Distributed randomized in-network localised load-balancing (DRILL) is one of the most commonly used protocols and relies on the “power of two choices” algorithm. However, this is a sub-optimal load balancing solution due to under-sampling and high number of candidate ports. Along with per-packet granularity, this sub-optimality increases the occurrence of out-of-order packets. Therefore, in this paper, we proposes a novel power of two choices-based load balancing protocol to overcome these shortcomings. Designated as state aware multiple-choice routing (SAMCR) for load balancing in data centers, it builds a Kalman filtering algorithm; to approximate queue length states using the developed processes and measurement models. Furthermore, it uses the estimated queue length states in an adaptive threshold-based covariance matrix to track the most optimal decision in the system. Mathematical simulations as well as an NS-3 discrete event network simulator were used to evaluate the superior performance of SAMCR over other state-of the art approaches in terms of system load, flow completion time and throughput for most types of flows.
The presented work is dedicated to the development of an AI agent that will perform automated CV sorting and evaluation to assist current hiring processes with a supervised machine learning model. A Random Forest Regressor containing 300 trees was trained on structured features based on the unstructured CV information including years of experience, number of related skills, education level, overlap of skills with the job description, and coverage ratio. The 50 actual CVs dataset was divided to 80 percent training set and 20 percent testing set and the ground truth target used in supervised learning was HR inspired rule based scoring. The model realized an impressive predictive power, as seen with the R² of 0.94 and Mean Absolute Error of 4.46 which signifies a significant agreement with human based ratings. The trained model is integrated to an AI agent that automates the ranking of the candidates, short listing and scheduling of interviews. The system is particularly effective at unloading HR staff and increasing the efficiency of the recruitment process; nevertheless, some of the key issues are the bias in the training data, the tendency of the system to treat different candidates with unequal consideration, and the limited interpretability of the model. The study highlights the importance of introducing feedback loops, reducing bias, and establishing transparency to ensure that AI-based recruitment systems were ethically, reliably, and humanly deployed.
Prompt engineering is emerging as an essential tradition to use generative AI in such domains as software, learning, health, and creativity. However, the field is yet to have a clear framework on assessing prompt quality, reliability and reproducibility. The comparisons and best practices are complicated by the fact that current efforts are most likely to be based on trial-and-error or task-specific benchmark. We offer A Systematic Assessment Framework for Evaluating Prompt Engineering (SAFE-PE), which implies standard measures and principles, including multi-dimensional evaluation numbers. It puts alongside use of quantitative data (accuracy, diversity, robustness) with qualitative data (interpretability, fairness, ethics) in order to give a holistic picture of prompt performance. We show that the framework is effective through case studies of the large language models Meta AI (LLaMA) in summarization, question and answer, and code generation. SAFE-PE provides a systematic assessment procedure that progresses the timely engineering as a scientific field and enables practitioners to obtain a useful means of using generative AI in a responsible manner.
The cybersecurity risk assessment methodologies have evolved over the previous decade, serving ever increasingly sophisticated and diverse networks and systems. Resilient cybersecurity systems have allowed businesses and governments to better defend themselves against threats from malicious entities. With the rising complexity of user-centric and heterogeneous cellular networks, the need for automated cybersecurity risk assessment has become an emerging research challenge. This challenge extends to the emerging technologies including Narrowband Internet of Things (NB-IoT), which is the focus of this paper. As the intricacies of NB-IoT networks increase, the need for security risk assessment and trust building also becomes dire. In this paper, we have developed a comprehensive NB-IoT network security risk assessment framework. We have presented a risk assessment taxonomy that encompasses three dimensional challenges including service management, application & cloud, and edge architecture. This developed framework defines the level and severity of security assurance in terms of risk management for a single point of failure. Through this investigation, we highlight, quantify, and present some key vulnerabilities pertaining to NB-IoT network security and make recommendations to manage such vulnerabilities in order to develop a trusted network.
