33760 research outputs found
Sort by
Metamorphoses: An Exploration of Scenic Design
No full textThis thesis is a record of the scenic design process for Metamorphoses, a play by MaryZimmerman, produced at the University of Maryland --- College Park by the School of Theatre,
Dance, and Performance Studies. This thesis contains documentation for the scenic design and production process for this
performance of Metamorphoses. These documents serve as the foundation of this scenic design.
This thesis includes research images, photographs of ¼” scale models, drafting plates, paint
elevations, a properties list, a properties book detailing hand props and furniture, and a reflection
on the production and design process
THE MOLECULAR BASIS FOR GAMMA PHAGE TARGETING
No full textBacteriophage γ is a highly specific virus that infects Bacillus anthracis, leading to its adoption bythe Centers for Disease Control and Prevention (CDC) as a presumptive anthrax detection tool.
This dissertation investigates the molecular basis of γ phage targeting, focusing on the interactions
of its receptor-binding proteins. Using a combination of artificial intelligence protein prediction
software (AlphaFold3), biochemical assays, and genetic analysis, we characterize key phage
proteins responsible for host recognition and infection. Our study identifies the previously reported
tail protein GP14 as an evolved distal tail (EvoDit) hexameric hub that connects the tape measure
protein to the receptor-binding protein (RBP) of γ phage. Additionally, we demonstrate for the
first time that GP15 is a trimeric RBP containing an intramolecular chaperone that undergoes selfcleavage to ensure proper protein maturation. Both GP14 and GP15 exhibit strong affinity
exclusively for γ phage-sensitive strains. Furthermore, we confirm that the previously identified
receptor for γ phage's endolysin PlyG, the secondary cell wall polysaccharide (SCWP), plays a
crucial role in γ phage adsorption, likely serving as the primary receptor. Structural and functional
analyses reveal that GP14 and GP15 form a receptor-binding complex, facilitating irreversible
attachment to B. anthracis. Additionally, we uncover that PlyG naturally exists as a dimer, with
monomeric mutants exhibiting significantly reduced activity, highlighting dimerization as a key
factor in its enhanced bacteriolytic function and therapeutic potential. We further explore the
translational potential of γ phage components in developing rapid bacterial diagnostics, which have
traditionally relied on antibodies. The binding domain of PlyG (CBD) is integrated into a lateral
flow assay for anthrax detection, offering significant advantages over the CDC’s γ phage infection
assay, which requires specialized expertise and takes several days to produce results. These
findings advance our understanding of phage-host interactions through tail protein characterization
and highlight the diagnostic potential of γ phage proteins. Future research will refine the structural
characterization of these proteins and enhance the design of phage-based rapid diagnostic tests. By
leveraging γ phage biochemistry, this work lays the foundation for innovative countermeasures
against B. anthracis and other bacterial pathogens
ENHANCED ETHANOL-BASED LOCOREGIONAL THERAPIES FOR TREATMENT OF SOLID TUMORS IN UNDERPRIVILEGED POPULATIONS
No full textCancer is quickly becoming the leading cause of death globally, with an estimated 19.3 million new cancer cases and 9.9 million cancer deaths in 2020. While surgical resection of solid tumors, chemotherapy, or radiation are common methods for cancer management, a paucity of medical resources to perform the procedures prevent underprivileged cancer patient populations from receiving adequate treatment and care. Minimally invasive ablative therapies have the potential to address these issues and thus are crucial to addressing the increasing cancer burden in low- and middle-income countries. Ethanol ablation is one such low-cost treatment that directly injects ethanol into tumors to induce necrosis and is capable of treating nodules up to 3 cm in diameter. Intratumoral injections of ethanol into non-encapsulated tumors may, however, experience ethanol escape and leakage from the injection site, resulting in the need for multiple treatments and an increased risk of collateral damage to surrounding tissues. To mitigate this shortcoming, ethyl cellulose (EC) was dissolved in the injected ethanol to form a solid gel upon contact with the tumor environment, promoting ethanol retention within tumors and subsequently improving ablative efficiency and efficacy. The central hypothesis of this dissertation is that EC confers substantial biophysical enhancements to ethanol-based locoregional therapies that allow for improved cancer management strategies. This work explores the capabilities of EC-ethanol in treating solid tumors, which includes 1) pairing the technology with ultrasound to visualize its local distribution, 2) using it in tandem with photodynamic therapy to enhance locoregional tumor ablation of more advanced disease, and 3)
