This paper presents a web-based environment, designed to provide researchers of Old Church Slavonic manuscripts with an intuitive, semantically sound framework and the respective tools for collection, description, categorization, analysis, and presentation of their respective findings. And while the digital repertorium described here concentrates mostly on the paleographic features of the South Slavic manuscripts from the fourteenth century, a period of considerable literary achievements in the Balkans, we feel that the incorporation of the state-of-the-art methods form the field of digital presentation and preservation of cultural heritage will contribute to the better understanding of a still under-researched topic.
This article deals with a new handwritten dataset, Kyrgyz-MNIST, and recognition of handwritten Kyrgyz letters through the Artificial Neural Network. There were collected 89 023 manuscript samples of 36 Kyrgyz letters. Since 33 of the 36 letters in the Kyrgyz language are similar to the Russian letters, the special letters “ң”, “ү”, and “ө” were studied with special emphasis. The system of handwriting recognition works with Image processing methods and Machine Learning algorithms helped to recognize Kyrgyz manuscripts and convert them into computer letters. As a result, the CNN method recognized the handwriting letters more accurately. In total 36 letters were recognized with an average accuracy of 99.1%. Newly studied, the special letters in Kyrgyz Language are also recognized with more than 99% correct.
ABSTRACT: The paper delves into the fundamentals of information systems security, providing an analysis of the security levels of both information and network resources. It describes various types of attacks on information systems along with their detection methods. Additionally, it characterizes systems based on anomaly detection methods in information system behaviors and outlines key aspects of constructing attack detection systems. The study further investigates network anomalies and provides detailed characterizations of attack types. Novel methods for detecting attacks on information systems have been developed, including a new approach involving the utilization of threat expectation reports for attack detection purposes. Moreover, a method for detecting anomalies based on behavior in information systems has been devised. Specifically, the presented algorithm focuses on establishing a model of regular or typical system/user functionality. Through this approach, the system compares current activity rates with the normal activity profile, identifying significant deviations as potential sources of attack or network anomalies.
Analysing dataset representing ten years (1999-2008) of clinical care at 130 US hospitals and integrated delivery networks which includes over 50 features representing patient and hospital outcomes, we have implemented various machine learning and deep learning models to achieve higher accuracy in classifying diabetic patients in order to predict the early re-admission cases. To overcome the sever challenge we have faced with class imbalance in our datasets, we have introduced SMOTE sampling technique. Having resampled our data and implemented our deep learning architecture, we have obtained an accuracy of 98%. The relatively higher performances which were obtained upon using both support vector machine and deep learning models are suggestive of implementing these algorithms in order to overcome challenges related to hospital diabetic re-admission cases in both managing the patents and preventing this devastating disease.
Abstract. In the conditions of rapid development of technologies for evaluation and growth of scientific research, the issue of effective organizational process of scientific competitions becomes especially important. This article is devoted to the analysis of prospects and innovations in the development of automated expert systems used in the organization of scientific competitions, with the main focus on the formation of topics of research works.
Non-Fungible Tokens (or, for short, NFT) establish the ownership and the authenticity of digital assets. Regardless the high security levels provided by blockchain, are not excluded the cases where the ownership of a digital object may be claimed without proper rights. Fraud and copyright infringement should be prevented through the insurance of the authenticity and the integrity of a digital object associated with an NFT. In this work we consider the case of NFT images under the aspect of being susceptible to duplication, theft, tampering, fraud, or in general any potential threat against their authenticity and integrity properties. To deal with such cases we propose an integrated system that provides an a priori embedding of authenticity and integrity related information by deploying robust authenticity information hiding in the frequency domain alongside fragile integrity information hiding in the spatial domain before the image is being publicly available as NFT. Moreover, an automated secure procedure for the development of smart contracts alongside the development of an NFT dual that serves as backup and is not publicly available through the marketplace is proposed in order to retain the initial digital image. We design a modular architecture that incorporates all the components proposed in this work and discuss its potentials on its deployment over different digital objects that can be considered as NFTs.
The main objective of this study is the prediction of the academic performance of students after graduating high school and pursuing a bachelor's degree so we could help them to choose the best suited degree regarding their performance in secondary school with the help of ML algorithms. We try to achieve this by proposing a new model based on machine learning algorithms to predict the grade of algebra and calculus in the first semester of university for first year students. The dataset chosen was real, collected from Polytechnic University of Tirana, the faculty of Information Technology. It was of the students distributed in 5 years. Core parameters analyzed include math, physics, elective math and high school average grade, along with Matura exam performance. The methodology includes data collection, employing diverse algorithms and evaluation methods like Random Forest, K-Nearest Neighbor, Naïve Bayes, and Artificial Neural Networks. The study highlights data analysis through graphical representation, employing programming languages like Python. This study showed that out of the ML models that were used Random Forest has the best performance with an accuracy of 90-93%. According to the study findings, the main factors that had the most effect were elective math, overall average math, and physics. We found out that you can indeed predict how well students will do using just a few key pieces of information with high accuracy.
