Scholar Commons - Institutional Repository of the University of South Carolina
Not a member yet
34867 research outputs found
Sort by
Hallucinations in Large Foundation Models: Characterization, Quantification, Detection, Avoidance, and Mitigation
No full textDeception is an inherent aspect of social interactions, with research indicating that most people engage in deceptive behavior at least once or twice daily . In parallel, advances in artificial intelligence have led to machines exhibiting deceptive tendencies. These deceptions can be categorized into two types: unintended and intentional. Unintended deceptions - often referred to as hallucinations - occur when generative AI systems produce plausible and convincing narratives yet are factually inaccurate. This phenomenon primarily results from the systems\u27 architectural design, extensive parametric memory, and reliance on statistical assumptions. In this thesis, we provide a comprehensive discussion on the characterization, detection, avoidance, and mitigation of hallucinations. Although recent research has observed early signs of deception, cheating, and self-preservation in top-performing reasoning models, these phenomena fall outside the scope of the current study. The author of The Coming Wave, co-founder of DeepMind, and current CEO of Inflection AI, Mustafa Suleyman, outlines three waves of AI: Wave 1: Classification and training. Wave 2: Generative AI, which creates new data. Wave 3: Interactive AI, where conversations serve as the interface and autonomous bots collaborate behind the scenes. We are currently in Wave 2 - Generative Artificial Intelligence (GenAI), which has profoundly impacted everyday life and accelerated AI development. Recent advancements in GenAI have demonstrated significant accuracy in generating high-quality text, images, videos, and even software code with minimal human intervention. Large Foundation Models (LFMs), like GPT and DALL-E, are accessible to the general public, enabling individuals to efficiently produce high-quality creative content on a large scale. For example, in healthcare, GenAI models help in drug discovery and medical imaging analysis, while in education, they enhance learning by creating adaptive and personalized content, among other applications. Nonetheless, the widespread adoption of GenAI has brought about substantial challenges concerning misinformation, safety, and ethical issues, highlighting the need for regulatory measures to mitigate its impact. A key challenge of LFM lies in its tendency to generate factually inaccurate, logically incoherent, or entirely fabricated outputs while maintaining an appearance of plausibility - a phenomenon referred to as ``hallucination\u27\u27. For instance, earlier last year, Air Canada encountered legal action following an incident in which its AI-powered chatbot provided inaccurate information regarding bereavement travel discounts. Moreover, the Cambridge Dictionary has declared ``hallucinate\u27\u27 as its Word of the Year for 2023. As GenAI systems, including large language and image or video generation models, are widely adopted across industries, hallucinations present a critical obstacle. In an interview with The Verge, Google CEO Sundar Pichai described AI hallucinations as an inherent feature of LLMs, calling it an ``unsolved problem.\u27\u27 In this dissertation, I examine six distinct components to address the challenge of hallucination. (i) Characterization: We developed a first-of-its-kind taxonomy for the systematic classification of hallucinations and introduced a large-scale benchmark called HILT. (ii) Quantification: We introduced novel evaluation metrics, including the Hallucination Vulnerability Index (HVI) and its automated variant HVI_auto, to assess and rank the hallucination of LLMs. We are confident that the dataset and the evaluation metrics will be valuable resources for future researchers studying hallucination behaviors in LLMs and developing effective detection and mitigation strategies. (iii) Detection: We introduced an innovative automated span-based hallucination detection method, referred to as Factual Entailment; this technique achieved a 30% increase in accuracy on the FACTOID benchmark compared to state-of-the-art TE methods. (iv) Avoidance: We introduced a new prompting technique, termed ``Sorry, Come Again?\u27\u27 (SCA), to avoid hallucinations through prompt analysis. [PAUSE] injection technique slows LLM generation to enhance comprehension. Using optimal paraphrasing combined with LDA improves performance in both the Number and Time categories. (v) Mitigation: We introduced radiant, Retrieval-Augmented entIty-context AligNmenT, a paradigm that combines RAG with alignment principles, enhancing the interaction between retrieved evidence and the model’s internal representations. (vi) Multi-modal: Lastly, we constructed a similar taxonomy of hallucinations and datasets, called VHILT and ViBe, for both (a) Image-to-Text and (b) Text-to-Video modalities, along with a preliminary analysis. These datasets will benefit researchers in the community by supporting further research. Overall, this dissertation offers a concrete approach to evaluating the content generated by language models and addressing hallucinations across all modalities
