-
Prospective Students
-
Faculty and Staff
-
Current Students
-
Alumni and Friends
Serendipity is a concept associated with accidental and unexpected discoveries that are valuable. In the recent decade, many researchers have advocated serendipity, as part of the “beyond accuracy” metrics, to encourage a recommender system to be an exploratory discovery tool instead of a narrowly focused machine. However, due to serendipity’s elusive and subjective nature, it is challenging to model. Collecting large-scale ground truth data is also a challenge. In this dissertation, I addressed both the challenges of serendipity model construction and the ground truth collection for recommender systems.
Leveraging the recent breakthrough in generative AI and large language models, I utilized three types of generative AI models: Large Language Models (LLMs), Transformers-based cross-domain models, and Diffusion Models (DMs), to construct a serendipity recommendation model. In addition, I used Large Language Models to collect serendipity ground truth data from large-scale e-commerce reviews data. The extensive experiments demonstrated the effectiveness of generative AI in modeling serendipity and ground truth collection. This dissertation advances the understanding and implementation of serendipity in recommendation algorithms, which will empower ordinary people with opportunities of bumping into unexpected but valuable discoveries.
Bulk RNA sequencing (RNA-seq) provides a cost-effective overview of gene expression but lacks resolution to identify cell-type-specific contributions in heterogeneous tissues. Computational deconvolution methods address this by estimating cell-type proportions from bulk data, enabling finer biological insights. This dissertation develops and applies statistical frameworks to improve the accuracy and interpretability of deconvolution results.
We begin by reviewing RNA-seq technologies and the impact of cellular heterogeneity. Deconvolution is then framed as a nonnegative matrix factorization (NMF) problem, with attention to challenges like non-uniqueness and noise sensitivity. Building on recent identifiability theory, we propose a geometric structure-guided NMF (GSNMF) that incorporates biological priors—such as marker genes—and local manifold structure to stabilize estimation.
To further enhance reference-free deconvolution, we introduce pseudo-bulk augmentation: a strategy that synthesizes single-cell-derived mixtures to enrich bulk data. This approach mitigates issues related to underdetermined solutions and improves robustness.
A comprehensive benchmarking study compares reference-based and reference-free methods using metrics like correlation, root mean squared error, and mean absolute deviation. Results show that while high-quality reference data can improve performance, augmented reference-free approaches like GSNMF are highly effective when reference data are scarce. We conclude with future directions and ongoing challenges.
This dissertation takes a liberatory socio-technical approach to explore how technology can empower Black undergraduate women in computing (BWIC) by fostering a sense of belonging, strengthening computing identity, and enhancing self-efficacy. Given the underrepresentation of Black women in STEM, this research challenges traditional approaches by centering their lived experiences and perspectives in the design of inclusive and supportive spaces. Unlike many existing interventions that are designed without direct input from those they aim to support, this study prioritizes active participation, agency, and decision-making among BWIC, ensuring that the solutions created truly reflect their needs. Led by a Black woman in computing, this research builds on personal and collective experiences to develop an authentic and affirming virtual community. Using a multi-modal (Discord) community as a technology-driven intervention, this study examines how virtual spaces can provide a sense of belonging, community, community support, computing identity, peer support, and networking opportunities that help BWIC navigate academic challenges in a predominantly white and male-dominated field. By leveraging qualitative and quantitative methods, this research investigates the ways in which participation in a virtual community influences engagement, confidence, and long-term persistence in computing. Findings reveal that the virtual community plays a critical role in mitigating feelings of belonging, isolation, providing access to a community, peer support, and fostering meaningful connections. While some participants engaged actively in discussions, others participated through passive engagement ("lurking"), yet both forms of involvement contributed to a greater sense of inclusion and identity in computing. Barriers to participation, such as academic workload, social anxiety, outside responsibilities and lack of awareness of available resources, underscore the importance of low-pressure, flexible engagement opportunities and targeted outreach to early-career students. Additionally, the presence of Black women in the community served as a powerful motivator, reinforcing participants' belief in their ability to succeed in computing. This dissertation contributes to computing education by providing evidence-based insights on the role of virtual communities in fostering diversity and inclusion. It highlights the potential of technology-driven interventions to break down systemic barriers and offers practical recommendations for institutions, educators, and policymakers to build sustainable, culturally responsive support networks. By demonstrating the impact of virtual communities in creating empowering and affirming spaces, this research emphasizes the need for intentional efforts to ensure Black women in computing are not just included, but supported, valued, and positioned for success in the field.
Most research on recommendation sources focuses on what happens before consumers make a purchase. This dissertation asks a different question: How do recommender types (AI vs. human) and product framing (utilitarian vs. hedonic) shape how consumers feel after the experience? Using Expectancy-Disconfirmation Theory (EDT) as a framework, this dissertation explores the influence of the variable, post-consumption satisfaction. It also examines whether consumer expertise moderates these effects.
