ICT Innovations 2017

Viral videos can reach global penetration traveling through international channels of communication similarly to real diseases starting from a well-localized source. In past centuries, disease fronts propagated in a concentric spatial fashion from the source of the outbreak via the short range human contact network. The emergence of long-distance air-travel changed these ancient patterns. However, recently, Brockmann and Helbing have shown that concentric propagation waves can be reinstated if propagation time and distance is measured in the flight-time and travel volume weighted underlying air-travel network. Here, we adopt this method for the analysis of viral meme propagation in Twitter messages, and define a similar weighted network distance in the communication network connecting countries and states of the World. We recover a wave-like behavior on average and assess the randomizing effect of non-locality of spreading. We show that similar result can be recovered from Google Trends data as well.

Penetration of Plug in electric vehicles (PEVs) is expected to rise in the next few years especially in areas with new deployed smart power grid systems. Charging and discharging of PEVs will introduce several challenges related to load stabilization and information security. In this paper, we discuss a secure discharging protocol where users can be protected from possible information security and privacy attacks. The protocol also incorporates required remote authorization and payment transaction mechanisms. Our protocol is developed based on the use of encryption mechanisms and the dual signature approach. Using the security protocol verification tool, Automatic Verification and Analysis of Internet Security Protocols (AVISPA), the security aspects of the proposed protocol are verified. Our approach is robust against misuse of electric vehicles and unfair payment issues as it allows for user-based authentication in addition to the authentication of associated electric vehicles.

This article presents a system for early detection and alerting of the onset of a heart attack. The system consists of a wireless and mobile ECG biosensor, a data center, smartphone and web applications, and a remote 24 h health care. The scientific basis of this system is founded on the fact that a heart attack can be detected at least two hours before its onset, and that a timely medical attention can dramatically reduce the risk of death or serious tissue damage.

So far, there are no commercial products matching the goals and functionalities proposed by this system, even though there are a number of proof-of-concept studies, and a number of similar products on the market. For the greater part, these currently offered solutions are specifically intended for conducting stress tests in modern hospitals, or as personal fitness devices. Most of them have limited battery power, do not use algorithms for heart attack detection, and/or require constant supervision by medical personnel.

Internet of Things (IoT) has revolutionized transportation systems by connecting vehicles consequently enabling their tracking, as well as monitoring of driver activities. Such an IoT platform requires a significant amount of data to be send from the on-board vehicle to the off-board servers, contributing to high network usage. The data can be send at regular intervals or in an event-based manner whenever relevant events occur. In interval-based approach, the data is send even if it is not relevant for reporting leading to a wastage of network resources, e.g., when the data does not change considerably compared to the previously sent value. In this paper, we investigate the possibility of using an event-based architecture to send data from the on-board system to the off-board system. The results show that our event-based architecture improves the accuracy of data available at the off-board system, by a careful selection of events. Moreover, we found that our event based architecture significantly decreases the frequency of sending messages, particularly during highway driving, leading to reduced average data transfer rates. Our results enable a customer to perform trade-offs between accuracy and data transfer rates.

Road safety, traffic congestion and efficiency of the transport sector are major global concerns. Improving this is the primary objective of intelligent transport systems (ITS). Having Internet of things (IoT) based solutions for ITS would enable motorists to obtain prior contextual guidance to reduce congestion and avoid potential hazards. IoT based solutions enabling collection of data from client nodes in a wireless sensor network in the transport environment implementing ITS goals is studied. The parameters to be monitored, type of sensors and communication related design parameters are identified to develop an effective IoT based solution. Road safety techniques studied include distance sensing, improper driving detection and accident prevention, weather related events and negligent driving detection and accident avoidance. Vehicle to vehicle communication and vehicle to infrastructure based channels are studied. Wireless communication technologies suitable for the channels are studied. Additional benefits and services that can be added to a system with the IoT approach are also studied. The effectiveness of such a system is studied with the use of validation framework. Multiple case studies of current and future IoT based ITS along with the challenges in the application is discussed.

We study a class of random walks on graphs with two mechanisms: local diffusion and random relocation. Such mechanisms are common in search algorithms, an example being the PageRank. The rate with which two mechanisms are mixed is called damping factor. It determines the first-passage time, defined as the time required for a walker starting from a source node to find a given target node. We provide bounds on the stationary distribution of random walks. Global mean first-passage time as a function of the damping factor is computed in closed form for a simple example. The results provide new insights on the search engines.