Supporting the College Athlete Amidst the Onset of Name, Image, and Likeness
No full textAthletic administrators often assume a distinct stakeholder role that has the most immediate and frequent impact on college athletes’ well-being and retention. The implementation of name, image, and likeness (NIL) represents a historic transformation in college athletics. Utilizing a sport development framework, particularly the retention stage, to examine elite athlete development, how athletic administrators socially support college athletes in the new context of NIL and its impact on administrators’ work experiences were empirically assessed. Semi-structured interviews were conducted with 18 athletic administrators at nine universities. The themes of Implementation Challenges, Lack of Uniformity, Holistic Benefits to Athlete Well-Being, and Mostly Unrealized Negative Outcomes identified best practices to supporting and retaining college athletes. Several practical recommendations are presented, focused primarily on the well-being benefits athletes have realized through NIL and the long-term impact NIL may have on athletic department personnel. For a more nuanced understanding of NIL, outcomes beyond financial gain must be recognized to better support holistic athlete development
Exploring Cultural Capital and Literacy Practices in Chinese Immigrant Families: A Critical Ethnographic Case Study
No full textThis dissertation explores the cultural capital and literacy practices of Chinese immigrant families residing in a Southeastern U.S. city. Using an ethnographic case study approach, the research examines two groups of Chinese immigrant parents: those with higher educational backgrounds and those who are restaurant owners. The study investigates how parents’ educational and cultural backgrounds influence their perceptions of and commitments to their children’s education, as well as their home literacy practices. It also explores their understanding of the model minority stereotype and their views on educational equity in the U.S. The findings reveal that both groups of parents value education and support their children’s learning but differ in their approaches to educational involvement and literacy practices. These differences are shaped by their distinct cultural models, reflecting their unique experiences and perspectives. The study concludes by emphasizing the importance of recognizing and understanding the diversity within Chinese immigrant families and advocating for culturally responsive educational practices that support their children’s development and academic success
Methods and Applications for Bayesian Semiparametric Survival Analysis
No full textSurvival analysis is a cornerstone of biomedical and clinical research and plays an important role in fields as diverse as engineering, actuarial science, and sociology. In this dissertation, we develop new semiparametric Bayesian methodology for three problems from survival analysis: 1) adjustment for treatment crossover in randomized controlled trials (RCTs), 2) multilevel modeling of clustered survival outcomes when the cluster size is also informative, and 3) divide-and-conquer Bayesian inference for massive survival data. Our methods are semiparametric in the sense that we assume the covariates have a linear effect with regard to the log-hazard or the log- survival time; however, we make minimal assumptions about the baseline hazard or the residual error distribution of the log-survival time.
In Chapter 2, we present a unified three-state model (TSM) framework for evalu- ating treatment effects in clinical trials in the presence of treatment crossover. Treat- ment crossover occurs when patients switch from their randomly assigned treatment to a different trial treatment. Researchers have proposed diverse methodologies to es- timate the treatment effect that would have hypothetically been observed if treatment crossover had not occurred. Our proposed TSM framework unifies existing methods, effectively identifying potential biases, model assumptions, and inherent limitations for each method. The TSM framework also facilitates the creation of new methods to adjust for confounding effects from treatment crossover. To illustrate this capability, iii we introduce a new Bayesian imputation method that falls under its scope. Using a piecewise constant prior for the hazard, our proposed method directly estimates the hazard function with increased flexibility.
