Masters Theses

Permanent URI for this collection

Browse

Recent Submissions

Now showing 1 - 20 of 23008
  • Zero-Shot Scene Graph Relationship Prediction using VLMs
    Dutta, Amartya (Virginia Tech, 2025-03-24)
    Scene Graph Relationship Prediction aims to predict the interaction between the objects in an image. Despite the recent surge of interest in open-vocabulary and zero-shot SGG, most approaches still require some form of training or adaptation on the target dataset, even when using Vision-Language Models (VLMs). In this work, we propose a training-free framework for the VLMs to predict scene graph relationships. Our approach simply plugs VLMs into the pipeline without any fine-tuning, focusing on how to formulate relationship queries and aggregate predictions from the object pairs. To this end, we introduce two model-agnostic frameworks: SGRP-MC, a multiple-choice question answering (MCQA) approach, and SGRP-Open, an open-ended formulation. Evaluations on the PSG dataset reveal that well-scaled VLMs not only achieve competitive recall scores but also surpass most trained baselines by over 7% in mean recall, showcasing their strength in long-tail predicate predic- tion. Nonetheless, we identify several practical challenges: the large number of potential relationship candidates and the susceptibility of VLMs to choice ordering can affect con- sistency. Through our comparison of SGRP-MC and SGRP-Open, we highlight trade-offs in structured prediction performance between multiple-choice constraints and open-ended flexibility. Our findings establish that zero-shot scene graph relationship prediction is feasi- ble with a fully training-free VLM pipeline, laying the groundwork for leveraging large-scale foundation models for SGG without any additional fine-tuning.
  • Enhancing Layout Understanding via Human-in-the-Loop: A User Study on PDF-to-HTML Conversion for Long Documents
    Mao, Chenyu (Virginia Tech, 2025-03-24)
    Document layout understanding often utilizes object detection to locate and parse document elements, enabling systems that convert documents into searchable and editable formats to enhance accessibility and usability. Nevertheless, the recognition results often contain errors that require manual correction due to small training dataset size, limitations of models, and defects in training annotations. However, many of these problems can be addressed via human review to improve correctness. We first improved our system by combining the previous Electronic Thesis/Dissertation (ETD) parsing tool and AI-aided annotation tool, providing instant and accurate file output. Then we used our new pipeline to investigate the effectiveness and efficiency of manual correction strategies in improving object detection accuracy through user studies, including 8 participants, comprising a balanced number of four STEM and four non-STEM researchers, all with some background in ETDs. Each participant was assigned correction tasks on a set of ETDs from both STEM and non-STEM disciplines to ensure comprehensive evaluation across different document types. We collected quantitative metrics, such as completion times, accuracy rates, number of wrong labels, and feedback through our post-survey, to assess the usability and performance of the manual correction process and to examine their relationship with users' academic backgrounds. Results demonstrate that manual adjustment significantly enhanced the accuracy of document element identification and classification, with experienced participants achieving superior correction precision. Furthermore, usability feedback revealed a strong correlation between user satisfaction and system design, providing valuable insights for future system enhancement and development.
  • Aircraft Anti-Icing Analysis: Water Droplet Dynamics Under High-Frequency Atomization and Superhydrophobic Effects
    Thomas Fernandez, Kevin (Virginia Tech, 2025-03-21)
    Structural icing is a significant engineering challenge that has prompted extensive research into thermal and mechanical preventive measures. Common solutions involve the spraying of de-icing chemicals and high-power consumption heating systems for larger aircraft that add to the weight. Still, complexities arise from water droplets freezing at supercooled levels. A novel approach uses the structure's vibration to induce atomization, a proposed active anti-icing method using high-frequency Piezoelectric Transducer (PZT) vibration and combines it with the passive method of surface roughness variation by fabricating superhydrophobic surfaces. The study analyzes the droplet impact at 3 speeds. The impact is recorded with high-speed imaging using selected resonant frequencies (between 6 kHz and 25.6 kHz) to determine the optimal range for atomization. The study of the active method of atomization involved adjusting the frequency applied (as single and a sweep of frequencies) to the transducer material attached to an aluminum flat plate at a constant AC voltage supply, and variation of droplet velocity parameters. The best actuators are selected and determined through the analysis of frequency response and the magnitudes of the amplitude of vibration that are generated. The effect of single and sweep frequencies on the droplet dynamics is studied by analyzing 3 quantities: the Spread factor, the Volume ejected per ms (Vatomized), and the total energy (Eatomized) of the atomized droplets. The combination of the three helps to determine three key outcomes: The dynamics of the droplet, the change in dynamics due to vibrations, and the most effective atomization. It is observed that during atomization, Wenzel state (Hydrophilic) pining becomes more prevalent in the droplet as opposed to a non-vibrating static surface. Vibration also promotes spreading, meaning thinner droplet lamella (droplet height on the surface) and more surface area contact, thereby higher wetting. Furthermore, the