Doctoral Dissertations

Permanent URI for this collection

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

Browse

Now showing 1 - 20 of 18441

Filtering and Domain Decomposition Techniques for Intrusive and Non-intrusive Reduced Order Models of Convection-Dominated Problems
Moore, Ian Robert (Virginia Tech, 2026-06-16)
Galerkin reduced order models (ROMs) offer significant computational savings for the simulation of partial differential equations, yet they often exhibit spurious oscillations and loss of physical fidelity in convection-dominated and multiscale regimes. This dissertation develops a framework for correcting the intrinsic spatial under-resolution of Galerkin ROMs through complementary regularization and hybridization strategies. Large eddy simulation-inspired filtering approaches are introduced, including the Ladyzhenskaya ROM and the approximate deconvolution Leray ROM (ADL-ROM), which incorporate spatial filtering to model the effects of truncated scales and enhance stability while maintaining efficiency. Rigorous numerical analysis is provided for both models, establishing stability and convergence results that strengthen their theoretical foundations. To address strongly localized high-gradient features, the OpInf-Schwarz ROM combines operator inference with a Schwarz domain decomposition framework, restricting full order modeling to localized regions while retaining ROM efficiency elsewhere. Finally, I apply spatial filtering concepts directly to OpInf to stabilize non-intrusive ROMs for convection-dominated systems.
Contributions to Machine Learning with Abstention and Surrogate Modeling with Complex Outputs
Zhang, Xinlei (Virginia Tech, 2026-06-16)
Machine learning and statistical modeling have become integral components of modern scientific and engineering analysis, providing powerful tools to model and understand complex phenomena. However, real-world applications frequently present unique challenges, such as data imbalance, fairness constraints, and high-dimensional structured outputs. This dissertation presents novel statistical methodologies to address these challenges, with a focus on classification problems under data imbalance and fairness constraints, efficient emulation of computer experiments with tensor-valued outputs, and simulation-based experimental design and analysis. Chapter 1 presents the motivation and research objectives. Chapter 2 develops methods for logistic classification with a rejection option, addressing challenges of data imbalance and fairness constraints with respect to sensitive attributes. The proposed approach generalizes existing methods by incorporating convex surrogate loss functions and fairness-aware optimization. Performance is evaluated through comprehensive simulation studies and real-world datasets, including a case study on Methicillin-resistant Staphylococcus aureus (MRSA) prediction using imbalanced electronic health record (EHR) data. Chapter 3 introduces a so-called Deco-GPT method, a Gaussian process-based emulator for computer experiments with tensor-valued outputs. The method applies tensor unfolding and higher-order singular value decomposition (HOSVD) to improve scalability and predictive accuracy. An extension for handling missing output values is incorporated via matrix completion. Performance is evaluated through simulations and case studies, demonstrating strong performance across diverse settings. Chapter 4 introduces a simulation framework using CARLA to generate structured datasets, including safety-critical driving events and a route completion dataset with mixed inputs. We consider deep Gaussian process models for analyzing such data with mixed inputs. More specifically, we handle qualitative factors via indicator-based representation and learned embeddings, and compare against standard Gaussian process and latent variable Gaussian process models. Chapter 5 summarizes the main contributions of this research and discusses potential directions for future work.
Understanding how urbanization alters the neuroendocrine system and behavior of adult and developing wild song sparrows (Melospiza melodia)
Fossett, Taylor Elaine (Virginia Tech, 2026-06-16)
Rapid environmental change, including urbanization, is shaping ecosystems around the globe. Animals often respond to changes in the environment through changing their behavior. In particular, urban-living animals are commonly found to exhibit vastly different behaviors compared to their rural-living counterparts, such as increased boldness (i.e., less fearful of predators and humans) and heightened territorial aggression. However, the brain mechanisms that are shaped by the environment and, in turn, alter behavior, are unclear. The neuroendocrine system plays a critical role in processing and responding to environmental cues through internal regulation of physiological processes and behavioral outputs. In this dissertation, I investigate the relationship between the environment, behavior, and multiple measures of the neuroendocrine system to better understand how urbanization alters the brain and behavior of a wild songbird, the song sparrow (Melospiza melodia), that reliably expresses differences in aggression across urban and rural habitats. Collectively, each chapter seeks to address the outstanding question, why are urban song sparrows are more aggressive than rural song sparrows? by asking four main questions: Chapter 1 how do changes in human activity, including altered human activity in response to the COVID-19 quarantine, influence aggression of urban and rural-living song sparrows across a span of 10 years of study? Chapter 2 how does arginine vasotocin (AVT, a neuroendocrine signaling nonapeptide that regulates social behavior) and its receptors AVT3R, AVT4R, contribute to aggression in urban and rural-living song sparrows, and how is the AVT system shaped by urbanization? Chapter 3 how do differences in resource availability that are associated with urbanization influence neuropeptide Y (NPY, a neuroendocrine signaling neuropeptide that regulates feeding behaviors) and its receptors NPY1R, NPY2R, NPY5R, and how does the NPY system relate to aggression in urban and rural song sparrows? Chapter 4 how do developmental stressors that are associated with urbanization, including brood parasitism by the brown-headed cowbird (Molothrus ater), organize the neuroendocrine system, including genes that regulate feeding behaviors, social behaviors, and the stress response: NPY and its receptors: NPY1R, NPY2R, NPY5R; AVT and its receptors: AVT3R, AVT4R; aromatase (ARO), androgen receptor (AR), estrogen receptor type 1 (ESR1), estrogen receptor type 2 (ESR2), corticotropin releasing hormone (CRH), mineralocorticoid receptor (MR), and glucocorticoid receptor (GR) in the hypothalamus and hippocampus (only MR and GR)) of nestling urban and rural song sparrows? First, I found that urban birds were reliably more aggressive than rural birds over those 10 years of study. Interestingly, both urban and rural birds were less aggressive