ABSTRACT. Electromagnetic (EM) fields/waves play critical and significant roles in our lives. Systems such as communication, remote sensing, integrated command/ control/surveillance, intelligent & fast transportation use EM waves. Also, EM waves are used in medicine, environment, education, marketing, and defense. Among many others, EM engineering deals with broad range of problems from antenna design to EM scattering, indoor–outdoor radiowave propagation to wireless communication (4G, 5G, 6G, …), radar systems, subsurface imaging, novel materials, EM compatibility, electroacoustic and/or electro-optical systems, etc. The frequencies vary from DC to Terahertz, even above, systems’ sizes vary from kilometers-large to nanometers-small. Most of these systems are broadband and digital, and they have to function close to many others therefore they create severe EM interference (EMI) and EM Compatibility (EMC) problems. As many others, EMC problems also must be taken into account by engineers from the earliest possible design stages. Establishing an intelligent balance among theory, practice, and modeling and numerical simulation has become essential.

ABSTRACT. We analyze the impact of the design space size and sampling approach of the datasets used for training of deep neural networks for antenna modeling. Three different datasets for modeling the gain of a dipole antenna are generated. Each dataset comprises 10,000 different antennas, obtained by full-wave simulations. The datasets differ in ranges of design variables and use different approaches (random values or values from the grid) for sampling the design space. Feed-forward neural networks based on multi-layer perceptrons of various topologies are trained using all three datasets, and the obtained results are compared in terms of accuracy and training time.

ABSTRACT. The paper presents design and analysis approach for an electro-magnetic biosensor for drug release modelling. The sensor resonant frequencies are analyzed using simple empirical formulae and compared to the results obtained using a full wave numerical solver. The sensor-antenna set up was then fabricated and tested and results compared with simulation.

ABSTRACT. Herein, the fabrication and measurement of a 1:3 scale wideband high-frequency (HF) antenna is discussed. The antenna is a combined TEM horn and loop antenna to provide directional radiation through combination of fundamental electric and magnetic dipole modes. Resistive terminations are applied to reduce the low-frequency turn-on at the expense of efficiency. Full size with image fits within a 3-m sphere, giving ka of 0.5 and 1 at 7.95 and 15.9 MHz, respectively. The impedance match is <2:1 over the entire extended HF band (2-30 MHz). The fabrication is a third of that for 1-m tall. Additionally, the metal sheets used in the design are modified to be a wireframe structure to reduce cost, weight, and impact of wind loading with minimal RF impact. Measurements include impedance with and without resistive terminations as well as qualitative characterization of the radiation characteristics by using the low-power weak signal propagation reporter (WSPR) protocol in the HF band.

ABSTRACT. A method based on a deep neural network for predicting the phases of excitations of the antenna array for the desired radiation pattern is presented. An example of a linear array consisting of four microstrip patch antennas is considered. A dataset comprising 10,000 array samples is generated using a 3D full-wave electromagnetic solver to take into account the coupling effects. Results demonstrate that, once it is trained, the deep neural network can successfully calculate the phases of excitations of the array elements based on the given radiation pattern.

ABSTRACT. Abstract 1

ABSTRACT. The flourishing of smart cities has introduced complex and mutually connected cyber-physical systems depending on IoT devices, making them extremely vulnerable to sophisticated cyber attacks. Efficient intrusion detection systems are essential to safeguard these environments, since traditional security solutions fall short in these circumstances. This study explores the integration of XGBoost, a powerful ensemble learning algorithm, with an adapted implementation of a well-known metaheuristic algorithm to improve the detection of network intrusions. Metaheuristics algorithm is employed to fine-tune XGBoost’s hyperparameters, improving classification acuracy and reducing computational overhead. Experimental assessment on benchmark intrusion detection dataset demonstrates that the metaheuristic-optimized XGBoost models significantly outperform baseline models in terms of accuracy, precision, recall, and F1-score. The introduced methodology presents a robust and scalable IDS framework appropriate for the complex and resource-constrained infrastructure of smart cities.

ABSTRACT. Environmental protection plays an increasingly important role in modern post-industrial society. Environmental factors affect everyday life across the world, from health to economic development. Nevertheless, certain byproducts of industry are unavoidable, especially in developing economies. It is therefore fundamental to meticulously monitor and control pollutant levels to ensure the future of our planet. Accurate forecasting of pollution can help mitigate and plan industrial production in a way that reduces environmental impact in a timely and effective manner. This work explores the potential of artificial intelligence (AI) algorithms, specifically, long short-term memory (LSTM) neural networks augmented with attention mechanisms—for pollutant forecasting using historical time series data. To ensure optimal performance, a modified metaheuristic algorithm based on the Bat Algorithm (BA) is introduced and used to tune the models’ hyperparameters. Simulations conducted on real-world data have demonstrated promising results, with R2 scores as high as 0.926548 achieved by models optimized using the proposed method.

