Doctoral Dissertations

Permanent URI for this collection

https://hdl.handle.net/10919/11041

Browse

Now showing 1 - 20 of 18294

VHealth Suite: A Unified, Secure, and Intelligent Patient-Centered Framework for Legacy System Integration in Virtual Hospital Ecosystems
Alsalamah, Sara Abdullah I. (Virginia Tech, 2026-05-11)
A virtual hospital (VH) is a distributed, digitally enabled healthcare ecosystem that extends clinical services beyond physical facilities, facilitating patient-centered (PC) care across geographically dispersed settings through interoperable infrastructures, telemedicine platforms, and hub-and-spoke coordination. However, legacy healthcare information systems remain fragmented, disease-centered, and operationally reactive, which limits secure data sharing, knowledge integration, and system-wide capacity awareness. These challenges are further exacerbated by rising demand, workforce constraints, and the need for predictive operational intelligence to enable efficient and scalable care delivery. To address these limitations, this dissertation proposes VHealth Suite, a unified, secure, and intelligent framework designed to modernize legacy healthcare information systems and seamlessly integrate them into VH ecosystems without requiring system replacement. The framework is implemented as a multi-component architecture that integrates secure data exchange, intelligent knowledge extraction, predictive operational intelligence, and human-in-the-loop interaction. First, the secure data exchange component is realized through VHealth-AC, a novel access control (AC) model that enables fine-grained and secure access to PC data across distributed and autonomous healthcare systems. The model employs a five-tier PC information classification scheme and operates as a neutral collaboration security domain, allowing clinicians to securely access patient data across institutional boundaries at the point of care. Second, intelligent PC knowledge extraction is achieved through VHealth-CNN and VHealth-MFusion. VHealth-CNN leverages a double-layer convolutional neural network (CNN) to extract and classify health-related features from biomedical data, achieving prediction accuracies of 91.3%, 93.5%, and 95% for obesity, hypertension, and diabetes, respectively. VHealth-MFusion introduces a hierarchical multimodal deep learning framework that integrates chest X-ray (CXR) images with structured clinical data, achieving 97.2% overall classification accuracy, improving robustness under class imbalance, and reducing misclassifications among clinically similar conditions. Third, predictive operational intelligence and clinical routing are addressed through VHealth-Routing, an AI-driven framework that combines clinical decision support with capacity-aware optimization. The framework integrates a clinical routing engine, a spatiotemporal prediction engine, and a constrained re-ranking mechanism to align clinical relevance with operational feasibility. It is evaluated using a large-scale real-world dataset from the Seha VH ecosystem in Saudi Arabia, comprising over 15 million records, with a representative subset of 1,006,111 appointments used for experimentation. Results demonstrate strong routing performance, with XGBoost achieving 73.2% Top-1 accuracy and 97.6% Top-3 accuracy, alongside effective demand forecasting and waiting time estimation, supporting improved workload distribution and reduced system inefficiencies. Finally, the human-in-the-loop component is implemented through VHealth-Bot, an AI-driven conversational platform that integrates natural language processing, diagnostic reasoning, and adaptive learning to support clinician–patient interaction. The system enhances real-time symptom assessment, personalized response generation, and collaborative decision-making, while maintaining clinician oversight to ensure safety and preserve clinical expertise. Evaluation results indicate improvements in diagnostic support, workflow efficiency, clinician–patient communication, and patient satisfaction. Overall, VHealth Suite provides a scalable, privacy-preserving, and intelligent architecture that unifies clinical intelligence with operational optimization. The proposed framework enables proactive, data-driven, and PC care delivery in large-scale VH ecosystems, improving clinical outcomes, enhancing operational efficiency, and fostering more responsive healthcare systems.
Identifying Novel Therapeutic Approaches for Individual Symptoms of Alcohol Use Disorder
Dong, Yuyang (Virginia Tech, 2026-05-11)
Alcohol use disorder (AUD) is a multi-symptomatic disorder which presents a continued challenge and burden to healthcare worldwide. While current Food and Drug Administration (FDA) approved treatments of AUD address some symptoms, they do not address others, notably alcohol dependence-associated pain; additionally, the existing treatments have limited effectiveness, leaving subpopulations of AUD patients underserved. Our work examined and summarized the current research on alcohol dependence-associated pain and provided a comprehensive overview of techniques to quantify nociception in rodent models, in the context of AUD research. We then examined different potential nociceptive signaling pathways for their involvement in dependence associated-pain using the chronic intermittent ethanol vapor (CIE) mouse model. We found that the alcohol dependence associated-pain mechanism is distinct from other mechanisms of chronic pain and that it is independent of endocannabinoid signaling pathways. As our prior work found association between plasma levels of pro-inflammatory lipid 15-Hydroxyeicosatetraenoic acid (15(S)-HETE) in AUD patients with their alcohol craving, and the 15-lipoxygenase (15-LOX) signaling pathway is involved in the development of chronic pain, we examined the relationship between this pathway and various symptoms of AUD. We found that 15-LOX signaling contributes to the escalation of alcohol intake characteristic of alcohol dependence and the development of craving-like behaviors. Overall, our findings highlight the importance and uniqueness of the different mechanisms that underlie different symptoms of AUD, with alcohol dependence associated-pain having a distinct mechanism different from other chronic pain mechanisms and powerfully implicate the 15-LOX signaling pathway in escalation of alcohol intake and craving.