during the COVID-19 quarantine than in prior years. However, this was part of a pattern of decreasing aggression in both urban and rural habitats over the entire 10-year period. These findings suggest that human activity has nuanced impacts on animal behavior independent of environment. Second, I observed high AVT3R mRNA signal in the lateral septum, medial bed nucleus of the stria terminalis, hippocampus, Nidopallium, cerebellum and optic tectum, with diffuse signal in the lateral hypothalamus and nucleus rotundus. There were no significant differences in the number of AVT3R transcripts in the lateral septum between urban and rural males. When I quantified mRNA expression for AVT and two of its receptors, I found that urban males had higher relative mRNA expression of AVT3R in the hypothalamus compared to rural males, and aggression immediately before capture was negatively associated with the expression of AVT and AVT4R. Thus, expression of AVT and its receptors has the potential to contribute to higher aggression in male song sparrows but may not explain habitat-related variation. Third, I found that urban males had greater mRNA expression of all three hypothalamic NPY receptors: NPY1R, NPY2R, and NPY5R. Moreover, while I found no evidence for habitat-related differences in the number of NPY-immunoreactive cells in the infundibular nucleus (IN) or paraventricular nucleus (PVN) of the hypothalamus, overall, there were significantly more NPY-immunoreactive cells in the IN than the PVN. Despite finding differences in NPY receptor hypothalamic mRNA expression, I failed to find a relationship between aggression and NPY system gene expression, nor NPY peptide abundance. Our results suggest that the NPY system is sensitive to urbanization but may not explain differences in aggression during the pre-breeding season or in acute aggressive responses to a simulated territorial intrusion. Finally, I found that rural nestlings had higher baseline corticosterone compared to urban nestlings, but no differences in breathing rate. I also found that there were no effects of habitat type or brood parasitism status on any of the neuroendocrine system genes that I measured. This contradicts our hypothesis that the brains of developing birds are impacted by habitat type. However, I did find that gene expression of all genes measured increases with age, except for AVT. Though AVT mRNA expression stayed consistent throughout development, AVT4R mRNA expression increased, suggesting increased sensitivity to AVT in urban and rural nestlings. Interestingly, female nestlings had higher mRNA expression of ESR1, while male nestlings had higher mRNA expression of NPY. Collectively, the results of my last chapter suggest that urban environments may not be as stressful for developing songbirds as we once thought, or that the effects of urbanization on the brain may not appear in the first 10 days of life. Together, my dissertation work advances our understanding of how urbanization shapes behavior through shaping the neuroendocrine system in adulthood, not in the first 10 days of life. Urbanization is a complex form of rapid environmental change and understanding its effects on the brain and behavior of wild animals requires a multifaceted approach that combines the use of field techniques, molecular tools, and cross-dispensary research.
Sex Differences in Protein Polyubiquitination During Fear Memory Consolidation
Brown, Brieann Mary (Virginia Tech, 2026-06-16)
Many neurological pathologies are a result of disruption in the normal memory process, with insufficient encoding and retrieval of memories being common in neurodegenerative disease. However, memory disruption does not always result in impaired memory, with pathologies like Post Traumatic Stress Disorder (PTSD) and anxiety disorders resulting from pathological strengthening or generalization of memories. Interestingly, these disorders show marked sex differences with females being over 2 times more likely to be diagnosed, highlighting the importance of researching memory in a sex specific manner. However, despite these known sex differences, the majority of research has been soley in males, leaving gaps in our understanding of how molecular mechanisms vary by sex. One potential modulator for these sex differences may be the protein ubiquitin. Previously, our group and others have highlighted the need for protein degradation in fear memory formation which relies on the ubiquitin proteasome system (UPS). In the UPS, K48 polyubiquitin chains mark proteins for degradation and we have shown that both males and females require K48 polyubiquitin to form fear memories, though targets vary by sex. We have also identified a role for the non-degradatory M1 polyubiquitin chain in fear memory formation though a similar pattern of sex specific utilization and targets once again emerges. However, this work was done at a single timepoint, making the temporal dynamics of these markers across the consolidation period yet to be elucidated. To address this gap, we utilized an unbiased polyubiquitin-type specific proteomic approach to identify protein targets of K48 and M1 in the amygdala 2 and 4 hours after contextual fear conditioning. Notably, we found that our previously reported sex differences persist for several hours after fear conditioning, with both the temporal dynamics of these changes and targeted pathways varying by sex. Building upon the pronounced sex differences in polyubiquitination in fear memory formation, we also investigated an additional non-degradatory polyubiquitin chain, K63. Our prior research identified K63 polyubiquitination in the amygdala as a uniquely female mechanism for forming fear memories unlike the previously mentioned chains. However, the mechanism responsible for the sex-specific need for K63 polyubiquitination was still an unknown. Interestingly, we had previously observed increased 5-hmC (transcriptionally active methylation marker) at the promoter of the ubiquitin gene Uba52. We hypothesized that perhaps this resting difference in Uba52 methylation could be responsible for our observed female specific usage of K63. To investigate we used a CRISPR-dCas9-TET1 targeting Uba52 in the male rat amygdala to recapitulate the female methylation state and saw that fear conditioning now resulted in male K63 polyubiquitin targets. We then combined this approach with a CRISPR-dCas13 to knock down K63 chain formation and saw that males now needed K63 polyubiquitination to form fear memories. Together this body of work supports the hypothesis that the molecular mechanisms for fear memory formation are sex specific with various ubiquitin chain being a main driver of these differences. Excitingly, we have also identified the underlying mechanism of previously reported sex differences in K63 polyubiquitination. This identified mechanism of increased 5-hmc at the Uba52 promoter may underly not only what we have seen in K63 but also other reported sex differences making it an exciting new direction to explore in memory research.
