Doctoral Dissertations (Winter 2014 to Present)
Permanent URI for this collection
New submissions to the University of Delaware Doctoral Dissertations collection are added as they are released by the Graduate College. In most cases the Graduate College deposits all dissertations from a given semester after the official graduation date.
Doctoral dissertations from 1948 to present are also available online through Dissertations & Theses @ University of Delaware. Check DELCAT to locate print or microform copies of dissertations that are not available online.
More information is available at Dissertations & Theses.
Browse
Recent Submissions
Item Morphological Control in the Design of Bio-Inspired, Stimuli-Responsive, Bilayer HydrogelsKlincewicz, Francis GregoryThe development of next-generation responsive polymeric systems requires materials with built-in practical functionality within the structure. Nature has provided many muses for the design of these “smart” stimuli-responsive materials, particularly hydrogels. For example, the pinecone has inspired bilayer hydrogel systems that curve upon application of a stimulus due to a contraction mismatch between interfaced responsive, “active” and non-responsive, “passive” hydrogels. However, as the stimuli-responsive polymer field has developed, there is a need for understanding the impact of hydrogel morphology on the actuation of bilayer hydrogels. In this thesis, the thermal response of bilayer hydrogels is linked to the morphological changes caused by nanofillers and the synthetic solvent in the component active, stimuli-responsive hydrogels. First, to study the effect of morphological changes induced by synthetic solvent, the thermal response of hydrogel bilayers fabricated via digital light processing (DLP) 3D printing is tuned using a cosolvent mixture. These bilayers are comprised of a thermally-responsive hydrogel, poly(N-isopropyl acrylamide) (pNIPAAm), and a non-responsive hydrogel, poly(2-hydroxyethyl acrylate) (pHEA). By printing these hydrogels from either ethanol or an ethanol-water solution, morphological changes including pore size are enacted in the pNIPAAm hydrogels, while the pHEA hydrogels are relatively unaffected. These morphological changes are correlated to a decreased thermal response with increased presence of water during printing due to the formation of a hydrophobic “skin” layer upon heating of the hydrogel. This skin layer is also observed in bilayer hydrogels formed by interfacing pNIPAAm and pHEA hydrogels. A comparison to an existing theoretical model is shown to have poor agreement with the experimental data, which is quantified using a “correction factor.” This research showcases precursor solution solvent as a handle to tune the morphology of bilayer hydrogels and presents limitations with existing theory. Extending these research findings, montmorillonite (MMT) clay is explored as a filler to potentially improve water transport to accelerate the thermal response of the bilayers fabricated in Chapter 1 using the solvent mixture. The introduction of surface hydrophilicity by the clay is hypothesized to mitigate the formation of the skin layer within the active pNIPAAm layer of the bilayers printed from the ethanol-water cosolvent mixture. Through exploration of single-layer active hydrogels, the addition of MMT is shown to cause faster actuation than the clay-free control, which is attributed to a decrease in skin layer formation. However, at higher MMT loading, the size of clay-rich domain structures is shown to correlate to a decreased initial rate of actuation and final magnitude of actuation. By interfacing these pNIPAAm/MMT hydrogels with pHEA hydrogels, bilayers are formed and shown to actuate considerably faster and to a greater curvature than the clay-free control. The increased bilayer actuation is related to the ability of the active layer to exert a contractile force on the passive layer during contraction, which is evidenced by improved prediction by the existing theoretical model which had shown poor predictive capability in Chapter 2. This research demonstrates the impact of inorganic, hydrophilic nanofillers for accelerating the response of bilayers through inhibition of skin layer formation and improvement of contractile force. Finally, the impacts of the loading and the structure of non-isocyanate polyurethanes (NIPUs) on the melt rheology and mechanical properties of blends with poly(lactic acid) (PLA) are investigated for potential use in 3D printing. Pendent methoxy groups on the lignin-derivable NIPU are hypothesized to improve blend melt strength and viscosity compared to the methoxy-free, petroleum-derived NIPU due to intermolecular interactions. The potential for plasticization or rubber toughening by the NIPUs is investigated through thermal and rheological characterization of the phase morphology. The inclusion of NIPU within PLA is shown to decrease the relaxation time and viscosity of the blend compared to the neat PLA due to the NIPU likely being incorporated into a singular phase with the PLA. The Young’s moduli, yield strength, strain-at-break, and toughness of the blends are found to be generally comparable to the properties of the neat PLA, demonstrating a lack of plasticization. However, in both the melt and the solid state, the properties of the lignin-derivable NIPU are shown to slightly exceed the petroleum-derived NIPU, potentially illustrating the increased intermolecular interaction by the pendent methoxy groups. This research serves as a preliminary study for potentially 3D-printable materials as well as how polymer processing aid structure and loading may affect polymer blend rheology and mechanical properties.Item Markedness Relations in the L2 Perception of English Obstruents by Arabic SpeakersAlkahtani, FaisalThis dissertation investigated the effect of markedness theory on second language (L2) speech perception. Markedness theory was developed based on the observation of speech output (i.e., speech production) of first language (L1) speakers. Research on speech productions showed that L2 speakers also exhibit similar patterns to L1 where the effects of markedness theory play a role in the learning process. Specifically, L2 speakers were found to have more difficulty with marked sounds than unmarked sounds. However, there is not a large body of research that focus on how markedness theory plays a role in the speech perception process. In this study, the perception of Arabic speakers learning English of 8 English obstruents were investigated to see if the effects of markedness would be exhibited. Specifically, the study attempted to see if voiced obstruents would be more difficult to perceive than voiceless obstruents and if sounds in word final position would be more difficult to perceive than in word initial position. Arabic was chosen as a case study since Arabic has voicing gaps in its obstruents phonemic inventory which runs counter to implicational universals. Specifically, while Arabic has the marked sound /b/, it lacks the unmarked sound /p/. An experiment was conducted where 31 Arabic participants and 31 native English speakers participated. They listened to nonce words containing the target sounds (/p/, /b/, /f/, /v/, /s/, /z/, /t/, and /d/), and they were asked to match auditory stimuli with visual representations of the nonce words that appeared on a screen. Results showed mixed patterns; Arabic speakers showed greater perceptual accuracy to sounds in initial position than final position confirming the principle that initial position is unmarked relative to final position. On the other hand, there was no significant difference between the perception of voiced and voiceless obstruents in initial