Holistic and Generalizable Evaluation of Generative Models
Liu, Minqian (Virginia Tech, 2026-05-11)
The rapid advancement of generative models, e.g., language models and multimodal models, has created a critical gap between their sophisticated capabilities and the evaluation methodologies used to assess them. While these models now excel at complex creative and interactive tasks, evaluation approaches remain anchored to task-specific metrics designed for constrained generation. This dissertation addresses the fundamental challenge of developing comprehensive, scalable, and generalizable evaluation methodologies that match the complexity of modern generative models. First, I introduce our efforts to advance the evaluation paradigms for assessing emerging generative capabilities. I develop instruction-tuned evaluators that generalize to unseen evaluation aspects without retraining, enabling adaptive assessment with new criteria. I then extend evaluation to the multimodal scenario, where I introduce a holistic framework for unified text-and-image generation and address the unique challenges of cross-modality assessment. Next, I extend the evaluation of generative models to scientific domains and propose the structured ideation space that decomposes scientific ideas into orthogonal conceptual dimensions, enabling fine-grained novelty assessments grounded in retrieved evidence rather than surface-level similarity. Beyond assessing generative capabilities, I present our work in investigating generative models' safety and how to deploy them responsibly. I develop a systematic framework for evaluating safety risks in goal-driven interactions, revealing critical misalignments between models' refusal behaviors and their deployment of unethical strategies. I also introduce a generalizable guardrail training methodology that enables AI guardrails to adapt to unseen policies and domains. Together, these contributions establish new evaluation paradigms that enable more principled development of generative models while ensuring alignment with human values and societal needs.
Digital Twins for Accelerated Qualification in Additive Manufacturing
Deshmukh, Kaustubh Vilas (Virginia Tech, 2026-05-11)
The goal of this research is to enable accelerated qualification of metal parts produced using a specific additive manufacturing process called laser powder bed fusion. Realizing the goal will lead to a Born Qualified Additive Manufacturing paradigm for part quality assurance. Born Qualified Additive Manufacturing entails predicting part quality as it is manufactured and implementing corrective actions in real time to cure defects, thus enabling rapid, in-situ, online, and non-destructive assurance of part integrity. Despite its demonstrated ability to elevate performance while simultaneously reducing manufacturing cost, complexity, and time, the adoption of additive manufacturing processes in safety-critical applications remains limited. The main barrier is quality – the prevalence of defects and inconsistent (heterogeneous) microstructures lead to large uncertainty in safety and performance. High costs (~50% of manufacturing cost) and time are spent on post-build quality assurance through destructive characterization and testing. Hence, rapid, online, non-destructive, in-situ, and real-time quality assurance is vital for the sustainability of additive manufacturing. In pursuit of Born Qualified Additive Manufacturing, the overarching objective of this work is to establish a novel physics- and data-informed hierarchical digital twin framework to understand (model), observe (monitor), predict, and control the causal process phenomena that govern the critical-to-quality aspects of additively manufactured parts. The central hypothesis of the digital twin framework is that combining rapid physics-based computational modeling, in-situ sensor-based monitoring, and machine learning enables accurate prediction and control of the solidified microstructure and part properties. The hierarchical digital twin framework is stratified into the following three parts. Part 1: Modeling the causal process phenomena and understanding their effect (correlation) on solidified microstructure and mechanical properties. Part 2: Prediction of solidified microstructure using physics and data-integrated machine learning. Part 3: Control of part thermal history through computational modeling and real-time in-process sensing for improved part quality. Collectively, this work provides new theoretical and practical foundations in modeling, monitoring, machine learning, and machine control toward Born Qualified additive manufacturing.
Understanding invasion of stream restoration projects and the resulting impacts to the soundscape
Ripa, Gabrielle Nicole (Virginia Tech, 2026-05-11)
Stream restoration is an important tool to address stressors of urban streams, such as flashy flows and urban runoff, that lead to channel erosion and poor water quality. However, the disturbance associated with stream restoration can leave space and resources available for invasive plants to establish. To understand the dynamics between stream restoration and invasion, I examined the vegetation communities of 46 stream restoration projects in the Chesapeake Bay watershed and paired unrestored stream reaches. I found that restored stream reaches were more invaded than their unrestored pairs and the increased invasion, though related to increased resource availability (e.g., soil nutrients, photosynthetically active radiation), was not explained by differences in availability of those resources between reaches (i.e., restored vs. unrestored) or time since restoration. Utilizing the variation in restoration outcomes, I assessed the importance of resource availability, land use, project attributes, planting design, and project monitoring variables in predicting invasive plant cover. The most important variables were resource availability variables, such that increasing light or soil nutrient availability correlated with increased invasion. Therefore, recommendations to restoration practitioners on how to limit invasive plant establishment include preserving overstory trees and limiting use of fertilizer within the limits of disturbance. Additionally, projects that used a reference site, either in restoration design or for project monitoring post-restoration, had lower invasive cover than those that did not. Though important to ecosystem function and recovery, wildlife responses to restoration are rarely assessed as part of post-restoration monitoring. Therefore, I deployed autonomous recording units across 20 of the 46 paired streams for one year to examine the impacts of stream restoration and invasive plants on soundscapes, or the sum of all sounds in the environment (e.g., birds, frogs, insects). I observed significant seasonal variation between restored and unrestored streams and between high and low invasion streams, as determined through established soundscape indices, such as increased bioacoustic activity in the winter on high invasion streams. Soundscape differences between high and low invasion streams could be due to phenological differences between invasive and native plants whereas differences due to restoration could be due to changes in geomorphology and hydrology. Given the documented negative impacts of invasive plants on native ecosystems, my work provides an understanding of how stream restoration affects invasion, methods to limit invasion of stream restoration projects, and the first application of passive acoustic monitoring to assess effects of restoration and invasion on soundscapes.
