Data exposure in software services engineering is critical because leakages of confidential or sensitive information are frequent, but developers often struggle to prevent them due to diverse causes like hardcoded secrets, storage misconfigurations, insecure logging, improper data transmission, or unsafe deserialization. Modern software development practices can exacerbate these risks. Existing methods for detecting data exposure, often using regular expressions or static pattern analysis, frequently generate many false positives or lack the deep contextual understanding required for reliable detection across varied programming languages. To address these limitations, this thesis presents an effective approach to improve data exposure detection in software services using a Large Language Model, enhanced via low rank adapters (LoRA) for efficient specialization and few-shot learning for ambiguity refinement. Unlike detection systems relying solely on static patterns or simple heuristics, the Large Language Model-powered framework presented provides deep contextual code analysis, enabling highly accurate identification of exposures. The evaluation demonstrates performance significantly surpassing existing tools and alternative techniques in both precision and recall. Key capabilities include efficient model adaptation through LoRA, ambiguity resolution using few-shot learning based on optimized thresholds, and precise line-level localization of identified exposures, ensuring more reliable results and facilitating faster remediation. The integration of a feedback loop for continuous learning further distinguishes the framework as an accurate, scalable, and intelligent solution suited for detecting data exposures in complex software service environments.

The rise in anti-trans legislation has been linked to increased mental health disparities among transgender and nonbinary (TNB) individuals, including elevated levels of anxiety and depression. Despite growing concerns, the association between anti-trans legislation and vicarious stress, defined as the emotional distress experienced when learning about violence, discrimination, or harmful policies affecting other TNB individuals, has not been examined quantitatively. This study investigated the association between awareness of anti-trans legislation and vicarious stress, as well as the potential moderating roles of community connectedness and hope. A sample of 575 TNB participants completed measures assessing anti-trans legislation awareness, vicarious stress, community connectedness, and hope. Results indicated a significant positive association between anti-trans legislation and vicarious stress (β = .11, p < .01). Neither community connectedness (β = −.01, p = .83) nor hope (β = −.02, p = .73) moderated this association. These results corroborate existing evidence that anti-trans legislation contributes to vicarious stress within TNB communities and add to the mixed literature on resilience factors as moderators. Implications for clinical practice and future directions for research are discussed.

This study examines the driving forces and behavioral choices of young people involved in religious tourism, focusing on Nanshan Temple as a case study. As religious tourism evolves beyond traditional spiritual pursuits, young tourists demonstrate a range of motivations, including spiritual support, cultural exploration, stress relief, personal growth, and curiosity. Through the utilization of semi- structured interviews and participant observation, the research identifies self-actualization as a key driver, showing how religious tourism addresses broader psychological needs. The findings emphasize that young tourists are attracted to experiences that blend spiritual, cultural, and aesthetic elements, often favoring immersive, short-term, and sustainable travel. The study extends existing tourism motivation theories by integrating psychological frameworks like Maslow’s hierarchy of needs, providing a nuanced understanding of young people's engagement with religious tourism. The research not only identified self-actualization as a key driver among young tourists, but also found that push factors like stress relief and pull factors such as the cultural and natural attractions of religious destinations significantly influence their travel decisions. It also offers practical implications for tourism management, suggesting strategies for enhancing religious destinations’ appeal through interactive, personalized, and environmentally conscious experiences. Ultimately, this research contributes to the theoretical and practical understanding of religious tourism as a dynamic platform for holistic personal development among younger generations.