A 2x2 between-subjects experiment (N = 500) tested the impact of recommendation source and product framing using a common stimulus—a jazz clip—framed as either utilitarian or hedonic, and recommended by either an AI or a human. Results revealed that consumer expertise was the most consistent and powerful predictor of post-consumption satisfaction across all conditions. The type of recommender and product framing had little to no direct effect. Of the anticipated effects, only a modest increase in satisfaction for AI-recommended utilitarian products, approached statistical significance.
Contrary to expectations, the post-consumption experience is less about who recommended the product and more about how confident the consumer feels in the category. These findings challenge assumptions about matching recommender type to product type and suggest a shift toward tailoring recommendation strategies based on consumer expertise.
ABSTRACT
Ryan Chester
ATHLETIC IDENTITY AND THE CAREER DEVELOPMENT EXPERIENCES OF DIVISION II BLACK STUDENT ATHLETE BASKETBALL PLAYERS
(Under the direction of Dr. Mark D’Amico)
The purpose of this qualitative study was to explore the lived experiences of Black men and women, NCAA Division II basketball players regarding their athletic identity and career development experiences. The goal of the study was to identify better ways for institutions to equip this population with the support, guidance, and the skill set they will need for their inevitable retirement from their athletic playing careers. The study was guided by three research questions, (1) How do Black student athletes, who compete in men’s and women’s basketball at the Division II level, perceive their athletic identity? (2) What are the career development experiences of Black student athletes who compete in men’s and women’s basketball at the Division II level? (3) What is the potential relationship between athletic identity and the career development experiences of Black student athletes who compete in men’s and women’s basketball at the Division II level? The study gauged participants level of athletic identity using the abbreviated 7-item version of the Athletic Identity Measurement Scale (AIMS) (Brewer & Cornelius, 2001) and utilized the Interpretative Phenomenological Analysis (IPA) approach to analyze the data of eight individual interviews. Findings outlined the importance of exposure for Black student athletes and demonstrated how a student athlete’s professional athletic playing desires can impact their career-related decision making. Findings also explored the specific challenges that Black student athletes face while on their college campus.
Generative Artificial Intelligence (AI) has emerged as a transformative force in marketing, transitioning from traditional technological tools designed to execute predefined tasks to intelligent agents capable of learning, creating, deciding and adapting autonomously. Unlike other technologies, AI possesses the ability to analyse data, generate insights, and autonomously determine optimal courses of action in real time. However, this technological progress is accompanied by a significant gap in understanding AI’s full potential and limitations, leading to widespread misconceptions on how AI agents can be used effectively and ethically. These uncertainties highlight the need for a comprehensive framework to guide Generative AI integration into marketing practices.
The Primary challenge addressed in this dissertation is driven by the lack of a marketing-specific theoretical framework to guide the integration of AI logic into marketing practices. To bridge this gap, this study proposes a novel theoretical framework based on Hunt’s inductive realist approach, specifically designed for AI-driven marketing practices. By positioning AI as both a creative and decision-making agent, the framework highlights the necessity of iterative refinement and ethical alignment to ensure AI applications resonate with societal values and address evolving consumer expectations.
This research adopts a novel two-step methodology, grounded in the indigenous theory development inductive realist approaches to construct an initial theoretical framework for AI in marketing. The approach emphasizes foundational premises and iterative propositions, providing a structured yet adaptable model ideal for addressing the complexities of emerging research domains. Further an empirical study is designed to identify perceptions of AI, uncovering key themes above. Cognitive maps are constructed to visualize the relationships among these themes, providing insights into how they interact and influence marketing outcomes. This empirical analysis designed to further assist theoretical advancements, offering a robust foundation for future research and practice of AI-driven marketing strategies.
T
As heterogeneous parallel architectures grow increasingly complex, achieving high performance and effectively teaching parallel programming have become more challenging. Benchmark suites are powerful tools for illustrating and evaluating optimization techniques in practical, performance-critical scenarios. But commonly used parallel benchmark suites (e.g., SPEC OMP and Rodinia) are primarily designed for performance assessment purposes only. They are not intended for performance optimization training or educational instruction in parallel programming. Furthermore, their complexity in configuration and deployment often limits their accessibility, reducing their practical utility for researchers and students in educational settings. To address these limitations, this dissertation presents two novel benchmark suites, NeoRodinia and CUDAMicroBench, that support not only performance evaluation, but also the exploration of optimization strategies. These suites are further augmented with educational features, such as integration with large language models (LLMs) for optimization guidance and interactive, browser-accessible execution environments.
NeoRodinia features a structured three-level parallelization model (P1, P2, P3) across CPU worksharing, GPU offloading, SIMD, and tasking. It provides standardized execution workflows, automated performance evaluation scripts and visualization tools. Additionally, NeoRodinia integrates AI-assisted analysis, allowing LLMs to offer optimization recommendations and debugging insights. CUDAMicroBench is a modular microbenchmark suite targeting key GPU optimization challenges such as memory hierarchy usage, warp divergence, and concurrent kernel execution, serving as a practical reference for GPU performance tuning.