In Chapter 3, we introduce a Bayesian method to jointly model the time-to- event outcome and cluster size for multilevel survival data. We use Dirichlet process mixtures (DPMs) for both the cluster size model and the survival model, allow- ing us to infer covariate effects and predict survival probabilities without imposing strong assumptions about the underlying data distributions. The two models are joined through a shared random effect, thus capturing potential informative cluster size (ICS). ICS arises when the cluster sizes are not independent of the measured outcomes. To the best of our knowledge, this is the first Bayesian semiparametric method for handling ICS.
In Chapter 4, we propose new distributed inference methods for Bayesian survival analysis based on piecewise exponential (PWE) models. To alleviate the computa- tional burden of fitting these models when sample size is large, we propose to use divide-and-conquer (D&C) Markov chain Monte Carlo (MCMC). We consider D&C MCMC for both the PWE model with independent observations and the mixed effects PWE model for clustered observations. For clustered survival data, we propose a novel data partitioning scheme to handle imbalanced cluster sizes. Our D&C MCMC algorithms greatly reduce the cost of running MCMC on large survival datasets, thus facilitating more efficient posterior inference
Naturally Creative: A Qualitative Study of Adolescent Literacy in an Outdoor Classroom
No full textThis qualitative action research study examines how teens respond to literacy lessons taking place outdoors in a wooded setting. Specifically, the study investigates the influence of outdoor lessons on students’ writing practices, their beliefs about writing and their writers’ identities, and their attitudes toward the natural world. The study features responses from 26 high school students enrolled in standard English language arts (ELA) classes and whose literacy abilities have been labeled in deficit terms because of past failures to meet standardized benchmarks of performance and behavior. Participants took part in nine outdoor lessons that each followed the same format: a shared reading of a piece of literary text; an artifact prompt, which asked students to complete a physical task in the wooded setting; and an open-ended writing prompt that allowed for creative responses in multimodal and multigenre formats. Findings revealed that students’ writing practices expanded while outside, with special growth in students’ storytelling skills, imaginative writing, and conscious rhetorical choices. Student responses also indicated growing affective connections to the natural world. Examined through the lens of embodied literacies and artifactual literacies, and informed by research from outdoor education scholars, the study shows the potential for the development of outdoor classrooms that are conducive to authentic writing experiences through sensory engagement, especially for marginalized populations, while also serving to counter the negative effects of an overly standardized ELA curriculum. Findings also point to the value of carefully selected mentor texts; the inclusion of literary texts in the secondary ELA classroom; and freedom to move and talk while creating
Blackmore\u27s Prince Arthur
No full textSir Richard Blackmore, a court physician under William III, was a prolific and controversial writer. His first published poem in English was the heroick epic Prince Arthur, which was hugely popular in his own day and went through 3 editions in 2 years. Today, however, no modern edition of Prince Arthur exists, as the last edition was published in 1714. This new critical edition presents a reading text of the poem and Blackmore’s Preface based on the first 3 editions. To aid the modern reader, an introduction gives a brief biography of Blackmore and situates the poem within both its poetic and political contexts, as well as laying bare the allegory celebrating William III that lies at the heart of the poem
Analyzing Factors Affecting Retention and Attrition Rates of Teachers in the Pee Dee Region of South Carolina
No full textThe purpose of this qualitative action research study was to determine the factors and variables that affected the retention and attrition rates of novice and seasoned teachers in the Pee Dee region of South Carolina. Utilizing three frameworks, the Self-efficacy theory, the Social Identity theory, and the Multidimensional Theory of Burnout, the researcher conducted a virtual survey of 33 respondents and interviewed seven participants to collect data. Data was thematically coded to discover commonalities amongst the participant responses.
