Abstract: Obtaining a 3D surface model of the human pubic bone with sufficient accuracy is a difficult task due to its complex geometry. The task can be further complicated by the fact that some parts of the bones are often missing or damaged, resulting in the lack of complete volumetric data. Therefore, our idea is to create a parametric pubic bone model as the base for building 3D surface model with optimal number of parameters, using prediction techniques and artificial neural networks (ANN). The study is conducted at the sample of 32 polygonal models of the male right pubic bone. At the each bone 9 anatomical landmarks are located, and 12 parameters as linear distances between these landmarks are determined and their values are measured on the samples. These values represent the dataset from which different combinations as input and output variables are taken. Three-layer back-propagation architecture of Neural Networks (NN) is chosen, with 1, 2, 3 or 4 neurons in the input layer, while the hidden layer has 5, 10 or 20 neurons. Levenberg - Marquardt (LM) and Bayesian Regularization (BR) algorithm are used for training 15 different NN architectures. Correlation coefficients and Mean Square Errors as performance indicators for all NN architectures are measured and compared with the aim to select optimal number of input parameters and optimal NN architecture. The best results are obtained for 3 input parameters (d1, d3 and d4), 5 neurons in hidden layer and BR training algorithm. The results of the study show that it is enough to localize 4 points and to measure the values of only 3 parameters, in order to get subject specific predictive bone model.

Abstract: Due to the transparency and open government initiatives, a large amount of open government data has been published on the open data portals. In order to provide data reusability, open data portals have different search mechanisms which depend on the quality of datasets metadata. Therefore, within this paper we present the results of the analysis of metadata changes on 40 government open data portals over a two year period. One of the main dataset search resources on the open data portals are datasets categories and tags used for datasets description. Therefore, the focus of our analysis is on tracking changes in the number of datasets on the portals, and changes that consequentially occurred in categories and tags available on the portals. We present different valuable results that can be used for tracking the trend of metadata changes and growth of the metadata that can affect their findability.

Abstract: In less than a decade, blockchain technology has seen a rise in popularity due to its innate security properties and overall disruptive potential. Surpassing its initial use in fintech and cryptocurrencies, blockchain and similar distributed ledger technologies have been used in healthcare, supply chain management, and within the public sector. However, recent studies show that blockchain-based technologies have found uses in software engineering (SE) as well. Namely, blockchain technologies can be used in all phases of the Software Development Life Cycle - software requirements, the engineering/development process, software testing and quality assurance, as well as software maintenance. In this paper, by utilizing existing literature regarding blokchain technologies and SE, we provide insight on which type of blockchain technology could be beneficial for each of the use-cases in SE, highlighting the advantages and potential disadvantages. We discuss various consensus mechanism support, smart contracts technology, as well as storage solutions, tu finally give recommendations for identified use-cases.

Abstract: Dams are vital infrastructure objects which must be maintained properly to avoid catastrophic failures or major accidents. Recent development of dam monitoring systems capabilities provides massive datasets that are overwhelming for traditional approach to dam safety analysis, which usually relies on expert’s abilities to identify and undertake necessary measures in a timely manner based solely on observed variables. In most cases, the variables that may not be directly measured are of the biggest importance for structural safety (e.g., damage criticality, load capacity, etc.). These variables can be estimated using various numerical methods, such as finite-element method. The concept of “digital twin” applied to dam structures provides tools for the experts to assess various safety parameters of the dam in near real time by coupling numerical methods with live data from monitoring systems. This paper presents development of a dam safety management system, which serves to improve the maintenance, safety, and functionality of a dam. Dam safety management system is based on elaborate monitoring system and detailed FEM models of the dam and power plant which are used to provide estimations of structural behavior and implementation of various “what-if” analysis. The system also uses MLR models of various dam responses for quick checks of observed values. The integration of data from monitoring system with FEM and MLR models is achieved through efficient data management framework. Quality control of data is performed through fully automated process to provide inputs for data assimilation in numerical models. This way a digital twin of the dam is created that is used for dam safety assessment and evaluation of reactive and proactive measures.

Abstract: This paper focuses on a problem of providing real-time biofeedback to a swimmer through a haptic user interface. This interface is integral part of a special wearable device that can be used during swimming. Wearable devices have been proved to be very efficient for providing feedback during various activities, including sports and particularly swimming. We considered several modalities for this interface and finally decided for the haptic modality as it has not been studied in this context before and can, according to our experiences, provide promising results. The proposed haptic biofeedback interface consists of a wearable device attached to the swimmer's lower back and equipped with sensors, a processor, and vibrotactile motors in the belt. The selected hardware configuration allows multiple modes of operation depending on the requirements of the biofeedback application. The wearable device is designed to be part of a larger biofeedback system or to be operated independently.