investigating the local and systemic antitumoral effects of EC-ethanol to cause distant tumor regression. The results from this dissertation demonstrate the versatility of the EC-ethanol technology in treating a variety of solid tumors and can be built upon existing clinical workflows worldwide to improve cancer treatment outcomes for all patients
Towards Trustworthy AI: Methods for Enhancing Robustness and Attribution
No full textCurrent deep learning systems demonstrate remarkable performance across diverse computer vision tasks, ranging from image classification to generative modeling. However, these models remain vulnerable to subtle adversarial manipulations and pose substantial challenges in auditing and interpreting their predictions. In this dissertation, we explore two fundamental challenges crucial for deploying trustworthy AI systems: robustness and attribution.
In the first half, we focus on robustness against adversarial perturbations—small, imperceptible changes that significantly alter model behavior.In the first chapter, we investigate the geometric properties of activation functions on adversarial training, a defense mechanism for imperceptible perturbations. We identify that activations that exhibit low curvature, mitigate overfitting thus improving generalization and mitigating double descent phenomenon.
In the second chapter, we investigate the influence of shift-invariance, a critical property of convolution neural networks on adversarial attacks. We prove theoretically for simple datasets that invariance to circular shifts can also lead to greater sensitivity to adversarial attacks. We then empirically verify this for real datasets and realistic architectures, showing shift invariance reduces adversarial robustness
In the third chapter, we propose a new approach to make datasets “unlearnable” by adding imperceptible noise to the training data. Our approach named autoregressive perturbations, is a novel dataset-agnostic poisoning strategy capable of generating imperceptible yet potent data poisoning attacks resistant to adversarial training and strong data augmentations.
Complementing the study of robustness, the second half delves into attribution i.e understanding predictions from the lens of training data. In the fourth chapter, we propose a simple mechanism for understanding how sensitive model's predictions are with respect to the training data. We show that in contrast to prior work which uses a lot of computational power, and a large ensemble of models, a single self-supervised model can serve as a baseline for how the training data influences model predictions.
In the fifth chapter, we focus on memorization, a special case of attribution for generative models. We rigorously examine memorization in diffusion-based text-to-image architectures like Stable Diffusion, quantifying substantial replication of training data in generated outputs. To counteract such memorization, we devise practical interventions effectively reducing copying without compromising the generative quality of these models.
Collectively, this thesis provides novel insights and practical methods, contributing towards the development of more reliable and trustworthy AI systems
Testudog: Autonomous Quadruped Robot for Unstructured Terrain Navigation
No full textRobotic autonomy is a field that has attracted significant interest, as robots become more capable of independent problem-solving. Quadruped robots have risen in popularity for their ability to navigate complex terrain. However, robust autonomous navigation, especially through unknown and variable environments, remains a challenge. Platforms like Spot from Boston Dynamics are currently being deployed for a diverse range of commercial purposes, but often possess a significant upfront cost. To address these challenges, Robotics at Maryland (R@M), the University of Maryland's largest competitive student robotics group, has been developing Testudog.