position. This indicates that markedness theory doesn’t apply to L2 speech perception as it does to L2 speech production. In fact, it seems that phonetic factors played more prominent role on the participants' behavior. This research has significant implications for theory and practice that are discussed in the study in light of the existing knowledge of the problem under investigation.Item Non-Local Game HomomorphismsHoefer, GageIn the past few decades, the connections between non-local games arising in quantum information theory and the theory of operator algebras have undergone a phase of significant development. Operator algebras provide a particularly fruitful framework for approaching questions of non-locality in quantum systems, as the input-output behavior of measurements on bipartite quantum systems can be encoded through noncommutative operator algebras and their state spaces. This thesis is the compilation of a series of papers, written by the author in collaboration with Ivan G. Todorov, utilizing this framework by using the theory of operator algebras and completely positive maps to study questions involving non-local games. Using the simulation paradigm from information theory, we define a quantized version of homomorphisms and isomorphisms between classical hypergraphs, generalizing quantum homomorphisms and isomorphisms of graphs from the literature. We show that these quantum homomorphisms and isomorphisms constitute pre-orders and equivalence relations, respectively. We use quantum homomorphisms of hypergraphs to provide multiple examples of strict separation between correlation classes of varying types. Specializing to the case when our underlying hypergraphs arise from classical non-local games, we define quantum non-local game homomorphisms and isomorphisms. We also show how the existence of a homomorphism or isomorphism of some fixed type between games is reflected in a comparison of optimal and asymptotic game values with respect to this type. These non-local game homomorphisms are defined via a new class of non-signalling correlations which we introduce here. We develop the multivariate tensor product theory in the category of operator systems, which we then use to provide a characterization of game homomorphisms in terms of state spaces for various tensor products of canonically associated operator systems. We then apply these results to the study of synchronous non-local games, defining jointly synchronous correlations and showing how they correspond to tracial states of tensor products of C*-algebras canonically associated to each game party. We then move to a second level of quantization by proposing a "quantized" notion of hypergraphs (where "quantum" here is in the sense of a non-commutative analogue for a discrete combinatorial object), and introduce quantum homomorphisms between these quantum objects. We provide an initial foray into the properties quantum morphisms between quantum hypergraphs display, showing they satisfy analogous properties to quantum homomorphisms between classical hypergraphs. We also indicate initial connections these quantum homomorphisms between quantum hypergraphs have to the study of quantum input-output games. We end the thesis showing that homomorphisms of a local type between quantum hypergraphs is closely related, and in some cases identical, to the TRO equivalence of finite-dimensionally acting operator spaces canonically associated with each hypergraph. This suggests a quantum information theoretic approach to Morita equivalence in the category of operator spaces.Item ENHANCING SUICIDE PREVENTION: EVALUATING THE IMPACT OF UNIVERSAL SUICIDAL IDEATION SCREENING IN PRIMARY CARESelimov, Alyssa ValentinaBackground: Universal suicidal ideation (SI) screenings mitigate suicide risk through identification, early intervention, and monitoring. Despite the literature supporting the efficacy of universal SI screenings in primary care, many clinics indirectly assess suicide risk with depression screenings without supportive evidence. Purpose: This project explores whether direct screening for SI in primary care yields an accurate identification of SI, with secondary aims of increasing suicide risk mitigation interventions and evaluating the involved primary care providers’ (PCPs) perceptions of universal SI screening. Methods: A prospective quantitative analysis of universal SI screening was completed over eight weeks in a single primary care clinic. Four weeks of baseline suicide risk data using the Patient Health Questionnaire-2 (PHQ-2) preceded four weeks of universal SI screening with the Colombia Suicide Severity Rating Scale (C-SSRS). Descriptive statistics, chi-square tests, and Fisher’s exact tests were used to evaluate SI identification, risk mitigation intervention use, and PCPs’ perceptions of project feasibility and utility. Results: There was a significant decrease in the identification of SI with the C-SSRS compared to suicide risk identified with the PHQ-2 (?2 =7.11, ? = 0.008, ? = 0.114), highlighting the precision of the C-SSRS. Additionally, suicide risk mitigation interventions increased by 26% from baseline to intervention phase. All the involved PCPs believed the screenings to be moderately beneficial or very beneficial. Conclusions & Implications: This study supports the evidence of the applicability of universal SI screenings in primary care settings and highlights the nuances present in SI screening.Item Hardware-Based Multi-Platform Architecture for Super-Lattice Light Emitting Diode Infrared Scene ProjectorsLassiter, TianneInfrared scene projectors(IRSP) are critical laboratory tools used for the setup, calibration, and testing of Infrared(IR) imaging systems within a hardware-in-the-loop environment. These IRSPs are used to display dynamic user-defined video scenery that is then detected by an IR sensor in real-time simulations. IRSP technology until now has been dominated by resistor arrays that emit a heat signature to produce a desired image. A great deal of research has been put into a new IRSP system that can replace these arrays. In 2014, the CVORG research group, led by Dr. Kiamilev, and the University of Iowa research group, led by Dr. Thomas Boggess, produced the first IR Light Emitting Diodes(IRLED) scene projectors through SLEDs(Super-Lattice Light Emitting Diode’s) technology. Since then, the projectors have gone through different iterations, with each version improving greatly upon the prior array technology. SLEDs IRSP systems are designed to be modular and scalable to accommodate the demand for higher resolutions and faster frame rates for future-generation arrays. As existing IRSP technology advances, IRSP hardware capabilities need to expand in tandem. The current state-of-the-art projectors use commercial off-the-shelf Dewar products that cannot be easily modified and run at frame rates around 120 Hz, while also only being able to support up to 1K x 1K sized arrays. These limitations, in addition to physical limitations such as the fixed window size of the Dewar enclosure being limited to 1.6” and 100 built-in transmission line Input/Output(I/O) signals, put forth a central open question for IRSP technology: how do we improve performance such that kilohertz frame rates can be attained at larger resolutions? One of the major avenues would be to expand the hardware architecture to be able to accommodate this, giving room for the software structures to develop without the limitation of hardware resources. This research will focus on the design of a multi-platform approach to expand the hardware architecture, making the system more flexible and adaptable, by not only bypassing the limitations of the current cryogenic Dewar package but