Abstract: Digital Twins (DTs) are among the most popular and quickly evolving technologies, particularly within Industry 4.0. The rapid expansion of a new generation of information technologies, such as the Internet of Things (IoT), artificial intelligence (AI), and cloud computing, results in new requirements for DT modeling. Consequently, a five-dimensional DT model is emerging, focusing on servitization, but with limited consideration of standardization for integration of the services. In this paper, we propose a new, business context-based approach that enables accurate and fast integration of DT services, contributing to the standardization of DT integration specifications. Furthermore, we present an initial validation result from the aerospace domain. Keywordsdigital twin; services integration; business context.

Abstract: Relational Concept Analysis (RCA) is a multirelational data mining method that aims to extract knowledge from multiple formal contexts (i.e., objects, attributes, and a binary relation between them) and the relations between them. One of the problems RCA has is the lack of the possibility of extracting knowledge directly from data that is represented with ternary relations. While there are some existing solutions towards this problem, either they require complex preprocessing of the input data, or they lose some capabilities of RCA such as the different meanings of the relations between concepts (∃, ∀, etc). In this work, we present an intuitive extension to RCA to be able to use it with data directly represented with ternary relations. As an example of its usage, we apply it to a dataset called Knomana which includes ternary relations.

Abstract: Coping with acute stress in sports and overtraining syndrome are extremely important concepts related to sports success. The aim of this paper is to consider the possibility of development and validation of tests within an online platform OpenClick that would assess these phenomena of athletes. Such tests would be reliable, fast, simple, undemanding, but above all, they would provide quality and usable information to coaches and athletes.

Abstract: This paper deals with the concept of digital platforms that serve as the online collaboration channel. We have created the SMART2M digital platform for innovations, http://innovate.smart2m.eu/, to facilitate the solution of challenges and different competitions that may be used by the academia, as well as the industry. The detailed description, accompanying content and specification will reinforce the concept of approving the solution, which can be a solution to homework in case of academia, or the actual product, technical solution or service in case of industry. Academia can reward the students with points for a successfully submitted solution, while organizations can offer monetary incentives or cooperation agreement for the winning ideas. Challenges are presented to a wide group of solvers as a public Call to solve problems - innovate. Users who represent potential solvers send ideas to the entities that published the problem. The final solutions are owned by the entity that announced the Call, or can be further negotiated with the applicants through the direct connection.

Abstract: This paper presents an application used to recognize personal data in textual documents written in Serbian and English languages. Recognition is performed using SpaCy and Classla named entity recognition models. The results of personal data recognition are presented and analyzed. Keywordspersonal data, named entity recognition, SpaCy, Classla

Abstract: Successful application of data localization laws is possible only if the country has an IT infrastructure mature enough to store all sensitive data, which is usually not the case in developing countries. This paper proposes a model for building local cloud infrastructure in developing countries compliant with data localization laws. The infrastructure will be developed gradually by adding location-aware commodity hardware nodes placed within the country’s borders over a more extended period. Each node is considered a volunteer for a period defined by the participation policy. This kind of policy describes the formation of a cluster of geographically close nodes that might all be disposed of and replaced after the defined period expires. The government authorities take responsibility for motivating national universities and firms that do business inside the country’s borders to take the key role in provisioning computing and storage nodes for a more extended period specified by the participation contracts and policies. In return, authorities guarantee a set of benefits such as tax reductions and funds for R&D. The authorities should make laws and regulations that follow the gradual development of the local cloud infrastructure by introducing soft data localization first. Hard data localization will replace soft data localization at the moment of infrastructure maturity. Keywordslocal cloud infrastructure, developing countries, data localization regulations, data localization laws, location-aware nodes, participation policies, participation contracts

Abstract: This paper tackles the problem of handwritten mathematical expressions recognition using an algorithm from the field of optical character recognition. The segmentation of mathematical symbols was done with the help of numerical methods of vertical and horizontal histogram projections while for the classification of characters, a convolutional neural network was used. Measurements were done on CROHME’s 2011 competition database, but the neural network was trained using Kaggle’s database of math symbols. The success rate of the algorithm appears in Levenshtein distance and in the number of correct predictions. The neural network reached a success rate of 18%, while on the same database without symbols like sin, cos, tan, i, j, ..., and ÷ 30% was reached.

Abstract: The digital transformation is impacting every aspect of our everyday life. The recent social-distancing and isolation measures accelerated the migration of many very traditional processes to online operation. Different vendors develop their own applications for diverse purposes, which expose end-users to significant security risks. Source code protection is crucial for making the internet world a secure environment. This paper discusses code obfuscation, white-box encryption, tamper-proofing, and diversification techniques. The recommendations and discussions in this article contribute toward ensuring a secure digital world. Following them and comprehending their advantages and disadvantages will enable the delivery of code that end users will trust and use.