Bulgarian Water Utility operators must monitor flow and pressure on 15-minute intervals and report that data to Bulgarian Energy and Water Regulatory Commission for each of their water intakes and district metering/pressure management zones [10]. Currently this is happening either manually by an operator or automatically through a SCADA (Supervisory Control and Data Acquisition) system with data loggers with 2G/GSM network transmission. LoRa as an LPWAN (Low Power Wide Area Network) technology claims to be low cost, low power, long range secured technologies for data transmission. The research question of this case study is does LoRa (Long Range modulation) as a representative of the emerging IoT (Internet of Things) LPWAN technologies cover that water utility requirement considering the specifics of the use case and the regulatory limitations of ISM(Industrial Scientific and Medical) frequency range usage and more specifically is it possible to transfer as much data as required by the regulator within the limited duty cycle and fair access policy of the LoRa/LoRaWAN operators. The business benefit of this study for the water utility operator is that they will be able to lower their costs for network monitoring by deploying and maintaining their own LPWAN networks in the ISM frequency range. In addition, they will be able to build coverage and get not just monitoring data but also implement smart metering which is another big use case for them. The technical benefit would be that water utilities and other users will be able to high tune their LoRa/LoRaWAN equipment to fulfill specific use case requirements and still comply with ISM limitations.
This article inquires how blockchain technology can be utilized to improve validation and certification of skills in education to ensure a transparent, secure, and verifiable record of the knowledge that students acquire. Blockchain technology is introduced followed by an analysis of previous works using it in education. The paper presents key benefits for using blockchain technology for educational certification and proposes an architecture how it can be utilized in the diploma awarding process for graduates that makes validation accessible and secure. The architecture utilizing blockchain in education provides an efficient and secure way to verify academical skills, eliminating possibilities for fraud and irregularities in the graduation process.
Air quality stands as a pivotal factor influencing public health and well-being, shaping urban planning strategies, health management practices, and environmental policies. Having meticulously scrutinized PM2.5 concentration data from 42 monitoring stations across Bishkek, we have employed meteorological factors, including temperature and humidity from these stations to model air quality. We have thus implemented two distinct prediction methodologies namely, Random Forest Regression (RFR) and Long Short-Term Memory (LSTM). Our findings unequivocally favored RFR as the superior predictor for PM2.5 values, achieving accuracy levels of up to 90%.
The Felder-Solomon Index of Learning Styles is a questionnaire designed to provide a solid foundation for understanding student learning differences and preferences by addressing individual characteristics to drive into effective learning. This work aims to validate the Felder-Solomon Index of Learning Styles questionnaire originally created in English, in the Mozambican context for higher education students, considering the specific cultural characteristics of this context. The students involved were from initial programming learning classes and we wanted to verify their predominant learning styles. To verify its viability, it was necessary to carry out the cultural adaptation to the Portuguese language with the help of language experts. Subsequently, a pilot study was conducted to test the questionnaire's effectiveness in a convenience sample of 205 students. Cronbach's alpha coefficient was also used to measure the reliability of the questionnaire by checking its internal consistency. The Kaiser-Meyer-Olkin criterion was used for factor analysis and testing the overall consistency of the data. In order to verify the predominant learning style, it was necessary to evaluate the tendency of the answers, verifying the highest percentage of students with a certain preference. The Cronbach Alpha and the KMO values obtained were within the established reference ranges. The results also indicate that the questionnaire is valid for assessing students' learning preferences in higher education in the Mozambican context, and most students in the study were identified with a mild preference.
In this paper we have identified several good/promising practices from the literature of using Social Humanoid Robots (SHR) as which can contribute to the education, social integration and improvement of lifestyle of individuals with Autism Spectrum Disorder (ASD). Our interest is on the commercial SHR currently available on the market. We provide analysis and evaluation of the effectiveness, advantages and drawbacks, accessibility, user feedback and expert opinions, ethical considerations and future directions. The analysis proof that SHR are promising computer based Assistive Technology (AT) for individuals with ASD.
The inherent complexity of Introductory Programming contributes to the persistently high failure and dropout rates among Higher Education students. Several studies have been conducted to find possible solutions to this problem. Through this study, a systematic review was carried out where factors that influence the learning of programming concepts were analysed, as well as techniques to anticipate student's aptitude and practical strategies to support student's learning process to contribute to a deeper understanding of this problem and help in the search for a solution, or a path to reach a solution. The systematic review extracted 525 candidate papers, of which 49 were analysed, as the rest did not meet the established inclusion and exclusion criteria. The analysis of the results suggested that self-efficacy and prior knowledge are the most common factors identified that influence student performance. Data mining emerges as the most relevant aptitude prediction technique. At the same time, active learning, pair programming, and flipped classes appear to be promising strategies to help in the process of learning to program. These findings have significant implications for educators and professionals involved in teaching programming in higher education institutions, offering valuable insights for improving the quality and effectiveness of these courses.
