Abstract: We present a method for image tagging, i.e. assigning a set of tags/labels to an image. Three popular architectures of deep convolutional neural networks were used: VGG16, Inception-V3, and ResNet-50, which were pretrained on the ImageNet data set for a classification problem, and then fine-tuned on the HARRISON data set for the image tagging problem. The final model consists of an ensemble of these three convolutional neural networks, whose outputs were combined by different methods: averaging, voting, union, intersection and by two-layer feedforward neural network. We verified these models on the hashtag recommendation for images from social network task, with a predefined set of 50 possible hashtags. All of the models were evaluated using the following metrics: precision, recall and F1-measure.

Abstract: This paper consists of the formal specification and implementation of the system for simultaneous localization and mapping based on video content analysis using one video camera, without usage of any additional sensors. Lack of sensors lowers the system price, but software complexity becomes very high. Shi-Thomasi corner detection method is used in system implementation, as well as Lucas-Kanade optical flow method, RANSAC algorithm for optical flow vector filtering and Kalman filter for correction of the camera position.

Abstract: This work contains an analysis of the impact that named entities have on thematic text classification. The categorization has been conducted on documents acquired from well-known datasets such as the Reuters 21578, 20 Newsgroups and the WebKB set. Multiple classifiers have been trained so as to eliminate the chance of any results being a direct consequence of the chosen algorithm. Results from multiple papers in this field that are based on the same datasets and the same, or similar, data splits, are used to compare the achieved results. Besides that, for each dataset and each classification algorithm a control model, that does not include named entity recognition, is created to confirm the validity of the results. Acquired results do not show any significant improvement when using named entities, and in some cases even show worse performance.

Abstract: Collaborative Filtering technique is a state-of-theart method in recommender systems. This technique has proved to be successful both in research and industry. However, it is challenging to find an adequate algorithm for calculating user similarity, which enables the recommender system to generate the best recommendation for given domain and data set. In this paper, we proposed an architecture which allows users to additionally personalize the recommendation by letting them choose algorithm for calculating user similarities, or to implement their own algorithm on a distributed service. Online tests on implemented application for movie recommendation show that architecture proposed in this paper gives users the ability to personalize recommendation for more efficient results.

Abstract: The deregulation of the electricity market is a process which is currently very hot topic in the Southeast European region. Namely, few of the countries have their own power exchanges from long time ago, but few of them have just formed ones and the remaining countries need to decide in the near future the direction they are going to follow towards resolving this problem. With the introduction of the power exchanges and the fact that half of the power markets in the Southeast European region exist less than one year, forecasting the electricity price on those markets is becoming very attractive research area and is of great importance. In this paper, 24-hours ahead forecasting of the electricity price in these newly formed power exchanges is. To this end, an artificial intelligence models, specifically neural networks are used in this paper, which as an input use all information that are relevant for the corresponding power exchange price forecasting. The results show that among the newly formed power exchanges in the region, the price in Bulgarian power exchange is the most unpredictable one, while, on the other hand, the price in the Serbian power exchange is the most predictable one. Additionally, the results present in which hours of the day and in which days in the week the prices have the highest variations.

Abstract: Since application of information and communications technologies (ICT) to beekeeping is far from the potential, we are trying in this paper to systemize beekeeping process from the perspective of system control theory and detect appropriate points of possible innovations in beekeeping by application of modern ICT. We find that current - precision beekeeping – approach, focusing on monitoring and mostly treating sound factors of beekeeping process, neglects a number of aspects of the process, which can certainly be improved by ICT application.

Abstract: A rapidly increasing and vast number of “things” will be connected to the Internet, in all sectors of society (people, places, sensors, homes, industry, government services, etc.). The urgency of finding sustainable solutions requires “things” and services of the overall system to display autonomic intelligent behavior. The ability of cloud infrastructure to orchestrate the fine-grained and agile control of “things” is limited. This mandates an alternative approach intelligently moving control to the “things”. Thereby we minimize the reliance on cloud infrastructure, and are able to build more agile, intelligent and effective solutions in various application areas, such as Health, Transport and through Automation. We provide examples of such novel solutions tested in “Urban ICT Arena”, a largescale smart city testbed in Stockholm.

