Saudi Cultural Missions Theses & Dissertations
Permanent URI for this communityhttps://drepo.sdl.edu.sa/handle/20.500.14154/10
Browse
25 results
Search Results
Item Restricted Partial Learning for MIMO Detection(Saudi Digital Library, 2024) Babulghum, Abdulaziz; ElHajjar, Mohammed; Ng, Soon Xin; Xu, ChaoReliable and efficient multiple-input multiple-output (MIMO) detection remains a central challenge in modern wireless receivers. Optimal maximum-likelihood (Max-L) detection delivers the best performance. However, its exponential complexity is prohibitive, while linear schemes such as zero-forcing (ZF) and minimum mean square error (MMSE) are computationally attractive yet suffer from poor performance. Fully learned detectors improve robustness but introduce substantial parameter counts and computational complexity. Building on prior work on partial learning (PL), this thesis contributes a unified detection framework based on PL that addresses these trade-offs by applying learning only where it yields the most benefits: a subset of the weakest symbol streams, with the remaining streams detected using low-complexity linear detection. The first part of the thesis designs a soft-output PL demapper implemented with a small fully connected neural network (FCNN) for quasi-static channels and embeds it into an iterative detection scheme. The inner MIMO detector produces log-likelihood ratios (LLRs) that are exchanged with an outer convolutional decoder. EXIT charts and decoding trajectories are used to analyze convergence. Across representative 2×2 and 4×4 quadrature phase-shift keying (QPSK) systems, the iterative PL (Iter-PL) technique closes most of the gap to iterative Max-L and full-learning detectors while operating at a fraction of their complexity. Operation counts are reported and related to the number of learning-assisted streams *d*, demonstrating an explicit performance versus complexity trade-off. The second part extends Iter-PL to time-varying channels while also considering channel state information (CSI) error. The same FCNN-based soft demapper is trained using CSI errors. Results show that Iter-PL retains its iterative gains under 5% CSI error and remains markedly superior to purely linear detection. An adaptive PL strategy is further introduced to select *d* based on the average received signal-to-noise ratio (SNR), thereby achieving a near-constant target bit error rate (BER) with reduced average complexity. The final part addresses scalability in dynamic multi-user uplinks. A graph neural network (GNN)–based PL detector is proposed, where an approximate message passing (AMP) frontend supplies soft symbols and variance estimates to the GNN. The GNN then detects only the d weakest users, while ZF detects the remaining users. By operating on user graphs, the model generalizes across changing activity masks without requiring retraining and maintains a low parameter count. Simulations over multiple activity patterns consistently confirm low BER and favorable performance–complexity trade-offs. Overall, the thesis demonstrates that partial learning enables near-optimal soft detection with clear and quantifiable reductions in complexity, and that GNN-based partial learning offers the same benefits in multi-user scenarios. The proposed technique provides a practical approach to scalable, low-latency MIMO detection, making it suitable for evolving wireless systems.5 0Item Restricted INVESTIGATING NOVEL ANALYSIS APPROACHES FOR STRUCTURAL CONDITION ASSESSMENT USING ULTRASOUND AND INFRARED DATA(Saudi Digital Library, 2025) Alqurashi, Inad; Catbas, NecatiAging civil infrastructure, particularly reinforced concrete bridges, is experiencing progressive deterioration that threatens safety, serviceability, and long-term performance. Traditional inspection methods such as visual examination and hammer sounding are limited in their ability to detect subsurface defects and are prone to subjectivity. This dissertation develops and validates an integrated, multi-modal structural condition assessment framework that combines rapid Infrared Thermography (IRT), high-resolution Ultrasound Tomography (UT), Artificial Intelligence (AI)-driven anomaly detection, and immersive Digital Twin (DT) visualization to overcome these limitations. The research advances three main areas: (1) a dual-mode IR–UT workflow exploiting the complementary strengths of each modality, enabling rapid surface screening with IRT and in-depth defect characterization with UT; (2) optimized deep learning (DL) models tailored to each modality, with a transformer-based Grounding DINO model applied to raw Infrared (IR) imagery for automated detection