Analytical and Numerical Methods in Quantum Software and Hardware
Scott, Ryan Elliott (Virginia Tech, 2026-06-16)
There are many platforms proposed for quantum computing, and various applications of quantum information to consider. In this dissertation, we explore the control of a large class of quantum platforms subjected to dynamical tunability, and we consider applications on those platforms which give natural insight into quantum many-body simulations. We consider many-body ultracold dipolar systems subjected to Floquet control as a platform for computing, algorithms which measure many-body energy gaps using hybrid methods to speed up computation, questions of simulation complexity in many-body systems, and alternative hardware platforms in photonic systems and how they can generate entanglement.
Bridging Gaps: Evaluating Strategies to Improve Vaccine Uptake and Strengthen Clinical and Translational Research Recruitment Among U.S. Adult Populations
Hensley, Amanda Andress (Virginia Tech, 2026-06-16)
Background and Rationale: Despite decades of federal policy mandates and extensive public health infrastructure investment, critical gaps have persisted in the effectiveness and equity of two core activities in U.S. health research: vaccine promotion and research recruitment. Effective vaccination strategies and representative clinical and translational research enrollment have been essential to population health, yet both have continued to fail disproportionately in reaching the populations with the greatest need. Enacted in 2010, the Affordable Care Act (ACA) eliminated financial barriers to recommended vaccines, yet adult vaccination rates have remained well below Healthy People 2030 goals across populations. Simultaneously, more than 30 years after the NIH Revitalization Act mandated inclusion of underrepresented populations in federally funded research, racial and ethnic minority groups, low-income individuals, rural residents, persons with disabilities, and other structurally marginalized communities have remained systematically underrepresented in clinical and translational research samples. These twin failures have shared a common, underexamined cause: researchers have continued to design strategies for populations of convenience rather than populations of need, generating an evidence base that has reflected and reinforced existing health disparities rather than addressing them. Objectives: This dissertation pursued two parallel systematic synthesis objectives and one framework development objective. Project 1 aimed to synthesize the post-ACA peer-reviewed literature (2010-2025) on strategies designed to increase vaccine intention and improve vaccine uptake among U.S. adult populations, and to characterize the types, settings, populations served, and effectiveness of those strategies using a multiple linear regression. Project 2 aimed to synthesize and evaluate the characteristics and effectiveness of recruitment strategies used in NIH-funded clinical and translational research to engage underrepresented U.S. adult populations (1993-2025), focusing on peer-reviewed studies that evaluated strategies through comparative or controlled designs. Project 3 built upon the findings of Project 2 to develop ReFrame, a theory-informed, multi-domain framework for planning, implementing, evaluating, and reporting research recruitment strategies with an explicit health equity orientation. Methods: Both systematic reviews followed PRISMA 2020 guidelines and were registered on the Open Science Framework (Project 1: OSF 10.17605/OSF.IO/SM7YD; Project 2: OSF 10.17605/OSF.IO/BYS5G). Comprehensive multi-database searches were conducted across PubMed, Web of Science, Ovid/MEDLINE, and EBSCOhost, including CINAHL and PsycInfo, supplemented by Google Scholar gap checks and forward/backward citation chasing, with final searches completed in December 2025. Dual independent reviewers conducted screening and data extraction at each stage, with a third reviewer resolving conflicts. Study quality was assessed using the Effective Public Health Practice Project (EPHPP) Quality Assessment Tool, and overall evidence certainty was evaluated using the GRADE framework. For Project 1, strategies were categorized according to the WHO Behavioral and Social Drivers of Vaccination (BeSD) Model, and a multiple linear regression was conducted using vaccine uptake rate as the dependent variable across 79 eligible studies. For Project 2, a multiple linear regression was conducted using enrollment rate as the dependent variable across 43 eligible studies (118 strategy arms). ReFrame was developed through the convergence of systematic review findings with established theoretical models from health behavior science, implementation science, and community-based participatory research (CBPR) principles. Results: Project 1 (Vaccine SRMA): Of the 11,219 records identified, the vaccine systematic review yielded 92 eligible studies (79 vaccine uptake, 30 vaccine intention, 17 overlapping). In the meta-regression (Adjusted R2 = 0.2285), strategies prioritized specific populations (b=0.121, p=.027), using service quality improvement approaches (b=0.226, p=.040), and recruiting through community-based partner organizations (b =0.585, p=.005) or healthcare organizations (b=0.230, p=.040) were associated with significantly higher vaccine uptake rates. Onsite vaccination was associated with significantly lower uptake rates (b=—0.185, p=.033), an effect that a sensitivity analysis confirmed was COVID-19-context-specific. Narrative synthesis identified multi-component strategies, community partnerships, and health equity-focused designs as consistently associated with more favorable outcomes across both uptake and intention domains. GRADE certainty was LOW to MODERATE across most findings. Project 2 (Recruitment SRMA): Of 5,589 records identified, 43 studies (118 strategy arms) met eligibility criteria. The median enrollment rate across all strategy arms was 32.1% (range: 0%-93%). In the meta-regression (Adjusted R2=0.385), community-based sample pool sources (b=0.331, p=.001) and EHR/registry-based sources (b=0.282, p=.003) were associated with significantly higher enrollment rates relative to other source types. Larger sample sizes were associated with significantly lower enrollment rates (b=—0.0000072, p=.026). Individual study quality was predominantly Weak by EPHPP global rating (58%), with confounders and blinding as the most frequently Weak-rated domains. Project 3 (ReFrame): ReFrame organized recruitment strategy evaluation across five interdependent domains: Strategy Planning and Design; Implementation and Adaptation; Outcome Assessment; Equity and Inclusion Analysis; and Knowledge Sharing and Reporting. The framework integrated the Health Belief Model, Social Ecological Model, CFIR, RE-AIM, CBPR principles, and Diffusion of Innovations Theory, and proposed standardized metrics, practical tools, and minimum and enhanced reporting standards applicable across diverse research contexts. Conclusions: Across both systematic reviews, a consistent pattern emerged: strategies that embedded outreach within existing relational infrastructure, whether clinical (EHR/registry-based) or community (CBPR), outperformed convenience-based approaches that relied on broad, impersonal outreach to undifferentiated audiences. For vaccination, service quality improvement and community-based recruitment pathways were robust predictors of higher uptake. For research recruitment, community-engaged and EHR/registry-based sourcing were the only statistically significant predictors of enrollment success. These findings converged on a single, equity-weighted interpretation: the participant reach and engagement have been determined primarily by where researchers have looked, and the populations most frequently overlooked were those who have borne the greatest burden of health conditions that public health and clinical research have been designed to address. ReFrame has provided an evidence-grounded framework to operationalize this insight into systematic, evaluable, and equitable recruitment practice.
Geometric and Category-Theoretic Structures in Generalized Symmetries: Gauging and Decomposition
Perez Lona, Alonso (Virginia Tech, 2026-06-16)
The concept of symmetry has been a crucial guiding principle in the development of modern physics. Recently, its mathematical characterization has been significantly expanded under the framework of "generalized symmetries," leading to conceptual, theoretical, and practical advancements in quantum field theory and related fields. Simultaneously, this has generated a fruitful collaboration with contemporary mathematics. This thesis explores several aspects of generalized symmetries: on the one hand, their relation with several branches of mathe- matics, notably category theory, geometry, and algebra; on the one hand, their implications for physical phenomena such as gauging and decomposition. The present work is divided into two main parts, following the two-fold generalization of symmetries in the generalized sym- metries framework: that of higher-form symmetries, and that of non-invertible symmetries. The first part delves into higher-form symmetries and their relation with their mathematical counterpart, higher geometry. We provide rigorous mathematical constructions making this connection concrete, using the theory of higher principal bundles with adjusted connections in cohesive ∞-topos theory. Then, we concentrate on a specific example, that of three- dimensional σ-models involving quotients by higher finite groups. In this case study, we derive decomposition statements and work out the role higher cohomology groups play. The second part concentrates on non-invertible symmetries, specializing in two-dimensional theo- ries. Following the same model, we first develop a general mathematical construction, in this case concerning the calculation of partition functions of theories with gauged non-invertible symmetries. Then, we turn to applying this formalism to describe decomposition. Along the way, we constantly highlight the natural connections with non-commutative algebra, chiefly with Hopf algebras and their representation categories.
Beyond the Relay: Mediodorsal Thalamic Regulation of Prefrontal Cortex in Cue Detection
Runyon, Kelly Elizabeth (Virginia Tech, 2026-06-16)
The mediodorsal thalamus (MD) is a critical node within thalamocortical networks supporting prefrontal-dependent cognition, yet whether its projections to distinct prefrontal subregions constitute functionally specialized pathways has not been directly examined. Here we combined retrograde anatomical circuit mapping with projection-specific fiber photometry to characterize the structural and functional organization of MD connectivity with the prelimbic cortex (PRL), anterior cingulate cortex (ACC), and orbitofrontal cortex (OFC) in mice. Retrograde tracing revealed that MD neurons projecting to PRL and ACC originate from anatomically segregated subpopulations within the lateral MD subdivision (MDL), organized along its anterior-posterior axis and forming distinct reciprocal loops with each cortical target. Projection-specific calcium imaging during Pavlovian cue-reward conditioning revealed divergent dynamics across pathways: MD-ACC terminals underwent pronounced learning-dependent temporal sharpening, with response precision increasing across acquisition and trial-by-trial dynamics becoming coupled to behavioral performance, while MD-PRL terminals showed stable, sustained cue-evoked responses that did not reshape across learning. MD-OFC terminals showed no learning-related changes during Pavlovian conditioning, consistent with OFC's established role in action-outcome rather than cue-association learning, motivating a targeted examination of MD-OFC dynamics during an instrumental task, in which the animal must explicitly use the cue to guide behavior. During instrumental performance, MD-OFC terminals exhibited robust trial-phase-specific modulation that was absent during Pavlovian conditioning, including history-dependent pre-trial baseline signals that predicted upcoming behavioral performance, post-outcome signals encoding reward identity, and across-trial carryover effects linking outcome experience to future circuit state. Together, these findings demonstrate that MD is a functionally heterogeneous nucleus whose anatomically segregated projections support complementary aspects of cue-guided learning and flexible behavior, positioning MD as an active regulator of prefrontal computation rather than a passive thalamocortical relay.
