Virginia Tech - Wake Forest University School of Biomedical Engineering & Sciences
VTech Works (Virginia Tech)Not a member yet
122049 research outputs found
Sort by
Graph-Based Computational Approaches for Modeling Viral Evolution
No full textModeling viral evolution is essential for understanding how pathogens adapt, spread, and generate new variants of concern. Yet, it remains challenging due to high mutation rates, minimal sequence divergence, and the scale of modern genomic data. Most phylogenetic trees enforce a strictly bifurcating structure that struggles to represent recurrent mutations, recombination, convergent evolution, and intra-host diversity. In contrast, quasispecies theory describes viral populations as clouds of closely related mutants evolving within a high-dimensional sequence space, where evolutionary relationships are more naturally captured by graphs than trees.
In this dissertation, I develop a sequence of graph-centered frameworks that integrate viral fitness, mutational distance, and mutational dynamics to model viral evolution from algorithmic and data-driven perspectives. First, ViraFit introduces a proof-of-concept model that couples epidemiological spread on contact networks with evolutionary dynamics on fitness landscapes, demonstrating how mutation, selection, and network structure jointly shape adaptive trajectories. Second, the Variant Evolution Graph (VEG) provides a scalable graph-based representation of SARS-CoV-2 evolution derived from mutational distances, allowing multiple ancestral relationships and capturing virus-specific evolutionary patterns that are difficult to represent with phylogenetic trees. A derived Disease Transmission Network further supports inference of likely transmission pathways and superspreaders. Finally, the Ancestor-Joining algorithm extends this representation into a predictive framework, Mutation Learning Graph (MLG), by inferring intermediate ancestral variants and enabling graph neural network–based lineage classification and mutational link prediction across geographically diverse SARS-CoV-2 cohorts.
Together, ViraFit, VEG, and MLG form a unified methodological progression that links mechanistic modeling, evolutionary reconstruction, and predictive graph learning, providing a scalable, mutation-centric view of viral evolution that complements traditional phylogenetic approaches and supports future variant forecasting.Doctor of PhilosophyViruses such as SARS-CoV-2 evolve rapidly, generating many closely related variants as they spread through a population. These small genetic changes can influence how easily a virus spreads, how severe the disease becomes, and whether existing vaccines or treatments remain effective. Because of this, understanding how viruses change over time is essential for public health and pandemic preparedness.
This dissertation develops new computational approaches to study viral evolution by using graphs, where each viral genome is represented as a node and edges show how one strain may have changed into another. Graphs offer a flexible way to capture the many possible paths a virus may take as it mutates, including patterns that traditional phylogenetic methods often miss.
The first part of this work introduces ViraFit, a simulation framework that models how viruses mutate and how disease is transmitted simultaneously. It shows how the structure of human contact networks and the fitness of different viral strains work together to shape which variants become dominant.
The second part presents the Variant Evolution Graph (VEG), a new method for organizing real viral genome data. VEG makes it easier to detect important virus-specific evolutionary events, such as repeated mutations, recombination, and diversity within infected individuals, that may not be clearly evident in standard evolutionary trees.