Rapid Characterization of Material Heterogeneity through Acoustic Resonance and Physics-informed Artificial Intelligence
Wu, Xiaofeng (Virginia Tech, 2026-05-11)
Metal additive manufacturing is reshaping modern industrial production by enabling the high-throughput fabrication of geometrically complex components with reduced material waste and shortened lead times; as production volumes scale, the ability to rapidly and reliably qualify every as-built component through non-destructive evaluation becomes critically important for ensuring consistent product quality, yet remains an unresolved challenge, particularly in resource-constrained and extreme environments where conventional characterization approaches such as X-ray computed tomography are prohibitively slow and costly for large-volume inspection. To address this gap, this dissertation proposes a novel physics-informed artificial intelligence framework that integrates laser acoustic resonance spectroscopy with physics-based simulation, active learning, Bayesian calibration, and deep learning surrogates to enable rapid, contact-free quality assurance of metal additive manufacturing components. To achieve our research goals, we first establish the theoretical foundation by demonstrating that laser acoustic resonance spectroscopy signals are uniquely sensitive to mesoscale material heterogeneity and by developing an active learning–enhanced framework for bi-directional inference between heterogeneity descriptors and resonance spectra. Subsequently, we extend this framework to industrially relevant additive friction stir deposition conditions through Bayesian calibration of unknown material parameters and a two-stage data augmentation pipeline that generates physically consistent simulation ensembles from a limited experimental dataset. We then characterize a structurally distinct multi-layer specimen cohort deposited on contaminated surfaces, establishing quantitative correlations between mesoscopic flaw features and interfacial mechanical performance that confirm the framework's applicability to realistic manufacturing conditions. Furthermore, we synthesize all preceding components into a fully non-destructive quality assurance pipeline that infers ultimate tensile strength from vibrational measurements alone, achieving cross-process generalization without costly measurement or destructive testing. In conclusion, this dissertation makes a significant contribution to the field of non-destructive evaluation and metal additive manufacturing quality assurance by establishing a new physics-informed paradigm for rapid, low-cost component qualification. Our research findings have direct implications for the industrialization of autonomous manufacturing in aerospace, defense, and space exploration sectors, where reliable and efficient component certification is indispensable for structural integrity and mission success.
Trustworthy AI for Smart Systems: Ensuring Resilience, Sustainability, and Responsibility in Autonomous Decision-Making
Chen, Dian (Virginia Tech, 2026-05-11)
The rapid growth of smart systems, such as those deployed in agriculture and healthcare, has underscored their transformative potential in enhancing efficiency, sustainability, and decision-making. However, this proliferation also raises pressing concerns around trustworthiness, privacy, resilience, and ethical use of AI technologies. This dissertation investigates the foundational principles and practical implementations of trustworthy AI in smart environments, with a multidisciplinary focus on sustainability, fairness, and transparency. To this end, the research explores three key tasks. First, it develops sustainable and resilient AI for smart farm systems using techniques such as transfer learning, deep reinforcement learning, and federated learning to address energy constraints, cyber threats, and operational uncertainty in real-world deployments. Second, it designs fair and privacy-preserving AI frameworks for smart healthcare systems by incorporating uncertainty-aware and bias-mitigating mechanisms into federated learning models for equitable disease detection, particularly Alzheimer's disease. Third, it creates explainable AI solutions for smart animal welfare systems through interpretable Bayesian models that provide causal reasoning and robustness against sensor noise and adversarial disruptions in complex environments. The key contributions include the development of a solar-powered, energy-adaptive monitoring framework; a novel TL-DRL approach to enhance farm system efficiency; and the first integration of evidential neural networks into federated learning for fair, uncertainty-aware medical diagnosis. The proposed methods are evaluated on real-world datasets, demonstrating improvements in system reliability, fairness, and predictive accuracy; an uncertainty-aware Bayesian network that unifies deep and interpretable features with feature-level uncertainty modeling and propagation for structured inference and explanation. This dissertation advances the field of trustworthy AI by addressing critical gaps in resilience, fairness, and interpretability, laying the foundation for ethical and robust deployment of AI in high-stakes, resource-constrained smart systems.
