the Development of Digital Society, Zagreb, Croatia) National Language Technology Platform  3:30 PM - 4:30 PM Machine Learning Applications Chair: Alan Jović  1. M. Mitreska (iReason LLC, Skopje, Macedonia), K. Mishev, M. Simjanoska (iReason LLC, Ss. Cyril and Methodius University, Faculty of Computer Science and Engineering, Skopje, Macedonia) NLP-Based Typo Correction Model for Croatian Language  Spelling correction plays an important role when applied in complex NLP-based applications and pipelines. Many of the existing models and techniques are developed to support the English language as it is the richest language in terms of resources available for training such models. The good occasion is that few of the methodologies provide the opportunity to adapt to other, low-resource languages. In this paper, we explore the power of Neuspell Toolkit for training an original spelling correction model for the Croatian language. The toolkit itself comprises ten different models, but for the purposes of our work, we use the leverage of pre-trained transformer networks due to their experimentally proven spelling correction efficiency in the English language. The comparison is performed over different pre-trained Subword BERT architectures, including Bert Multilingual, DistilBERT, and Roberta-XLM, due to their subword representation support for the Croatian language. Furthermore, the training is done as a sequence labeling task on a newly created parallel Croatian dataset where the noisy examples are synthetically generated, and the misspelled words are labeled with their correct version. Finally, the model is tested in-vivo as part of our originally developed speech-to-text model for the Croatian language. 2. M. Kovac, M. Brcic, A. Krajna, D. Krleza (University of Zagreb, Faculty of Electrical Engineering and Computing, Zagreb, Croatia) Towards Intelligent Compiler Optimization  The future of computation is massively parallel and heterogeneous with specialized accelerator devices and instruction sets in both edge- and cluster-computing. However, software development will become the bottleneck. To extract the potential of hardware wonders, the software would have to solve the following problems: heterogeneous device mapping, capability discovery, parallelization, adaptation to new ISAs, etc. This systematic complexity will be impossible to manually tame for human developers. These problems need to be offloaded to intelligent compilers. In this paper, we present the current research that utilizes deep learning, polyhedral optimization, reinforcement learning, etc. We envision the future of compilers as consisting of empirical testing, automatic statistics collection, continual learning, device capability understanding, multiphase compiling – precompiling and JIT tuning, and classification of workloads. The benefits of intelligent compilers are time savings for the economy, energy savings for the environment, and greater democratization of software development. 3. A. Kovač, S. Seljan, I. Dunđer (Filozofski fakultet, Zagreb, Croatia) An overview of machine learning algorithms for detecting phishing attacks on electronic messaging services  Information security, as one of the most important segments of information technology particularly in terms of personal data protection, has become one of the most important and controversial topics in the last few years. The reason is that almost all tasks are carried out with help of technology, and this has greatly changed the way people exchange information, interact with each other, and finally share their oftentimes sensitive and personal data. Online services such as social media platforms, instant messaging applications and services for e-mail sharing allow the mentioned possibilities of information exchange and communication, and as such participate significantly in human everyday life. These services enable fast sharing of information, but more often are controversial in terms of personal data security, especially in times of worldwide attacks that have recently become more frequent and dangerous. In this paper these attacks refer to a process of identity and sensitive data theft called “phishing”, which is considered a serious breach of privacy and is colloquially referred to as the “crime of the 21st century”. However, there are several machine learning algorithms that are more or less successful in combating phishing attacks. The aim of this paper is to present basic machine learning approaches for the detection of phishing attacks on electronic messaging services. 4. A. Ivanda, L. Šerić, M. Braović, D. Stipaničev (Fakultet elektrotehnike, strojarstva i brodogradnje, Split, Croatia) Application of Cogent Confabulation Classifier to Bathing Water Quality Assessment Using Remote Sensing Data  Cogent Confabulation is a comprehensive and simple method for data classification, but is unjustly neglected in modern machine learning. Cogent confabulation uses multiple evidence to classify data items, requiring less computation than Bayesian classifiers. Earth observation with remote sensing provides researchers evidence of various events and processes encoded in the reflectance of the surface in multiple wavelength bands. Decoding which bands provide information on which event is in the focus of many scientists. In this paper we are presenting the preliminary results on using the Cogent Confabulation Classifier on Sentinel-3 OLCI satellite data to predict the status of bathing water quality. Measuring bathing water quality is an important activity to protect human health, animal health and the environment. It is based on in situ measuring bacteria and categorizing quality based on the EU Directive 2006/7/EC. Study area used in this paper is Kaštela Bay and Braˇc Channel located in the Republic of Croatia. Data set is constructed of satellite data (bands values) and ground truth water quality based on in situ measurements. We developed a classifier that was trained on the constructed data set and is able to classify two distinct classes of bathing water quality based on satellite images. Results of the applied classifier are described, analyzed, and compared with other commonly used classification approaches in terms of accuracy and computational performance. 