more it spreads, the larger the volume of water is ejected. It was observed that the total energy (sum of Kinetic and potential energies) of ejected droplets have an inverse relation with the increase in Reynolds number. As the droplet speed increases in Re from ≈ 548 to ≈4797, the Eatomized reduces. Most notably, due to pinning, suggesting an increase in surface energy that promotes hydrophilic behavior and also the higher energy required to eject a droplet from a wider cross-section area (as the spreading increases with increase in Re). This research examines droplet interaction using parameters from both single-frequency and swept-frequency atomization, including the spread factor, Vatomized and Eatomized, to study droplet interaction. Here, swept frequencies exhibited less spatial dependency on droplet deposition while maintaining atomization rates, volumes, and energy levels comparable to those of single frequencies. Additionally, it explores the effects of combining atomization with a superhydrophobic surface, further improving the anti-icing characteristics. The study also establishes protocols for Abaqus FEA to simulate the frequency response of a PZT attached to a flat plate and outlines the design and construction of a supercooling chamber.
  • Gait Phase Estimation and Foot Trajectory Prediction During Dynamic Walking Using Gated Recurrent Units
    Wisman, Hayley (Virginia Tech, 2025-03-21)
    In the field of assistive robotics and exoskeletons, foot trajectory prediction has the potential to play a pivotal role in improving the functionality and user experience of worn devices. Rather than operating as a reactive system which only responds to user movement, a de vice which predicts future foot position can anticipate an action before it occurs, reducing latency and moving with the wearer for a more natural, uninhibited motion. While previous studies have focused on predicting continuous motion, they often overlook critical transitions between walking and standing, which are essential for natural locomotion. We propose in this study a foot trajectory prediction approach which leverages a recurrent deep learning architecture to make predictions based on sequential walking data. The first of the two ma chine learning models predicts the gait phase as a value between 0 and 1, while the second model leverages the gait phase prediction output to predict foot position in three dimensions. The models were trained and evaluated on IMU sensor data collected from three subjects instructed to walk on a treadmill at speeds varying from 0.5 mph to 1.5 mph. The result ing mean absolute error on gait phase percentage across subjects and velocity was 1.92%. For foot trajectory prediction, the cross-subject trained model achieved mean distance er ror of 2.85±2.89 cm, 3.29±2.82 cm, 4.15±4.12 cm, 5.33±5.46 cm, and 6.92±6.56 cm with prediction horizons of 0.1s, 0.25s, 0.5s, 1s, and 2s, respectively.
  • Virginia Logging Business Economic Sustainability Survey Including Perspectives from Across the Forest-based Supply Chain
    Barkman, Rebecca Ann (Virginia Tech, 2025-03-20)
    Logging businesses are an essential component of the forest-based supply chain. They are the connection between forest landowners, who grow the raw materials, and forest product mills that produce primary forest products. They are confronted with many operational challenges and issues that can make operating sustainably, producing a profit, and obtaining long-term economic viability seem unattainable. Although other businesses have similar operational challenges, logging businesses are somewhat unique in that they have minimal influence over delivered prices or the cost of stumpage, so changes in variable input costs can have large impacts on businesses' economic sustainability. Logging business operational challenges include increasing input costs such as equipment purchase costs, fuel costs, and equipment maintenance and repair costs. The economic sustainability of logging businesses affects the entire forest-based supply chain because one segment cannot function successfully without the others. This project evaluated operational characteristics, challenges, and issues related to the economic sustainability of logging businesses in Virginia from May through July 2023. A comparison was conducted using opinions and perspectives from professionals in other segments of the forest-based supply chain, on their outlook for the logging industry in Virginia concerning economic sustainability. Mail questionnaires, following the Dillman Method, were used to collect data from the survey populations which included logging business owners, consulting foresters (landowner representatives), and mill owners or procurement representatives. The response rates for logging businesses, mills, and consultant foresters were 27, 40, and 69 percent respectively. The top two challenges logging businesses faced in Virginia were fuel related. The number one challenge reported by logging businesses was fuel costs for in-woods harvesting equipment followed by fuel costs for trucks. Only 32.9% of logging businesses reported they were profitable in the past year. Only 26.1% of businesses had an outlook that their business was economically sustainable while 38.8% reported that their business was not sustainable. A greater percentage of mills (56.3%) and consultants (68.9%) reported their outlook for logging businesses was not economically sustainable. There were many neutral perspectives from all populations, however small positive changes in market conditions could move responses to the positive side of neutral. This study identifies the challenges in the industry as well as perspectives on the future of the forest industry's economic sustainability. The results of the study should be used as a catalyst encouraging segments of the industry to work together to address challenges and find solutions.