Steps Toward Open-ended Reasoning and Discovery with Language Models
Shojaee, Seyedeh Parshin (Virginia Tech, 2026-06-16)
Scientific discovery -- the process of distilling nature's complexity into compact, transferable knowledge -- has historically relied on human creativity, expertise, and intuition. Recent advances in large language models (LLMs), trained on vast amounts of scientific literature, raise a fundamental question: can these systems move beyond recovering existing knowledge to meaningfully participate in discovery? This thesis investigates this question across four research directions, progressively developing the capabilities necessary for open-ended discovery. First, we show that effective discovery systems require both broad scientific knowledge and systematic search. We introduce LLM-SR, a framework for scientific model discovery that combines LLM knowledge with evolutionary search, where LLMs guide the mutation and crossover of candidate hypotheses. Our results show that LLM-SR substantially outperforms state-of-the-art baselines. The second study examines limitations in current evaluations of LLM-driven discovery. We show that many benchmarks overestimate discovery capabilities because tasks are contaminated by training data. To address this, we introduce LLM-SRBench, a multi-domain benchmark designed with synthetic novel components to test models beyond memorization in the task of scientific model discovery. Results on LLM-SRBench show significant performance drop across existing methods, highlighting the importance of rigorous evaluation protocols for discovery. The third study investigates the role of adaptation. While humans continuously learn and adjust when facing unfamiliar environments, most existing LLM-based systems rely primarily on their pretrained knowledge during the search process. Motivated by recent advances in test-time training and reinforcement learning, we introduce DecAEvolve, a framework that enables models to adapt dynamically during evolutionary search with feedback obtained from the environment. We show that DecAEvolve substantially improves performance on out-of-distribution settings, establishing adaptation as a core requirement for discovery. Finally, the last study examines the role of exploration and diversity. We find that current LLM-based discovery systems often converge to narrow regions of the hypothesis space, limiting creativity and hindering stronger solutions in open-ended tasks. To address this, we introduce EvoDiverse, a framework that promotes diversity during evolutionary search. Across multiple scientific discovery tasks, EvoDiverse enables broader exploration and uncovers more promising regions of the search space, highlighting the importance of systematic exploration in open-ended discovery. Taken together, this thesis suggests that LLMs can actually become effective engines of discovery when equipped with principled search, rigorous evaluation, continuous adaptation, and diversity-preserving exploration -- four properties that we believe together define the path towards open-ended reasoning and discovery with language models.
Design and Development of a Social Network Analysis-Based Visualization Tool to Facilitate Asynchronous Online Discussions in Large-enrollment Classes
Zhang, Jianqiang (Virginia Tech, 2026-06-16)
Interaction is fundamental to meaningful learning, but it can be especially challenging to support and monitor in large-enrollment courses. Asynchronous online discussions (AODs) can facilitate instructor-to-student and student-to-student interaction, and patterns of interaction can inform decisions about facilitation strategies. However, revealing interaction dynamics in large-scale AODs remains challenging. Social network analysis (SNA) offers a useful approach for visualizing these interaction patterns. This study used design and development research (DDR) to design, develop, and evaluate the Canvas Discussion Analysis Tool (CDAT), an analytic tool intended to help instructors in large-enrollment courses identify and visualize interaction patterns among AOD participants. Evaluation results from expert review and usability testing suggested that CDAT was easy to use and particularly effective in identifying interaction patterns such as star, interconnected web, and cyclic patterns among AOD participants. The findings also identified areas for further refinement, including participant-profile identification, content-level analysis, accessibility, and broader transferability. Overall, the study contributes a literature-informed and formally evaluated prototype for supporting instructor-facing analysis of large-enrollment AODs.
Examining Lord's Paradox from Causal Inference Perspective: A Simulation Study
Zhu, Xiao (Virginia Tech, 2026-06-16)
In two-wave nonequivalent control group designs, Change Score Analysis (CSA) and Analysis of Covariance (ANCOVA) are commonly used to estimate treatment effects. However, these analytical models can yield conflicting results in both direction and magnitude – a phenomenon known as Lord's Paradox – raising concerns regarding analytical method selection in non-experimental research. This dissertation examines Lord's Paradox from a causal inference perspective to clarify the conditions under which these discrepancies arise. Specifically, the study aims to identify the conditions under which CSA and ANCOVA yield unbiased estimates, to determine when their results converge or diverge, and to assess how measurement error in the pretest affects their performance. To address these objectives, a general data-generating model is employed that incorporates unobserved confounding variable. Analytical derivations are conducted to express the CSA and ANCOVA estimators as functions of structural parameters, enabling a precise characterization of bias. These results are complemented by simulation studies that evaluate estimator performance in several metrics. The findings indicate that CSA yields unbiased estimates only under restrictive conditions, particularly in the absence of dynamic selection or symmetric confounding (i.e., when the effects of confounding variable on the pretest and the posttest are equal). In contrast, ANCOVA achieves unbiasedness under broader conditions, although it may also produce biased estimates when its assumptions are violated. Divergence between CSA and ANCOVA results is shown to be common in the presence of dynamic selection and asymmetric confounding, and may involve not only differences in magnitude but also reversals in the direction of the estimated effect. The analysis further demonstrates that measurement error in the pretest substantially affects estimator performance, particularly for ANCOVA, as declining reliability leads to increased bias and reduced inferential accuracy.