also providing a more direct connection to the array pads themselves. Using the multi-platform approach on the Close Support Electronics(CSE), the goal of the multiple CSE setup is to replace a single CSE with various multi-CSE configurations to increase the frame rate.Item Leveraging geospatial analysis and econometric methods to evaluate the impacts of events on community wellbeing: a dichotomy of objective and subjective metrics(University of Delaware, 2025) Tabatabaei, FarhadEvent tourism is a vital part of the tourism industry, known for its economic impact and potential to support community development. However, its broader effects on residents’ wellbeing are less understood. Existing research often emphasizes economic benefits based on residents’ perceptions, overlooking measurable impacts. There's also limited spatial and temporal analysis, and few studies explore causal links between hosting events and wellbeing, especially the interaction between residents’ objective data and subjective experiences. This study aims to address these shortcomings through a multi-method approach, combining geospatial analysis, temporal causality testing, and configurational methods. Focusing on four counties in Hawaii from 2000 to 2022, it draws on Core-Periphery Structure Theory and Spillover Theory to analyze spatial patterns, causal pathways, and gaps between perception and reality. The findings reveal notable spatial heterogeneity between core areas which tend to benefit more than peripheral ones, and complex temporal dynamics, including evidence of bidirectional Granger causality between events and socio-economic indicators. The research also highlights discrepancies between residents’ subjective wellbeing and objective data, offering a more nuanced picture of how event tourism affects communities.Item School readiness profiles of Head Start children: stability and within-profile variation(2025) Cuccuini-Harmon, Cara M.Kindergarten entry is a critical time for young children that helps to shape their academic, social-emotional, and cognitive developmental trajectories. Differences in the competency levels of school readiness contribute to starting-gate inequalities in education. These entry-level differences can be attributed to children’s sociodemographic backgrounds and prekindergarten experiences. Understanding children’s development of school readiness and the variation in their skill sets highlights the need for a nuanced understanding of school readiness development in marginalized populations. Previous person-centered school readiness studies find varying patterns of school readiness. Some studies observe stability and changes in these profiles within the same school year or across grades. This study contributes to this research as one of the first to include literacy and executive function skills as part of the school readiness profiles and to observe the stability of these profiles across two time points within a low-income population. Using the Head Start CARES (Classroom-based Approaches and Resources for Emotion and Social Skill promotion) restricted secondary data, the aim of this study is (1) to observe how school readiness skills (emotion knowledge, social problem-solving, social skills, interpersonal skills, executive functioning, behavior problems, learning behaviors, and preacademics) combine in unique ways within four-year-old Head Start children, and (2) to observe the stability of the school readiness profiles over the course of the prekindergarten school year. Using a latent class analysis model, this study identifies five distinct school readiness profiles with varying levels of school readiness functioning—overall high needs, average social skills/behavior problems/learning behaviors, mixed levels of functioning, strong emotion knowledge/executive function/preacademics, and overall strength. These profiles indicate variations in school readiness skill patterns for Head Start children. Using a latent transition model, this study finds five distinct school readiness profiles at the beginning and end of the prekindergarten school year, but with qualitative changes for the average profile and the strong profile in the spring. Ultimately, most children were more likely to move to a different qualitative profile (65 percent) throughout the prekindergarten school year, with the exceptions of the strength profile (78 percent) and the high needs profile (55 percent), where children were more likely to stay. The findings from this project will contribute to the knowledge of across- and within-group school readiness skill set differences. This will allow early identification of those who need more support, a more targeted approach to teaching, and to inform curricular, instructional decisions, and intervention to better align with children’s unique skill sets and contributing demographic and background factors.Item The effects a novel mutation in the mitochondrial unfoldase protein ClpX and a ClpX deficiency have on the regulation of the erythroid heme biosynthesis pathway(University of Delaware, 2025) Gillis, Samantha M.Mutations in genes involved in the heme biosynthesis pathway can cause various hematologic disorders. Previous research determined that heme synthesis is regulated by the mitochondrial AAA+ unfoldase ClpX through the activation of ALAS which initiates the first step of the pathway. A reported dominant mutation in the ATPase active site of human ClpX, p. Gly298Asp, showed the phenotypical accumulation of the intermediate protoporphyrin IX of erythropoietic protoporphyria (EPP) and effects on ALAS in heme synthesis (Yien, 2017). We created a CRISPR knock in murine cell model to further investigate the effects of this CLPX G298D mutation on the mitochondrial steps of the pathway in addition to identifying how iron status can act as a possible therapeutic for the mutation’s clinical effects. In addition, previous work investigating the role of CLPX in heme synthesis via a Clpx-/- MEL cell line is expanded upon using a more primary, physiologically relevant cell line, G1E-ER4 cells, where we are seeing less severe phenotypical results that may lead to how this loss of Clpx will act in a mouse model (Rondelli, 2021).Item Graph and hypergraph learning: theory and applications in computational network analysis(University of Delaware, 2025) Wang, YifanNeural networks, the computational models inspired by the human brain as powerful and advanced machine learning models, have played a central role in artificial intelligence since the 1980s. Given the power of being universal approximators, learning with neural networks has achieved remarkable performance on wide-range tasks, excelling across numerous applications such as computer vision, natural language processing, and graph-structured data analysis. In particular, modern architectures of neural networks applied in graphs and hypergraphs, combined with scalable optimization techniques and the availability of large-scale data, have further amplified their practical effectiveness across diverse domains. However, understanding their generalization capabilities, which refers to a model’s ability to perform accurately on previously unseen data, remains a fundamental challenge, particularly when modeling complex higher-order structures and competitive dynamics inherent in many real-world scenarios, i.e., social network analysis. ☐ This dissertation addresses these challenges by providing a rigorous generalization and learnability analysis of neural networks explicitly designed for higher-order relational data, with a specific application on computational network analysis. In particular, we combine the theoretical analysis with the conditions and mechanisms that enable