  • Role of PERM1 in the Development of Insulin Resistance and Diabetic Cardiomyopathy During High-Fat Diet Feeding
    James, Amina N.'Kechi (Virginia Tech, 2025-03-20)
    Heart failure is a leading cause of death in the United States, impacting approximately 6.7 million people. Several comorbidities are associated with heart failure, contributing to adverse clinical outcomes. Among these comorbidities, diabetes is highlighted as a prominent risk factor for heart failure, with approximately 20-40% of heart failure patients having type 2 diabetes. As the prevalence of heart failure continues to rise, there is a need for novel therapeutic methods to address this concern. PPARGC1 and ESRR Induced Regulator in Muscle 1 (PERM1) is a striated muscle-specific regulator of mitochondrial bioenergetics, predominantly expressed in skeletal and cardiac muscle. Our group has previously demonstrated that PERM1 is downregulated in both human and mouse failing hearts, and that Perm1-knockout mice exhibit reduced cardiac contractility and energy reserve. However, the role of PERM1 in cardiac dysfunction in diabetes remains unknown. We hypothesized that loss of PERM1 increases vulnerability to metabolic insults and exacerbates diet-induced insulin resistance and cardiac dysfunction. To test this, C57BL/6N male wild-type (WT) and Perm1-knockout (Perm1-KO) mice were fed either a normal diet or a high-fat diet (HFD; 60% calories from fat) for up to 43 weeks. We found that PERM1 expression was upregulated in the hearts of WT mice after 8 weeks of HFD feeding, coinciding with an increased level of carnitine palmitoyltransferase 2 (CPT2), a key enzyme involved in mitochondrial fatty acid uptake. Importantly, both WT and Perm1-KO mice exhibited similar increases in total body weight, fat mass, and fasting blood glucose levels throughout 43 weeks of HFD feeding, suggesting that loss of PERM1 did not accelerate the development of either obesity or diabetes. Echocardiographic assessments showed that WT mice maintained systolic and diastolic function, despite moderate cardiac remodeling, manifested as a subtle but significant increase of left ventricle posterior (LVPW) wall thickness. Unexpectedly, 8 weeks HFD feeding partially restored systolic function in Perm1-KO mice with no change in LVPW thickening. These findings show that while HFD feeding induced obesity and insulin resistance, its effect on cardiac function was relatively moderate and neither was exacerbated by the loss of PERM1. Unexpectedly, this study suggests that HFD feeding in Perm1-KO mice could partially compensate for cardiac dysfunction.
  • Analysis of Plane Strain Deformations of Linearly Elastic Strain-Gradient Materials by the Finite Element Method
    Dahiya, Akshay (Virginia Tech, 2025-02-27)
    At small scales, numerous experimental studies have shown that material behavior strongly depends upon the specimen size. Classical theories are unable to explain this size dependence, whereas a strain gradient continuum theory has intrinsic length scales and may well describe mechanical deformations of small size bodies. In the current contribution, we develop a numerical software based on the finite element method (FEM) to analyze infinitesimal deformations of strain-gradient dependent materials by introducing auxiliary variables to enable the use of simple low order polynomials as basis functions. We use Lagrange multipliers to satisfy the non-classical boundary conditions pertinent to strain gradients. To verify the developed software, we analyze plane strain deformations of a clamped, transversely isotropic beam. The obtained stresses and displacements compare well with the analytical solutions available in the literature, thus verifying the numerical solution. The Method of Manufactured solutions (MMS) was used to further verify the developed code as the assumed displacements and the resulting stresses were successfully reproduced. To study the effect of the material characteristic length (lc) in isotropic and transversely isotropic materials, we numerically study some known problems of plane strain elasticity and compare the classical and strain-gradient solutions. As lc is increased, the beam becomes stiffer as evidenced by a decreased tip deflection under the same loads. This numerically predicted stiffening reflects the experimental findings. We also observe that as the beam thickness becomes much larger as compared to lc, the strain-gradient solution approaches the classical solution.