The HTTP adaptive streaming (HAS) is a popular mechanism for delivery of live and on-demand video contents encoded with different qualities and divided into segments with equal length. The mechanism adapts the requested segment qualities to the quality of the link, providing uninterrupted service even in congested network conditions. In this work, we analyze the HAS for delivery of Video on Demand (VoD) contents from server performance point of view for different segment lengths and different network conditions. For that purpose, we created an environment for real-case measurements of the server performance and measured performance parameters like CPU utilization, generated in-bound and out-bound traffic and number of established TCP connections. From the analysis of the obtained data, we conclude that streaming of shorter video segments generates more appropriate and predictable traffic pattern, but requires more CPU power and TCP connections. Therefore, the shorter contents are suitable for streaming in networks with very low packet losses. Longer video segments, on the other hand, tend to require more resources only at the beginning of the streaming session, which they release before the end of the session, and hence, alleviate the network equipment. The main advantage of using long segments is that they can achieve uninterrupted streaming experience even in harsh network environments such as congested wireless networks.

We developed a sensor node to support inter-node distances of up to 1 km, low-power consumption and low cost hardware. The design is modular and application specific sensors can be attached. It provides a flexible solution to a wide range of applications. An optimized MAC and routing layer (based on RPL and Trickle) supports interoperability among heterogeneous networks and low energy consumption over long distance transmissions of up to 1 km.

FPGA acceleration of compute-intensive algorithms is usually not regarded feasible because of the long Verilog or VHDL RTL design efforts they require. Data-parallel algorithms have an alternative platform for acceleration, namely, GPU. Two languages are widely used for GPU programming, CUDA and OpenCL. OpenCL is the choice of many coders due to its portability to most multi-core CPUs and most GPUs. OpenCL SDK for FPGAs and High-Level Synthesis (HLS) in general make FPGA acceleration truly feasible. In data-parallel applications, OpenCL based synthesis is preferred over traditional HLS as it can be seamlessly targeted to both GPUs and FPGAs. This paper shares our experiences in targeting a demanding optical flow algorithm to a high-end FPGA as well as a high-end GPU using OpenCL. We offer throughput and power consumption results on both platforms.

Searching of target based on random movements in space is an interesting topic of research relevant in different fields. For searching in complex networks besides the classical random walk, various biasing procedures have been applied for reducing the searching time. We propose one such biasing algorithm that favors movements towards more distant nodes, while penalizing going backward. Using Monte Carlo numerical simulations we demonstrate that the proposed algorithm provides lower Mean First Passage Time for several types of generic and real complex networks.

Weed detection from images has received a great interest from scientific communities in recent years. However, there are only a few available datasets that can be used for weed detection from unmanned and other ground vehicles and systems. In this paper we present a new dataset (i.e. Carrot-Weed) for weed detection taken under variable light conditions. The dataset contains RGB images from young carrot seedlings taken during the period of February in the area around Negotino, Republic of Macedonia. We performed initial analysis of the dataset and report the initial results, obtained using convolutional neural network architectures.

The rule induction algorithms generate rules directly in human-understandable if-then form, and this property is essential of successful intelligent classifier. Similar as crisp algorithms, the fuzzy and rough set methods are used to generate rule based induction algorithms. Recently, a rule induction algorithms based on fuzzy-rough theory were proposed. These algorithms operate on the well-known upper and lower approximation concepts, and they are sensitive to different T-norms, implicators and more over; to different similarity metrics. In this paper, we experimentally evaluate the influence of the T-norm Algebraic norm on the classification and regression tasks performance on three fuzzy-rough rule induction algorithms. The experimental results revealed some interesting results, moreover, the choice of similarity metric in combination with the T-norm on some datasets has no influence at all. Based on the experimental results, further investigation is required to investigate the influence of other T-norms on the algorithm’s performance.

Analyzing human trajectories based on sensor data is a challenging research topic. It has been analyzed from many aspects like clustering, process mining, and others. Still, less attention has been paid on analyzing this data based on hidden factors that drive the behavior of people. We, therefore, adapt the standard matrix factorization approach and reveal factors which are interpretable and soundly explain the behavior of a dynamic population. We analyze the motion of a skier population based on data from RFID-recorded ski entrances of skiers on ski lift gates. The approach is applicable to other similar settings, like shopping malls or road traffic. We further applied recommender systems algorithms for testing how well we can predict the distribution of ski lift usage (number of ski lift visits) based on hidden factors, but also on other benchmark algorithms. The matrix factorization algorithm showed to be the best recommender score predictor with an RMSE of 2.569 ± 0.049 and an MAE of 1.689 ± 0.019 on a 1 to 10 scale.

The amount of online courses and educational content available on the Internet is growing rapidly, leaving students with large and diverse number of choices for their areas of interest. The educational content is spread into diverse e-learning platforms, making its search and comparison even more challenging. Classifying educational content into a standardized set of academic disciplines or topics can improve its search, comparison and combination to better meet students’ inquiries. In this paper we make use of well-known techniques from Information Retrieval to map course descriptions into two common sets of topics, one manually created and well-controlled, i.e. CIP, and one collaboratively created, i.e. Wikipedia. We then analyze and compare the results to see how the size of the topic schemes and their associated data, such as textual descriptions, affect the accuracy of the end results.