efficient learning, which are then complemented by extensive empirical studies to validate these theoretical insights. Specifically, our work first study the learnability of the competitive threshold model from a theoretical perspective. We demonstrate how competitive threshold models can be seamlessly simulated by artificial neural networks with finite VC dimensions, which enables analytical sample complexity and generalization bounds. Based on the proposed hypothesis space, we design efficient algorithms under the empirical risk minimization scheme. For worm dissemination across the wireless sensor networks, we design a communication model under various worms, which are considered vulnerable to attacks by worms and their variants. We learn our proposed model to analytically derive the dynamics of competitive worm propagation. We develop a new searching space combined with complex neural network models. Finally, we develop margin-based generalization bounds for four representative classes of hypergraph neural networks, including convolutional-based methods (UniGCN), set-based aggregation (AllDeepSets), invariant and equivariant transformations (M-IGN), and tensor-based approaches (T-MPHN). Through the PAC-Bayes framework, our results reveal the manner in which hypergraph structure and spectral norms of the learned weights can affect the generalization bounds, where the key technical challenge lies in developing new perturbation analysis for hypergraph neural networks, which offers a rigorous understanding of how variations in the model's weights and hypergraph structure impact its generalization behavior. ☐ By utilizing PAC learnability and PAC-Bayesian frameworks, this work offers theoretical foundations for analyzing the relationship between the neural network architectures and their generalization performance, which shows potential to utilize the theoretical results for model optimization. In addition, with empirical validation on both synthetic and real-world datasets, this work reinforces the applicability of the theoretical findings and demonstrates the models’ ability to handle complex high-order data and competitive dynamics. Altogether, these contributions advance the foundational understanding of neural networks and offer practical methodologies for improving their generalization in structured, competitive environments.Item Deep learning algorithms for biomedical image segmentation in low-data scenarios(University of Delaware, 2025) Zid Alblwi, Abdalrahman HmodAutomatic segmentation via deep learning plays a major role in biomedical imaging, enhancing diagnostics by dividing images into regions of interest. This procedure helps medical experts understand disease characteristics, lesion sizes, and other crucial details. Despite its potential, deep learning-based automatic segmentation often relies on large annotated data to accurately predict lesions and other critical regions. Among imaging modalities, ultrasound, widely used for its accessibility, real-time capabilities, and effectiveness in detecting lesions, remains inadequately investigated due to the inherent challenges in medical imaging, such as data availability and privacy concerns. This work identifies key research gaps in ultrasound imaging segmentation to address these challenges and contributes to advancements in this critical area. ☐ This dissertation focuses on three key areas for advancing ultrasound image segmentation and improving biomedical image analysis. First, it aims to improve supervised learning-based architectures for tumor segmentation, particularly U-Net models, which, despite their success in biomedical segmentation, often lack reliability for clinical use, especially when tested on out-of-dataset samples. Second, it addresses the challenges posed by limited annotated ultrasound data, which restricts the performance of supervised models. Finally, it addresses the scarcity of ultrasound datasets paired with corresponding masks, a significant issue caused by data privacy concerns, the lack of datasets from various countries, and the high costs of expert-level annotations. ☐ This dissertation introduces an improved supervised model based on a refined U-Net architecture incorporating ReSidual U-blocks (RSU) and Attention Gates to address segmentation challenges in scenarios with limited data. These enhancements improve the model’s ability to capture critical features and long-range dependencies, improving lesion segmentation performance in ultrasound images. Building on this, we integrate a Denoising Diffusion Probabilistic Model (DDPM) with the RSU architecture to create a deeper network capable of handling the high variability and noise in ultrasound datasets. This combination enhances segmentation mask accuracy and addresses challenges posed by samples with diverse characteristics, such as size and shape variations of regions of interest. ☐ Next, we improve data augmentation by enhancing the Mixup technique to address limited data scenarios in image segmentation. Using K-means clustering, ultrasound images are grouped into clusters of similar samples, and Mixup applies within clusters. This approach has the potential to reduce randomness, avoid mixing unrelated regions like tumors and dark backgrounds, and ensure more effective augmentation. It also diversifies the dataset by generating new samples and masks, mitigating data scarcity. Building on this contribution, we extend the application of Cluster Mixup to unsupervised segmentation. The goal is to leverage unlabeled ultrasound images by augmenting healthy samples with Cluster Mixup, followed by unsupervised learning to detect suspected tumors. This approach could show the potential to qualitatively and quantitatively improve the segmentation of regions of interest and enhance diagnostic capabilities. ☐ Additionally, we build upon Cluster Mixup by proposing a variant of D-DDPM, a diffusion-based model, to learn the distributions of combined images and masks, enabling the simultaneous and joint generation of synthetic images and annotations. This technique expands the dataset with a large number of image-mask pairs. We involve medical experts in evaluating the synthetic dataset, ensuring the selection of relevant samples, and improving dataset quality. Statistical analysis obtained from medical experts shows the reliability of our approach and its potential application to real-world problems. ☐Item Research on molecular beam epitaxial growth of gallium selenide thin films(University of Delaware, 2025) Yu, MingyuBroadening the variety of two-dimensional (2D) semiconductors is crucial to the semiconductor development roadmap since their van der Waals (vdW) properties benefit the fabrication of heterostructures with multiple functions. Layered chalcogenides have garnered significant attention as emerging 2D materials, owing to their diversity and versatile properties. GaSe is a prominent member of this family, valued for its potential in optics, electronics, and optoelectronics. GaSe is well-suited for compact heterostructure devices due to its facile fabrication into atomic-scale ultrathin films. It also exhibits remarkable properties, including a bandgap transition from an indirect 3.3 eV in a single layer to a direct 2.1 eV in bulk, p-type conductivity, nonlinear optical behaviors, and high transparency across 650-18000 nm. These make GaSe a promising material for transistors, photodetectors, and photovoltaics. However, challenges persist in achieving wafer-scale synthesis. ☐ This study investigated the molecular beam epitaxy (MBE) synthesis of GaSe and achieved high-quality wafer-scale GaSe thin films. We explored GaSe growth on c-plane sapphire and GaAs (111)B substrates and examined