4:30 PM - 5:00 PM Break  5:00 PM - 6:15 PM Machine Learning Applications Chair: Darko Huljenić  1. Z. Lavrić, K. Osman (Zagreb University of Applied Sciences, Croatia) A Machine Learning Approach to Modelling and Simulating the Behaviour of Systems with Deformable Particles  2. T. Matijašević, T. Antić, T. Capuder (Faculty of Electrical Engineering and Computing, University of Zagreb, Zagreb, Croatia) Voltage-based Machine Learning Algorithm for Distribution of End-users Consumption Among the Phases  Distribution networks are poorly observable, which is especially true for low voltage (LV) networks, where observability is affected by the reduced number of smart meters and the lack of network data. The main role of smart meters is the energy consumption calculation, while many other possibilities remain unexploited and some important data remain unused, including the end-user phase connection. This aggravates the problem of phase identification for energy utilities, which consequently complicates the numerous calculations required for the smooth operation of the distribution network. Machine learning (ML) methods are increasingly used in solving these problems, and therefore various authors have presented the ML-based phase detection using voltage or end-user consumption measurements. However, this paper develops an ML model for the LV distribution network phase identification, which is applied to both types of measurements. The model is tested on two networks: a smaller network is a test network that allows easier understanding of the developed algorithm, while a larger network is a real-world network with over 100 nodes and more than 50 end-users. Results show the effectiveness of the phase identification model in both networks using voltage measurements, while phase detection using end-user consumption gave significantly poorer results. 3. T. Kovačević, S. Goluža, A. Merćep, Z. Kostanjčar (Faculty of Electrical Engineering and Computing, Zagreb, Croatia) Effect of Labeling Algorithms on Financial Performance Metrics  Machine learning models are increasingly common in predicting financial market movements. Unlike some other research areas, labels of financial time series are not unambiguously determined, which is why multiple labeling algorithms were proposed. The effect of a particular labeling algorithm on a trading strategy is often overlooked as most existing research uses only one type of labels to develop a machine learning model used for trading. However, different labeling algorithms lead to different generalization errors that may impede financial performance of a strategy. This paper examines the relationship between the classification performance of a model and the financial performance of a strategy based on the same model. The relationship is examined for two commonly used labeling algorithms: fixed-time horizon and triple-barrier method. Although the results for both labeling algorithms confirm a statistically significant correlation between classification and financial performance, the correlation coefficient itself has a low value. 4. A. Fotouhi , H. Guicheniti , H. Chour, M. BenMabrouk (Capgemini Engineering, Meudon, France) Human Daily Activities: From Detection to Prediction  While the number of connected objects in the world of Internet of Thing (IoT) is increasing, an efficient and intelligent solution to exploit the huge amount of generated data from home appliances does not exist. Smart homes powered by IoT devises are able to automate and monitor the every day activities of home owners, and improve the life quality especially for elderly and disabled people. In this paper, we take advantage of deep learning and machine learning techniques and propose a comprehensive solution to tackle this problem. Our proposed approach, is able to analyze raw and unlabeled IoT data from the connected devices in a smart home and provide data-driven services. After analysing and processing raw data, and then human activity detection and recognition in real time, we predict the upcoming activity and its occurring time using LSTM (Long Short-Term Memory). Our simulation results show that the developed models can predict a wide range of human activities from unlabeled data with the accuracy of 85%. 5. G. Mirceva, A. Naumoski, A. Kulakov (Faculty of computer science and engineering, Ss. Cyril and Methodius University in Skopje, Skopje, Macedonia) Classification of protein structures using deep learning models   Protein molecules are very important in the organisms because they participate in different processes. Understanding the way they interact in the cells is of high importance. In order to understand that, solving the task of protein classification could be really helpful thus providing valuable knowledge about the similar proteins that belong so same class. In this paper we focus on solving the task of protein classification. First, we extract some features of the proteins thus obtaining feature vectors, and then by using deep learning architecture, we create prediction model that could be used for classifying protein structures. We present some experimental results of the obtained classification models. Friday, 5/27/2022 10:00 AM - 11:00 AM, Camelia 1, Grand hotel Adriatic, Opatija 10:00 AM - 11:00 AM