Testudog is a quadruped robot platform that seeks to provide a low-cost and fully autonomous navigation solution. Built completely from scratch, Testudog features a lightweight, 3D-printed frame with four modular legs, actuated by Quasi-Direct Drive motors and capable of 12 degrees-of-freedom movement. With advanced control systems, integrated torque sensing, and terrain-mapping sensors, Testudog aims to deliver an autonomous quadruped platform competitive with commercial systems, while remaining affordable through the use of modular, resourceful, and robust design strategies
Advancements in Hybrid Multilevel Circuits: Leveraging Joint-Phase Redundancy for Enhanced Performance Metrics
No full textEfficiency, size, and cost are crucial performance metrics in high-powered electrical systems required for transportation electrification, renewable energy integration, and grid modernization. Improvements in these metrics facilitate further adoption of green technologies over combustion-based alternatives. The power converter is central to these electrified systems, thus the research and optimization of power electronics converter circuits are essential for overall performance improvements. Multilevel converters are the preferred choice for applications exceeding a kilovolt because of their ability to achieve high efficiency and high-quality waveforms while withstanding the voltage stresses required for medium voltage applications. However, traditional multilevel converter topologies rely on many bulky circuit components, leading to high converter cost, size, and weight. Despite this, traditional multilevel circuit configurations remain widely used in contemporary power systems because alternatives often increase complexity without delivering clear performance benefits.
This dissertation focuses on one alternative, the hybrid neutral point clamped (NPC) converter, a less commonly used class of power electronic circuits. This topology integrates features from well-established and traditional topologies, namely the neutral point clamped converter, flying capacitor converter, and cascaded H-bridge converter. However, as with other alternative multilevel converter designs, performance benefits are unclear in existing literature. This work demonstrates through comprehensive analysis, simulation, and experimentation that the hybrid NPC outperforms the traditional multilevel designs across the key performance metrics, thus establishing the hybrid NPC as a viable solution for future high-powered electrical systems.
This work also proposes several innovations to the hybrid NPC topology that further improve the topology's performance. The key to these innovations is the introduction of `joint-phase redundancy,' which uses common mode voltages to reveal additional system states using are leveraged to increase overall converter performance. Joint phase redundancy can both enhance the cost, size, and efficiency of existing hybrid multilevel designs, and enable novel configurations that are otherwise not possible with traditional per-phase redundancy. This joint-phase redundancy specifically allows the use of lower voltage devices, which results in lower converter cost and size while increasing output waveform quality and efficiency. The proper application of joint-phase redundancy also enables operations in an extended voltage range, thereby increasing the power rating of the converter without hardware design modifications.
Additionally, this work addresses the complexity often associated with these multilevel topologies involving floating capacitors. These circuits are often cited for their increased control complexity and the need for large capacitances. The work addresses these challenges with a novel model predictive control strategy designed to minimize switching losses while ensuring high-frequency regulation of the capacitors, thus maintaining low capacitance requirements. Experimental results validate the practicality of this approach, showing its potential for use in industrial applications while offering substantial improvements in efficiency and control over previous designs. The extensibility of this analysis and control framework is used to compare the hybrid NPC class of circuits against competing designs.
This dissertation demonstrates that the hybrid NPC converter is a promising alternative to traditional multilevel converter topologies for high-powered electrical systems. The introduction of joint-phase redundancy and the associated analysis and control methodologies further the value of the hybrid NPC converter. The findings from this work provide valuable insights and practical guidance for future designers to adequately consider hybrid NPC circuits, offering an effective path toward optimizing power converter designs for next-generation, high-performance electrical systems
Convex geometry and asymptotics of stability thresholds in algebraic geometry
No full textThis thesis investigates stability thresholds in algebraic geometry, particularly the - and -invariants, which play a central role in various areas, including the existence of K\"ahler-Einstein metrics on Fano manifolds.