the impact of growth parameters, including substrate temperature, flux ratio, and growth rate. Structural and optical properties of GaSe thin films were characterized. First-principles calculations were employed to analyze the GaSe growth and GaAs (111)B wafer processing mechanisms. Machine learning was used to build Bayesian interface models for guiding and predicting MBE synthesis experiments. ☐ We obtained GaSe single-crystal films (about 32 nm thick) on c-sapphire with a root mean square (RMS) roughness of 1.82 nm using optimized growth parameters. We further developed a three-step mode to fabricate 3-layer GaSe films with enhanced crystallinity and surface morphology, achieving an RMS roughness of 0.61 nm. On GaAs (111)B substrates, we systematically explored the growth window for GaSe and observed the gamma'-GaSe polymorph for the first time. We also demonstrated the respective advantages and limitations of 2D substrates (e.g., sapphire) and 3D substrates (e.g., GaAs) for GaSe growth, revealing distinct mechanisms of vdW and quasi-vdW epitaxy. Using experimental databases, we developed machine learning models to predict the crystallinity and surface morphology of GaSe films based on input growth parameters. In addition, we provided comprehensive insights into GaAs (111)B wafers, including deoxidation, passivation, and preservation. ☐ This study advances the wafer-scale production of high-quality GaSe single-crystal thin films. The discovery of gamma'-GaSe, with its centrosymmetric unit layer structure, opens avenues for exploring unique properties such as enhanced optoelectronic performance. Furthermore, our work provides valuable insights into the MBE growth of both 2D/2D and hybrid 2D/3D heterostructures, broadening material potential for device applications and establishing a foundation for integrated quantum photonic devices. Ongoing research aims to further develop GaSe as a platform for quantum technologies and extend the machine learning-based automated MBE synthesis platform to more vdW chalcogenide materials.Item Exploring advanced binder and electrolyte system for high-performance lithium-sulfur batteries(University of Delaware, 2025) Taiwo, Gbenga SamuelLithium-Sulfur (Li-S) batteries have received significant attention as promising alternative to lithium-ion batteries due to their high theoretical specific capacity (1675 mAh·g-1Sulfur) and energy density (2500 Wh·kg-1). However, there are fundamental challenges impeding their commercialization. Notorious among those challenges is the “polysulfide shuttle” consisting of the dissolution into the electrolyte and subsequent crossover to the negative electrode of long-chain, high-order sulfides (LiPSs). The mitigation of this polysulfide shuttle through electrode engineering and electrolyte design is key to realizing durable Li-S cells. In this thesis, successive studies are conducted with the goal of contributing novel insights and approaches for overcoming this fundamental barrier. ☐ The lesser-known role of the conventional binder (polyvinylidene fluoride, PVDF) in the instability and degradation of sulfur electrodes in Li-S batteries, as well as potential strategies for its mitigation were investigated. In addition, the influence of various electrolyte systems on the reaction pathways governing polysulfide dissolution and migration has been systematically investigated. Based on these findings, a biphase electrolyte system is proposed to eliminate LiPSs-shuttle and enhance the electrochemical stability of Li-S batteries. ☐ The first study investigates a critical failure pathway in PVDF-based sulfur electrodes that have not been widely recognized. It is revealed that 1,3-dioxolane, a common solvent in Li-S electrolytes, falls within the Hansen solubility sphere of PVDF, leading to the gradual dissolution of PVDF. Consequently, this dissolution compromises electrode structural integrity, increases electronic impedance, and causes rapid loss of active sulfur via the LiPS shuttle. To address this issue, a novel double crosslinked starch (DCS) binder was synthesized via two cross-linking reactions and contrasted against PVDF to mechanistically highlight the impact of such PVDF-dissolution. Compared to PVDF, DCS remains insoluble in the electrolyte, preserving the structural integrity of the sulfur electrode and exhibiting 2.5 times lower impedance after 200 cycles. Operando Raman spectroscopy, visual cell observations, and ex-situ UV-vis analysis confirmed superior polysulfide retention of the DCS-bound electrode. It achieved a discharge capacity of 522 mAh·gs-1 at 0.1 C rate after 200 cycles, representing a capacity fade of 0.14% per cycle beyond the initial 20 cycles. Under similar conditions, PVDF-bound sulfur electrode delivered a lower discharge capacity of 434 mAh· gs-1, and a higher decay rate of 0.24% per cycle after stabilization. This study highlights the limitations of PVDF binder in S-electrodes and investigates an alternative candidate (DCS) binder, which improves the structural integrity of S-electrodes and demonstrates superior electrochemical performance. ☐ The second study explores the influence of different electrolyte paradigms on the dissolution and diffusion of LiPSs in Li-S batteries. These electrolytes are categorized into fully solvating electrolytes (FSEs) and sparingly solvating electrolytes (SSEs). A comprehensive literature meta-analysis reveals that the most significant benefit of sparingly solvating electrolytes in Li-S batteries is the improvement in coulombic efficiency. Optical operando Li-S cell and ex situ UV-vis analysis were used to investigate polysulfide speciation in FSE and SSE. Experimental optical imaging enables real-time visualization of polysulfide dynamics, while UV-vis spectroscopy facilitates the identification of soluble polysulfide species within the electrolyte at the completion of cycling. Optical imaging and UV-vis characterization reveal that an increase in lithium salt concentration in the electrolyte, which renders it more sparingly solvating, induces a shift toward the formation of shorter-chain polysulfides. The transition to shorter-chain polysulfides indicates a reduction in the polysulfide species participating in LiPS-shuttle, thereby enhancing the coulombic efficiency. Under similar conditions, FSE follows the well-known reaction pathway, where sulfur initially converts to long-chain LiPS before transitioning to short-chain LiPS. This study demonstrates that SSEs minimize LiPS-shuttle, which is a major contributor to capacity fading and poor coulombic efficiency in FSEs. ☐ The final study proposes a novel biphase electrolyte based on the phase separation between solvents with different polarity. The biphase electrolyte decouples the cathode electrolyte (catholyte) from the anode electrolyte (anolyte), forming a selective interfacial membrane between them. This membrane blocks LiPS crossover from the catholyte to the anolyte, while allowing Li+ ion transport. Hansen Solubility Parameters of solvents are used to identify non-aqueous solvent pairs that exhibit and maintain phase separation during the cycling of Li-S batteries at room temperature. Candidate solvent pairs exhibited significant differences in their partial polarities. The high-polar solvent was used to prepare the catholyte due to its strong LiPS solvation power, while the low-polar solvent was used to prepare the anolyte. As a proof-of-concept, cycling an operando visual cell with biphase electrolyte confirms that the selective interfacial membrane effectively confines LiPSs to the catholyte. Electrochemical cycling further demonstrates that the biphase electrolyte maintains a more stable coulombic efficiency after the initial cycles compared