Abstract: This paper is interested in unraveling the dynamics of a network of chess players constructed by observing 20 000 online chess matches played on Lichess.com. A complex network of players is built based on match data and social network analysis techniques are applied. The results are compared to over-the-board chess and other previously studied online chess platforms, to determine if and how the differences in implemented matchmaking algorithms impact network structure and topological properties of the network models. We show that the Lichess social network follows a power-law degree distribution and is not a small world network. It is highly disassortative and clustering is weakly expressed, making Lichess.com more similar to other online chess platforms than over-the-board chess. We employ assortativity analysis to explore the correlations between the difference in player ratings and show that the matchmaking algorithm remains fair, matching players of vastly differing skill level in less than 3% of the matches.

Abstract: This research is developed in order to describe the behavior present in the market and Limit Order book dynamics, using the concepts of supervised and unsupervised learning. The main mathematical object of interest is the limit order book, whose job is to keep track of all incoming and outgoing orders. There is a wide variety of possibilities to be explored for how to use machine learning techniques to get insights into market behavior. More precisely, in order to develop a statistical arbitrage strategy, the leverage of machine learning techniques can be employed. Furthermore, the concept can be enhanced with the feature that interprets the relationship of different features previously extracted from the limit order book data. The main idea is to employ a Graph Neural Network in order to describe the relationship between different features, and that relationship can be seen as a new feature that is potentially informative and possesses the power to uncover hidden and unknown knowledge from the data set. This work studies the ability to use Graph Neural Networks in order to get more insights from the stock market data. More precisely, this work investigates the ability to use Graph Neural Networks to label the stock market data into one of the labels from the set S={sell, buy, idle}. The obtained results are examined by using the F-score measure and compared with the results obtained by using the recurrent neural networks. This study discusses the potential for using GNNs for stock market data.

Abstract: The transition to clean (green) production is one of the biggest topics facing the whole world. The manufacturing industry is a major polluter, so any effort to reduce pollution is welcome. Smart sustainable production with minimal pollution and energy consumption is no longer just an idea, but a necessity. Smart production include state-of-the-art technologies such as the Cyber-Physical Systems, Big Data, Cloud Computing, Internet of Things and Artificial Intelligence. The aim of this paper is to find and analyze the application of machine learning methods that are mostly used to optimize manufacturing processes with a greater focus on machining processes, i.e., turning and milling processes. The paper gives a brief overview of machine learning methods that have been identified as the most used in data processing for machining processes optimization.

Abstract: Prevention of traffic accidents in transport using new technical achievements is one of the main topics in academic and industrial circles, especially due to the accelerated development of autonomous vehicles. Special attention is paid to the safety of transport of hazardous substances, such as hazardous waste. This paper describes one of the methods to improve communication between vehicles and vehicles and infrastructure in order to reduce traffic accidents during the transport of explosives and toxic substances. The possibility of complete monitoring of the transport of dangerous goods using Unmanned Aerial Vehicles (UAVs), which also enable communication between vehicles in areas of poor or no network signal coverage, is discussed. Simulation results indicate the reduced interrupt probability of communication in the network with UAVs.

Abstract: In this paper we examine the scientific production of Serbian scientists in the area of computing, published between year 2008 and year 2020, extracted from the Web of Science (WoS) database. Our dataset includes papers published in journals with impact factor, which are classified in at least one out of seven WoS categories with the prefix "Computer Science". The work includes identification of unique authors' and institutions' identity from the dataset, as well as creation of a database used for bibliometric and coauthorship analysis. As a result, we have created coauthorship network using Gephi, identified the leading authors and institutions, as well as the most popular journals, WoS categories and collaborators of Serbian authors in the field of computer science for the analyzed period.

Abstract: Compliant grippers are used in the macro and micro world for capturing and manipulating objects, when it is necessary to manipulate fragile objects, such as surgical instruments, in biomedicine for manipulating individual cells, etc. In addition, grippers with a rigid contact between the grippers and the rigid objects can lead to damage to the objects, so in these cases, a control system is required to control the contact forces. Development of compliant finger-like gripper for soft touch application is described in this paper. This paper provides an answer to the question how to design, model and implement such compliant gripper. The developed compliant fingers are inspired by the real human fingers. Machine learning methods where proposed for realization of non-distractive grasping of fragile objects where feedback information is acquired by computer vision system while intelligent algorithms where considered for motion control and grasping. The complexity of lays not only in non-destructive grasping but also in adequate force implementation in order to avoid dropping the grasped object. Hence, nonlinear simulation of compliant gripper, by means of finite elements method (FEM), is provided in this paper. Several experiment where done to provide real life conformation of such analysis.

Abstract: This research aims to propose a novel framework for the assessment of the consequences of hazardous events on a water resources system using dynamic resilience. The two main types of hazardous events considered are: a severe flood event, and an earthquake. Given that one or both hazards occur, this framework utilizes a digital twin based on a system dynamics (SD) model, backed by an Artificial Neural Network (ANN) to estimate the dynamic resilience. The ANN was trained using a large, simulated dataset ranging from very mild to extreme hazard combinations. The ANN’s efficacy was quantified using the average relative error metric which equals 2.14% and 1.77% for robustness and rapidity, respectively.