Virginia Principals and School Law
McClure, Amy Newcomb (Virginia Tech, 2026-06-15)
This quantitative study examined the level of school law knowledge possessed by secondary principals and assistant principals in Virginia public schools, including its relationships to the type, length, quantity, and recency of their law-related professional preparation and training, and years of administrative experience. The following questions guided this study: 1) What is the relationship, if any, between the principals' actual knowledge of school law and their self-reported confidence in their understanding of school law? 2)What is the relationship, if any, between the principals' knowledge of school law and the kind of school law preparation received? (College course, school system workshop, non-school system seminar, other)? 3) What is the relationship, if any, between the principals' knowledge of school law in relation to the length/quantity of preparation received? (semester, quarter, three weeks minimum, one day or less, other)? 4) What is the relationship, if any, between the principals' knowledge of school law and the recency of school law preparation? 5)What is the relationship, if any, between the principals' knowledge of school law and the years of administrative experience held? Findings that came out of this research include: 1) Virginia secondary principals and assistant principals demonstrated moderately high overall school-law knowledge. 2) Virginia high and middle school secondary principals and assistant principals demonstrated major overconfidence on two legal topics. 3) Virginia secondary principals and assistant principals reported relatively high confidence in their school-law knowledge, yet this confidence showed little relationship to actual performance. 4) Virginia secondary principals and assistant principals demonstrated good alignment on the specific topic of warrantless searches of cell phones, with low knowledge scores corresponding to appropriately low confidence levels. 5) The type of school-law preparation received was not significantly related to knowledge scores. 6) Due to extreme skew in the data, the relationship between the length or quantity of school-law preparation and knowledge scores could not be meaningfully tested. 7) Neither the recency of school-law preparation nor years of administrative experience was significantly related to knowledge scores.
Managed Internalization: Global Refugee Governance Norms and Local Script Ecologies at Ethiopia's Frontiers
Mekibib, Daniel Assefa (Virginia Tech, 2026-06-15)
Ethiopia engages with the international refugee governance regime by adopting emerging norms into the country's laws. However, the same national legal framework produces varying governance outcomes across refugee population groups, ranging from conditional inclusion to spatial quarantine. Drawing on Amharic-language parliamentary records, secondary sources, and interpretive constructivism and process tracing, this dissertation investigates why. The study builds on existing scholarship that conceptualizes refugee governance as a performative act of statehood while introducing two analytical concepts. The first is managed internalization, which is observed through the national law-making process. It illustrates how states exploit structural gaps in global governance regimes to adopt international norms while preserving discretionary space for sovereign action. The second is script ecology, which involves identity narratives influenced by bilateral relationships, co-ethnicity, and domestic and regional political structures, shaping how those norms are applied at each frontier. The frontier is conceived of as a site of ongoing contested statecraft, not merely a refugee-hosting geographic location. The state's performative act targets the international community, the domestic legislature, and those on the frontiers. These elements produce a governance architecture that explains the pledge-practice gaps and variation of outcomes across refugee populations. This research contributes to three scholarly areas: pledge-practice studies in refugee research, norm diffusion in international relations, and state-making at the margins in African and Global South scholarship.
Breadth of Protection: Evaluating Long‑Term Immunogenicity, Cross‑Protection, and Dissemination Strategies of the ARPV/ZIKV Vaccine
Tanelus, Manette (Virginia Tech, 2026-06-15)
Zika virus (ZIKV) is a mosquito-borne Orthoflavivirus that is distributed throughout the Americas, Africa, and Asia. Zika virus has also been known to co-circulate with other Orthoflaviviruses such as West Nile, Dengue, Japanese encephalitis virus, and yellow fever virus. Our central goal is to produce a robust immunogenic vaccine against ZIKV. ARPV/ZIKV (Aripo-Zika) is a chimeric virus containing the prM and E proteins of ZIKV and the nonstructural proteins of Aripo virus (ARPV). Herein, we explore ARPV/ZIKV cross-protection against other Orthoflaviviruses, durability of protection and best dissemination strategies. Immune competent and immune compromised mice were vaccinated with ARPV/ZIKV and challenged with SPOV. Our data indicates SPOV is cross-protective against SPOV in both mouse models, but does not completely protect against all symptoms of disease. We also assessed the durability of protection afforded by ARPV/ZIKV in immune competent mice over ten months and assessed the best dissemination strategies for delivering the vaccine. Our results indicate ARPV/ZIKV completely protects against a ZIKV challenge ten months post vaccination. Also, ARPV/ZIKV was not compatible with Sf9 cells, which have been used for propagation of other vaccines; and the strategies used to lyophilize ARPV/ZIKV did not produce a uniform stable powder product. However, we did determine that ARPV/ZIKV can be concentrated and stored long-term for vaccine research and transport. Overall, our data indicate the chimeric vaccine platform continues to be a promising avenue for developing a vaccine for ZIKV and other Orthoflaviviruses.