  • Does Shame Amplify the Relations Between Perfectionism and Eating Disorder Symptoms? Cross-Sectional and Longitudinal Tests
    Patarinski, Anna Gabrielle G. (Virginia Tech, 2025-02-06)
    Shame and perfectionism are associated with and longitudinal predictors of eating disorder (ED) symptoms. Because shame represents painful emotions that result from negative self-evaluation and perfectionism involves unrealistically high standards for oneself, proneness to shame may intensify the relation between perfectionism and ED symptoms. The current study aimed to examine relations between perfectionism, shame, and ED symptoms (binge eating, body dissatisfaction, and excessive exercise) cross-sectionally and longitudinally across three months. College students [N = 259; 78% women; 71% White, mean (SD) age = 19.21 (1.24)] completed an online baseline survey in August 2022 and a follow-up survey three-months later. Participants completed measures assessing binge eating, body dissatisfaction, excessive exercise, shame, and perfectionism. Data were analyzed using path analysis in Mplus and significant interactions were probed using the Johnson-Neyman technique. Baseline shame was positively associated with baseline levels of binge eating and body dissatisfaction while baseline perfectionism predicted follow-up excessive exercise. There were not main effects of baseline perfectionism on any baseline ED symptoms nor of baseline shame on any follow-up ED symptoms. Interactive effects revealed that baseline perfectionism was negatively associated with follow-up levels of binge eating and excessive exercise for participants with average and high, but not low, levels of shame. There was not an interactive effect between baseline shame and perfectionism in predicting body dissatisfaction. Clinically, among individuals low in perfectionism, binge eating and excessive exercise interventions should address shame.
  • The Effects of Low Earth Orbit Satellite Constellation Altitude and Inclination on Simulated Non-Terrestrial Network Performance
    Downs, John Steven (Virginia Tech, 2025-03-12)
    Through the history of satellite communications, non-terrestrial networks (NTNs) have varied in scale and architecture. Satellite orbital elements have been a key feature of the ongoing development and design of modern constellations. Orbital altitude and inclination have impacted the types of service and extent of ground coverage each network can provide. The large scale of low Earth orbit (LEO) mega-constellations has allowed them to provide reliable internet access to a variety of communities and landscapes. When compared to their higher altitude counterparts, the high velocity of low-altitude satellites adds a layer of complexity to each system's network topology. The development of simulation-based testbeds has allowed for further analysis of these complexities. In the study, three LEO mega-constellations were designed, varying in altitude and inclination. The three constellations were simulated using a space network testbed and network latency measurements were taken using Mininet. Of the three constellations, the design most closely resembling existing NTNs with a lower altitude and moderate inclination was found to be the most efficient in terms of network performance. However, the other two designs still provided some unique advantages, which are discussed.
  • On the application of variational mechanics in modeling the flow around a cylinder in ground effect
    Zelaya Solano, Hever Jonathan (Virginia Tech, 2025-03-10)
    For high Reynolds number flow over a cylinder near a flat moving surface, a potential flow model can be used to represent flow over the leading edge. However, the potential flow solution requires knowledge of the circulation around the cylinder. This circulation value can be found with an auxiliary condition using energy methods. Two choices for this variational condition exist at present. Gol'dshtik and Khanin (1978) postulated an ad-hoc variational approach that looks only at the velocity on the cylinder boundary. The other approach is guided by the novel work of Gonzalez and Taha (2022), an extension of Gauss' principle that considers the entire velocity field. The first approach was calculated by Petrov and Maklakov (2022), whereas the second approach has not yet been applied to a cylinder in ground effect. These two models are applied to modeling a cylinder in `ground effect', and the predictions of these two models are compared with a computational fluid dynamics (CFD) simulation by considering the pressure distribution and forces on the cylinder as a function of the cylinders proximity to the wall. For gap-to-radius ratios approximately between 1 and 6, it is demonstrated that gauss' principle provides an acceptable auxiliary condition that gives an accurate potential flow representation of the leading-edge flow. The model is also able to calculate the lift-coefficient given an approximation of the trailing-edge pressure distribution based on experiment.