Development of Novel Phosphine- and Copper-Catalyzed Alkyne Functionalization Methods
Buchbinder, Nicklas (Virginia Tech, 2026-05-08)
Organoboron compounds find immense application in organic synthesis and medicinal chemistry. The activation of boron-carbon bonds has enabled the formation of carbon-nitrogen, carbon-oxygen, carbon-sulfur, carbon-hydrogen, and carbon-carbon bonds, making them highly valuable precursors to complex molecules. Unlike many other organometallic reagents used in cross-coupling reactions, organoboron compounds display notable stability and low toxicity, further cementing their value in pharmaceutical science and agrochemical production. Furthermore, medicinal chemists have leveraged the electrophilic character of boron for the development of reversible covalent inhibitors. Five of these compounds have received approval for clinical usage by the Food and Drug Administration (FDA) for various indications. These applications underpin the value of novel reactions that selectively forge carbon-boron bonds. 1,3-Enynes are highly unsaturated substrates that present significant challenges for selective functionalization. Herein, we describe a synthetic protocol for installing a boronic ester (Bpin) into 1,3-enynes with excellent chemo-, regio-, and stereoselectivity. Utilizing a copper catalyst and pinacolborane (HBpin), the boronic ester was delivered to the internal alkyne carbon, a selectivity which had only been previously achieved with the use of directing groups. Both (Z)- and (E)-1,3-enynes are tolerated without isomerization of the alkene, highlighting that (Z,Z)- and (Z,E)-2-boryl-1,3-dienes are accessible with our methodology. The 2-boryl-1,3-diene products were further derivatized into useful functional groups, highlighting their synthetic utility. In a follow-up study, the mechanistic intricacies of the previously disclosed hydroboration reaction were investigated. A kinetic analysis revealed that the 1,3-enyne has first-order kinetics, HBpin has zeroth-order kinetics, and the catalyst (CuOAc and Xantphos) has a positive fractional rate order. These results indicate that hydrocupration (copper-hydride insertion into the alkyne) is rate-limiting and important for governing the observed selectivity. A positive fractional rate order in the catalyst suggests that the copper-hydride catalyst is in rapid equilibrium with an off-cycle complex that is catalytically incompetent. 11B NMR studies involving copper(I) acetate, Xantphos, and HBpin revealed a new peak in the 11B NMR spectra (8.9 ppm), which may be the off-cycle complex. DFT calculations indicate that a cyclic CuHBpinOAc species serves as the off-cycle complex, which is consistent with our experimental data. DFT calculations also corroborated that hydrocupration was rate-limiting and that the transition state leading to internal alkyne borylation was the most energetically accessible. Finally, we developed a mild and convenient approach for the synthesis of α,β-dehydroamino acids using an inexpensive trialkyl phosphine catalyst. α,β-dehydroamino acids are non-canonical amino acids that contain unsaturation between the α- and β-sidechain carbons. This motif appears in many bioactive natural products and the FDA-approved dehydropeptidase inhibitor Cilastatin. The synthesis of α,β-dehydroamino acids has been heavily explored, but most previously established syntheses result in N-protected dehydroamino acids, which have very limited application in the synthesis of complex α,β-dehydroamino acids. The problem arises when attempting to deprotect the N-terminus, resulting in a primary enamine which tautomerizes into the corresponding imine before hydrolyzing into the α-keto ester. To circumvent this issue, we report a straightforward synthetic method that couples complex amides to alkynoates directly, negating any requirement for N-terminal deprotection at the unsaturated amino acid. The reaction relies on a substoichiometric amount of n-tributylphosphine, an inexpensive commodity. Several dehydroamino acid-containing dipeptides were accessed using our method, without epimerization of the chiral center. The α,β-dehydroamino acids were chemically modified into two natural products, one of which (scutianene M) had never been synthesized before.
Measuring LGBTQ+ Client-Reported Discrimination in Psychotherapy: Scale Development and Initial Validation
Winter, Samuel Laurence (Virginia Tech, 2026-05-08)
This project developed a novel self-report scale designed to assess LGBTQ+ client experiences of discrimination in psychotherapy. Existing literature documents the impact of minority stress on LGBTQ+ mental health and LGBTQ+ clients report experiences of discrimination in psychotherapy, yet no comprehensive, self-report scale exists to assess these experiences. Guided by minority stress theory and grounded in critical realism, this project views psychotherapy discrimination as a significant form of minority stress and emphasizes the importance of assessing these experiences from the perspectives of clients. This project included two phases. Phase I (theme and item development) included a meta-ethnographic literature review to propose 15 preliminary themes, which were refined in collaboration with an expert advisory panel. These results informed the generation of a scale item pool. Phase II involved pilot testing the scale to a sample (N = 51) of LGBTQ+ individuals who had concerns about the psychotherapy services they previously received related to their LGBTQ+ identity. This phase generated preliminary data to assess item performance, scale structure, item clarity and relevance, and free-response feedback regarding scale comprehensiveness. The pilot study produced a 50-item instrument with a 15-factor structure. The scale demonstrated convergent and discriminant validity and excellent internal consistency (α = .97). Results support the GSI-PDS as a robust tool for measuring discriminatory distal stressors in psychotherapy. Findings were used to develop recommendations for future revisions of the scale.
Constructing Guidelines for Incorporating Flexibility in Online Assessment Strategies in Higher Education
Barua, Lumbini (Virginia Tech, 2026-05-08)
While research consistently advocates for flexible learning environments and documents the benefits of flexible assessment practices, comprehensive and structured guidance for implementing flexibility specifically within the assessment domain remains limited. In response, this study aimed to develop and validate a set of research-based guidelines for incorporating flexibility into online assessment strategies in higher education. Using Design and Development Research and grounded in Moore's Transactional Distance Theory, the guidelines draw on a systematic analysis of 26 empirical studies published between 2015 and 2024. The resulting tool addresses four areas: tasks and formats, weighting and grading, deadlines and attempts, and feedback. A planning section precedes these areas to ensure alignment and contextual consideration. An expert panel of five instructional design and online learning practitioners evaluated the guidelines using a mixed-methods survey. Reviewers confirmed clarity, relevance, theoretical grounding, and practical value. Their feedback informed iterative revisions that improved usability, coherence, and accessibility. The study offers a validated framework for learner-centered assessment in online higher education.