Modeling the Role of Water in Protein Structure and Function
Mondal, Ronnie (Virginia Tech, 2026-06-15)
Water is a solvent with high static dielectric constant, substantial dipole moment, significant electronic polarizability, and a dynamic hydrogen-bonding network. As such, water plays a pivotal role in protein structure and function, the molecular mechanisms of which remain challenging to determine. This dissertation develops and applies computational frameworks to rigorously integrate water effects on protein structure and function. We first examine how stronger protein–water interactions in the classical AMOEBA force field produce more realistic deformation behavior in collagen mimetic peptides (CMPs). In particular, our simulations of (PPG)$_n$ CMPs ($n=5,12,25$) under physiological conditions capture the experimental observation that shorter CMPs deform more strongly than longer ones. To better characterize these CMPs, we developed textsc{HeliXplore}, an open-source Python package for analyzing single- and multi-strand deformations. We further report that the length-dependent deformation we observe is associated with anisotropic translational and rotational relaxation dynamics of surrounding water molecules. Overall, we show that AMOEBA offers a substantial improvement over traditional force fields that often underestimate protein–water interactions. We next investigate the diffusion of solvated sodium ions through voltage-gated sodium channels (Na$_{mathrm{v}}$s). To do so, we develop a Continuous-Time Random Walk (CTRW) model that describes microscopic diffusion as a function of spatial and temporal disorder at the molecular scale. In addition, we establish a framework that incorporates this molecular disorder using information from polarizable MD simulations, enabling a consistent bottom-up approach. Using the CTRW model, we show that increased temporal disorder of the ion can accelerate diffusion in disordered media. More broadly, our results suggest that structural dynamics may help explain how Na${_mathrm{v}}$s achieve both high selectivity, through strong sodium binding, and rapid ion diffusion across the membrane.
Statistical Inference for Image Classification, Language Reasoning, and Counterfactual Estimation
Ye, Youhui (Virginia Tech, 2026-06-15)
This dissertation addresses three interconnected challenges in statistical inference and machine learning: uncertainty quantification under distribution shift, robust causal estimation in panel data, and reliable confidence estimation for large language models. The first contribution introduces DPI-RG, a distribution-free framework for constructing predictive sets for high-dimensional data under distribution shifts. Using a conditional Wasserstein round-trip generative model, DPI-RG maps data to a lower-dimensional latent space where test statistics asymptotically follow a Chi-square distribution, yielding theoretically guaranteed p-values. The framework exhibits strong robustness to outliers and target shifts without requiring outlier labels during training. The second contribution presents CONCORD, a confidence estimation framework for large language models that represents responses as semantic distributions over constituent clauses rather than single embedding vectors. By combining Sentence Mover's Distance with Symbolic Answer Match, CONCORD captures both structural divergence and discrete answer agreement, producing calibrated confidence scores that substantially outperform logit-based and embedding-based baselines on mathematical reasoning benchmarks. The third contribution proposes an $L_{infty}$-regularized Synthetic Control method that bridges the sparse weighting of traditional synthetic control and the denser, more robust weighting of Difference-in-Differences. By distributing weights more evenly across control units while remaining data-driven, the method improves robustness and interpretability while increasing the likelihood of satisfying the parallel trends assumption.
"Everybody's Got Their Stuff They Bring as a Sexual Partner, Mine is My Disability": Navigating the Intersections of Disability and Sexuality
Toman, Madelyn Mae (Virginia Tech, 2026-06-15)
Although sexuality development is a universal process, sexual rights and relationships of individuals with disabilities have historically been denied and oppressed due to ableist policies, societal norms, and misrepresentation of what living with disabilities can mean. People with disabilities are just as likely to have positive and enjoyable intimate relationships as people without disabilities. Positive sexual relationships and experiences contribute to increased sexual well-being which increases quality of relationships and life. Using the Crip Theory of Sexuality and the Skill-based Model for Sexuality Development, this study investigated the ways that individuals with different disabilities experienced the intersections of disability and sexuality development and the supports and barriers to disability sexuality development and well-being. I used reflexive thematic analysis to analyze 18 semi-structured interviews with individuals with different disabilities (age range = 23 - 52; 44% non-binary, 39% women, 22% men; 72% White, 22% Jewish, 11% Mexican/Hispanic), and generated six themes: learning about the intersection of disability and sexuality through experience, diverse impacts of disabilities ranging from barrier to superpower, creating a "toolkit" of diagnoses and identities, power of the partner (dynamics), encroaching ableism and internalized ableism, and "it's a long, long process" – navigating time and context. Individuals with disabilities are navigating the ways their disabilities influence their sexualities through experiences and interactions with partners. As they navigate their sexuality across their lives, they are also experiencing supports and barriers that impact their sexuality development and sexual well-being. Identifying barriers and disrupting ableist influences are essential for healthy sexuality development for people with disabilities.