The data revealed administrative influence, teacher working conditions such as autonomy, time, and control, and workload were the most frequent elements of responses and discussion. In correlation with the frameworks, the study also concluded that teachers surveyed and interviewed possess an innate vocational trait that allows them to overcome challenges and struggles of the profession. They posited that they remain in their position for the love of teaching and the overall well-being and growth of their student
Awareness and Practice Methods for Musicians with Invisible Disabilities: Insights Gained from Playing the Flute with Ehlers-Danlos Syndrome
No full textEhlers-Danlos Syndrome (EDS) is a connective tissue disorder that presents significant physical and mental challenges for musicians, particularly in instrumental performance. These challenges may include chronic pain, joint instability, fatigue, anxiety, and depression. While such symptoms are associated with EDS, they are not exclusive to it and are also common among individuals with other invisible disabilities, such as fibromyalgia, postural orthostatic tachycardia syndrome (POTS), celiac disease, attention-deficit/hyperactivity disorder (ADHD), generalized anxiety disorder (GAD), and chronic fatigue syndrome (ME/CFS). This study explores the experiences of flutists with EDS and other invisible disabilities, focusing on the coping strategies and adaptations they use in their practice routines. A qualitative methodology, including semi-structured interviews and autoethnography, is used to capture individual experiences and reflect on the author’s personal journey as a flutist with EDS. The research aims to develop adaptive practices to support musicians with disabilities, inform music educators, and promote inclusivity in music education and performance. The findings will be analyzed thematically to identify key strategies and insights, contributing to a more supportive environment for musicians with invisible disabilities
An Evolving Concept of Well-Being in the Digital Era: Digital Well-Being in Tourism and Hospitality Experiences
No full textIn the context of increasing digitalization across the tourism and hospitality sector, digital technologies have become essential components shaping the contemporary tourist experience. While these technologies offer enhanced convenience, personalization, and connectivity, they also pose risks such as cognitive overload, distraction, and digital fatigue. Despite growing concerns, the concept of digital wellbeing remains underexplored. This dissertation addresses this gap by conceptualizing digital well-being, identifying its dimensions, and developing a validated measurement scale tailored to the tourism and hospitality settings.
By employing a multi-phase research design, this study includes a systematic literature review, a three-round Delphi study involving tourism and hospitality experts, and empirical scale development and validation. Study 1 conceptualizes digital wellbeing in the tourism and hospitality context and highlights its dynamic and multifaceted nature. Study 2 establishes a consensus-based definition and identifies key domains, indicators, facilitators, and barriers through expert input. Study 3 develops and validates a multidimensional scale encompassing eight dimensions: temporal control, hedonicutilitarian value, physical-cognitive strain, social connectivity, destination connection, self-discovery, usability and accessibility, and safety and security. The final measurement scale demonstrates strong reliability and validity across multiple samples. Further empirical testing confirms the nomological validity of the model by showing that digital literacy significantly enhances digital well-being, which in turn positively influences tourists’ satisfaction with technology usage during travel. This mediating role positions digital well-being within a broader relational network, highlighting its potential as a central mechanism linking various antecedents to technology-related travel experiences and outcomes.
This dissertation contributes to the well-being, psychology, and tourism and hospitality literature by establishing a context-specific conceptualization of digital wellbeing and bridging the gap between abstract theory and empirical application. It also provides practical implications for technology developers, tourism service providers, and policymakers. These include adopting wellness-oriented design principles, implementing digital well-being assessment tools, and promoting balanced digital engagement strategies. By fostering healthier and more meaningful digital experiences, this dissertation contributes to the sustainable development of tourism destinations and the broader well-being of tourists in the digital age
Episode 90: Mr. South Carolina: Walter Edgar
No full textFor nearly 40 years, Walter Edgar taught the history of South Carolina as a member of USC\u27s history department faculty and continues to illuminate the history of the Palmetto State on his podcast, Walter Edgar\u27s Journal. In this conversation, he remembers his early days as a graduate student and young faculty member at the university.https://scholarcommons.sc.edu/rememberingthedays/1091/thumbnail.jp