The final part of the dissertation expands these ideas using machine learning. The Mutation Learning Graph (MLG) combines graph representations with advanced neural networks to learn patterns in viral evolution, predict relationships that have not yet been observed, and help anticipate how future variants might emerge. Together, these methods provide a more detailed and flexible picture of viral evolution, offering tools to support genomic surveillance, early detection of new variants, and improved understanding of viral evolution. Although developed using SARS-CoV-2 data, the approaches are general and can be applied to many other rapidly evolving viruses and biological systems
CLAW: A Vision-Language-Action Framework for Weight-Aware Robotic Grasping
No full textVision-language-action (VLA) models have recently emerged as a promising paradigm for robotic control, enabling end-to-end policies that ground natural language instructions into visuomotor actions. However, current VLAs often struggle to satisfy precise task constraints, such as stopping based on numeric thresholds, since their observationto- action mappings are implicitly shaped by training data and lack explicit mechanisms for condition monitoring. In this work, we propose CLAW (CLIP-Language-Action for Weight), a framework that decouples condition evaluation from action generation. CLAW leverages a fine-tuned CLIP model as a lightweight prompt generator, which continuously monitors the digital readout of a scale and produces discrete directives based on task-specific weight thresholds. These prompts are then consumed by π0, a flow-based VLA policy, which integrates the prompts with multi-view camera observations to produce continuous robot actions. This design enables CLAW to combine symbolic weight reasoning with high-frequency visuomotor control. We validate CLAW on three experimental setups: single-object grasping and mixed-object tasks requiring dual-arm manipulation. Across all conditions, CLAW reliably executes weight-aware behaviors and outperforms both raw- π0 and fine-tuned π0 models. We have uploaded the videos as supplementary materials.Submitted versio
Decision-Making during Elementary Technological/Engineering Design-Based Learning
No full textThis research examines the decision-making subskills and informed decision-making strategies employed by third-grade STEM students during a technological and engineering (T/E) design challenge implemented through the Technological/Engineering Design-Based Learning (T/E DBL) pedagogical framework. Using a quasi-mixed method embedded multiple case study design, this study analyzed student responses to structured 2-1-1 prompts and unstructured Interactive Engineering Journals (IEJs) that documented their design process. Post-intervention, a stratified sample of 18 students—six from each of three third-grade classes—was selected for IEJ analysis, determined by the Spring 2025 Virginia Measures of Academic Progress Growth Reading Assessment (MAP) scores.
This study aimed to enhance our understanding of the decision-making skills and strategies utilized by third graders during T/E design challenges. Findings revealed that third grade students demonstrated a wide range of decision-making subskills and informed decision making strategies. These results underscored the role of T/E DBL in fostering decision-making skills, providing valuable insight into how young learners develop and apply decision-making in STEM contexts.Doctor of PhilosophyThis study explored how third-grade students used decision-making subskills and strategies while completing a technological and engineering (T/E) design challenge. The project used a teaching approach called design-based learning, where students identify a problem, plan possible solutions, test their ideas, and revise their work. To understand how students made decisions throughout this process, the study examined their Interactive Engineering Journals (IEJs), in which they recorded sketches, ideas, and explanations of their thinking and responded to researcher-provided prompts.
After the T/E Design Challenge, 18 student IEJs were selected for detailed analysis—6 from each of 3 third-grade classrooms. These students were chosen based on their spring 2025 reading assessment scores so the study could capture how different learners approach decision-making during a design task.
The results showed that third graders used a wide variety of decision-making subskills and strategies during the design challenge. Students asked questions, described the problem, considered different design options, gathered information, sketched ideas, and explained how and why they made certain choices during the design process. They also demonstrated decision-making strategies such as reflective thinking, collaboration, testing, and revising their designs, as well as using evidence to support their decisions.
The study found that the T/E design challenge naturally encourages students to engage in decision-making. These findings highlight how T/E design challenges can help young learners build the problem-solving and reasoning skills they will need throughout their education
Microbial and Immune Signatures of Stress and Antidepressants Across Sex and Generations
