In the emerging context of immersive technologies, using Virtual Reality (VR) in education provides innovative ways to experience and understand complex concepts such as chemical reactions. This project demonstrates how VR can revolutionize how students interact with reactants and reaction products, allowing them to observe and manipulate chemical reactions in a controlled and fully immersive environment. By creating a virtual laboratory, participants can directly experience the effects and dynamics of various chemical reactions without the risks of handling real substances. This approach not only increases student engagement and curiosity but also improves information retention and a deep understanding of chemical processes. This project illustrates how the combination of Artificial Intelligence and Virtual Reality can transform STEAM education by providing a more interactive, safe, and accessible educational method. Our solutions present significant advances in educational practices, with the technologies used increasing student engagement exponentially. In other words, by incorporating Artificial Intelligence algorithms within virtual environments, we can take the field of eLearning to a higher level.
This study analyzed EEG signals from patients with spinal cord injuries by examining five different hand-wrist movements. The signal processing steps designed for an automatic and robust EEG-based BCI system were applied sequentially. Initially, 37 different features in the time, frequency, and time-frequency domains were extracted from the preprocessed signal. After trying widely used Manifold Learning (ML) methods in the literature, including t-Distributed Stochastic Neighbor Embedding (t-SNE), Local Linear Embedding (LLE), Multi-Dimensional Scaling (MDS), and ISOmetric Mapping (ISOMAP), we attempted the Spectral Embedding method, which has not yet been utilized in EEG signal analysis. The signals were then classified using three different machine-learning algorithms. The study compared classification performance using the accuracy metric. A multi-class classification method was employed specifically the triple classification method. The most successful performance was achieved by using the ISOMAP machine learning method and kNN classifier for the Pronation-Palmar Grasp-Hand Open combination, with an accuracy of 0.993 +/- 0.016. Other methods used were t-SNE, MDS, LLE, and Spectral Embedding, respectively. Regarding classifiers, the kNN, SVM, and Naive Bayes algorithms were found to be successful in that order. Based on these results, we propose a suitable methodology for designing a robust BCI system.
After a natural disaster, it is very important to be able to make a building damage assessment quickly and accurately. Considering real-world scenarios, one of the most important issues related to the deep learning based computer vision models used in this field is their generalization ability. Ideally, we would like to have a global model which is capable of making damage assessment successfully in all disasters in the future. However, this is a difficult task as images of one disaster varies greatly from images of another due to reasons like different geographical region structure, structure of houses, position of clouds, etc. On the other hand, many damage assessment models are evaluated on an unrealistic in-domain (ID) test set, which contains data from the same disasters in the training and test sets. In this study, we focus on the generalization problem and compare the model performance in an ID test setting with that in an out-of-domain (OOD) test setting, which is more realistic since the test data is formed from disasters not in the training set. We show that the performance of the model is comparatively lower for the OOD test setting and there is a large generalization gap in this case. We conclude that the models should be evaluated by using the OOD test setting. Then, we propose to generate post-disaster images and masks by using generative adversarial network (GAN) models and use the generated data to fine-tune the damage assessment model. It is seen that the generalization ability of the damage assessment model is increased by using this approach.
Finding a job these days is challenging because of the size, diversity, and goals of the market in a society impacted by pandemics, economic crises, or military hostilities. Trust is the most crucial factor in the job domain after performance expectations. It is particularly significant for women, less active job seekers, and people who did not experience job recommendations. Since recommender systems (RS) are one of the most frequently encountered human-centered and online applications in our daily lives, it is important to note that sound principles of trusting the environment of Artificial Intelligence (AI) systems are also required to characterize the trustworthiness of recommender systems. Otherwise, inadequate advice, high expectations and bad interpretations could lead to making bad choices or to demotivating job seekers. This paper expands on previous research, highlighting the point of view of trustworthiness in job recommender systems (JRS) and providing an overview of the dimensions of AI trustworthiness for the job domain. The purpose of this study is to investigate how trustworthy and suggestive outputs can improve the communication between a job mediator and a job seeker by enhancing the credibility of the information provided to job applicants and increasing customer satisfaction.