Abstract: The pervasive diffusion of Information and Communication technologies (ICT) and automation technologies are the prerequisite for the preconized fourth industrial revolution: the Industry 4.0 (I4.0). Despite the economical efforts of several governments all over the world, still there are few companies, especially small and medium enterprises (SMEs), that adopt or intend to adopt in the near future I4.0 solutions. This work focus on key issues for implementing the I4.0 solutions in SMEs by using a specific case example as a test bench of an Italian small manufacturing company. Requirements and constraints derived from the field experience are generalised to provide a clear view of the profound potentialities and difficulties of the first industrial revolution announced instead of being historically recognised. A preliminary classification is then provided in view to start conceiving a library of Industry 4.0 formal patterns to identify the maturity of a SME for deploying Industry 4.0 concepts and technologies.

Abstract: Cyber-Physical Systems (CPS) lead to the 4-th Industrial Revolution (Industry 4.0) that will have benefits of high flexibility of production, easy and so more accessible participation of all involved parties of business processes. The Industry 4.0 production paradigm is characterized by autonomous behaviour and intercommunicating properties of its production elements across all levels of manufacturing processes so one of the key concept in this domain will be the semantic interoperability of systems. This goal can benefit of formal methods well known various scientific domains like artificial intelligence, machine learning and algebra. So the current investigation is on the promising approach named Formal Concept Analysis (FCA) to structure the knowledge and to optimize the CPS interoperability.

Abstract: The main issue with smart home/building management is that products from different manufacturers are incompatible with each other, and even products of the same brand are unlikely to really interoperate. To provide a fully interoperable solution or solution which is inherently leaned towards improved interoperability, a robust architectural approach to design an underlying system is compulsory. This paper presents one of possible architectural approaches for an integrative management platform of different smart home and building automation systems. Proposed approach is devised upon serviceoriented deployment principles to mediate and transform messages across a variety of systems, services and APIs. It addresses open interoperability standards and exploits the existing semantic data models to support different integration and interoperability patterns. At its core, the proposed approach is leveraged by an open software Smart Home Gateway responsible for communication with proprietary solutions and vendor data formats. Furthermore, it envisions the Canonical Data Model which serves as unified (XML-based) messaging format across the smart home management platform providing an abstraction of proprietary data semantics.

Abstract: The overall increase in number of intelligent devices and home appliances, effectively taking part in the IoT world, has led to the development of numerous innovative solutions across various domains. This paper introduces an ICT based solution for energy End User engagement towards more efficient behaviors and sustainable lifestyles. The proposed solution focuses on integration of comprehensive energy consumption monitoring platforms and/or Home Automation solutions with advanced energy services through a highly flexible cloud-based IoT platform and a service based approach. Employed energy services range from consumption data analytics and profiling over to performance evaluation and consumption prediction towards demand benchmarking and optimization. The ability to develop offer advanced energy services independently, i.e. as functional upgrade of an existing IoT platform, is used to formulate innovative business models and deliver a cost-effective solution.

Abstract: Agribusiness industry and food production are dealing with perishable products, unpredictable production and supply chain management variations, tendency to maximize waste reduction and adherence to strict regulations related to food safety. The Internet of Things (IoT) includes several technologies that can be put into practice, in order to respond to all these demands and resolve challenges, as it allows remote control of the production conditions and distribution of products. In this paper, an architecture for IoT-based information systems in seed-to-fork food production is proposed. The architecture includes IoT from the first to the last phase of production, distribution and even feedback consumers. The authors propose a hybrid solution that combines the IoT and Future Internet

Abstract: In this paper we propose a simple method for measuring the similarity of interests between Facebook users and grouping them with this measure in mind. The ultimate goal is to build a web application which will integrate with Facebook and offer users the ability to find other users similar to them, starting from the premise that users are similar as much as their interests match. A user of this application will be able to look through a list of similar users (sorted by descending degree of mutual similarity) and see the particular shared interests with each of those similar users.