The paper presents a hybrid algorithm for mathematical modeling of a photoplethysmographic signal using two Gaussian functions generating the basic waveforms of the signal. The presented algorithm makes it possible to synthesize series with variable maximum amplitude and different shape of the main waves of the PPG signal. Through the use of the ARIMA method, predictive modeling of the signal has been carried out and realization of long-term PPG signals in which the influence of circadian cycles is embedded. Studies of the statistical and frequency parameters of the variability of the synthesized signal were performed and time series histograms and Power Spectral Density were made, which were compared with a real PPG signal. The synthesis time of PPG series of different length is investigated.
Asynchronous machines are essential components that drive critical systems across industrial, trading, and residential sectors, powering heating units, pumps, and various appliances. Yet, ensuring their reliable process is paramount to prevent costly defects and maintain productivity. Notably, failures in the rolling element bearings (REB) account for about forty percent of motor failures, underscoring the urgency of early detection to mitigate operational risks and financial losses. To address this challenge, this paper proposes an innovative smartphone-based diagnostic technique for detecting bearing faults in induction machines. Leveraging the common availability and computational capabilities of smartphones, the approach utilizes the devices’ audio recording functionality to capture motor audio signals. Audio data collected from rotating machines with various fault types is used to train a 1D Convolutional Neural Network (1D CNN), and the trained model is then deployed on a smartphone for real-time fault diagnosis. Embedding this approach into a user-friendly mobile application enhances accessibility and usability, offering a cost-effective solution for fault diagnosis in induction machines.
In the conditions of digitization, electronic educational resources are important to support the learning and development process. The accessibility of learning materials for disabled people plays a key role in ensuring universal education and ensuring equal access to knowledge for all, regardless of their individual needs and capabilities. The paper examines the application of natural language processing and text summarization technologies to improve the accessibility of electronic educational resources. The research focuses on the use of automated text processing methods to facilitate the access to information in electronic learning environments. Through the application of these technologies, better understanding and quick access to key information is achieved, thus improving the overall effectiveness and inclusiveness of e-learning resources. The research examines different approaches to text processing and summarizing, emphasizing the technical possibilities for optimizing the learning process and supporting students with different needs. The obtained results demonstrate the potential of integrating these technologies in e-learning to improve the accessibility and effectiveness of learning resources.
Preserving the authenticity and provenance of artifacts is crucial for our cultural heritage. This study explores the potential of Decentralized Identity on Blockchain for artifact identification and verification. A review of related work revealed a great body of standards and guidelines for identification, documentation and verification of cultural heritage artifacts, as well as various use cases for the application of Decentralized Identity on Blockchain in different industries. However, this is the first study that demonstrates that Decentralized Identity on Blockchain can be successfully used for artifact identification and verification. In order to gather pertinent information about underlying processes, attributes, credentials, roles, and behaviors and validate the applicability of the proposed model, we conducted face-to-face and phone interviews with experts from relevant businesses and institutions. We further proved our hypothesis by implementing a range of methodologies - comparative analysis, experimental research, functional mapping. Finally, we put the proposed use case to the test by implementing an experiment using existing tools and analyzing the results. We conclude the study by showing the potential challenges, limitations and future work.
Sleeping posture recognition attracts much attention due to its crucial role in healthcare applications. Many approaches are proposed to address this task using in-bed pressure sensor data, including deep neural network models with their robust capabilities and neuromorphic computing for energy efficiency. This paper proposes a hybrid model by combining a RANC-compatible Spiking Neural Network model with a CNN-based feature extractor to classify 17 sleeping postures on a pressure dataset. Our model achieves a state-of-the-art accuracy of 93.0 % using the Leave-One-Subject-Out (LOSO) validation scheme, while the model's size is only 3.9 MB. This result significantly surpasses the previous SNN-based solutions and reveals the vast potential of combining SNN and CNN for various applications.
Limited ability for artists and designers to create outline images on vertical surfaces such as walls. Methods, such as hand painting, are time-consuming, labor-intensive, and limit the creativity of professionals and art enthusiasts. Picasso Robotics System aims to revolutionize this process by combining advanced printing technology with artistic creativity, reducing the time required to create each painting while maintaining precision. The main motivation of the project is to bridge the gap in the efficiency and accuracy of traditional methods by integrating design selection and remote control through a dedicated mobile app with artificial intelligence capabilities expanding creative control and simplifying workflow. The mobile application is based on an innovative image processing system based on contour vector elements. This technology provides high accuracy and efficiency in creating designs and also allows you to easily scale and adapt images to different formats.
Determining the distance between objects is an important requirement for autonomous systems operating in a complex non-deterministic environment. This task can be solved with expensive sensors. This research will investigate approaches that are based on input data obtained from a single RGB camera. For this purpose, methods of computer vision and artificial intelligence will be used to convert an image from an RGB image to a depth image. Two of the considered approaches use the depth image as an input and detect the objects directly on it. They can be applied directly to real images but are algorithms that require additional tuning of their parameters. For comparison, a third artificial intelligence approach was applied, which searches for the objects directly on the color image while the distance information is extracted from the depth image. Experiments have been conducted in a real setting to compare the effectiveness and applicability of the approaches when the task is to determine the distance to other road vehicles by using RGB camera mounted to inside of a car.