Migration within Pakistan's districts have different patterns which are influenced by various factors. However, a systematic analysis of these migration flows is missing, which limits understanding necessary for effective planning and policy formulation. This research aims to bridge this gap by employing complex network analysis to model and analyze district-to-district migration in Pakistan over an eight-year period. Using a one-mode weighted network approach, each district is represented as a node, with migration flows between districts represented as weighted links. This model will enable us to quantify and visualize migration connections, uncovering structural properties and regional disparities within the network. Both local and global network measures will be applied to understand structural properties while community detection will be done to examine its structural organization. Migration reasons are incorporated to capture the socio-economic factors of migration. The findings revealed key structural patterns: Karachi, Rawalpindi and Lahore emerged as major connectivity hubs with widespread links, while Karachi, Lahore and Peshawar stood out as volume hubs handling the largest migrant flows. Districts such as Awaran, Hattian Bala and Khizer were found to be the most accessible, whereas Larkana, Kashmore and Korangi acted as bridging districts linking different regions. At the global level, the migration network appeared sparse but structured, with a clustering coefficient of 0.6 indicating local group formation and a density of 0.3 suggesting that flows are concentrated along specific routes. Community detection revealed five distinct communities, two of which were dominated by Punjab districts, highlighting the province’s central role in shaping migration dynamics. Beyond these structural patterns, the analysis of reasons showed that family-related factors—particularly marriage and living with parents or spouse—accounted for the majority of migration, while job-related and return migration contributed comparatively less. Overall, internal migration in Pakistan follows structured hubs and regional clusters rather than random flows, providing valuable insights for policymakers to design targeted interventions, allocate resources efficiently, and promote balanced regional development.
Classification and regression datasets are utilized in tandem with deep learning approaches to ascertain the characteristics of individuals, such as their gender and age. Stagewise or phase by phase in terms of time, the activities conducted are given below: data used, training algorithm applied, use of accuracy of 89% and 92% for age and gender classification, respectively. Through the work, it can be derived that it has primary components that can be useful for preceding research, and, of course, it’s good for demonstrating that deep learning can be applied in facial recognition jobs. Besides that, a confirmation section will be included to enable those who wish to dig deeper into this topic to verify the information source.
This research paper details the design and implementation of an Aligned Domain Knowledge Graph, focusing on constructing a subject-based taxonomy for academic disciplines. The primary objective is to enhance information retrieval tasks, such as indexing, cataloguing, and searching digital documents within repositories, by providing a structured subject system. The study identifies shortcomings in current taxonomies, particularly the lack of a clear, unique identification for each subject. An ontology-driven methodology is employed, involving the characterization of subjects using traits like Philosophy, Logic, Method, Pure, Applied, Soft, and Hard, drawing inspiration from models such as Biglan Classification and Hegelian typology. A knowledge base is populated where each subject is assigned a persistent and opaque IRI with a unique GUID fragment for unambiguous identification. Utilizing a reasoner, the hierarchical taxonomy is automatically inferred based on defined ontological relationships, classifying subjects into disciplines and subcategories. Furthermore, a user-friendly website was developed using Agile methodology to manage and provide access to the taxonomy via search and visualization, addressing user needs for long-term sustainability
This research aims at developing an android-based mobile application to support the early childhood education teaching-learning process in preschools. Digital resources are essential in the changing landscape of early childhood education to support learning objectives and enhance literacy. This research investigates the how early children learning development within the educational system is affected by the ChildBook android application. Utilizing a mixed-methods approach, the study looks at the quantitative results and the qualitative information that preschool-aged children (3-5 years old) and teachers provided throughout the period of a school year. Quantitative tests compare individuals using the ChildBook app to those getting traditional literacy training by measuring phonics awareness, vocabulary acquisition, alphabet recognition, and reading comprehension skills before and after the intervention. Qualitative data which includes educator interviews and observations emphasizes opinions about the app effectiveness, level of engagement, and curricular integration. According to preliminary results, kids using the ChildBook app show significant improvements in vocabulary acquisition and phonetic understanding when compared to their control group peers. Additionally, qualitative data emphasizes educators' favorable feedback about the app ability to improve student engagement and support personalized learning experiences. Recommendations for effectively introducing educational technology into curricula for young children, encouraging collaborative learning environments, and supporting in the professional development of educators are among the effects for educational practice. This research contributes to the increasing body of research on digital literacy tools in early childhood education by providing insight on how they might improve traditional methods of instruction and enhance young students' development of learning skills.