of thermal anomalies, and a lightweight You Only Look Once (YOLO)-v8n model applied to UT volumetric slices for detecting internal delaminations, voids, ducts, and rebar, both trained on large, segmentation-assisted, color-standardized datasets to ensure robust performance under diverse field conditions; and (3) integration of Unmanned Aerial Vehicle (UAV)-based Light Detection and Ranging (LiDAR), photogrammetry, and multi-modal non-destructive testing (NDT) data into a geo-referenced Virtual Reality (VR) environment to support real-time, collaborative decision-making. Laboratory testing on engineered specimens with embedded defects and field deployment on multiple in-service bridges, including the NASA Causeway Bridge, achieved high detection accuracy (mAP@0.5 up to 0.93 for UT using YOLOv8n and 0.80 for IRT using Grounding DINO), strong localization (Average IoU ≈ 0.80–0.90), and significant efficiency gains through targeted UT scanning. The VR-based DT enabled inspectors to seamlessly review thermal anomalies, volumetric UT slices, and 3D geometry in a single immersive scene, reducing defect confirmation time from several minutes to approximately one minute per location. By fusing complementary NDT modalities with AI models purpose-built for each data type and immersive visualization, this research delivers a scalable, repeatable, and field-validated methodology for rapid, objective, and data-rich condition assessment of reinforced concrete structures, with potential for broader application to other infrastructure types to enable proactive maintenance strategies and improved lifecycle management.13 0Item Restricted Automated Synthetic Lung Tumor Generation for Training a U-Net Model on Lung CT Slices(Saudi Digital Library, 2025) AlJoher, Sarah; Blumensath, ThomasThis thesis presents an automated pipeline for generating synthetic lung tumor CT images and corresponding segmentation masks to improve deep learning–based tumor segmentation in low-data settings. Real tumor regions are extracted from annotated CT scans and inserted into healthy lung slices using a 2D Tukey window and Poisson image blending to preserve realistic texture and boundaries. Ground truth masks are generated automatically using the Segment Anything Model and refined through morphological operations. The synthetic and real images are used to train a 2D U-Net segmentation model, which is evaluated across multiple experimental trials on an external dataset composed entirely of real pathological CT scans. Results show that models trained with carefully curated synthetic data match or outperform models trained on real data alone, demonstrating improved generalization and robustness. This work highlights the potential of automated synthetic data generation to reduce reliance on large, manually annotated medical imaging datasets.20 0Item Restricted Computational Approaches for Drug Repositioning and Target Discovery in Alzheimer’s Disease(King Abdullah University of Science and Technology (KAUST), 2024) Alamro, Hind; Gao, XinAlzheimer’s Disease (AD) presents significant challenges to global healthcare systems due to its complex and progressive nature. Despite extensive research, the underlying mechanisms of AD lack clarity, and current treatments only alleviate symptoms without halting disease progression. Consequently, there is an urgent need for computational approaches that can accelerate research efforts and aid in the development of more effective treatments for AD. In this thesis, we address these critical challenges by developing computational and AI-based methods to improve the early detection of AD, identify novel biomarkers, and explore new therapeutic strategies through drug repositioning. To begin with, we focus on identifying key biomarkers associated with AD using gene expression datasets and then expand it to the identification of biomarkers through exploring the association between AD and its comorbidity, resulting in the discovery of new hub genes and miRNAs. Next, we examine the potential for drug repositioning by mining biomedical literature to uncover associations between drugs, targets, and diseases. This task was fulfilled by developing a systematic pipeline to extract valuable information from a curated collection of AD-related literature. The resulting data is subsequently used to construct a disease-specific knowledge graph, which is employed for drug repositioning using advanced graph-based techniques. Overall, this thesis contributes to AD research by employing computational methods, multi-data integration, and literature mining to provide new insights and therapeutic strategies. This work identifies key participants in AD progression and presents a pathway to accelerate the discovery of treatments through computational approaches.10 0Item Restricted Malignant Transformation of Oral Epithelial Dysplasia: Precision Diagnostics Utilizing a Deep Learning and Spatial Transcriptomics Predictive Modeling Approach.