Emotions are psychological phenomena present in the living beings. However, there is still no consensus for a more precise definition of the emotions. As a complex concept, the emotions can be part of a robot model for different usage in variety of robots. This paper presents the state of the art of robot behavior models that use emotions in a chronological order. The aim of this paper is to formalize the usage of emotion in robotics. Thus, a definition and explanation of a set of distinct parts in robotic models influenced by emotions is proposed. On the other hand, the properties of robots that use emotions, the so-called emotional robots, are also explored. As a result, the most important different properties that emotional robots should possess are retrieved and explained in detail. The aim of the formalism presented in this paper is to give tribute to the work already done and to give possible directions for the future work with emotional robots.

The aim of the paper is to present image retrieval for Alzheimer’s Disease (AD) based on brain atrophy pattern captured by the SPARE-AD (Spatial Pattern of Abnormality for Recognition of Early Alzheimer’s Disease) index. SPARE-AD provides individualized scores of diagnostic and predictive value found to be far beyond standard structural measures. The index was incorporated in the image signature as a representation of the brain atrophy. To evaluate its influence to the retrieval results, Magnetic Resonance Images (MRI) provided by the Alzheimer’s Disease Neuroimaging Initiative (ADNI) were used. For this research, baseline images of the patients with diagnosed AD and normal controls (NL) were selected from the dataset, including 416 subjects in total. The obtained experimental results showed that the approach used in this research provides improved retrieval performance, by using semantically precise and powerful, yet low dimensional image descriptor.

Traditional recommendation systems rely on past usage data in order to generate new recommendations. Those approaches fail to generate sensible recommendations for new users and items into the system due to missing information about their past interactions. In this paper, we propose a solution for successfully addressing item-cold start problem which uses model-based approach and recent advances in deep learning. In particular, we use latent factor model for recommendation, and predict the latent factors from item’s descriptions using convolutional neural network when they cannot be obtained from usage data. Latent factors obtained by applying matrix factorization to the available usage data are used as ground truth to train the convolutional neural network. To create latent factor representations for the new items, the convolutional neural network uses their textual description. The results from the experiments reveal that the proposed approach significantly outperforms several baseline estimators.

In this paper, we propose a new algorithm for clustering multi-dimensional time series (MDTS). It is based on the predictive clustering paradigm, which combines elements of predictive modelling and clustering. It builds upon the algorithm for predictive clustering trees for modelling time series, and extends it to model MDTS. We also propose adequate distance functions for modelling MDTS. We apply the newly developed approach to the task of analyzing data on forest growing stock in state-owned forests in Slovenia. This task of high importance, since the growing stock of forest stands is a key feature describing the spatio-temporal dynamics of the forest ecosystem response to natural and anthropogenic impacts. It can be thus used to follow the structural, functional and compositional changes of forest ecosystems, which are of increasing importance as the forest area in Europe has been growing steadily for the last 20 years. We have used two scenarios (quantitative and qualitative) to analyze the data at hand. Overall, the growing stock in Slovenian forests has been increasing in the last 40 years. More specifically, the growing stock of the three tree-size has progressive dynamics, which indicates that Slovenian state-owned forests have balanced structure.

Random Codes Based on Quasigroups (RCBQ) are cryptcodes, so they provide (in one algorithm) a correction of certain amount of errors in the input data and an information security. Cut-Decoding and 4-Sets-Cut-Decoding algorithms are proposed elsewhere and they improve decoding of these codes.

In the decoding process of these codes, three types of errors appear: more-candidate-error, null-error and undetected-error. More-candidate-errors can occur even all bits in the message are correctly transmitted. So, the packet-error (and bit-error) probability can be positive for very small bit-error probability in the noise channel. In order to eliminate this problem, here we define new decoding algorithms (called Fast-Cut-Decoding and Fast-4-Sets-Cut-Decoding algorithms) that enable more efficient and faster decoding, especially for transmission through a low noise channel. We present several experimental results obtained with these new algorithms. Also, we analyze the results for bit-error and packet-error probabilities and decoding speed when messages are transmitted through Gaussian channel with different values of signal-to-noise ratio (SNR).

A large number of emerging services expose their data using various Application Programming Interfaces (APIs). Consuming and fusing data form various providers is a challenging task, since separate client implementation is usually required for each API. The Semantic Web provides a set of standards and mechanisms for unifying data representation on the Web, as well as means of uniform access via its query language – SPARQL. However, the lack of data protection mechanisms for the SPARQL query language and its HTTP-based data access protocol might be the main reason why it is not widely accepted as a data exchange and linking mechanism. This paper presents an authorization proxy that solves this problem using query interception and rewriting. For a given client, it solely returns the permitted data for the requested query, defined via a flexible policy language that combines the RDF and SPARQL standards for policy definition.