Complex networks are widely applied in many research fields for modeling the complex relationships found within real-world observations. Such networks find various potential applications in biological, social, technological, and chemical sciences, among many others. The subject of complex networks is especially very important in modeling epidemic diseases to study their dynamics. The COVID-19 pandemic has left deep marks on humankind globally, including quite extensive effects in Pakistan. Since the outbreak of COVID-19 cases in Pakistan, efforts have been made to analyze patterns of disease diffusion. Following this, in this study, we examine the diffusion of COVID-19 cases in Pakistan's Balochistan province. We introduce a network model with location-location associations, which acts as an intermediary for disease diffusion. In this model, weeks and locations are nodes, and the heterogeneous weighted links are used to show the frequency of confirmed cases. The dataset of COVID-19 cases is modeled and analyzed, taking the confirmed cases from every location and forming a weighted two-mode network. We perform an analysis of this two-mode network based on global and local measures to understand its structure from the point of view of Balochistan. In addition, we project the two-mode network based on well-known projection techniques. The findings indicate that the transmission of COVID-19 in Balochistan has adversely affected some localities, which have a high number of confirmed cases.
Realistic and fascinating digital characters in video games, animated movies, and Virtual Reality (VR) / Augmented Reality (AR) experiences all depend on facial animation. Creating real facial emotions and speech synchronization historically needed time-consuming manual keyframing or costly motion capture. This research investigates the Carnegie Mellon University (CMU) Pronouncing Dictionary-based text-to-viseme system to automate facial animation. The system generates a rule-based algorithm built using the Python notebook to produce facial animation sequences. These sequences are applied to 3D models using a Blender addon. Using ARKit's 52-blendshape system, phonemes are mapped to visemes, and a proprietary dataset is created. This dataset is improved by manual adjustments. The framework is a potential character animation solution as it can automatically generate facial animations from text input. The proposed automated facial animation framework empowers animators, even those with limited expertise, to create quick and efficient animations with ease. It is designed to minimize the need for extensive refinement, streamline the animation process, and enhance the accessibility for users across varying skill levels.
Smart agriculture represents a burgeoning concept revolutionizing traditional farming by seamlessly integrating crucial technologies with sensory and internet-enabled devices. This transformation is realized through the harmonious amalgamation of diverse technological components, including Wireless Sensor Networks (WSN), Internet of Things (IoT), robotics, drones, and robust computing infrastructure. Our study presents a novel framework and comprehensive infrastructure for smart agriculture, covering from sensing (physical) layer to end-user (application) layer. It includes communication technologies, IoT, WSN, autonomous vehicles, computing, and data processing. We also analyze the disparities between traditional and smart agriculture across various parameters. We conclude the paper by suggesting some recommendations, which can assist in the revolutionary change in agriculture for worldwide adoption.