Other artificial intelligence topics Chair: Marko Horvat  1. N. Netkovska, G. Mirceva, A. Naumoski (Faculty of computer science and engineering, Ss. Cyril and Methodius University in Skopje, Skopje, Macedonia) Spatial Relationship between Schools Population and Book Related Services in Skopje  During these pandemic times, people are bored with digital devices and reading digital books, and more and more they want to get a hand on real printed books. This is even more true, when it comes for high school kids, that most of the time they are spending in front of digital devices. In this direction, our paper project focuses on the investigation of the spatial relationship between high schools’ population and book related services, mainly photocopying and large printing services. Using GIS analytical software we will use the gathered data, not only on the location of these objects, but also the number of secondary school kids, different types of services provided by the book related services as well as the rating score for each store in specific radius. The interpolation spatial map with combination of the ring buffer analysis, provides useful insight into possible improvement for new business opening locations of these services. In future we plan to further advance this analysis by providing more associated objects and their properties related to book services. 2. N. Almujally (Princess Nourah bint Abdulrahman University, Riyadh, Saudi Arabia), H. Alrashidi (Ministry of Education, Farwaniya, Kuwait), M. Joy (University of Warwick, Coventry, United Kingdom) To Share or Not to Share? Perceptions of University’s Faculty Members Regarding the Sharing of their Teaching-related Knowledge  While the concept of knowledge management (KM) has been widely discussed and implemented in a large number of commercial organizations within the Saudi context, the topic of applying KM to effect in higher education institutions (HEIs) has received limited attention. This is despite the fact that there is a recognition of the importance of managing knowledge in such HEI environments. Thus, this research was designed to identify academics’ perceptions about the sharing of teachingrelated knowledge within Saudi universities. An investigative study was conducted by collecting qualitative data via 22 semi-structured interviews with academics from different Saudi universities to capture their perceptions. The qualitative data show that the academics have clear ideas about several potential benefits of managing teachingrelated knowledge, despite the challenges they have faced when managing their knowledge using the currently existing KM approaches. This study holds considerable promise in relation to developing an effective web-based KM approach that fits the academics’ needs. 3. M. Horvat (University of Zagreb, Faculty of Electrical Engineering and Computing, Department of Applied Computi, Zagreb, Croatia), A. Stojanović, Ž. Kovačević (Zagreb University of Applied Sciences, Department of Computer Science and Information Technology, Zagreb, Croatia) An Overview of Common Emotion Models in Computer Systems  Emotions are an omnipresent and important factor in the interaction and communication between people. Since emotions are an indispensable part of human life, it would accelerate the progress of artificial intelligence and other fields of science that require data about emotions if they could be adequately described by computer systems. Today there are many different theories of affect, but few of them are used in affective computing. Other areas of computing also benefit from structured and expressive data models of the affective domain, such as human-computer interaction and brain-computer interfaces. Typical tasks include automated recognition and analysis of emotional states, mental fatigue, individual motivation, vigilance and stress resilience. In this paper four often used models of emotion and cognitive behavior are listed and their properties explained: discrete, dimensional, appraisal and action tendency models. For each model, algorithms are provided for similarity measures that can be used to determine the relatedness between different stimulation and estimation artefacts in their respective emotion spaces. The goal of this article is to help professionals find the optimal emotion model for their research and quickly become familiar with data modelling of affective states. 4. A. Ivanov, D. Orozova (Burgas free university, Burgas, Bulgaria) Intelligent Personnel Assistant for Field Researchers  The subject of the research are specialized intelligent personal assistants as tools designated to facilitate and automate the activities of scientists and researchers engaged in fieldwork. Most researchers still follow traditional data collection and processing methods, taking notes first on paper and then digitizing them long after the fieldwork is completed. Another common practice is using several digital devices such as video cameras, ultrasound detectors, and various sensors in combination with paper and laptops. These practices often cause long delays, data transfers between analog and digital media often lead to technical errors, and valuable notes or observations may be lost during the period from information gathering to digitization. Field researchers who use software systems in their work do not have a tool designed especially for this type of activity but are forced to adapt and combine different existing solutions. A specialized intelligent personal assistant is offered, which serves as a unified system for collecting, storing, and processing field data. The aim is to improve the quality of the collected field data and optimize the analysis. Good cooperation in gathering information is crucial for research processes.