In this paper we consider an error-detecting code based on linear quasigroups. We give a proof that the code is linear. Also, we obtain the generator and the parity-check matrices of the code, from where we obtain the Hamming distance of the code when a linear quasigroup of order 4 from the best class of quasigroups of order 4 for coding, i.e., the class of quasigroups of order 4 that gives smallest probability of undetected errors is used for coding. With this we determine the number of errors that the code will detect for sure.

Boolean functions are mappings \(\{0,1\}^n\rightarrow \{0,1\}\), where n is a nonnegative integer. It is well known that each Boolean function \(f(x_1,\dots ,x_n)\) with n variables can be presented by its Algebraic Normal Form (ANF). If (G, F) is an algebra of order \(|G|, \ 2^{n-1}\le |G|< 2^n\), where F is a set of finite operations on G, then any operation \(f\in F\) of arity k can be interpreted as a partial vector valued Boolean function \(f_{v.v.}:\{0,1\}^{kn}\rightarrow \{0,1\}^n\). By using the function \(f_{v.v.}\) and ANF of Boolean functions, we can characterize different properties of the finite algebras, and here we mention several applications. We consider especially the case of groupoids, i.e., the case when \(F=\{f\}\) consists of one binary operation and we classify groupoids of order 3 according to the degrees of their Boolean functions. Further on, we give another classification of linear groupoids of order 3 using graphical representation. At the end, we consider an application of Boolean representation for solving a system of equations in an algebra.

The initial idea for this research is to study the testing metrics throughout different software development methodologies, their importance and the way they influence the software quality. For that purpose, the research is conducted within an international IT company, where the teams are formed on geographically different locations. The main problem investigated is whether the team is committed to flawless and impeccable test management, regardless of the methodology used and how the clients’ requirements can affect it. The conclusion is distributed among the case studies and it is based on how the testing metrics can help monitor the development and testing process.

There are many approaches to risk management, and the practice shows that they are not suitable for IT-centric Micro Small and Medium Enterprises (MSME), but more targeted to large and complex organizations. At the same time, the existing approaches in isolation are aimed at a particular type of risk, not taking into account that MSMEs need a more integrated approach, constraint on time and resources and availability of data. Based on the field research of over 150 organizations, the initially proposed risk management framework was revised generally in the area of scope of the risk, duration, risk management team and risk valuation model. Various risk models were reviewed for appropriateness. The development of IT and its use in organizations allows for preference to data-driven models, but the limitation of MSMEs with resources, and understanding of complex data-driven model limits their use. The field research showed that MSMEs prefer a hybrid method for assessment of risks, as they couldn’t sustain a fully quantitative approach and as managers feel more confident with qualitative estimates.

This paper describes the results of first phase of our original research about implementing gamification in the university e-courses based on the learning styles. Main research aims were to conduct comprehensive review of literature and current practice in implementation of gamification concept in higher education and to identify game elements that can make a positive impact on a specific learning style (LS). Result of first research phase is conceptual model for including game elements in e-learning courses in higher education based on Felder-Silverman Learning Style Model (FSLSM). The implementation and evaluation of the proposed model is planned as a future work. Our model presents extension of FSLSM model providing not only information about the learning styles, but also about their semantic groups in game-based learning contexts. The proposed model can be used as theoretical framework for further research on relationships between student’s learning style, achievements and behaviors in Virtual Learning Environments (VLE).

Behavioral Targeting (BT) is a technique used by online advertisers to increase the effectiveness of their campaigns, and is playing an increasingly important role in the online advertising market. Although it’s a technique that has been used for more than 10 years in online advertising, and there are surveys of its effects on developed markets, there is little evidence for the results of behavioral targeted campaigns in the developing countries as Macedonia. In this paper we give a short review of what behavioral targeting is, we list the forms of behavioral targeting, its advantages and disadvantages, and we present a comparative analysis of two advertising campaigns for a Macedonian food company - using behavioral targeting vs. using only geo-targeting (only users in Macedonia) to show the difference in results. From this analysis we may conclude that the targeted advertising provides far more than non-targeted at similar conditions. The difference in clicks is huge, even with slightly fewer impressions, and if we consider that the cost of the campaign are usually dependent on impressions (more impressions - more expensive campaign), then the cost of the campaign in targeted advertising would be lower, and finally the overall CTR is almost 5 times higher for targeted advertising. This paper tries to contribute to the scientific-research field and offer a motivation for the digital marketing agencies in developing countries and their clients to start using more behavioral targeting in their campaigns and accomplish higher results.