Advancing Yield Predictions in Pinus taeda (L.): An Artificial Intelligence (AI) Approach Leveraging LiDAR-derived Individual Tree Crown (ITC) Metrics, Competition Indices (CI), and Satellite Remote Sensing Indices
Barua, Gunjan (Virginia Tech, 2026-05-08)
This study assessed high-resolution UAV-LiDAR and multi-sensor satellite time series (Sentinel-1 and Sentinel-2) utilizing non-parametric machine learning and deep learning architectures across diverse planting densities (618, 1,236, and 1,853 trees ha-1) and thinning regimes to: 1) evaluate the accuracy of individual tree yield predictions over a 4-year interval using LiDAR-derived structural metrics and competition indices (see Chapter 1); 2) assess the temporal transferability of a single-date LiDAR acquisition for forecasting annual yield over a 7-year horizon (see Chapter 2); and 3) evaluate the efficacy of fusing continuous optical and synthetic aperture radar (SAR) time-series data using deep learning sequence models for plot-level yield estimation (see Chapter 3). Chapter 1 results showed that machine learning models significantly outperformed traditional parametric methods for medium-term yield prediction. Support Vector Machine (SVM) achieved the highest individual tree-level accuracy (normalized root mean square error (nRMSE) of 9.59%, R2 of 0.59) and underestimated stand-level volume by only -1.50%, while Random Forest (RF) achieved an nRMSE of 10.86% (R2 of 0.48) and overestimated stand volume by 1.53%. Chapter 2 results demonstrated the temporal transferability of a single age-8 LiDAR acquisition to predict annual growth from age 9 to 15. The RF model maintained high stability across the 7-year horizon (R2 greater than or equal to 0.83). Permutation feature importance revealed a biological shift where early-year predictions relied on individual structural metrics (tree top height increased MSE by 161.84%), while later-year predictions were dominated by distance-dependent competition indices (up to a 200% increase in MSE). Chapter 3 results showed that deep learning sequence models (Gated Recurrent Unit (GRU) and Long Short-Term Memory (LSTM)) and RF successfully scaled predictions to the plot level (R2 of 0.49). GRU yielded the lowest overall error (RMSE of 60.38 m3ha-1, MAE of 34.28 m3ha-1). However, a distinct U-shaped error trend emerged across stand densities; the RF model achieved its lowest error in medium-density stands (618 to 1,236 trees ha-1; RMSE of 36.66 m3ha-1), while error rates sharply increased in high-density stands (greater than 1,237 trees ha-1; RMSE of 88.44 m3ha-1) due to signal saturation above 125 m3ha-1. Three primary conclusions come from this research: 1) non-parametric machine learning models utilizing individual tree crown metrics and competition indices accurately predict tree-level yield without violating the homoscedasticity assumptions that limit traditional linear models (see Chapter 1); 2) a single-date LiDAR acquisition is temporally transferable and captures fundamental ecological shifts from individual size-dependent growth to distance-dependent competition as canopy closure intensifies (see Chapter 2); and 3) deep learning architectures effectively fuse continuous SAR and optical satellite data for landscape-scale estimation, though sensor signal saturation remains a critical bottleneck in dense, mature plantations (see Chapter 3). From these findings, we hypothesize that deploying machine learning and deep learning models to integrate multi-scalar remote sensing data: 1) overcomes the spatial and logistical constraints of traditional plot-based field inventories, which 2) translates to highly accurate, continuous yield forecasting across varying silvicultural regimes, and 3) enables dynamic, data-driven precision forestry management, provided that sensor saturation thresholds in high-biomass stands are properly identified and mitigated.
"I No Longer Saw Students Daydreaming or Sleeping in Class": Exploring Perceptions of Active Leaning Among Elementary Teachers of Arabic in Beirut
Al Dirani, Hajar (Virginia Tech, 2026-05-08)
Active learning is widely associated with improved student achievement, engagement, motivation, and higher-order thinking. However, limited research has examined how active learning is understood and implemented in Lebanese schools, particularly in Arabic language instruction. This qualitative study investigated how elementary teachers of Arabic in Beirut, Lebanon, conceptualize active learning, the instructional practices they reported using, their perceptions of its effectiveness in fostering student engagement and learning, and the contextual factors influencing implementation in K–6 classrooms. Ten in-service teachers participated in semi-structured interviews conducted via Zoom. Data were transcribed and analyzed using inductive thematic analysis. Four themes emerged: (a) teachers conceptualized active learning as a student-centered process emphasizing participation, interaction, discovery, and knowledge construction; (b) teachers reported using collaborative learning, experiential activities, questioning, differentiated instruction, and selective lecture; (c) teachers perceived these strategies as increasing student engagement, confidence, and meaningful learning; and, (d) teachers identified contextual influences, including curriculum demands, workload, resources, school leadership, socioeconomic inequalities, and sociocultural attitudes toward Arabic. Findings suggest that teachers value active learning but implement it within broader institutional constraints.
A Dual Basis for the Equivariant Quantum K-theory of Cominuscule Varieties
Summers, Kevin Barrett (Virginia Tech, 2026-05-08)
The equivariant quantum K-theory ring of a flag variety is a Frobenius algebra equipped with a perfect pairing called the quantum K-metric. It is known that in the classical K-theory ring for a given flag variety, the ideal sheaf basis is dual to the Schubert basis with regard to the sheaf Euler characteristic pairing. We define a quantization of the ideal sheaf basis for the equivariant quantum K-theory of cominuscule flag varieties. These quantized ideal sheaves are then dual to the Schubert basis with regard to the quantum K-metric. We prove explicit type-uniform combinatorial formulae for the quantized ideal sheaves in terms of the Schubert basis for any cominuscule flag variety. We also provide an application utilizing the quantized ideal sheaves to calculate the Schubert structure constants associated to multiplication by the top exterior power of the tautological quotient bundle in the equivariant quantum K-theory ring of the Grassmannian. We go on to give a conjectural formula for this quantized ideal sheaf basis for a generalized full flag manifold, G/B. This conjecture is supported with examples and is consistent with the relationship that exists between G/B and the affine Grassmannian.
