Transactions on Machine Intelligence2821-16933420201201A Novel Approach to Reducing Energy Consumption, Economic Savings, Service Quality Enhancement, and Resource Utilization in Cloud Data Centers19921020907810.47176/TMI.2020.199ENM. H.MahmoudianStudent of Digital Electronics Department, Faculty of Electrical Engineering, Amirkabir University of Technology, Tehran, IranH.TaheriAssociate Professor, Faculty of Electrical Engineering, Amirkabir University of Technology, Tehran, IranA.Beik-MohammadiStudent of Digital Electronics Department, Faculty of Electrical Engineering, Amirkabir University of Technology, Tehran, IranJournal Article20200706Cloud data centers often provide the infrastructure for millions of virtual machines in dynamic environments. Virtual machine deployment is a process where it is determined which virtual machines should be executed on which physical machines within the virtualized infrastructure. Given the randomness of customer requests, the virtual machine deployment issue must be formulated as a dynamic optimization problem. On the other hand, providers must be able to respond to virtual resource requests in complex dynamic cloud computing environments, considering service elasticity and overbooking physical resources. In this work, five experiments were designed and conducted for the two-stage optimization of such issues. In these experiments, after evaluating online heuristic algorithms, various overbooking protection coefficients, and different scaling methods, a non-deterministic formulation was considered for optimizing four objective functions (energy consumption, economic savings, service quality, and resource utilization). The experimental results, considering 96 different scenarios, show that two of the online phase heuristic algorithms, taking into account a memetic algorithm for the offline phase, setting the overbooking protection coefficient to 0.75, and scaling the four objective functions based on the shortest Euclidean distance to the origin, yield the best performance.https://www.tmachineintelligence.ir/article_209078_40044acbbcf49ef2c6321d63c35fdc9b.pdfTransactions on Machine Intelligence2821-16933420201201Fully Automated Brain Tumor Segmentation in MRI Images Using a Modified Level Set Method21121720908110.47176/TMI.2020.211ENM.GheymatgarDepartment of Electrical and Electronics Engineering, Faculty of Electrical and Computer Engineering, Semnan University, Semnan, IranJournal Article20200702Gliomas are among the most common types of brain tumors found in adults, originating from glial cells and infiltrating surrounding brain tissues. Accurate identification and segmentation of these tumors are crucial for diagnosis, treatment planning, and patient monitoring. Despite significant advancements in medical imaging and computational analysis, glioma detection remains challenging due to the high variability in tumor shape, size, and location across different patients. Conventional segmentation methods, particularly level set approaches, often require manual intervention, limiting their efficiency and reproducibility in clinical settings. In this study, we propose a fully automated glioma segmentation method based on a modified level set framework. Unlike traditional semi-automatic level set techniques, our approach eliminates the need for manual initialization, thereby improving consistency and reducing operator dependency. The proposed method enhances boundary detection and region refinement, leading to more accurate segmentation results. To evaluate the effectiveness of our approach, we conducted extensive experiments using the standard BraTS 2017 dataset. Performance was assessed through both quantitative and qualitative evaluation metrics, including the Dice similarity coefficient. Our method achieved an average Dice coefficient of <strong>79%</strong> for the entire tumor, demonstrating its reliability and effectiveness compared to conventional techniques. The fully automated nature of this approach offers promising potential for integration into clinical workflows, aiding radiologists and medical professionals in the early detection and precise delineation of gliomas.https://www.tmachineintelligence.ir/article_209081_3ace6fc07847de02e3ae600620f0d236.pdfTransactions on Machine Intelligence2821-16933420201201Design and Simulation of a Single-Phase Grid-Connected Microinverter for Photovoltaic Systems21822520910410.47176/TMI.2020.218ENH.HassanpourElectrical Engineering, K.N. Toosi University of Technology, Tehran, IranM.EhsanianAssociate Professor, Department of Electronics, K.N. Toosi University of Technology, Tehran, Iran0000-0001-7859-7132Journal Article20200803In single-phase grid-connected photovoltaic (PV) systems, maintaining a constant input power while dealing with a pulsating output power is a crucial challenge. This isolation is typically achieved using an energy storage element, most commonly a high-voltage DC link capacitor. The AC array structure of such systems consists of a single solar array connected to the grid through an inverter. Since the inverter's operational lifespan should align with that of the solar array, the choice of capacitor plays a critical role in system reliability. Electrolytic capacitors, despite their cost-effectiveness, exhibit a significantly shorter lifespan than solar arrays, necessitating the use of high-cost film capacitors with enhanced durability. A key challenge in designing the DC link voltage controller is mitigating voltage fluctuations in the capacitor and suppressing second harmonic ripple, both of which are inversely related to capacitor capacity and, consequently, system cost. Traditional approaches often require large capacitors to reduce these fluctuations, leading to increased cost and bulkiness. This paper proposes a digital control strategy for stabilizing the DC link capacitor voltage in single-phase grid-connected PV systems. The proposed method employs a low-pass finite impulse response (FIR) filter, which effectively suppresses voltage oscillations