No full textThe gut, brain, and immune system form an integrated and dynamic network that shapes development and health outcomes across the lifespan. Increasing evidence indicates that this axis is highly sensitive to stress exposure, displays robust sex-dependent properties, and may transmit biological effects across generations. This dissertation investigated how variable stress or perinatal antidepressant exposure alters gut microbial composition, immune signaling, and behavioral phenotypes in a sex-specific and multigenerational manner. To address these questions, we utilized 16S rRNA or whole-genome shotgun metagenomic sequencing in tandem with behavioral assays and multiplex cytokine profiling in rat and mouse models of chronic stress or perinatal SSRI exposure. Across studies, we demonstrate that the gut microbiome exhibits innate sex differences that are present early in life and continue to diversify with age. Aging was associated with increased alpha diversity and marked shifts in microbial metabolic pathway representation as well as a reversal of the Firmicutes–Bacteroidota ratio. Chronic stress induced distinct microbial alterations in males and females, with males showing a greater magnitude and diversity of microbial alterations. We further observed that female offspring more closely mirrored the microbial community structure of their dams, and that perinatal citalopram treatment produced stronger and more persistent effects in female offspring than in males. These findings reveal that stress and antidepressant exposure interact with sex and developmental stage to shape gut microbial composition and immune function. Microbial alterations following stress appear context-dependent, serving either compensatory or maladaptive roles depending on biological state and environmental challenge. Collectively, this work advances our understanding of how the gut–brain–immune axis integrates life experience across sex and generations and highlights the microbiome as a dynamic mediator and potential therapeutic target in stress-related neuropsychiatric vulnerability.Doctor of PhilosophyThis dissertation explores how factors such as stress or antidepressant use can influence the community of microbes living in the gut, and how these effects differ between males and females and can even extend across generations. The gut plays a key role in digestion, the immune system, and communication with the brain. Because of this, changes to gut microbes may help explain why stress affects people differently and why some individuals are more vulnerable to anxiety, depression, or other stress-related health concerns. In this research, stress and antidepressant exposure were modeled in rodents to understand how these experiences shape the gut and immune system over time. We found that males and females naturally have different patterns of gut microbes, and these differences continue to change throughout life. Stress changed the gut in both sexes, but males showed larger and more widespread changes than females. We also observed that female offspring tended to inherit gut microbe patterns more similar to their mothers, suggesting that some stress-related biological effects may be passed from one generation to the next. When pregnant animals were treated with a commonly prescribed antidepressant, their female offspring showed stronger changes than males, indicating that early-life medication exposure may shape development in sex-specific ways. Overall, this work shows that the gut microbiome is deeply connected to how the body responds to stress and medication, and that these responses are influenced by biological sex and family history
International Journal of Hospitality Management
No full textThis study introduces the concept of Social Dynamic Capabilities to address the growing need for hospitality organizations to align business strategies with social sustainability goals. Employing a grounded theory approach, we develop propositions and a conceptual framework delineating social dynamic capabilities from the broader concept of dynamic capabilities. Our research was conducted in two phases: an exploratory phase to identify core skills and practices for social dynamic capabilities and a confirmatory phase to refine the framework. The findings underscore the importance of stakeholder collaboration and community involvement in the operationalization of social dynamic capabilities, leading to the concept of Collaborative Social Transformation. Our study advances the dynamic capabilities framework by extending it into the social domain and providing a practical framework for implementation. This study provides a new perspective on integrating social sustainability issues into long-term organizational practices.Published versio
Modeling and Stability Analysis for Multiphase Constant On-Time Control With Phase Overlapping
No full textMultiphase buck voltage regulators (VRs) powering today's high-performance processor loads must regulate output voltage within a small tolerance window and handle a large steady-state current. For output voltage regulation, multiphase current-mode and V² constant on-time (COT) control techniques are widely used because these control methods can help increase control bandwidth, minimize output impedance, and achieve fast transient response with less output capacitance. To supply large current demands, the phase count of buck VRs continues to increase. With increasing phase count, steady-state phase overlapping becomes inevitable in the practical duty range. However, currently, there is no good small-signal model to analyze stability and design high-bandwidth control loops for multiphase COT buck converters operating under phase-overlapping conditions. To address this issue, this dissertation presents a systematic small-signal modeling approach for multiphase COT buck converters operating with phase overlap, utilizing the describing function method.
Prior works extended the small-signal models of single-phase COT control to design multiphase COT buck converters. These extensions rely on the assumption that multiphase and single-phase COT buck converters are equivalent from a small-signal perspective. This equivalence applies when there is no phase overlapping or when the duty cycle D 1/N (with phase overlap). In multiphase COT buck converters, total current ripple is used to modulate each phase duty cycle and achieve automatic phase interleaving. When the duty cycle D 1/N, then the total current ripple will exhibit variable on-time and off-time modulation after introducing a perturbation. Due to this difference in modulation principle, single-phase COT control models cannot be used for multiphase COT control design when the duty cycle D > 1/N.