  • Efficient Lateral Lane Position Sensing using Active Contour Modeling
    Smith, Collin Mitchell (Virginia Tech, 2025-03-10)
    As research into autonomous vehicles and Advanced Driver Assistance Systems (ADAS) has grown, research into computer vision techniques to detect objects and lane lines within images has also grown. The heavier computational load of modern techniques involving neural net- works and machine learning limits the ability to downscale to cheaper, less computationally- capable platforms when needed. The goal of the project is to develop a robust and computationally efficient method to estimate vehicle position within a lane. A clothoid lane line model based in real-world coor- dinates is projected into the image pixel-space where a novel approach to image segmentation and active contour modeling is performed. Another novel approach presented is the use of velocity as an input from a source outside the algorithm into the process to predict the initial conditions of the model in the next frame, rather than using the algorithm to produce an estimate of the velocity as an output to other systems. Validation is performed using the TuSimple dataset using both ideal and realistic scenarios to evaluate the performance of the various aspects of the algorithm against the current state-of-the-art methods.
  • Improved 2D Camera-Based Multi-Object Tracking for Autonomous Vehicles
    Shinde, Omkar Mahesh (Virginia Tech, 2025-03-06)
    Effective multi-object tracking is crucial for autonomous vehicles to navigate safely and efficiently in dynamic environments. To make autonomous vehicles more affordable one area to address is the computational limitations of the sensors, therefore, cameras are often the first choice sensor. Three challenges in implementation of multi-object tracking in autonomous vehicles are: 1) In these vehicles, sensors like cameras are not static, which can cause motion blur in the frames and make tracking inefficient. 2) Traditional methods for motion compensation, such as those used in Kalman Filter-based Multi-Object Tracking, require extensive parameter tuning to match features between consecutive frames accurately. 3) Simple intersection over union (IoU) metric is insufficient for reliable identification in such environments. This thesis proposes a novel methodology for 2D multi-object tracking in autonomous vehicles using a camera-based Tracking-by-Detection (TBD) approach, emphasizing four key innovations: (1) A real-time deblurring module to mitigate motion blur, ensuring clearer frames for accurate detection; (2) deep learning-based motion compensation module that adapts dynamically to varying motion patterns, enhancing robustness; (3) adaptive cost function for association, incorporating object appearance and temporal consistency to improve upon traditional IoU metrics; (4) The integration of the Unscented Kalman Filter to effectively address non-linearities in the tracking process, enhancing state estimation accuracy. To maintain a Simple Online and Realtime (SORT) framework, we enhance detection by fine-tuning YOLOv8 and YOLOv9 models using autonomous driving datasets like BDD100K and KITTI, which are specifically tailored for these scenarios. Additionally, we incorporate a non-linear approach using the UKF to better capture the influence of various tracking dynamics, further improving tracking performance. Our evaluations show that the proposed methodology significantly outperforms existing state-of-the-art methods while maintaining the same inference rate as the baseline SORT model. These advancements not only improve the accuracy and reliability of multi-object tracking but also reduce the computational burden associated with parameter tuning and motion compensation. Consequently, this work presents a robust and efficient tracking solution for autonomous vehicles, making it viable for real-world deployment under both computational and cost constraints.
  • Channel Sounding for D-Band Measurements
    Frietchen, Samantha Michelle (Virginia Tech, 2025-03-06)
    With the advent of new technologies introduced with each cellular generation, there is need to characterize a variety of different communications links. Areas, such as software defined radios, have been explored to fill flexibility needs for dynamic sounding. Also of heavy interest is exploring the terahertz frequency band for communication potential in 6G. However, numerous channel sounding measurements must be collected to properly support channel models for this region. The work detailed in this thesis aims to address this current research areas, with three main contributions: (1) detailing a flexible software define radio channel sounding architecture for easy, configurable channel sounding, (2) a comparison of sounding waveforms within a software defined radio framework, and (3) a detailed D-Band channel sounding framework and short-range path loss measurements. In the first contribution, a low cost radio (Ettus B210) is used as the channel sounding transmitter with a frequency retuning software to overcome the small instantaneous bandwidth of the low cost transmitter. In the second contribution, an upgraded version of the SDR channel sounder transmitter from the first contribution is used to compare different sounding waveforms. Each of the waveforms were tested within the same channel sounder architecture and the results were compared to make recommendations about which waveform to use in a variety of circumstance. In the third contribution, a new channel sounder, with sub-THz up and down conversion, was used to collect path loss measurements at D-Band. In these contributions, we target addressing two prominent areas of channel sounding research: use of low-cost radios for channel sounding and (sub-)terahetz frequency channel characterization.