Mn+1Xnenes are the layered Titanium Carbides and/or Nitrides where M is an early transition metal and X is Carbon and/or Nitrogen. Due to its excellent electrical conductivity, Ti3C2Tx, a 2D metal, with surface functionalized group Tx= O, OH and F, is emerging as a potential electrode for optoelectronic devices on conventional semiconductors. Much of the physics of Schottky barrier, which controls the current across the Schottky-interface at the MXene-semiconductor junction is not well-understood till date. This dissertation presents an experimental analysis of Schottky barrier formed at Ti3C2Tx and n-type GaAs interface from temperature-dependent current-voltage relationship. Schottky barrier using two different types of MXene Ti3C2TX: disordered multilayer nanoparticles and azimuthally aligned single-layers of MXene used for this study shows different Schottky barrier heights on n-GaAs. This study leads to the understanding of the factors dominating the Schottky barrier, and hence the current conduction across MXene-based metal-semiconductor interfaces.First, MXene Ti3C2Tx was synthesized in multilayer powder-form with optimal synthesis parameters from parent MAX phase Ti3AlC2 and delaminated into single-layers by intercalation. Characterization of as-synthesized MXene was performed by X-Ray diffraction (XRD), scanning electron microscopy (SEM), electron energy dispersive X-Ray spectroscopy (EDS) and atomic force microscopy (AFM).
Second, two types of Schottky-interfaces using multilayer and delaminated Ti3C2TX were formed on n-type GaAs by facile drop-casting method. The current-voltage relationship across the Ti3C2TX/n-GaAs interface were found rectifying for both samples. Schottky Barrier heights at Ti3C2TX/n-GaAs were determined from current-voltage relationships which showed lower barrier heights for the delaminated MXene/n-GaAs Schottky-interfaces. While the multilayer MXene/n-GaAs interface showed an average barrier height of 0.75 eV, the average barrier height at delaminated MXene/n-GaAs interface was found to be 0.63 eV. Based on cross-sectional scanning transmission electron microscopy (STEM), Kelvin probe Force microscopy (KPFM) and annealing experiments, Schottky barrier heights in MXene/n-GaAs systems has been explained by reduced Fermi level pinning (FLP) and interlayer trapped water.
Third, inhomogeneity in Schottky barrier height were analyzed by assuming Gaussian distribution model using temperature-dependent current-voltage data.
Lastly, self-biased photodetection of both MXene-based diodes was demonstrated under 785nm wavelength. Delaminated MXene-based device shows superior photoresponsivity compared to multilayer MXene-based device.

Cadmium Telluride (CdTe) is a direct bandgap semiconductor with a bandgap of 1.5 eV. According to the Shockley-Queisser limit, the theoretical efficiency of CdTe solar cells is 33%. Currently, the highest laboratory-recorded power conversion efficiency for CdTe solar cells reaches 23.1%. While the short-circuit current (JSC) is optimized, the open-circuit voltage (VOC) and fill factor (FF) have less utilization. Doping is an important strategy to enhance VOC. Additionally, due to the high electron affinity of CdTe, implementing a back contact between CdTe and the electrode is necessary to enhance the fill factor.Using an AsCl3 vapor annealing doping approach, arsenic-doped CdSeTe devices have achieved approximately 18% efficiency, much higher than CdSeTe devices fabricated without vapor annealing. Besides the significant enhancements of efficiency, this vapor annealing approach led to a longer carrier lifetime of over 72 ns and VOC of 850 mV.
The As2Te3 and As2Se3 solutions were synthesized using DI water and ammonium sulfide as solvents at room temperature. Subsequent experiments and characterizations show that these arsenic chalcogenides can serve as dopants and back contacts. In the case of the As2Te3 solution, the formation of tellurium promotes hole transport. Compared to the As2Ses doped CdSeTe device, the fill factor (FF) of the CdSeTe device doped with As2Te3 increased from 70.44% to 73.09%.
Antimony chalcogenides can potentially serve as dopants. Initially, Sb2S3 films were synthesized, and the impact of precursor processing ambient on crystal growth behavior was investigated. This project demonstrates the feasibility of synthesizing antimony chalcogenides, which can be used as dopants and back contacts in CdSeTe solar cells. In the next section of the chapter, Sb2Se3 and Sb2Te3 solutions, using en and edtH2 as solvents, will be synthesized and can be applied to the CdSeTe substrate. In addition to their role as dopants, Sb2Se3/Sb2Te3 could also work as back contacts. However, compared to Sb2Se3, the conduction band offset between CdTe and Sb2Te3 is larger, which leads to electron transport to the back contact and results in electron-hole recombination at the back contact. The FF increased from 67.68% to 70.70% when switching from Sb2Te3 to Sb2Se3 as the dopant sources.