In this dissertation, we first developed a general describing function model for multiphase current-mode COT buck converters. As for single-phase current-mode control, the total current feedback in these converters makes the whole system dynamically non-linear. Hence, we treated the entire current loop and power stage as one entity. Unlike prior works, we derived a continuous-time frequency-domain model for multiphase current-mode COT control by performing a Fourier analysis directly on its perturbed waveform. In this way, we account for the change in total current modulation principle between the no-phase and phase-overlapping regions. The proposed model is quite general and applies to arbitrary phase numbers and phase overlapping numbers. We have also extended this model to multiphase current-mode COT buck converters with coupled inductors.
Next, the proposed model for multiphase current-mode COT control was used to study the total current loop stability issue under phase overlapping conditions. Also, the minimum external ramp required for stability with different number of phase overlapping is derived. With two overlapping phases, the critical external ramp slope equals half of the sensed total current rising slope. However, the critical ramp slope with three overlapping phases equals the sensed total current rising slope and increases further with an increase in the number of overlapping phases.
After that, we developed a general small-signal model for multiphase V² COT control, including the capacitance voltage ripple and phase overlapping effects. Using this model, it is demonstrated that direct output voltage feedback is inherently unstable in phase overlapping regions. In other words, unlike single-phase V² COT buck converters, even large ESR OSCON capacitors cannot ensure stability with phase overlapping. Hence, external ramp compensation is required to ensure stability when phase overlapping occurs. However, an external ramp can ensure stability only if the output voltage ripple has sufficient current strength. Hence, a physical current loop is required to ensure stability with low ESR ceramic capacitors. The critical limits derived were verified using SIMPLIS simulation and experimental results.
In sum, this dissertation provided a mathematical framework to derive small-signal models for multiphase COT control methods based on phase manager under phase overlapping conditions. Small-signal models were derived for the popular multiphase current-mode and V² COT control under phase overlapping conditions. Using this model, we identified that total current loop of multiphase current-mode COT control becomes unstable without sufficient external ramp under phase overlapping conditions. Also, the stability boundaries with different number of phase overlapping are provided. The method proposed here can also be extended to other variations of multiphase COT control using phase manager.Doctor of PhilosophyToday, Artificial Intelligence (AI) is being increasingly adopted to enhance the computing performance and reasoning capabilities of large-scale data center servers, as well as handheld smartphones. These systems run on high-speed integrated circuits called "microprocessors," which require sub-1V input voltage and consume thousands of amperes of current. Also, these processors run on GHz clock frequencies and utilize multiple small processing units (called 'cores') in parallel to improve data processing speed. When these AI processors switch tasks from one core to another, the current consumed by processors rapidly increases or decreases. During this current transient, the power supply for these processors must ensure that the processor core voltage stays within a small tolerance window to prevent processor damage. With new advancements in AI processors, this tolerance window is becoming smaller and smaller. However, the current transients introduced by processors are becoming larger and faster. These trends introduce new challenges in designing power supply for these microprocessors.
To power these multi-core AI processors, the power supply for these processors connects multiple small buck converters in parallel to supply a large current with good efficiency. Each buck converter is called a phase, and the power supply is collectively known as a multiphase buck converter or multiphase voltage regulator (VR). To regulate output voltage during the fast processor current transients, multiphase VRs widely use an advanced control method called Constant On-Time (COT) control to achieve a high-control bandwidth. This control method uses total output current ripple of multiphase buck converter as ramp signal to generate Pulse-Width Modulation (PWM) signals for each phase. Multiphase buck converters together with COT control form a non-linear system. Hence, a small-signal model is essential for performing system design using frequency-domain design tools familiar to practicing engineers, such as Bode and Nyquist plots.
In multiphase VRs, the PWM signals of two subsequent phases are delayed by Tsw/N, where Tsw is the switching period of each phase's PWM signal. At present, to handle increasing processor current, the multiphase VR phase count (N) continues to increase. Due to increasing phase count, the constant-on time (Ton) of each phase becomes larger than the delay between phases, i.e., Tsw/N, and PWM signals of two or more phases begin to overlap at steady-state. This phenomenon is known as phase overlapping. In the past, a lot of research has been directed towards developing linear models and providing design guidelines for single-phase COT buck converters. This is because single-phase linear models can be extended to design multiphase COT buck converters without phase overlapping.