Environmental drivers of greenhouse gas dynamics in temperate and tropical wetlands
Lopez Lloreda, Carla De Lourdes (Virginia Tech, 2026-06-12)
Wetlands play important roles in carbon cycling, contributing to both carbon storage and carbon emissions. Surface waters in wetlands are particularly important sources of methane (CH4) and can also function as net emitters of carbon dioxide (CO2) due to unique hydrologic and biogeochemical conditions in wetlands. Wetland carbon cycling can have high variability compared to other freshwater ecosystems because of the many underlying processes influencing the production, consumption, and transport of carbon. However, wetlands are frequently overlooked in studies of aquatic ecosystems due to their complexity. In addition, some wetland ecosystems have been difficult to incorporate into global wetland studies such as small, headwater wetlands and tropical wetlands. Due to unique local conditions, such as dynamic hydrology and warmer temperatures, headwater and tropical wetland ecosystems could potentially have larger contributions to carbon emissions than other wetland ecosystems. I used a combination of synoptic sampling, high-frequency sensors, and experimental manipulations to address questions about the spatiotemporal variability and environmental drivers of CO2 and CH4 in understudied wetland ecosystems. In small, headwater wetlands in the Delmarva Peninsula, I sampled 20 wetlands, which varied in their geomorphic features (size, perimeter:area, etc.) and characterized the spatial and temporal variability of CO2 and CH4 over the course of two years. I also used paired oxygen and CO2 high-frequency sensor data and estimated rates of aerobic metabolism to further identify dominant biogeochemical processes influencing carbon cycling in these wetlands during a one-year period. In freshwater coastal wetlands in Puerto Rico, I sampled seven sites over four years and performed an experimental manipulation using seawater to test the influence of salinization on CO2 and CH4 production. Wetlands across both Delmarva and Puerto Rico study sites were supersaturated in CO2 and CH4 with respect to the atmosphere, highlighting the role of wetland surface waters as likely sources of CO2 and CH4 to the atmosphere. Wetlands in Puerto Rico had particularly high CH4 concentrations, reaching up to 93μM. In both regions, spatial and temporal variability in CO2 and CH4 concentrations was high, particularly for CH4. In Delmarva, both gases were more spatially variable than temporally while in Puerto Rico, CO2 was more variable temporally and CH4 was more variable spatially. At the landscape scale, smaller wetlands with higher perimeter:area ratios in Delmarva had higher CO2 and CH4 concentrations. Larger wetlands also had higher rates of aerobic metabolism, while the smaller wetlands showed more influences of groundwater and anaerobic respiration. Comparing wetlands with other freshwater ecosystems globally, CO2 in Delmarva wetlands was more variable than in lakes and rivers. In Puerto Rico, CH4 varied across regions but CO2 did not vary spatially. In addition, increasing salinization decreased CH4 but increased CO2, with differences in the magnitude of response across wetland sites. This work highlights how geomorphologic characteristics such as size, govern patterns of CO2 and CH4 and their dominant processes in small, headwater wetlands while other local factors such as underlying geology could influence the spatial variability of CH4 in tropical coastal wetlands. Temporally, we find that CO2 increases with temperature and CH4 increases with decreasing oxygen in both regions. Ultimately, understanding the variability of CO2 and CH4 in wetlands, and how these might change with changing environmental conditions and across different wetland types, will continue to be critical to understanding the current and future role of wetlands in the global carbon cycle.
Programmatic Approaches to Preparing Mainstream Educators for Multilingual Learners: A Study of Teacher Preparation Through a Linguistically Responsive Teaching Framework
Skeen, Sarah Michele (Virginia Tech, 2026-06-12)
Multilingual learners (MLs) represent the fastest-growing student populations in the United States, yet many educators report feeling unprepared to meet their linguistic and instructional needs. The Linguistically Responsive Teaching (LRT) framework (Lucas and Villegas, 2011) outlines the dispositions, knowledge, and skills teachers need to support both content learning and language development in general education classrooms. While teacher preparation is frequently examined as a site for developing these competencies, relatively few studies have explored how ML-related content is structured across entire educator preparation programs (EPPs) particularly across multiple institutions. This qualitative study investigates how EPPs within a single state interpret limited policy requirements related to ML preparation and structure that content across required coursework in undergraduate elementary programs. Data were collected through interviews with elementary education program leaders and faculty, focusing on how LRT components are incorporated, how ML-related content is distributed throughout coursework, and what factors influence programmatic decision-making. Findings indicate that ML preparation is present across programs but varies widely in emphasis. This variation is shaped the interaction of policy expectations, institutional constraints, faculty expertise, and local program contexts.
Essays on Decision Making under Risk, Betrayal, and Manipulation
Anthwal, Pervesh (Virginia Tech, 2026-06-12)
This doctoral thesis consists of three chapters that study decision-making under risk, betrayal, and information manipulation. Chapter 2 investigates how preferences for income redistribution respond to differences in expected income and income volatility. Using a theoretical framework and a controlled experiment, the chapter shows that individuals with lower expected incomes choose higher redistribution, while those with higher expected incomes prefer lower redistribution. Exposure to greater income volatility increases support for redistribution, consistent with a self-insurance motive, whereas social preferences play a more limited role in shaping behavior under risk. Chapter 3 examines how betrayal aversion influences voting behavior. Using a vignette experiment, we show that voters are less likely to support otherwise preferred candidates when there is a possibility of betrayal. The results further indicate no significant differences across types of betrayal, implying that participants respond similarly regardless of the nature of the betrayal. Chapter 4 studies a model of information design in the presence of a strategic manipulator. An information designer selects a signal structure, while an intermediary can manipulate message probabilities, and receivers are naive to such manipulation. The analysis characterizes equilibrium information structures under varying degrees of alignment between the designer and the manipulator. A benevolent designer provides fully informative signals, whereas a compromised designer optimally reduces informativeness, with the extent of distortion depending on manipulation costs and degree of compromise.

Browse

Recent Submissions