The work is divided into four main parts. First, we provide a counterexample to Tian's stability conjecture for -invariants, demonstrating that these invariants do not always stabilize or become monotone for large . Second, we resolve the Cheltsov--Rubinstein problem for strongly asymptotically log del Pezzo surfaces by studying the last remaining pair and establishing necessary and sufficient conditions for the existence of K\"ahler-Einstein edge metrics. In the last two parts, we study the stabilization and asymptotics of toric - and -invariants, showing that while -invariants stabilize from in the toric setting, -invariants rarely do, and we derive their asymptotic expansions. The results combine techniques from convex geometry, Ehrhart theory, and birational geometry, offering new insights into the interplay between convex geometry and algebraic geometry
PALS 2025 : Deliverable 7
No full textThis report details the perspective and reflections of Artist Heather McMordie and her involvement with the Community Engagement Team in the UMD PALS 2025 Green Isles Project.Acknowledgements: School of Architecture and Urban Planning, Creative Placemaking Minor, PALS (Partnership for Action Learning in Sustainability), Purple Line Corridor Coalition, Takoma Langley Crossroads Development Authorityhttps://drive.google.com/file/d/1xLlEuYs0l9vc1VFA63Efy8qtGx1fsfCj/view?usp=sharin
SES AS A PREDICTOR OF COMMITMENT: THE ROLE OF COMMUNITY IN PREDICTING WORK AND LIFE OUTCOMES OF LOW SES WORKERS
No full textThis research aims to determine if a novel approach to attitudinal commitment is meaningful in predicting work and life outcomes across socioeconomic status (SES). I present
the concept of community commitment, an attitudinal commitment that is defined by cultural
experiences that elicit feelings of pride, obligation to others, and identity, and can therefore
affect an individual’s relationship to the workplace and life outcomes. Further, I explore how the
impacts of resource-based stress and cultural identity may help explain the relatively greater
importance of community commitment for low SES workers. This study found counter
hypothetical results, indicating community commitment is stronger as an individual’s SES
increases. By exploring the roles of community commitment, this research exemplifies the need
of better understanding low SES workers and the relational impact of their communities to build
a more resilient and inclusive workforce
Identifying Risk Scenarios of a Solid Oxide Electrolyzer System for Hydrogen Production at Nuclear Power Plants
No full textSolid Oxide Electrolysis (SOE) is a developing technology for the production of clean hydrogen. SOE is a key technology used in a high temperature electrolysis facility, named for the high temperature steam at the inlet of the electrolyzer stack. The required steam temperature of 750 C, can provided via thermal energy from a Nuclear Power Plant (NPP), which also reduces the electricity required to produce the hydrogen. At present, the U.S. government and industry are researching and testing SOE designs connected to NPPs to enable commercial-scale system hydrogen production at NPPs. Commercial SOE facilities have the potential to produce clean hydrogen at a high efficiency and more flexible operating paradigms for NPPs. The deployment of these systems requires a more robust understanding of the operational hazards of the SOE facility’s design. However, to date the published literature has not presented a detailed description of the relevant hazards. This work adds to the growing body of engineering knowledge about hydrogen production facilities by identifying a comprehensive list of failure modes, mechanisms, and consequences. We describe our approach for conducting this analysis and document the SOE system we analyzed. To understand the hazards, a failure modes and effects analysis (FMEA) was conducted on a high temperature electrolysis test facility with a maximum power input of 25 kW and hydrogen production rate of 0.726 kg/hr developed at INL. All identified risk significant scenarios leading to the consequences of membrane degradation, hydrogen and oxygen mixing, hydrogen release, or nitrogen release are discussed. We identified system components that contribute to the most high-risk scenarios and proposed mitigation strategies to reduce these risks. These results were used to develop fault tree structures at a high level of abstraction to identify significant combinations of failures within the system. We created an enumerated list of risk significant scenarios. This research will assist the hydrogen stakeholders to make informed design choices to ensure safety and reliability in the continued development and deployment of SOE technologies. In the future, these results have the potential to be scaled to commercial SOE facility designs. The research provides a starting point for a comprehensive quantitative risk assessment needed to establish the risk-informed regulatory foundations that will ensure the safe and reliable deployment of solid oxide electrolysis coupled to nuclear power plants