  • Investigation of the Aerodynamic and Acoustic Performance of a Scaled eVTOL Propeller in Axial and Non-Axial Flight
    Lundquist, Ryan David (Virginia Tech, 2025-03-04)
    With the recent emergence of Urban Air Mobility (UAM) as a potential solution to alleviate congested urban transportation, concerns have arisen regarding adherence to noise emission regulations and general public acceptance. With the design of new and innovative air vehicles utilizing electric Vertical Takeoff and Landing (eVTOL) propulsion systems for UAM applications, significant gaps remain in the understanding of their aerodynamic and acoustic performance, particularly when interacting with disturbances such as turbulence generated by buildings. To address safety, noise, and performance challenges, effective optimization methods must be developed. However, there is a lack of sufficient experimental data to support these advancements. This study investigates the aerodynamic and acoustic performance of a scaled eVTOL propeller operating in both axial and non-axial flight. A comprehensive summary of the experimental propeller's design is provided. Thrust, torque, and sound pressure data are acquired from wind tunnel testing of the experimental propeller operating with various blade pitch angles, yaw angles, and under several inflow velocities. The experimental results are subsequently compared to a custom-developed Blade Element Momentum Theory (BEMT) utility for low-fidelity predictions. The findings aim to provide baseline data for Computational Fluid Dynamic (CFD) validation, enhancing predictive tools for advancing safe and efficient urban air transportation. Experimental results exhibit positive correlations between thrust, torque, and acoustic intensity with increasing yaw angle. The acoustic profile of the propeller at large yaw angles features an increase in broadband noise, a characteristic feature of Blade-Wake Interaction. Additionally, BEMT calculations predict thrust and torque within 10% accuracy of the measured data across most conditions. Supplementary calculations of the induced velocity fields offer preliminary insights into the distortion effects for future studies on interactions between eVTOL propellers and turbulent flows.
  • Using Passive Acoustic Monitoring to Estimate Bird Community Response to Land Management in Southeastern Georgia
    Watson III, Daniel Hays (Virginia Tech, 2025-02-28)
    Working lands, such as mine reclamation and timber production sites, may be able to provide supplementary habitat for declining disturbance-dependent birds, such as Bachman's Sparrow (Peucaea aestivalis) and Northern Bobwhite (Colinus virginianus). However, habitat use is likely contingent on specifics of land-use practices, especially those that could alter understory vegetation. My first research objective was to use autonomous recording units (ARUs) and BirdNET algorithms to compare the relative abundance of eight focal bird species across site treatments representing land management types: surface mine reclamation, timber production, young savanna, and mature savanna. All sites were established in upland pine (Pinus spp.) habitat throughout the southeastern Coastal Plain region of Georgia, USA from May-June 2024. I hypothesized that the mine reclamation site would support similar focal species, but in lower abundances than timber production and both savanna sites, and vegetation characteristics would also influence relative abundance along with site treatment. Model selection showed site treatment influenced relative abundance for all species and explained more variation in relative abundance than measured vegetation characteristics. The mine reclamation site had similar relative abundances for focal species when compared to the timber production site, suggesting these treatments provide comparable habitat. The young savanna site exhibited the highest abundances for most species, whereas the mature savanna site had lower abundances, suggesting some focal species may prefer habitat lacking overstories. Focal species responded differently to vegetation characteristics; for example, Common Nighthawk showed a positive response to grass cover, whereas Prairie Warbler responded negatively. My results provide strong evidence that site treatment influences the relative abundance of all focal species and highlight the need for future studies to parse out the exact mechanisms underlying these differences. Additionally, this study highlights the potential for working lands to provide habitat for disturbance-dependent birds and the effectiveness of using ARUs to assess the effects of land management on bird relative abundances. My second research objective assessed optimal survey frequency when using Royle-Nichols (RN) models to estimate abundance or relative abundance from ARU data. Passive acoustic monitoring with ARUs can enable efficient monitoring of avian populations. RN models may be well suited for estimating abundance or relative abundance from ARU detection/non-detection data; as repeated surveys can easily be conducted with ARUs. Yet, optimal survey effort using these methods remains unexplored. Using ARU data from four site treatments in southeastern Georgia, I assessed how survey frequency and mean cumulative detection probability influenced estimates for Blue Grosbeak (Passerina caerulea) and Bachman's Sparrow from May–June 2024. A baseline dataset of 50 daily surveys was subsampled into reduced frequencies: 25 surveys every 2nd day, 17 every 3rd day, 13 every 4th day, 8 every 7th day, and 5 every 10th day. RN models were fitted to each subsample. Abundance estimates decreased with subsampling, showing survey frequency arbitrarily influences estimates and RN models should be viewed as relative, not absolute, abundance estimates. However, the specific order of relative abundance across site treatments remained consistent for both species during subsampling, indicating RN models can still reliably infer effects across sites. Mean cumulative detection probability decreased with subsampling yet remained >70% for both species. Subsampling reduced precision in relative abundance estimates for both species; particularly for Bachman's Sparrow, emphasizing species-specific sensitivity to survey effort. However, subsampling every 2nd day or every 3rd day resulted in moderate losses of precision (≤ 34%) for both species, suggesting reduced survey frequency may be a viable strategy for efficient data collection depending on species detectability and study goals. Together, these findings from my research objectives highlight the potential of working lands to support disturbance-dependent bird conservation and demonstrate how passive acoustic monitoring with ARUs can be an effective tool for the conservation and management of bird populations.