Abstract: Computational Systems, from transactional systems, as discussed in this work, provide information during the departure of any sport. This information will help the technician to decide about the possibilities of reorganization of the team and analyze situations that the team has deficiencies. In this way, it is possible to act in the deficiencies to improve the technical performance and the team's tactical. The study had as objective to present the development of a digital system for the review of game of modality of goalball Paralympic. The information of the scout technician brings the reality of facts and help the coach to modify the staff of the best way to win the game. In some cases, the player who performs a move wonderful, may have committed three or four imperiling errors, an illusion to believe that the player is well in the block. This model of system propitiated an improvement in the understanding of the game by players, aiding in the "feedback" between the technical commission and athletes and helping in the identification of problems that occur with players during drills and games. The financial statements at run time showed that it is possible to analyze the team during the departure. The basic charts generated offer a preview of the results of the team, in real time. Another point considered is volume of information acquired during the time of departure, due to ease of data management. We conclude that the results obtained indicate that the digital system for analyses goalball game is a means needed for the analysis technique and tactics, both individually and collectively, serving as a tool for the preparation of strategies for drills and games and helping in the decision making process.

Abstract: Although short texts, tweets and other types of comments that can be found on social networks, carry significant information about the person who wrote them and the object to which they refer when they are cumulatively observed. The interest for machine analysis of the feelings that are expressed by some text grows, but algorithms that perfectly determine emotion that text carries are still not found. Problems that occur in the machine analysis of the text are different, starting with the complexity of the language in which the text is written to the complexity of the feelings expressed by it. This paper provides an overview of some of the existing solutions for sentiment analysis and gives an overview of our solution. Sentiment analysis will be shown in the case of tweets written in the Serbian language. Although the method is still in the development phase, the resulting accuracy of 82% is encouraging at this stage of the method development.

Abstract: Although the concept of cloud computing is known for more than a decade, not all industries are equally ready to leverage its benefits, e.g. minimum upfront investment costs, significantly lowered hardware administrative burden, vast scalability. Most Industrial Control Systems (ICS) specifically designed and developed for the Smart Grid industry still follow the traditional architecture and are not leveraging the benefits brought by the cloud. One such ICS is the Advanced Distribution Management System (ADMS), which relies on vast network models to monitor, control and optimize the work of electric power distribution systems. The goal of this paper is to analyze the costs of shifting the ADMS from on premise to a cloud solution. In the discussion, we compare two cloud providers, Microsoft’s Azure and Amazon Web Services. We estimate the cloud computing costs for small, medium and large utilities and compare them to the traditional solution.

Abstract: Bringing “smartness” into tourism destinations requires dynamically interconnecting stakeholders through different concepts of Information and communication technologies (ICTs) with tourists. Traditionally, communication between stakeholders like the government authorities, retail business executives, retailers and tourists is realized as one-way communication pointed from stakeholders towards tourists, which implicates that stakeholders are mostly left without feedback. This research examines the smart tourism phenomena, with a focus on communication flows between stakeholders and tourist. An overview has been made on the traditional communication models, highlighting their disadvantages. We propose an architecture which supports two-way communication, in order to mitigate these problems, relying on current technical achievements such as internet of things (IoT), data mining techniques, high speed 3G/4G mobile networks, etc.

Abstract: Business process management (BPM) systems are becoming increasingly data-driven. At the same time IoT (Internet-of-Things) sensing devices are generating large quantities of data. Some of this data represents information about critical events. These events may be used to trigger and drive some automated BPM response procedures to minimize negative effect of critical event or to repair potential damage presented by those events. In this paper, we are proposing pairing of BPM driven application and IoT for sensor error discovery and recovery automation.

Abstract:

Abstract: Recently there is a rapid increase in application of software intensive systems in the domain of Electric Power Distribution (EPD). EPD companies use a significant number of field devices that are usually connected via a cloud infrastructure. While communicating with each other such devices produce enormous amounts of data that should be processed in order to be of any value for the EPD and the software that is deployed into its infrastructure. Systems that deal with processing of such data, especially in cloud-based environment are denoted as Stream Computing Systems. They usually have very strong quality requirements for reliability, performance, scalability and etc. In this paper we propose an architectural approach to manage dependability and more specifically – reliability of stream computing systems. For this purpose we apply the principles of autonomic computing to an architecture that is based on microservices.