This dissertation focuses on providing general small-signal models for multiphase COT buck converters. In the second chapter, we explained why single-phase COT control models cannot be extended to multiphase COT buck converter operating with phase overlapping. Then, we developed small-signal models for multiphase current-mode COT-controlled buck converters by using the describing function method as a tool. In the third chapter, we identified a potential instability issue in multiphase current-mode COT buck converters with phase overlapping. Also, the stability boundaries with different numbers of phase overlapping are provided. In the fourth and fifth chapters, we extend the describing function method to model multiphase V² COT buck converters and derive the stability boundary of this control with phase overlapping.
Nowadays, many commercial COT controllers that support multiphase VRs operating with phase overlap are designed using circuit simulation software without a theoretical basis. However, simulation-driven designs are iterative and could become very time-consuming if the phase count becomes too large in the near future. The proposed models in this dissertation will help in improving these controller's bandwidth and optimize transient performance with reduced design time
Electronics
No full textThe study presents a novel process to design lightweight, high-performance cooling manifolds for power electronics using generative design. The process begins with a baseline design that defines the constraints of the manifold with regard to the target cooling geometry and flow path. A flow optimization is then performed to optimize flow distribution and maximize convective efficiency. Once a final fluid volume is obtained, a structural optimization is conducted to minimize weight and material usage. The simulation results for the final design demonstrated a 40.1% increase in the average heat transfer coefficient, a 7.5% decrease in average chip temperature, a 76.6% improvement in temperature uniformity, and a 63.3% reduction in weight at the expense of a minimal 5.1% increase in pressure drop compared to the baseline design.Published versio
Analysis of Acceleration Techniques and Fast Nonlinear Solvers
No full textThis dissertation focuses on the analysis and development of acceleration techniques and fast solvers for nonlinear systems of equations. Building upon the fixed-point and extrapolation frameworks introduced in the early chapters, we explore structural connections between residual-based acceleration methods and Krylov subspace techniques.
The first main contribution is a unified algebraic framework establishing the equivalence between the Anderson Acceleration method and the CROP (Conjugate Residual with Optimal Trial Vector) algorithm. By formulating both methods within a common affine subspace representation, we show that their full, untruncated forms produce identical iterates, motivating new hybrid variants such as CROP-Anderson and real-residual CROP (rCROP) methods.
The second contribution is a perturbation analysis of Anderson-type variants, examining the effects of deterministic and stochastic errors on convergence. Numerical experiments confirm that acceleration efficiency depends critically on both the choice of update strategy and the nature of perturbations.
The third contribution extends this unified perspective to nonlinear Krylov subspace methods. Nonlinear extensions of GMRESR, GCRO, and LGMRES are derived, forming the nlKrylov family of algorithms, and analyzed in the context of inexact Newton solvers, with convergence results established under relaxed conditions on residual and Jacobian approximations.Doctor of PhilosophyThis dissertation studies mathematical methods that accelerate the solution of nonlinear equations, which arise in many areas of science and engineering. Traditional iterative solvers can be slow or may fail to converge on challenging problems. Acceleration techniques improve convergence by intelligently using information from previous iterations. This work focuses on developing and analyzing techniques that accelerate these computations, making them faster and more reliable.
The first major contribution shows a deep connection between two popular acceleration methods, demonstrating that, under ideal conditions, they produce identical results. This insight leads to new hybrid approaches that combine the strengths of both methods.
The second contribution studies how errors-either from approximations or random perturbations-affect the performance of these acceleration techniques. Numerical experiments show that careful design of the update strategies is essential for maintaining efficiency.
Finally, the dissertation extends these ideas to a broader family of advanced iterative methods, providing new algorithms and theoretical guarantees for solving large-scale nonlinear problems more effectively.