In this work we have experimented with the use of genetic programming as a feature selection method as well as a classifier to obtain business failure prediction models with different prediction temporal horizons. In the prediction models, a wide set of explanatory variables has been used, all of them based on the annual accounts of the company. In addition, an extended set of explanatory variables incorporating variables from the economic environment has been considered. Comparison of the prediction results between these alternatives shows a trend towards better results using the feature selection process, while there is no trend towards better results using economic environment variables.
Prominence is one of the most important measurements in topography and mountaineering. This paper describes an efficient, almost linear time algorithm for computing mountain prominence for all peaks on Earth using digital elevation models (DEMs). It builds on top of a classic algorithm and leverages the observation that only a few peaks have their prominence determined by a relatively distant other mountain. Thus, the classic algorithm can be adapted to memorize and use less information without the loss of correctness. The algorithm is demonstrated using 3 arcsecond real-life data from SRTM datasets. Its importance is underscored by the increasing accuracy of Earth mapping methods and the corresponding growth in the amount of data that must be processed to compute prominence.
Federated Learning (FL) is a rather new distributed machine learning paradigm based on a collaboratively decentralized privacy-preserving technology. It supports a range of multiple clients (starting from “simple” mobile devices but also including organizations, institutions, etc.) coordinated decentralized machine learning by one or more central servers. As it is a rather new approach various strategies for organization of multiple clients and implementation reliable environments have been developing. The efficiency of selected FL strategy for a particular problem is influenced by involved actors and organizational structure. Particular attention within FL should be paid to adoption of data privacy and security aspects. FL is powerful and widely applicable in areas like banking and finances, automotive industry, IoT and smart environments, health and medicine etc.. In this paper we will pay attention to several crucial aspects of FL, present its essential characteristics, and briefly illustrate several characteristic applications.
Adept data management, as well as analysis, hold an increasing significant role in fast progressing filed of financial modeling. This paper is based on major elements such as relational databases, big data and cloud computing, each of them is presented from a perspective concerning financial modeling. The proposed solution is based on our investigation into the usability of each technology, optimal contexts for their application, and the methods for effectively interconnecting them. Each element is described in terms of operational efficiency, integration and overall performance, with machine learning techniques as means of assessment. Besides the thorough presentation of these technologies, the paper includes an analysis of their implementation so as to improve financial modeling of financial institutions. The interconnectivity and the correct utilization of the application of this technologies may prove of utmost importance because of relational databases, big data, and cloud computing, which represent major data-driven elements for the business environment.
Automated news article classification is a method of categorizing textual data into predefined classes. Addressing this problem finds applications in diverse domains, including information retrieval, topic modeling, sentiment analysis and content recommendation systems. In Albanian, though there is a rapid increase of digital content, there is limited availability of text corpora, presenting significant obstacles for advancement of natural language processing research and applications. The contribution of this paper is twofold. First, we introduce a dataset consisting of 9600 news article titles spanning across various categories. Second, we utilize this dataset to assess the effectiveness of several machine learning algorithms for topic classification. Experimental results demonstrate the efficacy of recurrent neural networks in comparison to simpler classifiers and ensemble methods.
Amidst an era of speedy technological growth, fraud is a complex challenge. This article presents an innovative analytical method that uses utility consumption data to identify probable instances of fraud in utility-based services. This study evaluates the efficacy of neural networks, specifically Artificial Neural Network (ANN) — a subset of neural networks, compared to Light Gradient Boosting Machine (LGBM), a tree-based model. It uses data from the Zindi challenge to analyze their capability to detect fraudulent consumption patterns. The article focuses on optimizing features using Pearson correlation and discusses the difficulties associated with imbalanced data sets. The LGBM model has exceptional performance, as seen by its impressive ROC AUC score of 0.878242. This number highlights its remarkable ability to differentiate fraudulent actions from ANN, thus establishing future advancements in fraud detection in utility-based services.
The profession of Medical Bioengineer is a complex one because it is an interdisciplinary job, on the border between medicine and engineering. The professional training of students within the Medical Bioengineering specialization, especially the clinical engineering branch, requires knowledge both from a technical and functional point of view of all medical equipment in health facilities. In this study, we propose the development of a virtual reality platform adapted to biomedical training in the field of medical devices. It immerses students in realistic medical scenarios based on medical devices, integrating feedback and physical simulations for a dynamic learning environment. The platform’s advanced grading system accurately evaluates student interactions, considering task accuracy, technical finesse, and efficiency. It includes a robust progress-tracking component that allows educators to monitor individual progress over time, providing valuable insights into student learning trajectories. This project bridges the gap between traditional medical training and the revolutionary potential of VR. It aims to provide an immersive learning experience where students can interact with medical instruments and devices in a simulated environment. The application’s system and progress tracking feature provide an objective assessment of comprehension abilities, enhancing the adaptability of medical education. The resulting VR platform greatly enhances the educational experience and aligns with the faculty’s vision of progressive educational methodologies, as well as the standards for learning in VR [7]. Because in some situations, during the years of study, students cannot have access to all types of medical equipment, this platform allows them direct interaction with a wide range of medical equipment.