This thesis develops a synchronizer for non-synchronous dynamic networks under minimal assumptions. Our model allows continuous topological changes without any guarantee of eventual global or partial stabilization and assumes that nodes are anonymous. The resulting construction, named the Continuous-Dynamics Synchronizer or κ-Synchronizer, reflects two key characteristics: the network remains continuously dynamic throughout execution, and computations are performed locally relative to each node’s current set of at most ∆ neighbors, where ∆ denotes the maximum degree observed. By limiting dependencies to immediate neighbors and tolerating arbitrary edge changes, the κ-Synchronizer provides a robust mechanism for achieving synchronization in highly dynamic distributed systems. This deterministic synchronizer is the first to enable nodes to simulate a dynamic network synchronous algorithm for executions in a semi-synchronous dynamic environment under a weakly-fair node activation scheduler, despite the absence of a global clock, node identifiers, persistent connectivity or any assumptions about the edge dynamics (in both the synchronous and semi-synchronous environments). The κ-Synchronizer operates with memory overhead at the nodes that is linear on the maximum node degree and logarithmic on the runtime of the underlying synchronous algorithm being simulated.
Beyond the construction of the synchronizer, the thesis establishes fundamental impossibility results for communication models in dynamic environments. It is shown that neither the classic pull model nor the classic push model is sufficient to implement a synchronizer under arbitrary edge dynamics. These results identify inherent limitations of conventional communication models and motivate the need for more structured atomic synchronization mechanisms in adversarially dynamic networks.

Low birth weight (LBW) remains a critical public-health indicator, linked strongly with higher neonatal mortality, developmental delays, and lifelong chronic diseases. Using the 2021 U.S. Natality dataset (> 3 million births), this thesis develops a Bayesian, tree-based, nonparametric framework that models the full birth-weight distribution and quantifies LBW risk.

The raw dataset is condensed into 128 mutually exclusive classes defined by seven dichotomous maternal-infant predictors and 10 (or 11) birth-weight categories, comprised of 10\% LBW quantile categories and one additional aggregated normal birth-weight (NBW) category. The full and LBW-only models are grown to contrast and investigate how variable selection is altered based on the restriction the dataset. The models are Classification and Regression Trees (CART) using the marginal Dirichlet-Multinomial likelihood as the splitting criterion. This criterion is equipped to handle sparse observations, with the Dirichlet hyperparameters informed by previous quantiles from the 2020 dataset to avoid "double dipping."

Employing a two-tier parametric bootstrap resampling technique, a 10,000 tree ensemble is grown yielding highly stable prediction estimates. Maternal race, smoking status, and marital status consistently drive the initial LBW risk stratification, identifying black, smoking, unmarried mothers among the highest-risk subgroups. When the analysis is restricted to LBW births only, infant gender and maternal age supersede smoking and marital status as key discriminators, revealing finer biological gradients of risk. Stable and informative mean ensemble estimates are obtained with narrow 95\% percentile intervals.

The resulting modeling framework combines the interpretability of decision trees with a custom quasi-Bayesian splitting criterion, yielding delivering actionable, clinically relevant insights for targeting maternal-health interventions among the most vulnerable subpopulations.

In the traditional Von-Neumann architectures, such as Central Processing Units (CPUs) and Graphics Processing Units (GPUs), the performance gap (about 1000x) between the processor and memory (Memory Wall) has drastically limited the throughput of data-intensive applications. Additionally, the high power consumption of each data transfer (about 40x more than an arithmetic operation) over the memory bus of limited bandwidth (Power Wall), has severely diminished the energy efficiency of CPUs/GPUs. This degraded performance and energy efficiency of the CPUs/GPUs have been exacerbated by the meteoric rise of applications such as Artificial Intelligence (AI), Machine Learning (ML) algorithms, databases, bio-informatics, etc., that often process gigabytes to terabytes of data.