From measurement to policy: Sugar-sweetened beverages as a public health nutrition target
Downey, Haylee Montez (Virginia Tech, 2026-05-07)
Sugar-sweetened beverages (SSB), or beverages with added caloric sweetener, have become a staple of the American diet with the industrialization of food. However, consuming SSB is associated with increased risk of diet-related chronic diseases, including cardiovascular disease and type 2 diabetes. Given these harms, accurate measurement of SSB intake and understanding ways to reduce SSB intake are needed to improve diet quality and prevent disease. SSB taxes are an effective way to reduce SSB purchasing. To further understand SSB taxes and purchasing, I developed an experimental marketplace, or a virtual storefront, in study 1. Participants purchased less SSB when SSB were taxed, helping to validate the paradigm. In study 2, I used the experimental marketplace to examine how different tax bases, or sets of products taxed, influenced beverage purchasing. Taxes on SSB or both SSB and non-sugar sweetened beverages were similarly effective in reducing SSB purchasing. In study 2, I also examined if the effect of taxes on SSB purchasing depended on individual characteristics relevant to health equity, including household income and level of SSB intake, finding larger reductions for people who drink more SSB. In study 3, I examined the reliability of a tool to assess beverage intake across household income levels. Measurements were correlated across time for both income groups, but intake was significantly higher for the first measurement. Overall, this work contributes to better understanding of SSB as a target in public health nutrition, including development of measurement tools and evaluation of reduction policies.
The Impact of Controlled Diets High-In and Free of Ultra Processed Foods on Behavior and Brain in Emerging Adulthood
Leslie, Emma Henry (Virginia Tech, 2026-05-07)
Ultra-processed foods (UPF) comprise over half of the American diet and have been linked to poor health outcomes. Adolescence and young adulthood represent critical periods of executive function development, during which UPF consumption is highest. This crossover randomized controlled feeding trial examined the effects of UPF consumption on cognition, brain function, and eating behavior in individuals aged 18–25 years. Participants completed two 14-day controlled diet conditions in random order: a high UPF diet (81% energy from UPF) and a NonUPF diet (0% energy from UPF). Cognitive testing and functional magnetic resonance imaging (fMRI) were conducted before and after each diet to assess executive function, delay discounting, and brain response to milkshake. Following each intervention, participants completed an ad libitum buffet meal containing matched UPF and NonUPF foods to evaluate eating behavior. Removal of UPF from the diet improved inhibitory control as measured by the Flanker task. In adolescents (18–21 years), orbitofrontal cortex (OFC) response to milkshake consumption decreased following the UPF diet and increased following the NonUPF diet, whereas young adults (22–25 years) showed no changes. Habitual UPF intake was positively associated with OFC response independent of experimental diet condition. Additionally, entorhinal cortex response predicted subsequent energy intake at the buffet meal. Together, these findings demonstrate that UPF consumption may alter executive function, food valuation, and eating behavior in adolescents and young adults.
Maternal Nutrition Management in Late Gestating Beef Cattle and Its Impact on Offspring Performance
Alves Cruz, Vinicius (Virginia Tech, 2026-05-06)
This dissertation aimed to elucidate how different dietary supplements can positively impact dam's productivity and optimize calf performance through maternal nutrition. With that, two trials evaluated nutritional management of late-gestating beef cows to enhance offspring productivity. The goal of trial 1 was to evaluate the effects of rumen-protected omega-3 fatty acids (eicosapentaenoic (EPA) and docosahexaenoic (DHA)) supplementation to first-calf beef heifers during third trimester gestation on performance and physiological responses of the offspring. The goal of trial 2 was to evaluate the effects of three sources of Cu, Zn, and Mn (sulfate, organic, and hydroxychloride) supplementation to beef cows during third trimester on performance and physiological responses of the offspring. In the first trial, forty-four pregnant Angus first-calf heifers were ranked by initial body weight (BW) and body condition score (BCS), and assigned to receive a supplement containing 234 g/ heifer/ feeding of Ca salts of PUFA based on EPA and DHA acids (OMG, Strata; Virtus Nutrition LLC, Corcoran, CA; n= 22), or 2) or 234 g/heifer/ feeding of Ca salts of saturated and monosaturated fatty acids based on palmitic and oleic acids (CON; EnerGII, Virtus Nutrition; n= 22). From day 0 (beginning of the first trimester) until calving, cows were gathered and fed the treatments thrice a week. From day 11 (days of gestation 195 ± 5.1) to calving, cows were allocated to rangeland pasture. First-calf heifer BW and BCS were recorded (days −10 and −9), and blood was collected on day −10, and upon calving. Calves were weaned on day 260, and preconditioned from days 260 to 302, and remained on the feedlot from days 303 to 350. No differences were detected for heifer BW or BCS changes at the beginning of the trial and at calving (P ≥ 0.33 and P ≥ 0.65, respectively). A tendency for greater concentration of colostrum IgG (P = 0.08) was observed in heifers supplemented with OMG vs. CON cohorts. There were no differences (P ≥ 0.31) among treatments at calving for calving rate, birth BW, heart girth, and from OMG heifers had greater (P = 0.04) plasma IgG concentration compared to CON calves. No differences were observed (P ≥ 0.11) for weaning rate and age, birth to weaning average