This research work introduces a moving vehicle detection, classification and tracking framework based on a successful combination of machine and deep learning models. A vehicle detection solution combining a Gaussian Mixture Model (GMM)-based video foreground extraction to a level-set based frame segmentation is proposed first. A vehicle counting process is then performed on the obtained detections applying a Kalman filtering model. Next, the detected vehicles are recognized by their type applying a convolutional neural network (CNN) - based classifier and also by using the motion-based tracking results. Finally, the performed simulations are discussed here.
Defect detection is a crucial process in an industry that prioritizes quality. With the rapid rise in industrial automation, most companies are shifting their focus from manual to automatic detection methods. Manual inspections can be performed in a variety of ways, but these methods do have some drawbacks. They are time-consuming and require intensive labor. To solve this problem, we proposed deep learning algorithms to detect the defects automatically. Firstly, we used a 6-degree freedom manipulator to collect defect data, which were then identified and detected using various approaches. In this study, we compare our proposed model with another deep learning models such as YOLOv3, YOLOv4, and SSD. The proposed model outperforms these conventional deep-learning models using downsampling with dilation convolution to produce the high-semantic feature map. Combining the prediction from both upsampling and downsampling operations boosts the accuracy of the model. The accuracy of the SSD, YOLOv3, YOLOv4, and our proposed method are 76%, 63%, 62%, and 81%, respectively. The accuracy of our proposed model is 81% after evaluating these notable algorithms. The mean Average Precision (mAP) of the SSD, YOLOv3, YOLOv4, and our proposed method are 61%, 63%, 60%, and 65%, respectively. The mAP of our proposed model is 65% after evaluating various types of defects.
Network engineers are essential to the management of computer systems, as they configure services and devices on the network to guarantee effective data routing and improve computer networks which make easier to perform everyday tasks like sending and receiving emails as well as more sophisticated ones like cloud computing and online gaming. In addition, modern networks have to serve a rising number of IoT devices without compromising performance and handle more sophisticated cybersecurity threats. Network engineers must always be learning and adapting to new protocols and technologies if they are to successfully tackle these difficulties. Moreover, the introduction of Large Language Models (LLMs) that are available as open-source software has revolutionized technical innovation by enabling the automation of network setups and augmenting the capabilities of network administration. These developments represent an important step forward in the field of network engineering, with the goal of maximizing efficiency and guaranteeing strong network security and functioning.
In this contribution, we propose a method for providing User-Based Collaborative Filtering Recommendations using the emotions present in social media reviews. We use a deep learning model which identifies the review dominant emotion from the 6 primary emotions proposed by W.G. Parrott. For experiments, we use a dataset containing book reviews and movie reviews that we collected from Goodreads and IMDB websites using our customized web scrapers. Moreover, we represent the book and associated movie as an unique item in the dataset. We validated our recommender system using a set of system-unseen reviews that simulate a set of users seeking for recommendations. The top k recommendations received for each simulated user are then analyzed by our proposed performance measures.
Internet of Things (IoT) interference in our lives is based on the large amount of sensor data, from complex and various sources (smart cities, industrial sensors, wearable devices, Health Care IoT), collected over time but also gathered every moment. All this data came in an amazing variety of formats and values. What we can do with all this data ? There are, mainly, two directions 1) act when values thresholds were exceeded (heart rate, blood pressure, industrial critical values) or, 2) detect evolution patterns that can lead to dangerous situations (detect an increasing heart rate for a patient that can lead to critical situation in, say 5 minutes). Considering a large variety of sensors, up to past years, 2) was much more difficult to achieve but with greater impact (consider only the management of resources that can be directed to most critical situation). This paper proposes and discusses a system architecture using relational database (SQL) to store critical data, non-relational database (NoSQL) to store all data and IoT programmable gateways or a Critical Data Layer to send selected data to SQL database. Two strategies in detecting critical evolutions are discussed.