The SDN architecture supports the detection and mitigation of DDoS attacks as soon as possible, which is difficult in a conventional network. The SDN controller identifies DDoS attacks in their early phases and mitigates their impact on the entire network using proper identification patterns and detection schemes. This study presents a feature set for distinguishing DDoS attacks from regular traffic. It provides a system paradigm for detecting DDoS assaults using an SDN controller. The system model's detection module is built on an RBF neural network, which is then compared to other methodologies. The proposed system model is capable of identifying high network traffic from DDoS attacks and dealing with DDoS attacks in their early stages. To handle extremely large DDoS traffic flows in Software Defined Networking (SDN), this study presents a rapid and effective DDoS detection method. The purpose of this study is to introduce new approaches for addressing DDoS threats in SDNs and to assist SDN controllers in managing excessive malicious traffic. This research explains several scenarios and examples where these tactics can be applied, and explores various studies on DDoS attacks. Furthermore, three different algorithms have been compared: (i) Quasi-Newton Gradient Algorithm, (ii) Gradient Descent with Momentum Algorithm, and (iii) Variable Learning Rate Gradient Descent Algorithm. A Radial Basis Function (RBF) neural network has been trained using these algorithms. During this research work, the NSL-KDD dataset was used. The findings show that, for a shorter training period, the Gradient Descent with Momentum Algorithm produces results with higher accuracy.
To ensure that plant diseases are well controlled and managed that would reduce crop losses and ultimately improve on food security then diseases need to be identified correctly at the right time. This paper uses a CNN approach for identifying plant diseases using the Plant Village, web-based dataset that has on average 35 classes, with 29281 images, comprising of both healthy and diseased leaves. To improve the model performance further techniques like data augmentation, contrast enhancement, noise reduction techniques were used. The training results of the proposed CNN had a training loss of 0.0808 and a validation loss of 0.3330.The training as well as the validation accuracy achieved were 97.41% as well as 90.34% respectively. Other measures of evaluation of the presented model are precision of 0.9139; recall of 0.9034; the F1 score was 0.9019. Thus, in order to raise the accuracy in classification, other features like color, veins, roughness of the leaf surface etc., were also considered. It was also indicated how much effective the proposed model was for edge computing solutions as compared to other models including DenseNet121, ResNet50, Alex Net and VGG16. This work shows that contemporary agriculture could reliably and dependably utilize deep learning in the detection of plant diseases as a dependable and scalable tool.
People with visual impairments encounter many obstacles and challenges while navigating daily surroundings which typically affect their mobility, independence, and overall quality of life. In response to these challenges, this paper presents the development of the "Smart Hat," a wearable gadget that helps visually impaired people by offering real-time object detection and voice assistance. The module utilized in Smart Hat is ESP32-CAM used to record live video transmitted wirelessly to a mobile application. The TensorFlow framework is utilized by the device to provide precise and effective object recognition which is then transmitted through voice assistance via headphones. This solution provides an intuitive user interface along with auditory cues to help the blind person navigate their surroundings safely. With an accuracy rate of more than 91%, Smart Hat offers a lightweight, portable, and affordable substitute for traditional assistive technologies like guide dogs or costly electronic aids. The goal of Smart Hat technology is to empower individuals with visual impairment by giving them more independence and self-assurance in their surroundings.
This paper examines the switch from model-based to data-driven control systems, providing an overview of the challenges encountered and the potential solutions. It reviews the history, current progress, and future perspectives of data-driven control technologies. Additionally, it outlines the differences between model-based and data-driven control, compares various data-driven control approaches, and addresses essential issues in data-driven optimization. Finally, the paper delves into the data-driven control process and related research areas.
Approximately 900 million individuals globally are said to have diseases associated with the skin, thus becoming one of the prevalent diseases in the world. Common conditions in southern Punjab are acne, psoriasis, and fungal infections. Eczema has associated red tones and is usually marked by dryness and itchiness. Psoriasis creates reddish patches due to the scale formation and the occurrence of thick patches. There are common forms of fungi infections known as ringworm and jock itch and mostly fungi are favored when in a warm and damp environment. Only through quick detection and treatment can one prevent critical cases of skin problems. Skin diseases prove difficult to diagnose because every skin is different in either the types or textures they hold. Researchers have recommended various early detection methods for them. Their solutions through a range of machine learning algorithms like random forest, naive Bayes, logistic regression, kernel SVM, KNN, and CNN based on which they used detection methods for various skin diseases. Based on the detailed research, the real-time symptoms from an IoT and machine learning model can diagnose the skin diseases. Cameras will take pictures of the skin images. IoT will measure body temperature, and a deep learning algorithm and CNN algorithm predict the diseases. The output is provided on the GUI as an Android application, in order to help them understand whether they have skin diseases and allergies early.