(University of Maryland Baltimore, 2025) Alajaji, Shahd Abdullah; Sultan, AhmedOral squamous cell carcinoma (OSCC) remains a major global health burden with limited improvements in overall survival over recent decades. Most OSCCs arise from oral potentially malignant disorders (OPMDs), including oral epithelial dysplasia (OED), which is currently graded subjectively by histopathological examination. The urgent need for objective, biologically informed risk stratification tools has driven the integration of artificial intelligence (AI), spatial transcriptomics, and functional genomics in oral cancer research. This thesis tests the central hypothesis that deep learning and spatial transcriptomic approaches can objectively predict the malignant transformation of OED by identifying histomorphological patterns and immune-epithelial gene signatures associated with malignant transformation zones and cancer progression. To evaluate this, three specific aims were pursued: 1. Develop and compare AI models for predicting malignant transformation of OED based on lymphocyte distribution and tissue morphology. 2. Identify spatially informed predictive biomarkers in proliferative leukoplakia (PL) using spatial transcriptomic profiling. 3. Functionally assess the role of mEAK-7, a novel regulator of non-canonical mTOR signaling, in OSCC initiation using a gene knockout mouse model. In Aim 1, we trained and evaluated multiple machine learning and deep learning models, including classical regressors, state-of-the-art neural networks, and weakly supervised pattern-recognition networks using a multi-institutional dataset of annotated whole slide images (WSIs) of OPMD cases with known transformation status. In Aim 2, spatial transcriptomics (10x Genomics Visium HD) was performed on PL samples to identify gene signatures predictive of transformation, with a focus on immune–epithelial interactions. In Aim 3, a 4NQO-induced oral carcinogenesis model was applied to mEAK- 7 knockout mice to assess its functional role in OSCC development. AI models demonstrated that lymphocyte infiltration patterns can predict malignant transformation, with deep learning models achieving accuracies up to 83.4% in distinguishing transformed from non-transformed cases. Spatial transcriptomics revealed downregulation of epithelial barrier genes (FLG, CASP14) and immune activation signatures (S100A8, S100A9, CD74) in transformation zones, supporting a model of barrier disruption and neoantigen-driven immune remodeling. The mEAK-7 knockout study showed significantly reduced OSCC incidence, implicating alternative mTOR signaling in OSCC initiation and validating spatial findings through in vivo functional evidence. In conclusion, this thesis presents an integrated, multi-modal investigation into the malignant transformation of OED, providing evidence that AI and spatial biology can complement conventional pathology in predicting cancer risk. The combined findings offer a foundation for future precision diagnostics in oral cancer prevention and identify novel molecular targets for early intervention.32 0Item Restricted Pseudo-Labeling for Deep Learning-Based Side-Channel Disassembly Using Contextual Layer and Feature Engineering(Saudi Digital Library, 2025) Alabdulwahab, Saleh Sami S; Son, YunsikEmbedded devices face critical cyber-attacks due to their lightweight design and the sensitive data they handle. Integrating cloud and embedded systems increases the need for security measures against threats. Among these threats are deep learning-based side-channel disassembly attacks, which can expose sensitive information or steal software intellectual properties. Conducting a security test to evaluate the systems against these threats is essential. However, the main challenges include a comprehensive and refined dataset for training deep learning-based side-channel attacks and the lack of public datasets; labeling and profiling such attacks are costly and time-consuming. Additionally, accurately disassembling a single instruction is difficult due to the multiple classes representing each instruction and the obfuscation caused by dummy instructions. This study aimed to create an advanced side-channel evaluation methodology that performs three main deep-learning tasks: profiling using context-aware pseudo-labeling techniques at an instruction level, a