In the last few years, several research proposals and industry prototypes have demonstrated that processing in/near memory (PIM) using Dynamic Random Access Memory (DRAM) is an effective computing paradigm to improve the throughput and energy efficiency of executing data-intensive applications by orders of magnitude as opposed to CPUs and GPUs. Notwithstanding the plethora of proposed PIM architectures, they suffer from challenges such as DRAM intrusion, the lack of flexibility to support multiple applications, an interface with a host processor or a controller, and either the complete absence or availability of a primitive programming and compilation framework.

This dissertation tackles the memory and power bottlenecks in CPUs/GPUs and barriers to the widespread adoption of PIM architectures for edge and server computing systems. It proposes PIM designs based on runtime-configurable Neuron Processing Elements (NPEs) integrated into DRAM without altering standard operation or protocols, while meeting strict area and power constraints. Each NPE, built from programmable artificial neurons (ANs) with local storage, uses a radically different approach to logic design and is therefore significantly smaller than equivalent CMOS designs. NPEs implement multiple operations of varying data formats and precision. Runtime configuration is achieved with minimal overhead. The integration enables a unified platform for various data-intensive workloads, achieving up to a 10x throughput increase and 10–100x energy efficiency gains over CPUs, GPUs, and prior PIM solutions.

Erasing harmful or proprietary concepts from powerful text-to-image generators is an emerging safety requirement, yet current “concept erasure” techniques either collapse image quality, rely on brittle adversarial losses, or demand prohibitive retraining cycles. These shortcomings stem from a myopic view of the denoising trajectories that govern diffusion-based generation. EFlow introduces the first framework to cast concept unlearning as exploration in the space of denoising paths, optimized via GFlowNets with the trajectory-balance objective. By sampling entire trajectories rather than single end states, EFlow learns a stochastic policy that steers generation away from target concepts while preserving the model’s prior. It eliminates the need for handcrafted reward models, generalizes effectively to unseen concepts, and avoids hackable rewards—all while improving overall performance. Extensive empirical results demonstrate that EFlow outperforms existing baselines and achieves an optimal trade-off between concept removal and prior preservation.

In the era of digitalization, governments and public agencies have increasingly leveraged Artificial Intelligence (AI) technologies to enhance the efficiency and effectiveness of public service outcomes. Among various AI-driven tools, Facial Recognition Technology (FRT) has rapidly transitioned from specialized law enforcement applications into a broader spectrum of public service domains, including digital tax services, healthcare, education, and transportation. This swift and expansive deployment aims to streamline administrative processes, improve service accuracy, and achieve higher levels of citizen satisfaction.

However, despite its promising potential, reliance on inadequately tested or unreliable FRT—particularly systems trained on datasets lacking diverse representation—raises critical concerns. Such biased systems can exacerbate existing societal inequities, heightening risks of discrimination against marginalized communities and individuals. Additionally, the proliferation of FRT invites profound debates around privacy invasion, data security vulnerabilities, and the erosion of governmental accountability.

The resulting societal tension poses significant risks, potentially deepening social divisions, undermining democratic processes, jeopardizing fundamental public values, and creating unforeseen negative consequences. Navigating this complexity is foundational to the inquiry undertaken by this dissertation. Notably, while the urgency of addressing these challenges is clear, the current public administration literature suffers from a scarcity of theoretical frameworks and robust empirical data capable of guiding sound policymaking and responsible implementation of FRT in public services.

To bridge this gap, this dissertation systematically investigates the implications of adopting FRT within digitally delivered public services through the lenses of public administration and management. Structured as a three-essay research endeavor, each essay focuses on distinct public service scenarios, employing diverse theoretical frameworks and methodological approaches to critically analyze the societal impacts associated with government-led deployments of FRT.

As FRT use is likely to continue rolling out, the insights from this dissertation contribute significantly to the formation of interdisciplinary theoretical foundations, integrating perspectives from digital government research, information systems, and communication studies. Furthermore, the research offers actionable guidance for policymakers, public administrators, and technology practitioners, facilitating informed decision-making and responsible integration of intrusive surveillance technologies into public service frameworks.