  • Functions of Mediodorsal Thalamic Astrocytes in Cue-Based Learning
    Marschalko, Kathleen Rose (Virginia Tech, 2025-02-25)
    To successfully navigate daily life, organisms must be able to identify stimuli that are predictive of beneficial outcomes. A key thalamic nucleus involved in this process is the mediodorsal thalamus (MD), which bidirectionally communicates with the prefrontal cortex, facilitating cognitive and decision-making functions. Despite the MD's involvement in higher-order relays, the precise mechanisms underlying its astrocytic activity, its contribution to synaptic plasticity, and the subsequent effects on cognitive processing remain poorly understood. Emerging data highlights the pivotal role of astrocytes in regulating synaptic transmission, with astrocytic calcium activity being linked to gliotransmitter release. Abnormalities in astrocytic calcium activity have been found to impair learning and memory, thus insights into their mechanism during cognitive processes in the MD could reveal novel targets for investigating cognitive disorders. In this study, we investigated astrocytic activity during a cue-based learning task, uncovering notable differences in the timing of astrocytic calcium release between early and late stages of the task. To investigate plasticity-related changes between early and late stages, the density of astrocytes, glutamatergic nerve terminals, and astrocyte glutamate transporter proteins will be examined. We found that MD astrocytic calcium activity responds to the initial cue and the reward, suggesting that this activity mediates the temporal dynamics of synaptic plasticity, influencing how thalamic circuits adjust to both cues and outcomes during learning.
  • Impact of Interdependent Physical and Social characteristics on Housing Recovery Following Tropical Cyclones
    Haque, Anmol (Virginia Tech, 2021-09-03)
    The aim of this study is to explore the interdependent impact of existing social (median household income) and physical (percent damage) characteristics on housing risk of a coastal community as the percent chance of vacancy followed by tropical cyclones. We developed a housing risk assessment framework for an idealized hypothetical study area consistent with existing physical and social characteristics of Hampton Roads, Virginia, USA. The housing risk assessment framework was simulated for a time period of 10 years and the distinct trends in housing recovery were observed for variations in the physical and social variables. The unique feature of the framework is its ability to demonstrate housing recovery risk for single and consecutive multi-hazards (combined storm surge and wind hazard) with a consideration of both existing physical and social characteristics of a coastal community. The applicability of the framework further lies in user-defined scenarios like events of gentrification (lower income households being replaced by medium income households) and modified recovery rates. To distinguish between the trends we grouped the percent damage and median household income in high, low and medium classes. It was found that the highest damage and lowest income groups recovered the slowest with an expected residual chance of housing vacancy even after 10 years. Some major findings of the study included - multi-hazards caused an amplification in housing risk compared to single hazard and gentrification was found to reduce effects of multi-hazard and hence faster recovery than without gentrification. This framework therefore has promising implications in disaster resilience and risk management policies and planning for coastal multi-hazards as it can predict impacts of extreme scenarios along with contributions towards the need for immediate intervention post disaster.