disassembly model enhanced with moving log-transformed temporal interaction features, and a sequence labeling model for the detection of dummy instructions using natural language processing techniques. Utilizing gated recurrent units, the proposed pseudo-labeling model achieved 0.996 R2 in estimating the power trace for the assembly instructions. The proposed features improved the disassembly model's accuracy to 0.993, outperforming the related works. Additionally, the detection of dummy instructions using a long short-term memory model reached an accuracy of 0.979. This study provides valuable insights and methodology for measuring the software robustness against side-channel attacks.16 0Item Restricted RESEARCH ON ELECTRICITY CONSUMPTION PREDICTION BASED ON DEEP INTEGRATION MODELS(Harbin Institute of Technology, 2024) Alkhattabi, Moayad; Chen, YingWith the development of power system and the improvement of intelligence level, power consumption forecast plays a vital role in power system operation management, energy dispatching and market trading. Effective power demand forecasting is the key to power system planning and operation, and is of great significance to achieve safe, efficient and sustainable energy supply. However, the traditional deep learning model has the problem of falling into local optimal in the optimization process, which leads to the unstable performance of the model. To overcome this problem, the particle swarm optimization (PSO) algorithm is improved in this study to improve the performance of deep learning models in power consumption prediction tasks. This study collates and summarizes the challenges encountered when dealing with nonlinear, non-stationary, and high-dimensional data. To overcome these challenges, an improved particle swarm optimization (PSO) algorithm is proposed to optimize the parameters of deep learning models, thereby enhancing the model's fitting ability and generalization performance. The improved PSO algorithm in this study adopts dynamic weight adjustment and multi-stage optimization strategy, which effectively realizes the balance between global search and local search, and greatly improves the performance of the model in complex power systems. In the process of model construction, the stack ensemble learning method is adopted, and five machine learning methods including long short-term memory network (LSTM) are used to build a deeply integrated power consumption model prediction model. To verify the validity and applicability of the model, extensive experimental tests are carried out on real world power system data sets. The experimental results show that the PSO-Stacking model in this study has a root-mean-square error (RMSE) of 0.095, a mean absolute error (MAE) of 0.074, and an R square (R²) of 0.862, which are robust performance indicators. These results demonstrate the effectiveness of improved particle swarm optimization algorithm and stacked ensemble learning model in power consumption prediction tasks. Compared with the traditional deep learning model, the optimized deep learning model using the improved PSO algorithm shows considerable improvement in accuracy, stability and response speed.38 0Item Restricted AUTOMATED DETECTION OF OFFENSIVE TEXTS BASED ON ENSEMBLE LEARNING AND HYBRID DEEP LEARNING TECHNIQUES(Florida Atlantic University, 2025-05) Alqahtani, Abdulkarim Faraj; Ilyas, MohammadThe impact of communication through social media is currently considered a significant social issue. This issue can lead to inappropriate behavior using social media, which is referred to as cyberbullying. The accessibility and freedom of expression afforded by social media platforms enable individuals to share their emotions and opinions, but it also leads to cyberbullying behavior. Automated systems are capable of efficiently identifying cyberbullying and performing sentiment analysis on social media platforms. In this dissertation, our focus is on enhancing a system to detect cyberbullying in various ways. Therefore, we apply natural language processing techniques utilizing artificial intelligence algorithms to identify offensive texts in various public datasets. The first approach leverages two deep learning models to classify a large-scale dataset, combining two techniques: data augmentation and the GloVe pre-trained word representation method to improve training performance. In addition, we utilized multi-classification algorithms on a cyberbullying dataset to identify six types of cyberbullying tweets. Our approach achieved high accuracy, particularly with TF-IDF (bigram) feature extraction, compared to previous experiments and traditional machine learning