Overall, this dissertation provides a unified perspective on acceleration strategies and demonstrates how they can be applied to modern large-scale nonlinear computations
Bridging Security and Agility: A Comprehensive Approach to Integrating Security Practices in Agile Development through DAST, LLMs, and Automation
No full textEffectively integrating security practices within Agile software development is essential as software systems become complex and critical. While Agile methodologies are widely adopted for their responsiveness and efficiency, many security practices remain documentation-heavy and process-driven, creating friction with Agile's emphasis on frequent delivery, individuals, and interactions. This Ph.D. dissertation investigates the integration of security practices—particularly Dynamic Application Security Testing (DAST), used to identify critical real-time vulnerabilities in web applications—into Agile workflows, examining its perceived impact on development teams and processes. We first surveyed Agile practitioners to understand their perspectives on security integration, revealing both benefits and challenges in implementation. We then explored how Large Language Models (LLMs) could improve the comprehension of security testing outputs, demonstrating that LLM generated summaries enhance the accessibility and understanding of security alerts. Subsequently, an in-depth real-world case study of a Kanban-based Agile team integrating DAST into its Continuous Integration Continuous Development (CI/CD) pipelines uncovered practical obstacles—such as report complexity and workflow interruptions, alongside conditions that supported successful adoption, including increased automation and dedicated engineering support. Finally, the insights from these studies informed the development of SafeAIMerge, a CI/CD-based tool that integrates DAST scanning with LLM-generated summaries to deliver actionable, developer-friendly security feedback within pull requests (PRs). Practitioner evaluations indicate that the tool reduces cognitive and emotional workload during vulnerability remediation, enhances security report understanding, and supports software developers in more efficient resolution of security issues. Together, these studies form a cohesive body of evidence demonstrating how security practices such as DAST, when supported by automated workflows, LLMs, and guided by practitioner-centered design, can be effectively embedded into Agile development.Doctor of PhilosophyModern software is deeply embedded in everyday life, making security a critical concern. At the same time, many software teams rely on Agile development methods that emphasize speed and frequent updates. Traditional security practices are often difficult to fit into these fast-paced workflows, creating challenges for teams trying to remain both secure and efficient. This PhD dissertation focuses on how security practices can be effectively incorporated into Agile software development. We first examined how software developers perceive security, identifying common benefits as well as obstacles that make security practices difficult to adopt in Agile. Next, we investigated how LLM tools can help developers better understand security testing results, showing that LLM-generated explanations make security issues clearer and easier to address. We then studied a real-world Agile team integrating automated security testing—specifically Dynamic Application Security Testing (DAST), which checks running web software for security weaknesses—into their development process, uncovering both practical challenges and factors that supported successful adoption. Based on these findings, we developed SafeAIMerge, a DAST-based tool that provides clear and actionable security feedback directly within developers' existing workflows. Overall, this research demonstrates that developers perceive security can be effectively integrated into Agile development when it is automated, clearly explained, and designed around developers' needs, helping teams build secure software without slowing development
Violence Prevention Programming in Secondary Public Schools in Virginia
No full textAcross the United States violent behavior among youth, particularly adolescents, ages 10 to 19, is occurring at alarmingly high rates. Current literature and statistical data both concur that homicide is the leading cause of death for this age group. This phenomenon is not the result of a single factor or event, nor is it particular to any one subgroup or geographical location. It is the result of a multitude of socioecological factors that affect individuals and their behaviors. Schools provide an ideal venue for addressing these factors through preventative measures and interventions because of their ongoing relationship with adolescents. This cross-sectional interpretive collective case study answered three research questions through interviews with school and community leaders operating in urban areas within the state of Virginia by investigating how urban public secondary schools design and implement youth violence prevention programming, with a focus on how schools are collaborating with community organizations and local government. Data was collected through stakeholder interviews, analytical memos, local crime statistics, and research of current programming. All supplemental data was collected from publicly available resources.Doctor of EducationAmerican youth, particularly adolescents, ages 10 to 19, are participating in violent crimes at alarmingly high rates. Research shows that homicide is the leading cause of death for this age group. This phenomenon is not caused by a single factor or event, nor is it particular to any one subgroup or geographical location. It is the result of many factors that affect individuals causing them to act through violence. Schools are able to support at risk students through preventative measures and interventions because of their ongoing relationship with adolescents. This study explored this topic through interviews with school and community leaders operating in urban areas within the state of Virginia. The interviews were supplemented with analytical memos, local crime data, and research on current youth violence prevention programming to provide a comprehensive view of current youth violence prevention programming