In addition to benchmark-based contributions, this dissertation advances parallel programming education by introducing the Interactive OpenMP Programming book. By employing deliberate prompt engineering strategies, it effectively leverages large language models (ChatGPT-4, Gemini Pro 1.5, and Claude 3) to enhance the quality, relevance, and pedagogical value of the generated content. Delivered via a Jupyter-based environment, it enables real-time experimentation with OpenMP constructs, promoting hands-on learning and deeper understanding.
Collectively, these contributions form a unified educational infrastructure for modern parallel computing. By combining benchmarking, structured optimization guidance, and LLM-driven interactive learning, this work bridges performance engineering and pedagogy, providing a scalable and adaptable solution for educators and learners in today's heterogeneous HPC landscape.
This study examines the relationship between Enterprise Risk Management (ERM) maturity and firm performance, with an emphasis on how organizational complexity, industry type, and board engagement may moderate this relationship. Although prior research has broadly acknowledged ERM’s role in mitigating risks, few studies explore how ERM maturity contributes directly to firm performance outcomes. Additionally, the literature often treats ERM as a static process rather than a dynamic capability that evolves and adapts within different organizational structures and industry-specific environments. This gap highlights an important opportunity to examine ERM maturity as a strategic asset that enhances decision-making, optimizes resource allocation, and drives sustainable competitive advantage.
Drawing on the Resource-Based View (RBV), Agency Theory, and Contingency Theory, this study addresses these gaps through a quantitative survey of risk professionals from 111 publicly listed companies. By focusing on how ERM maturity impacts firm performance and controlling for variables such as firm size, market volatility, and leverage, this research seeks to provide nuanced insights into ERM's role as a driver of resilience and adaptability. The findings will contribute to the literature by offering a more comprehensive view of ERM maturity’s influence across diverse organizational and industry contexts, ultimately providing practical guidance for companies aiming to refine their risk management capabilities in pursuit of competitive advantage.
Policies and procedures in modern society have unintentionally perpetuated inequities for marginalized students, particularly African Americans, exacerbating math anxiety. Despite efforts to improve mathematics education, African American students continue to experience disproportionately high rates of math anxiety, negatively impacting academic performance, mathematical identity, and future opportunities. While there is extensive research on math anxiety among African American girls, there is a notable lack of studies focusing on African American, elementary aged, boys. Through a qualitative, case study design, the research utilizes a questionnaire, classroom observations, and student interviews to gain insights into the emotional and psychological challenges African American 3rd grade boys face in mathematics. The study investigates the role of race, gender, and socio-economic status in shaping African American 3rd grade boys' experiences with math, as well as the impact of instructional practices and teacher-student relationships on mitigating or exacerbating math anxiety. Findings from this research contribute to the broader conversation about inequities in mathematics education, highlighting the need for culturally responsive teaching strategies and support systems to foster confidence and resilience in students from marginalized communities. This dissertation aims to provide actionable recommendations for educators to recognize, address, and reduce math anxiety in young learners, with a particular focus on African American boys, to promote positive mathematical identity and long-term academic success.
Flooding is one of the most frequent and destructive natural disasters, posing significant risks to communities, infrastructure, and economies worldwide. In North Carolina, diverse flood types—driven by varied topographic and climatic conditions—necessitate adaptable forecasting methods. Traditional flood prediction models rely on hydrological and meteorological data but often struggle to incorporate the complexity of different flood types within a single framework. This research explored the application of machine learning (ML) techniques in flood forecasting, specifically evaluating how different ML models handle varying data requirements across different regions with flood types.
Focusing on the Upper Haw River and Cape Fear River watersheds in North Carolina, this study employed a Convolutional Neural Network (CNN) to predict flood stages using hydrometeorological data, including rainfall, elevation, distance from the river, soil type, land use/cover, wind speed, wind direction, soil water volume, lag feature, and gauge stages. Rather than comparing ML accuracy against traditional hydrological models, this study examined how ML models can adapt to diverse flood conditions and data constraints. The results indicate that CNN-based models effectively capture spatial dependencies and patterns, providing valuable insights into the role of different input features, such as lag effects and rainfall distribution, in flood prediction.
A Command Line Interface (CLI) was developed to enable real-time interaction with the model, enhancing its usability for decision-makers. The study highlighted the strengths and limitations of ML-based forecasting, demonstrating its potential while identifying areas requiring further refinement, such as incorporating additional meteorological variables and real-time data. Aimed to evaluate the feasibility of using ML for statewide flood prediction to encompass a range of flood causes, this research also contributed to determining the boundaries of coastal and piedmont regions and the type of data requirements to develop flood forecasting models in these regions.