algorithms applied to the same dataset. We employed two ensemble machine learning methods with the TF-IDF feature extraction technique, which demonstrated superior classification performance. Moreover, we used four feature extraction methods, BoW, TF-IDF, Word2Vec, and GloVe, to determine which works best with the ensemble technique. Finally, we utilize a multi-channel convolutional neural network (CNN) enhanced with an attention mechanism and optimized using a focal loss function.17 0Item Restricted Human Activity Monitoring for Telemedicine Using an Intelligent Millimeter-Wave System(University of Dayton, 2025) Alhazmi, Abdullah; Chodavarapu, VamsyThe growing aging population requires innovative solutions in the healthcare industry. Telemedicine is one such innovation that can improve healthcare access and delivery to diverse and aging populations. It uses various sensors to facilitate remote monitoring of physiological measures of people, such as heart rate, oxygen saturation, blood glucose, and blood pressure. Similarly, it is capable of monitoring critical events, such as falls. The key challenges in telemonitoring are ensuring accurate remote monitoring of physical activity or falls by preserving privacy and avoiding excessive reliance on expensive and/or obtrusive devices. Our approach initially addressed the need for secure, portable, and low-cost solutions specifically for fall detection. Our proposed system integrates a low-power millimeter-wave (mmWave) sensor with a NVIDIA Jetson Nano system and uses machine learning to accurately and remotely detect falls. Our initial work focused on processing the mmWave sensor's output by using neural network models, mainly employing Doppler signatures and a Long Short-Term Memory (LSTM) architecture. The proposed system achieved 79% accuracy in detecting three classes of human activities. In addition to reasonable accuracy, the system protected privacy by not recording camera images, ensuring real-time fall detection and Human Activity Recognition (HAR) for both single and multiple individuals at the same time. Building on this foundation, we developed an advanced system to enhance accuracy and robustness in continuous monitoring of human activities. This enhanced system also utilized a mmWave radar sensor (IWR6843ISK-ODS) connected to a NVIDIA Jetson Nano board, and focused on improving the accuracy and robustness of the monitoring process. This integration facilitated effective data processing and inference at the edge, making it suitable for telemedicine systems in both residential and institutional settings. By developing a PointNet neural network for real-time human activity monitoring, we achieved an inference accuracy of 99.5% when recognizing five types of activities: standing, walking, sitting, lying, and falling. Furthermore, the proposed system provided activity data reports, tracking maps, and fall alerts and significantly enhanced telemedicine applcations by enabling more timely and targeted interventions based on objective data. The final proposed system facilitates the ability to detect falls and monitor physical activity at both home and institutional settings, demonstrating the potential of Artificial Intelligence (AI) algorithms and mmWave sensors for HAR. In conclusion, our system enhances therapeutic adherence and optimizes healthcare resources by enabling patients to receive physical therapy services remotely. Furthermore, it could reduce the need for hospital visits and improve in-home nursing care, thus saving time and money and improving patient outcomes.28 0Item Restricted Cloud computing efficiency: optimizing resource utilization, energy consumption, latency, availability, and reliability using intelligent algorithms(The Universit of Western Australia, 2024) Alelyani, Abdullah Hamed A; Datta, Amitava; Ghulam, Mubasher HassanCloud computing offers significant potential for transforming service delivery with a cost-efficient, pay-as-you-go model, which has led to a dramatic increase in demand. The advantages of virtual machine (VM) and container technologies further optimize resource utilization in cloud environments. Containers and VMs improve application reliability by distributing replicated tasks across different physical machines (PMs). However, several persistent issues in cloud computing remain, including energy consumption, resource management, network traffic costs, availability, latency, service level agreement (SLA) violations, and reliability. Addressing these issues is critical for ensuring QoS. This thesis proposes approaches to address these issues and improve cloud performance.30 0
- «
- 1 (current)
- 2
- 3
- »