to conventional electrolyte, indicating effective suppression of LiPS-shuttle. This study presents a new pathway for advancing the development of high-performance Li-S batteries. ☐ Together, these studies highlight the importance of developing alternative binders for S-electrodes, elucidate the reaction mechanisms governing polysulfide speciation in an electrolyte that reduces LiPS-shuttle, and propose a novel electrolyte that effectively suppresses LiPS-shuttle.Item "Land to the tiller! Education for all!": constellating the Ethiopian Student Movement and the Black Radical Tradition(University of Delaware, 2025) Bekele, SolyanaThe 1960s Ethiopian Student Movement (ESM), a moment where the youth’s radical fervor and commitment to a socialist Ethiopia was a harbinger of the end of empire, is a significant moment in the trajectory of Ethiopian political history worth intervening and reconstituting as part of the Black Radical Tradition (BRT). Ethiopia’s presumed history of remaining uncolonized in the face of European domination has created a “picturesque medievalism of kings and queens” while masking the “overwhelming reality of the misery” of the Ethiopian people.1 And despite the sustenance of this misleading image, scholars have attempted to establish Ethiopia’s albeit differing coloniality and the emergent decolonial politic embraced by Ethiopian students and activists in the mid-century against Haile Selassie II’s feudalist and imperialist empire. This study aims to intervene here and contribute to the scholarship regarding the ESM to show how the Ethiopian students’ ideological posture on the socioeconomic and national questions of Ethiopia was a manifestation of the phenomenon that is the BRT. Resituating the ESM within this inherited tradition establishes Ethiopia’s radical past, more accurately, as part of the transnational Black revolt uninterested in reformation, but rather socialist world-building—a moment in Ethiopian history overlooked, simplified and relentlessly critiqued. 1 Ethiopian Students Union in North America, “Repression in Ethiopia,” Africa Research Group 5, (1971): 1.Item Neural network reorganization following early-life seizures: neuroprotective role of ACTH in preserving cognitive function(University of Delaware, 2025) Khalife, Mohamed RabiehAction potentials represent the fundamental mechanism by which neurons transmit and encode signals, governed by complex and interdependent neural coding schemes known as rate, temporal, and population coding. Disruptions to these schemes, particularly during early developmental periods, can lead to profound and enduring cognitive deficits. This thesis investigates how early-life neurological insults, specifically early-life seizures (ELS), disrupt neural dynamics and subsequently impair cognitive functions. We chose the medial prefrontal cortex (mPFC) as our primary region of interest due to its critical but less extensively studied role in cognitive processes compared to regions like the hippocampus. Utilizing electrophysiological, behavioral and computational approaches, including single-unit recordings, fear extinction learning, graph theoretical analysis, and Graph Attention Networks (GATs), we characterized disruptions in neuronal firing rates, neuronal spike-timing, and population connectivity in the mPFC following ELS. ☐ Our findings demonstrate that ELS significantly impairs cognitive function by inducing rigid temporal firing patterns, reducing firing rates, and disrupting population-level network dynamics essential for adaptive cognitive processing. Importantly, implementing a neuroprotective intervention using Adrenocorticotropic Hormone (ACTH), acting through melanocortin 4 receptor (MC4R) signaling pathways, effectively preserved neural coding dynamics and rescued cognitive performance post-ELS. This therapeutic approach highlights the centrality of maintaining flexible and dynamic neural network function as a critical determinant of cognitive resilience. ☐ The future direction of this dissertation extend beyond ELS, suggesting a convergent mechanistic model where disrupted neural dynamics underpin cognitive dysfunction across multiple neurological and neurodevelopmental disorders, including Alzheimer's disease (AD), Parkinson's disease (PD), Fragile X syndrome (FMR1), and traumatic brain injury (TBI). By emphasizing the preservation of neural network dynamics, this thesis advocates for a fundamental shift toward network-centric therapeutic strategies aimed at maintaining and restoring cognitive integrity across diverse neurological conditions.Item Exploring protein dynamics and stability with advanced neutron scattering techniques(University of Delaware, 2024) Donnelly, Róisín B.Understanding the stability of protein-based therapeutics, particularly monoclonal antibodies (mAbs), is essential for ensuring their efficacy and longevity in biopharmaceutical applications. This dissertation investigates the intricate relationship between protein dynamics and thermal stability, driven by the need to develop advanced methods to assess long-term stability. Using bovine serum albumin (BSA) and the NIST monoclonal antibody (NISTmAb) as model systems, this research employs advanced neutron scattering techniques—Small-Angle Neutron Scattering (SANS) and Neutron Spin Echo (NSE) spectroscopy—to provide novel insights into protein dynamics and their relationship with the thermal stability. ☐ One important contribution of this work is the development and validation of a technique that uses Small-Angle Neutron Scattering (SANS) to measure hydrogen-deuterium exchange (HDX) in proteins. HDX assesses protein dynamics by quantifying the exchange of solvent-accessible hydrogen atoms with deuterium, which reflects the protein's conformational stability. The application of SANS in this context, termed HDX-SANS, offers a non-invasive approach to observe the HDX of proteins in their folded state, formulated in their buffer solutions. HDX-SANS complements other HDX methods, like HDX mass spectrometry, which is destructive and can be sensitive to formulation conditions. BSA was used first to demonstrate the noninvasive and quantitative capabilities of HDX-SANS, including the measurement of temperature dependent exchange rates and the determination of an activation energy of the HDX for BSA, which is found to be 81 ± 1 kJ/mol. ☐ Building on these findings, HDX-SANS was applied to NISTmAb under various formulation conditions, using an anionic Hofmeister series of sodium salts as excipients, including sulfate (SO42⁻), perchlorate (ClO4⁻), and thiocyanate (SCN⁻). NISTmAb is a standard mAb widely used by industry, which is publicly accessible. Its structural similarity to many mAbs on the market ensures that these findings are broadly applicable to a wide range of therapeutics. Our experimental results show that different types of salts have a strong impact on the HDX. The ranked order of HDX dynamics is observed to be: Na2SO4 < NaClO4 < NaSCN, which is consistent with both the anticipated ranked order of stability associated with the Hofmeister series and the effects of these anions on protein thermal stability, measured by differential scanning calorimetry. This alignment between the ranked HDX dynamics of different NISTmAb formulations and their corresponding melting temperatures suggests that the HDX dynamics observed in this study are consistent with the thermal stability of NISTmAb across various formulation conditions. ☐ While HDX in proteins provides an indirect measurement of intraprotein domain dynamics, to further understand mAb stability in formulation, the internal domain dynamics of NISTmAb are directly measured using NSE spectroscopy. NSE is a powerful technique, uniquely