ABSTRACT. Click fraud poses a growing threat to digital advertising ecosystems, impacting advertisers, analysts, and platforms dependent on ad-generated revenue. Traditional countermeasures, such as reCAPTCHA, can mitigate basic automated attacks but may also hinder legitimate user engagement. To address this challenge, this work proposes a novel approach leveraging reservoir computing, specifically Echo State Networks(ESNs), to classify user click sequences and detect fraudulent behavior. Recognizing the critical role of hyperparameter tuning in time series classification, this work introduces a modified version of the chimp optimization algorithm (ChOA) optimizer to enhance model performance. Experimental results demonstrate that our optimization framework significantly improves detection accuracy, achieving up to .751853 and outperforming standard baseline models. This study highlights the potential of ESNs and tailored optimization strategies for robust, scalable click fraud detection.

ABSTRACT. The continuously increasing energy demands of the world and the finite nature of fossil fuels are just some of the challenges contributing to the modern energy crisis. Many look toward renewable sources of energy as a solution. However, certain challenges are intrinsic to sources such as solar energy. One notable issue is the sporadic nature of production. Nevertheless, accurately predicting production could help mitigate this issue by providing better insights and aiding in planning around energy demand, thereby helping to balance consumption with grid supplementation. This work explores the application of time series forecasting methods to predict power production using real-world solar farm data. However, the use of forecasting models based on recurrent neural networks (RNNs) is not without challenges. Selecting appropriate hyperparameters is vital for achieving desirable outcomes. To address this, the study introduces a metaheuristic optimization approach based on the Reptile Search Algorithm (RSA) to improve the hyperparameter selection process.

ABSTRACT. This study explores the utilization of Artificial Intelligence (AI) in the processes of brainstorming and market segmentation to optimize business performance. Through a survey involving 200 respondents, the results show that the majority have a positive view of AI usage in supporting innovation and business development. A total of 59.5% of respondents agreed that AI can accelerate and enhance the efficiency of brainstorming processes, while 63.5% felt that AI provides new perspectives in finding solutions. In addition, 64.5% of respondents believe that the consistent use of AI can improve their business competitiveness. The study also found that AI plays a crucial role in making more accurate business decisions, with 56.49% of respondents expressing confidence in AI’s benefits in this context. These findings indicate that the acceptance of AI in the business world is increasing and that it can serve as an effective tool for driving creativity and innovation.

ABSTRACT. Researching the competencies of IT leaders in the field of artificial intelligence (AI) and the Internet of Things (IoT) is key and necessitated by global changes in the labour market. It will allow to highlight of emerging skills and guide the development of the workforce, thereby supporting the growth of innovation. This article presents a labour market study of job openings in the AI and IoT field and the competencies required of candidates for the advertised jobs. The research includes tracking advertised job positions and analysing what specialists are in demand, as well as the skills and competencies they need to possess. The research questions are addressed, like a what kind of job position, what type of employment and education are required for the sought-after position, and what experience should people applying for such a position have. Also, what are the key skills, competencies (organisational, leadership, personal), technical knowledge, soft skills, and main responsibilities required for an IT business leader is needed. The conducted study and analysis provide an overview of the types of job postings available for IT specialists in the fields of AI and IoT. The data from it will give a clear picture of what specialists, in what field and at what working hours are sought by business organisations. In addition, it was analysed what education, competencies and skills for working with AI and IoT the job candidates should have. The analysis prepared in this way will be useful for students and novice specialists, working specialists who want a career transition or growth, HR teams and companies and start-ups and entrepreneurs. This analysis will also be useful for educational organisations, where it will guide them in what areas to organise future practical training in the field of AI and IoT.

ABSTRACT. This article explores and analyzes the impact of artificial intelligence (AI) on the educational process. The purpose of the research is to identify the positive and negative aspects of AI use in education and assess their implications. A SWOT analysis was conducted based on the results of three surveys carried out as part of an international project. These studies reflect the perspectives of lecturers, IT companies, and students regarding AI and IoT training aimed at developing future IT leaders. The differences in perception among the groups emphasize the need for enhanced cooperation to adequately align education with the dynamic demands of the labor market. A model for integrating AI into the university environment is also proposed.