  • Numerical and Data Analysis of a Portable Free Fall Penetremeter
    Moore, Jonathan Joseph (Virginia Tech, 2025-02-24)
    Coastal environments are among the most economically and environmentally significant regions on Earth. However, rising sea levels and increasingly frequent storms driven by climate change pose growing risks to these critical zones. Understanding and predicting the evolution of coastal environments requires advanced models and high-resolution geotechnical data to characterize the mechanical behavior of coastal soils. Portable free-fall penetrometers (PFFPs) are widely used for this purpose, offering a rapid and efficient means of collecting in situ soil data. Despite their utility, the interpretation of PFFP data remains uncertain due to the empirical nature of existing methods used to infer soil properties from impact acceleration measurements. This research aims to improve the reliability of PFFP-based soil characterization by advancing both numerical modeling and data processing techniques. To this end, a two-pronged approach was taken. First, efforts were made to refine and streamline numerical modeling techniques to work towards the creation of a digital twin of BlueDrop PFFP impacting into the soil. This included identifying the current limitations of the MPM framework to simulate PFFP impact and addressing some of these limitations. In particular, this thesis focuses on the mitigation of volumetric locking by means of the implementation of the B-Bar algorithm in quadrilateral elements and the availability of strain-rate advanced consecutive models by testing the accuracy and efficiency of different stress integration algorithms. In addition, a Python library was also developed to streamline the testing of soil constitutive models using the IncrementalDriver software. Second, the processing of BlueDrop field data was centralized, standardized, and automated through the development of a Python library integrated with an SQLite database. This ensures consistency and accessibility of PFFP datasets for broader scientific and engineering applications. By advancing data processing methodologies and improving numerical modeling capabilities, this research contributes to a more rigorous framework for interpreting PFFP measurements and understanding soil behavior during impact. These developments support broader efforts to enhance geotechnical modeling of coastal systems, ultimately aiding in the prediction and management of environmental changes affecting these vulnerable regions.
  • Performance in Multipath & High-Mobility Leveraging Terrestrial and Satellite Networks
    Ghafoori, Amirreza (Virginia Tech, 2024-12-17)
    High-mobility scenarios, such as those experienced by autonomous vehicles or users in transit, demand reliable and high-performance network communication. This thesis presents a comprehensive measurement study comparing the performance of terrestrial 5G networks (ATT, Verizon, T-Mobile) and the Starlink satellite network in high-mobility scenarios. The study evaluates key performance metrics, including throughput and latency, across six globally distributed server locations: Virginia, California, Paris, Singapore, Tokyo, and Sydney. Measurements were conducted using a carefully designed testbed while driving a total of 860 km across urban, suburban, and rural terrains. The results reveal that 5G networks, particularly Verizon, excel in urban regions with higher peak throughput and lower latency, while Starlink demonstrates consistent performance in rural and remote areas. The impact of vehicle speed on network performance was also analyzed, highlighting Starlink’s resilience to high speeds compared to terrestrial networks. Heatmaps and statistical analyses underscore the complementary strengths of these networks, suggesting their integration via multipath protocols (e.g., MPTCP, MPQUIC) could enhance reliability and performance in critical applications such as autonomous vehicles, video conferencing, and AR/VR. This work provides valuable insights into the behavior of 5G and satellite networks in real-world high-mobility scenarios and lays a foundation for designing robust and efficient communication systems.
  • Toward Transformer-based Large Energy Models for Smart Energy Management
    Gu, Yueyan (Virginia Tech, 2024-11-01)
    Buildings contribute significantly to global energy demand and emissions, highlighting the need for precise energy forecasting for effective management. Existing research tends to focus on specific target problems, such as individual buildings or small groups of buildings, leading to current challenges in data-driven energy forecasting, including dependence on data quality and quantity, limited generalizability, and computational inefficiency. To address these challenges, Generalized Energy Models (GEMs) for energy forecasting can potentially be developed using large-scale datasets. Transformer architectures, known for their scalability, ability to capture long-term dependencies, and efficiency in parallel processing of large datasets, are considered good candidates for GEMs. In this study, we tested the hypothesis that GEMs can be efficiently developed to outperform in-situ models trained on individual buildings. To this end, we investigated and compared three candidate multi-variate Transformer architectures, utilizing both zero-shot and fine-tuning strategies, with data from 1,014 buildings. The results, evaluated across three prediction horizons (24, 72, and 168 hours), confirm that GEMs significantly outperform Transformer-based in-situ (i.e., building-specific) models. Fine-tuned GEMs showed performance improvements of up to 28% and reduced training time by 55%. Besides Transformer-based in-situ models, GEMs outperformed several state-of-the-art non-Transformer deep learning baseline models in efficiency and efficiency. We further explored the answer to a number of questions including the required data size for effective fine-tuning, as well as the impact of input sub-sequence length and pre-training dataset size on GEM performance. The findings show a significant performance boost by using larger pre-training datasets, highlighting the potential for larger GEMs using web-scale global data to move toward Large Energy Models (LEM).