daily gain (ADG), weaning weight, liver enzymes, and serum antibodies against respiratory viruses. During the preconditioning, OMG calves had greater (P ≤ 0.05) final BW and ADG compared with calves born to CON heifers. These differences in performance remained (P ≤ 0.05) throughout the receiving phase. A treatment × day interaction was detected (P = 0.05) for plasma cortisol concentration, which was greater (P < 0.01) for calves born to OMG heifers on days 260 and 263 and lower (P < 0.01) on day 306 compared to CON cohorts. In the second trial, seventy-two nonlactating, pregnant Angus cows were ranked by pregnancy type (artificial insemination or natural service), BW, and BCS and assigned to receive a supplement containing: 1) Cu, Mn, and Zn sulfate source (INR; n = 24); 2) Cu, Mn, and Zn organic-complexed source (ORG; n = 24); or 3) Cu, Mn, and Zn hydroxychloride source (HDX; n = 24) from day 0 (beginning of the first trimester) until calving, cows were gathered and fed the treatments thrice a week. Cow BW and BCS were recorded, and blood was collected on days 10 and 11, upon calving, and at weaning. Liver biopsies were performed in all cows on day 10 and upon calving (cows and calves). Longissimus muscle (LM) biopsies were performed, and blood was collected in all calves upon calving. Calves were weaned on day 260, backgrounded for 99 d, and then sent to a commercial feedyard. Calves blood samples were collected on days 245, 260, 264, 268, 275, 280, and 288. No differences were detected (P ≥ 0.31) for cow BW and BCS changes among treatments during gestation, and mineral sources did not increase (P ≥ 0.16) mineral liver concentrations of Zn, Mn, Se, and Co at parturition. However, there was a tendency (P = 0.07) for HDX to have increased liver Cu concentration vs. ORG and INR cows. Cows fed HDX and ORG had increased (P ≤ 0.03) BCS at weaning, and BCS change (P ≤ 0.03) from parturition to weaning vs. INR cows. No treatment differences were detected (P ≥ 0.21) for calf birth measurements, IgG levels, mRNA expression of hepatic enzymes, or LM genes associated with muscle and adipose tissue development. At weaning and during back¬grounding, no treatment differences were detected (P ≥ 0.21) for offspring performance, health outcomes, blood hormones, or metabolites. However, a tendency for a treatment × day interaction was detected (P = 0.07) for haptoglobin con¬centrations, which was reduced (P < 0.01) in calves from cows supplemented with HDX vs. calves from cows supplemented with ORG and INR 15 d after weaning. No treatment effects were noted (P ≥ 0.35) for final BW, feedyard ADG, and carcass traits between treatment groups. Collectively, these results are suggestive of programming effects on postnatal offspring health and productivity resultant from omega-3 fatty acids, on the other hand, different sources of Cu, Mn, and Zn supplementation during last trimester of gestation had little to no effect on offspring performance. However, the true potential of trace minerals from different sources such as hydroxychloride and omega-3 fatty acids supplementation during late gestation on the offspring performance still needed to be further explored.
Nanostructured Adsorbents for Selective Lithium Recovery: Mechanisms, Fabrication, Performance Evaluation, and Applications in Desalination
Pan, Yanan (Virginia Tech, 2026-05-05)
The growing demand for lithium in energy storage systems necessitates the development of selective and scalable extraction technologies, particularly from complex saline water resources. This dissertation systematically investigates nanostructured Lithium/Aluminum-layered double hydroxide (Li/Al-LDH)–based adsorbents through mechanistic modulation, structural engineering, and process integration to enhance lithium recovery performance. Poly(acrylic acid)-modified LDH (PAA@LDH) was first developed to regulate surface electronic density and interfacial charge distribution. Under optimized conditions (2.5 wt% PAA, 333 K), the lithium adsorption capacity increased from 2.08 mg/g to 3.41 mg/g in low-concentration produced water, with rapid equilibrium achieved within 40 min. Density functional theory (DFT) calculations revealed enhanced electron cloud density and reduced equipotential charge, confirming a charge-transfer-driven adsorption mechanism. To overcome diffusion limitations of powder adsorbents, electrospun lithium porous nanosorbent fibers (Li-PNFs) were fabricated by embedding LDH into a polyacrylonitrile matrix. The optimized fibers exhibited a uniform diameter of approximately 546 nm, tensile strength of 2.48 MPa, and yield stress of 0.09 MPa, ensuring mechanical robustness. The hierarchical porous structure enabled a significantly enhanced static lithium adsorption capacity of 13.45 mg/g, reaching equilibrium within 60 min. DFT analysis further revealed strong Li⁺ binding energy of up to -5.72 eV, indicating favorable adsorption thermodynamics. To bridge material performance with scalable operation, a continuous fixed-bed system was implemented. Under optimized conditions, the Li-PNFs achieved a lithium capture efficiency of 23.83% in a single-pass continuous experiment. Breakthrough behavior was successfully predicted using the Clark and Thomas models, demonstrating reliable dynamic adsorption modeling. Finally, a dual-functional MoS2–LDH@Sponge composite was developed to integrate photothermal evaporation with lithium-selective adsorption. The bilayer architecture achieved broadband light absorption (>97%) and demonstrated a dynamic hydration-controlled adsorption mechanism. Photothermal stimulation initially enhanced ion transport and interfacial evaporation, followed by hydration-shell restructuring that modulated lithium mobility. This work establishes a multi-scale framework that couples electronic modulation, nanoscale architecture, continuous process engineering, and solar-driven energy input for sustainable lithium recovery.