capable of probing nanometer and nanosecond-scale dynamics—precisely the relevant length and time scales for capturing the individual domain motions of an antibody. The analysis of the NSE results indicate that internal domain motions increase as the NISTmAb formulations approach their thermal transition temperature. This finding suggests that internal domain dynamics likely play an important role in the thermal stability of mAbs. In summary, the observations discovered in this dissertation advance our understanding of how protein dynamics are linked to thermal stability. The novel techniques and detailed findings presented offer a robust foundation for future research that could help the development of more stable and effective protein-based therapeutics.Item Statistical divergences and density estimation for anomaly detection and generative modeling(University of Delaware, 2025) Liao, YalinAnomaly detection (AD) is the task of identifying instances that deviate from an expected distribution. Unsupervised anomaly detection does not require any examples of anomalous data, which, if available, are typically insufficient to comprehensively define all aspects of anomaly. We propose to estimate an unnormalized density function from a dataset via noise contrastive estimation (NCE), on top of a composite feature representation combining an auto-encoder's latent features with the auto-encoder's reconstruction loss values. As an alternative to an auto-encoder, a pretrained model followed by PCA can also be used to construct the composite feature from principal components and the loss values of the principal component reconstruction. To further enhance the effectiveness of the NCE framework for AD tasks, we introduce two strategies to adapt the NCE framework: augmenting the training data by varying the reconstruction features to reduce the false negative rate, and optimizing the contrastive Gaussian noise distribution to better approximate the data distribution. Experimental assessments on multiple benchmark datasets demonstrate that the proposed approach not only matches the performance of prevalent state-of-the-art anomaly detection algorithms but also exhibits enhanced robustness on multimodal training datasets. ☐ In the second work, we introduce the decoupled Jensen–Shannon (DJS) divergence, a novel statistical divergence family that extends the Jensen-Shanon (JS) divergence, and includes the Kullback–Leibler (KL), reverse KL, and Jeffreys divergence as limit cases. While NCE is optimized to approximate the JS divergence, a density estimator could be formulated using any $f$-divergence under the same framework, if the optimal critical function can provide the information to calculate the density ratio as in the standard NCE. However, the KL divergence and reverse KL divergence possess mode-covering and mode-seeking properties, respectively, that result in the overestimation and underestimation of the density function. Our proposed DJS divergence is a convex combination of skewed KL and skewed reverse KL divergences, designed to mitigate these estimation biases. To facilitate its application to anomaly detection and generative modeling, we derive a variational formula for the DJS divergence and obtain a statistically consistent estimator when limited to finite samples. We explore the application of DJS divergence in anomaly detection and generative modeling, such as generative adversarial neural networks (GANs) in high-dimensional image spaces.Item Towards multi-scale inter-frame attention to improve deep learning tasks(University of Delaware, 2025) Bhattarai, AshutaAccess to specialized medical screening remains a challenge for individuals with sickle cell disease (SCD), particularly those in low-income and rural communities, where advanced diagnostic tools and expert evaluations are limited. In ophthalmology, Sickle Cell Retinopathy (SCR) diagnosis relies on ophthalmologic evaluation, including Optical Coherence Tomography (OCT) scans, but the manual interpretation is prone to subjectivity, fatigue-induced errors, and inconsistencies across clinicians. Similarly, video-based event analysis—such as reconstructing crime scenes from fragmented surveillance footage—is a time-intensive process that requires manual ordering and interpretation of unordered clips. These challenges highlight the need for automated solutions that enhance medical diagnostics and video-based decision-making. ☐ To address these issues, we propose Multi-scale Inter-frame Attention (MIA), a novel framework that enhances deep learning models for processing volumetric and video datasets. Our approach leverages spatial and spatio-temporal attention mechanisms to improve feature extraction and representation learning. We integrate MIA into two specialized models: the Cross-Scan Attention Transformer (CSAT) for SCR detection and the Sequential Ordering of Frames in Time (SOFT) for video-based action recognition. Experimental results demonstrate that CSAT+MIA outperforms conventional object detection models in diagnosing SCR, while SOFT+MIA enhances action recognition, particularly in temporally shuffled scenarios. ☐ Beyond domain-specific improvements, our research aims to establish a unified deep-learning method capable of capturing both inter-frame and intra-frame relationships for broader applications in medical imaging, surveillance, and video understanding. By integrating multi-scale inter-frame attention, we advance the field of automated diagnosis and event reconstruction, paving the way for more efficient, reliable, and intelligent decision-making systems.Item Understanding and defending against use-after-free vulnerabilities(University of Delaware, 2025) Chen, ZeyuOver the past decades, use-after-free (UaF) vulnerabilities have become a critical and widely exploited security concern. To address such increasing threats, this dissertation advances defense in multiple aspects, including UaF vulnerability detection, UaF exploit defense, and UaF bug fixes. ☐ The first research direction delves into a comprehensive empirical study of UaF vulnerabilities. Utilizing a dataset of 150 real-world UaF cases randomly sampled from representative software suites such as the Linux kernel, Python, and Mozilla Firefox, this study seeks to unravel the commonalities, root causes, and recurring patterns in real-world UaF bugs. The findings highlight the diversity and non-uniform distribution of root causes among different software, emphasizing that a generic UaF detector or fuzzer might not be an optimal solution. By categorizing the root causes into 11 patterns, several of which can be translated into simple static detection rules, this work further introduces a static bug detector named Palfrey. Palfrey exhibits superior coverage and accuracy in UaF detection, while minimizing time and memory overhead, making it a promising tool to address UaF vulnerabilities. ☐ The second research direction investigates the relationship between fuzzing techniques and UaF vulnerabilities. Fuzzing has proven effective in rapidly generating faulty inputs to discover software bugs, but its efficiency in identifying UaF instances with specific patterns and characteristics remains uncertain. The study explores whether the efficiency of fuzzing is dependent on the code scope of UaF, particularly for bugs with larger code scopes. Additionally, the research addresses the challenge of detecting non-deterministic UaF bugs and explores the necessity of auxiliary tools like address sanitizers (ASan) to assist fuzzers in identifying UaF vulnerabilities. The study provides practical guidelines to enhance UaF detection using fuzzers, aiming to improve the effectiveness of dynamic testing techniques in