The measurement of aircraft noise is very important for the industry of air transportation, for residents and for the municipality.In the United States of America, the Federal Aviation Authority (FAA) has a managing tool for aircraft noise map called Aeronautical Environmental Design Tool (AEDT) (Boeker Fleming, 2008); in French Bruitparif is the tool which satisfies the European directive for noise map elaboration strategy. The noise map enables city authorities, residents, and the airport manager to control the noise produced by aeronautical activities in order to master, for instance, the building of infrastructures surrounding the airport zone - a dataset built with a relational data base management system obtained from secondary radar detection. These data, after having gone through a data cleaning process, will be brought and tested with functions such as Naive Bayes, decision tree and random forest. Based on confusion matrix as a metric, we assessed the provided result by the decision tree with metrics as “ 98% for precision, 97% for recall and 97% for f1-score”. This article emphasizes that on the one hand, the data used for machine learning have the same quality in terms of integrity and precision of the radar itself and on the other hand the performance resides on the confusion matrix.
Skin cancer is one of the most common types of cancer worldwide, and its incidence rate is increasing day by day. Melanoma, a type of skin cancer, represents a small proportion of all skin cancers, although it is responsible for half of all skin cancer-related deaths. Early staging is a factor that significantly affects survival in cancer treatment. However, traditional methods of examining tissue under a microscope and measuring the depth of invasion, the deepest point of the tumor, require substantial time and specialization. This can lead to delayed early diagnosis and some variability between specialists that affects the accuracy of the diagnosis. Automating this process could shorten the diagnostic time and improve diagnostic accuracy. Therefore, the first step to determine tumor stage in melanoma is to automate epidermis segmentation. This study conducted an analysis of model performance on 69 melanoma whole slide image (WSI) samples employing various architectures, including base UNet, UNet++, UNet3+, Efficient-UNet, Swin-UNet, and TransUNet. According to our findings, the Trans-UNet architecture achieved the highest success ratio among the 33 test WSI samples, with a Dice Coefficient Score of 92 %.
The current surge in AI trends has catalyzed a strong inclination among organizations to transition towards AI-driven solutions. However, a significant challenge arises from the prevalent monolithic nature of existing applications, which often impedes scalability and limits the potential for enhancement through agent-based interventions. This paper aims to investigate strategies for transitioning from monolithic applications to microservices-based architectures and explore the utilization of agents for control within microservices environments. Subsequently, drawing from existing literature and our own insights, we endeavor to formulate a comprehensive strategy for transforming original monolithic applications into intelligently controlled microservices-based systems. We conclude with an IoT use case in order to illustrate the application of this strategy and highlight the advantages that can be achieved.
Model rockets are the advanced mechatronic systems established through the design, manufacturing, and integration of the model-scale mechatronic sub-systems for a rocket mechanism capable of operating at low altitudes. The safe recovery and reusability of model rockets are prominent tasks to address. Various actuator systems are used in model rockets to solve the safe recovery problem by safely releasing and recovering the payloads that they carry at specific altitudes, allowing the rockets to be recovered. Traditionally, model rockets have employed recovery units powered by gunpowder. This research aims to design a novel hybrid actuator system integrating mechanical components with pressurized gas and servo motor components as an alternative to traditional gunpowder systems to enhance reliability and safety. The objective is to develop a prototype of a hybrid recovery actuator system using CO 2 tubes and servo motors. The principle behind the designed recovery system relies on the release of CO 2 gas, which facilitates the separation of the rocket motor body from the main body. This principle is based on the mechanism within the designed actuator system, which functions by releasing the potential energy stored in compressed springs. Servo motors triggered by signals from the rocket’s flight computer release the springs, allowing the CO 2 gas to escape. The released gas separates the rocket bodies, ensuring a safe descent. Within the scope of this research, the efficiency and effectiveness of the CO 2 based hybrid recovery actuator system are evaluated through the production of a prototype. The experimental results unveiled that the development of the CO 2 based hybrid recovery actuator system enables safe and successful recovery of model rockets. To achieve the timely release of the CO 2 gas in the recovery system, the required pressure is determined via experimental studies for the separation of rocket bodies, ensuring a safe descent.