In-Vehicle Information Systems: Design Trends and Their Impact on Driver Behavior
Anderson, Gabrial Trail (Virginia Tech, 2026-04-27)
Introduction: Humans have a limited supply of resources to apply to a given task. Concurrent tasks that demand similar resources will cause interference and task performance will degrade. In-vehicle information system (IVIS) interactions are common while driving and can interfere with performing the dynamic driving task. Recent IVIS design trends have replaced physical analog controls with digital controls on touchscreens. Lack of feedback when interacting with digital controls may demand more of a driver as opposed to analog controls that may be used via tactile feedback alone. This dissertation investigated differences in driver distraction, driving performance, and duration of interactions between analog and digital IVIS use while driving. Method: Data from three naturalistic driving studies were used. Second Strategic Highway Research Program Naturalistic Driving Study (SHRP2 NDS) represented vehicles with analog IVIS (26 makes; model years [MY] 1996 – 2009), while Virginia Connected Corridor 50 NDS (VCC50 NDS; (Teslas; MY 2015 – 2017), and VTTI L2 NDS (Teslas; MY 2018 – 2022; Subarus; MY 2017 – 2021) represented digital IVIS. Driver eyeglances towards the IVIS were manually coded using video data to measure driver distraction. Standard deviation of lane position (SDLP), standard deviation of speed (SDS), and maximum X--Y accelerations were calculated for driving performance. Interaction duration was coded for each IVIS interaction and was used to represent duration. The distraction and driving performance studies used the same subset of data with representation across various driver ages (M = 39.30; SD = 16.35) and gender (Male = 616; Female = 494). Duration used a different subset of data with driver age (M = 43.05; SD = 13.66) and gender (Male = 219; Female = 163) similar to the other studies. Results: Driver distraction was higher for drivers interacting with a digital IVIS compared to analog IVIS. Driving performance, particularly SDLP, was worse for drivers interacting with digital IVIS compared to analog IVIS. Interaction duration did not differ between digital IVIS and analog IVIS. Conclusion: Interacting with digital IVIS had more impact on driver behavior than using analog IVIS. This difference is likely driven by using a touchscreen during interactions compared to physical controls with differences in interaction duration being ruled out as a confounding influence. Interpretation of these results should be limited to only the vehicles included in this study. Future research should include more digital IVIS models to better represent that population as only two manufacturers' approaches were in this dataset. Practical Implications: For the vehicles included in this study, digital IVIS interactions had greater impact on driver behavior than using analog IVIS. Previous research suggests digital IVIS can be designed to limit driver behavior impacts. Manufacturers considering an analog-to-digital IVIS transition should investigate the optimal design of digital IVIS to minimize impact on driver behavior.
Deep Learning for Enhancing Human and Environmental Health
Choi, Joung Min (Virginia Tech, 2026-04-21)
Ensuring human and environmental health is a growing global priority and a fundamental challenge at the intersection of computer science, biology, and medicine. Advances in high-throughput sequencing technologies have enabled comprehensive characterization of biological systems across multiple omics layers, offering unprecedented opportunities to support precision medicine and environmental risk prevention. These data have been widely used for disease understanding, patient stratification, and monitoring of microbial communities in both clinical and environmental settings. In recent years, deep learning has emerged as an approach for modeling nonlinear relationships from high-dimensional and noisy omics data, demonstrating improved performance over traditional machine learning methods across various tasks. However, its practical application remains fundamentally constrained by key challenges arising from omics data scarcity and heterogeneity, including (1) limited availability of labeled samples, (2) batch effects across datasets, (3) the prevalence of missing values, and (4) the need for efficient and robust learning under limited data conditions. This work proposes a series of deep learning frameworks to address these challenges and enhance the practical applicability of omics-based analysis. To mitigate the scarcity of labeled data and batch effects, BCtypeFinder and CancerSubminer are presented as cancer subtyping methods that leverage both labeled and unlabeled datasets while correcting batch effects, resulting in improved robustness and generalizability. To address missing data in longitudinal studies, DeepMicroGen is developed as a generative adversarial network-based imputation framework that captures temporal dependencies and accurately reconstructs incomplete observations, thereby improving downstream predictive performance. Furthermore, to enable efficient and robust learning under limited data conditions, ARGfore is proposed as a forecasting framework for predicting antibiotic resistance gene abundances from time-series omics data, achieving improved predictive performance with reduced computational cost. Collectively, the proposed methods help to advance the applicability of deep learning in omics research by addressing fundamental omics data-related challenges. This work contributes to more robust disease characterization and improved predictive modeling and forecasting, thereby supporting the broader goals of precision medicine and environmental risk prevention.
Toward Cybersecurity Evaluation-by-Design: Implications for Evaluation in Complex Sociotechnical Systems
Adeoye, Samson Olajide (Virginia Tech, 2026-04-21)
Cybersecurity is a multifaceted, sociotechnical phenomenon shaped by the dynamic interaction of people, processes, and technologies within increasingly complex organizational environments. Evaluation plays a critical role in understanding how well organizations safeguard the confidentiality, integrity, and availability of critical assets, yet prevailing approaches continue to privilege technical performance while under-examining human and organizational dynamics. This study advances the evaluation of cybersecurity practice as a sociotechnical endeavor through an exploratory inquiry that connects extant literature with practitioner-informed framework development. A scoping review of 34 studies drawn from 1,944 records examined how people, process, and technology (PPT) are conceptualized and evaluated in contemporary cybersecurity literature. The synthesis identified three dominant interaction patterns (people-process-technology, people-technology, and people-process) and four cross-cutting themes: human-centric sociotechnical synergy; the need to re-conceptualize evaluation for cybersecurity; the opportunities and ethical tensions associated with artificial intelligence; and technology as a complement rather than a substitute for human judgment. The review revealed a paradox: while cybersecurity discourse increasingly adopts sociotechnical language, evaluation practice remains fragmented and disproportionately focused on technical performance metrics. Building on these insights, the study develops the Cybersecurity Evaluation-by-Design Framework (CEDF) and used concept mapping with practitioners from critical agriculture and life sciences organizations to operationalize it. Guided by a critical realist retroduction approach, findings reveal that practitioners view cybersecurity as a complex adaptive system requiring continuous learning, evaluative thinking, and anticipatory capacity. Practitioner-generated clusters and importance-feasibility ratings informed a dynamic baseline—akin to progress markers—for assessing evolving organizational capabilities and developmental maturity. These findings offer foundational guidance for integrated, context-sensitive cybersecurity evaluation in complex and rapidly evolving environments as well as potential applications beyond cybersecurity, such as health emergency preparedness, climate adaptation, and disaster risk management contexts.

Browse

Recent Submissions