identifying UaF vulnerabilities. ☐ The third research direction explores how Large Language Models (LLMs) perceive and detect UaF vulnerabilities compared to traditional static analysis tools. Unlike deterministic analyzers, LLMs rely on probabilistic language reasoning and are influenced by prompt phrasing, input granularity, and abstraction level. To enable systematic evaluation, this work constructs a benchmark of 50 curated UaF cases specifically adapted for LLM-based analysis. These samples are distilled from the broader set of 150 real-world UaF cases and carefully reduced to single-file, self-contained programs that preserve critical semantic features while remaining within LLM token limits. The benchmark is used to evaluate LLM performance across prompt styles and input formats, including source code and LLVM IR. Novel evaluation metrics—such as Scalability Index, Context-Aware Precision, and Prompt Sensitivity Score—are proposed to capture the unique behaviors of LLMs in code analysis. The study reveals a semantic gap between LLM and static tool reasoning and highlights opportunities for hybrid approaches. ☐ In conclusion, this dissertation seeks to advance the state of the art in UaF vulnerability mitigation by comprehensively understanding real-world UaF characteristics, introducing innovative static analysis techniques, and enhancing the efficiency of fuzzing for UaF detection. These research directions collectively contribute to the goal of addressing the critical issue of UaF vulnerabilities and enhancing the security and reliability of software systems.Item Social support and communicative disenfranchisement: extending the theory of motivated information management within a contested illness context(University of Delaware, 2025) Edwards, TimothyThe way that individuals with contested illnesses seek out information about their illness plays a major role in determining their quality of life. This study used the Theory of Motivated Information Management to examine how individuals with contested illnesses deal with uncertainty and the factors that influence the usage of information-seeking strategies and how that influences their quality of life. The results reveal that there is a negative relationship between avoiding information and quality of life for individuals with contested illnesses. The results also demonstrate that for individuals with low perceived social support, there is a negative relationship between direct information-seeking and quality of life whereas for individuals with high perceived social support, this relationship is positive. The results also show that for individuals with low perceived communicative disenfranchisement, there is a positive relationship between direct information-seeking and quality of life, whereas for individuals with high perceived communicative disenfranchisement, this relationship is negative.Item In-depth stability characterization and engineering of bacterial N-terminal motifs and their protective tags(University of Delaware, 2025) Sen, SabyasachiProtein degradation plays a pivotal role in the maintenance of the cell by recycling abnormally expressed proteins and regulating stimuli-responsive protein lifespans. A protein degradation pathway of particular interest is the N-degron pathway, where short N-terminal motifs known as N-degrons can modulate protein half-life from two minutes to over ten hours. Initially, substrate specificity in this pathway was thought to depend solely on the identity of the first amino acid; however, recent studies have revealed that residues up to the fifth position from the N-terminus can significantly influence protein stability. Consequently, there remain open questions about how precisely the amino acid sequence of N-terminal regions governs protein stability as well as how generalizable are previously observed sequence trends. Clarifying substrate preferences within this pathway would enable biological engineers to finely control protein half-life and prevent unwanted degradation of recombinant proteins. ☐ To monitor protein degradation in vivo, we adapted the ubiquitin fusion technique to create a dual fluorescent reporter, allowing us to differentiate destabilizing degrons and stable sequences over three orders of signal magnitude. Utilizing this assay, we screened novel pathway candidates including highly destabilizing sequences derived from human and plant N-degrons. Using these imported sequences as templates, we established a high-throughput screening platform combining DNA library generation, fluorescence-activated cell sorting (FACS), and next-generation sequencing (NGS), allowing simultaneous analysis of numerous sequences. After validating our platform with a 60-member library screen, we scaled the platform to analyze combinatorially mutagenized sequences across the first five N-terminal amino acids, generating detailed sequence-specificity maps from over 800,000 sequences at a depth of at least 20 reads per sequence. ☐ This extensive dataset revealed previously undiscovered trends in the bulk data, such as the destabilizing impact of glutamine and the stabilizing effects of glycine and proline residues downstream of the N-terminal position. It has also revealed trends that are specific to certain sequence contexts, such as how two bulky residues at positions two and three can occasionally convert a sequence with a canonically stable N-terminal residue, such as Nt-Cys, into an N-degron. Furthermore, we have characterized sequence stability trends for synthetic N-termini commonly used for small molecule or protein ligation, such as Nt-Cys, Gly, and Ser. Leveraging these insights, we developed N-FIVE, a machine learning model capable of predicting and recommending N-terminal sequences based on desired stability profiles. Collectively, these findings represent the most comprehensive characterization of the Escherichia coli N-degron pathway to date and provide practical insights for protein lifespan modulation. ☐ Despite its potential, the N-degron pathway remains underutilized in a biological engineering context, partly due to limited methods for dynamically exposing neo-N-termini. Addressing this challenge could unlock valuable applications, including dynamic protein lifespan switches. To resolve fundamental obstacles in neo-N-termini generation, we first evaluated the stability of widely-used protective tags. Notably, we observed unexpected cleavage of ubiquitin and SUMO fusions in common E. coli strains. By identifying and knocking out four candidate deubiquitinases in BL21, we engineered the Zero observable Ubiquitin cleavage (sUbZero) strain, significantly enhancing ubiquitin-fusion stability and improving yields by over 50%. Next, we developed a SUMO protease (Ulp1) that is dependent on the nonstandard amino acid o-methyltyrosine for expression. Using this engineered protease, we created an inducible protein stability switch. We demonstrated its ability to conditionally remove protective tags, enabling degradation of N-degron-tagged proteins. Subsequently, we have utilized this technology to regulate the lifespan of a toxin, showing a proof of concept that the toxin’s function can be negated in the presence of our engineered protease. Future directions and optimization strategies for conditional cell death in a biocontainment application are also discussed. ☐ In summary, this thesis advances our understanding and application of how N-terminal protein motifs govern protein stability. By screening and analyzing millions of N-degron candidates and overcoming key barriers related to engineered protective tag removal, we provide new insights and practical tools that empower biological engineers to precisely tune protein half-life.