Digital forensics requires potential and insightful analysis of the Internet of Things (IoT) environment as compared to the traditional methodology of digital forensics. A variety of technology and technical devices including sensors, cloud computing, and RFIDs used in IoT environments. The utilization of such heterogeneous technology results in a huge volume of data raising security challenges in the IoT environment. Cloud resources face challenges such as identification and preservation related to digital evidence. In that manner, cloud forensics gets new technical and legal challenges in the provision of digital computing threats. Cloud forensics has faced significant complications in evidence collection, investigation, and extraction. On the other hand, IoT digital forensics faces both technical challenges and non-technical challenges based on the IoT environment and devices. In this paper, a succinct identification of issues and key challenges in IoT digital forensics design will be discovered. Also, this research paper focuses on the broad spectrum of challenges that are blockage in the provision of IoT forensics. Secondly, this study discusses the importance of digital forensics and -specific research domains that would help to improve digital forensic processes and make them more effective for IoT scenarios.
Automatic speech recognition is a process of using computers to convert voice signals produced by human speech into reasonable format i.e. text or command that conveys the same meaning as the speaker intended to do. Many researchers are working on various languages including English and other European languages like Spanish, German, and French etc. to develop an automated system for speech recognition (ASR). However, researchers on the development of ASR for the Urdu language have put very little effort. We have developed an Urdu speech recognition system using Deep Neural Network (DNN) on our developed corpus that contains some of the most frequently used words in Urdu like digits, season names, and month names. The accuracy rates of our ASR are very encouraging because 72% accuracy is achieved for 26 words and 92% accuracy is achieved separately for names of seasons.
Handwritten Character recognition falls under the domain of image classification that has been under research for years. The idea is to make the machine recognize handwritten human characters. The language focused in this research paper is English while using offline handwritten character recognition for identifying English characters. There are many publically available datasets from which EMNIST is the most challenging one. The main idea of this research paper is to propose a deep learning CNN method to help recognize English characters. This research paper proposes a deep learning convolutional neural network that is tested and compared with renowned pre-trained models using transfer learning. These parametric settings address multiple issues and are finalized after experimentation. The same hyper-parametric settings were used for all the models under test and E-Character with the same data augmentation settings. The proposed model named the E-Character recognizer was able to produce 87.31% accuracy. It was better than most of the tested pre-trained models and other proposed methods by other researchers. This research paper further highlighted some of the problems like misclassification due to the similar structure of characters.
Action recognition in videos is one of the essential, challenging and active area of research in the field of computer vision that adopted in various applications including automated surveillance systems, security systems and human computer interaction. In this paper, we present an in-depth comparative analysis of five CNN-RNN models based on pre-trained networks such as InceptionV3, VGG16, MobileNetV2, ResNet152V2 and InceptionResNetV2 with recurrent LSTM units for action recognition on Anomaly-5 dataset. The performance of these models is analyzed and compared in terms of accuracy, precision, recall & F1-scores and computational efficiency. The CNN-RNN architectures we considered for analysis in this paper, the ResNet152V2 based CNN-RNN model exhibits better performance and achieved highest accuracy, precision, recall and F1-score equal to 92.20% due to its ability to capture more complex spatial features. This comparative analysis may guide the researchers in selecting appropriate models for real-world applications for action recognition. In addition of this, a new dataset is developed called Anomaly-5 that can helps as a valuable resource for training and evaluating action recognition algorithms.