while maintaining system efficiency. The digital implementation of this controller enhances flexibility, reliability, and cost-effectiveness compared to conventional analog controllers. Simulation and experimental results demonstrate the effectiveness of the proposed approach in reducing voltage ripple and improving the stability of the DC link, thereby extending the lifespan of the inverter and enhancing overall system performance.https://www.tmachineintelligence.ir/article_209104_bfd4d7dc0b06de9783b4035221b0ecac.pdfTransactions on Machine Intelligence2821-16933420201201Energy Demand Prediction in Smart Grids Using Neural Networks Based on Optimization22623920910610.47176/TMI.2020.226ENA.Mohammadi-PourPower Group, School of Electrical Engineering, Shahid Beheshti University, Tehran, IranM.Setayesh-NazarAssociate Professor, School of Electrical Engineering, Shahid Beheshti University, Tehran, Iran0000-0003-4127-8458Journal Article20200614Demand prediction plays a crucial role in the real-time operation of electrical systems, particularly for monitoring, planning, and optimizing the operation of electrical devices. Accurate demand prediction ensures effective coordination between consumers and power companies, which is essential for efficient power grid management. This paper presents a novel approach for energy demand prediction using a neural network combined with an optimization-based method. Initially, a conventional neural network is employed to predict the required energy demand based on historical data. However, to improve prediction accuracy, a genetic algorithm (GA) is introduced to adjust the neural network’s weights automatically. This optimization method fine-tunes the network, enabling it to achieve better performance in predicting short-term energy demand. The integration of the genetic algorithm helps in overcoming the limitations of traditional training methods, such as slow convergence or local minima. Experimental results, based on real-time data randomly selected from various sources, demonstrate that the genetic algorithm-based neural network outperforms conventional approaches in terms of prediction accuracy and computational efficiency. The proposed method is validated through extensive testing, showing its potential for accurate short-term load forecasting in dynamic and complex energy systems. This research highlights the effectiveness of optimization algorithms in enhancing the predictive power of neural networks for energy demand forecasting.https://www.tmachineintelligence.ir/article_209106_4faeb2dfca1f707a2c9b432d0b9a320c.pdfTransactions on Machine Intelligence2821-16933420201201Load Balancing of Servers in Cloud Computing Using the Lion Optimization Algorithm24025020910710.47176/TMI.2020.240ENA.Esmaeili MoshiranDepartment of Computer Engineering, Faculty of Engineering, Tabriz Branch, Islamic Azad University, Tabriz, IranSh.BabaeiAssistant Professor, Department of Computer Engineering, Faculty of Engineering, Tabriz Branch, Islamic Azad University, Tabriz, IranJournal Article20200629The growing demand for rapid processing of large and heavy computational tasks, coupled with the advancements in networks and distributed systems, has led to the development of cloud computing systems. In cloud computing, all user needs are provided in a shared environment as services on-demand and as required. One of the significant challenges in cloud-based systems infrastructure is solving the problem of optimal and efficient load distribution. In other words, assigning more virtual machines to a single physical server leads to issues such as non-optimal resource allocation and load imbalance. Hence, the load among physical servers and, consequently, the entire system load must be balanced. It has been proven that the load balancing problem in cloud computing falls under the NP-hard category, and thus, various researchers have used heuristic, metaheuristic, greedy, and other algorithms to solve it. In this paper, the load balancing problem in cloud computing is considered as an optimization problem, and the Lion Optimization Algorithm (LOA) is used to solve it. The proposed method is simulated in MATLAB software and compared with Genetic Algorithm (GA) and Simulated Annealing (SA). The experimental results indicate that the proposed approach significantly improves load balancing in the allocation of virtual machines compared to GA and SA algorithms.https://www.tmachineintelligence.ir/article_209107_8e2bcb2fc10b6abf05ac3c5e1a6dc129.pdfTransactions on Machine Intelligence2821-16933420201201A Method to Hide Information in Image Based on Selected Pixels25125920910810.47176/TMI.2020.251ENS.KhosraviDepartment of Computer, Payame Noor University, PO BOX 19395-3697, Tehran, IranJournal Article20200724There are various methods for concealing information or transmitting it covertly, with steganography being a prominent approach for secret communication. Unlike watermarking and fingerprinting, steganography aims to hide information in a way that its presence remains undetectable. Steganography is commonly applied to electronic media, particularly audio and image files. This paper provides an overview of image steganography, its applications, and different techniques. The objective is to outline the key features of an effective steganography algorithm. The ultimate goal is to develop a cross-platform solution capable of concealing a message within a digital image file. An image comprises numerous pixels, each with three color values, and is composed of millions of such numbers. Modifying a few color values in specific pixels typically results in an image that closely resembles the original. This paper introduces a technique that involves altering selected pixel color values based on the intensity criteria of colors in the image. The method strives to minimize image quality degradation, maintain information security, and, whenever possible, avoid changes in the image size.https://www.tmachineintelligence.ir/article_209108_36fe531236a533327a830c79b47b62cd.pdf