33760 research outputs found
Sort by
Investigating Conversational Access to Historical Content
No full textSince the inception of writing systems, people have left behind written traces of their lives. Modern media systems further extend these traces to include spoken and digital records, raising the possibility that future systems could reconstruct aspects of a person’s life. This dissertation supports that vision by investigating conversational interaction with representations of historical figures that are grounded in historical content, commonly referred to as virtual immortality. This dissertation leverages recent advances in retrieval methods for conversational search, large language models (LLMs), and prompting techniques to propose a conceptual framework for a conversational agent, comprising three stages: retrieval, contextualization, and style transfer. Rather than building a full end-to-end system, the dissertation begins with an investigation into each component individually.
First, foundational techniques for conversational search are developed through participation in TREC CAsT 2021, using web-scale document collections. Experiments demonstrate that automatic query rewriting improves retrieval performance and user-perceived utility. Second, the work examines a domain-specific setting using a curated Ronald Reagan collection---diaries, interview transcripts, and public papers. A retrieval-based prototype for single-turn interactions supports semi-structured interviews with experts from libraries, archives, and museums (LAM). The study reveals implications for immersive experiences, archival assurance, and engagement of inquiring visitors. To help bridge this gap, the third part of the dissertation focuses on text rewriting. Specifically, the task of contextualization aims to generate clause-level elaborations for uncontextualized mentions. Reagan’s letters are used as input due to their brevity, and human annotations are collected to investigate system performance. Results show that prompting LLMs with detailed task instructions and few-shot examples can achieve near-human performance, confirmed by manual inspection. Finally, the dissertation investigates text style transfer---rewriting historical content to match a conversational style reflective of Reagan’s public representation. This is framed as a data-driven task, in which LLMs are prompted using comparable examples drawn from texts with topical overlap. Experiments across four sets of paired corpora show that single-turn, rather than chain-of-thought prompting, combined with short comparable examples, yields the best results. Automatic measures for entailment align well with human assessment of content preservation, while style classifiers struggle to capture the subtleties of stylistic variation when used to assess style strength. In sum, this dissertation offers an empirical investigation into components that could support conversational interaction with historical figures. It concludes with a discussion of the opportunities and limitations of such systems and aims to inspire future interdisciplinary work with born-digital historical collections
Navigating linguistic complexity in acquisition: From syntactic dependencies to referring expressions
No full textThis dissertation explores how young learners navigate complex linguistic structures and form generalizations under two challenging conditions: sparse input and unclear input. I focus on two case studies, the comprehension of instrumental "with"-questions and the production of definite "the"-expressions.
The first case study examines how 15-month-olds comprehend instrumental wh-questions with a stranded preposition "with" ("What did she break the stick with?"). I ask whether infants’ resolution of these prepositional object gaps in wh-questions will fail due to usage factors—infrequent question types and verb-based expectations—or if they will succeed via detecting a missing argument, signaled by the complement-selecting preposition "with". Despite input sparsity and unfavorable verb expectations, infants responded correctly to these questions, suggesting that they are pragmatically prompt in responding to wh-questions before fully acquiring their full syntax.
The second case study concerns the acquisition of definite NPs. Naturalistic data suggests children’s referential choices aligned closely with their mothers’, with low miscommunication and high guessability. An interactive elicited production task shows that even 3- to 4-year-olds adapted their referring expressions appropriately, indicating adult-like knowledge of "the". Their reported overuse of "the" in prior production studies likely reflects task artifacts. Longitudinal video corpora reveal that anaphoric cues ("a ball...the ball") in input to 14-month-olds were available but inconsistent, whereas situational cues were widely available. Crucially, input with variable uses at 14 months did not mislead learners’ later use of "the" at 20 months, implying that they integrate both anaphoric and situational cues in recognizing the familiarity presupposition of "the".
Altogether, this dissertation contributes to our understanding of how language learning unfolds as a dynamic, inferentially rich process. Young learners are remarkably ingenious in generating structure and meaning from limited input at the very beginning. Furthermore, this work highlights the role of methodology in revealing linguistic competence in early development
Institutional Agents with Marginalized Identities Working With Student Activists
No full textThis study explores the experiences of student affairs professionals at large public research universities when directly engaging with student activists with whom they share marginalized identities. For the purpose of this study, these student affairs educators are institutional agents while they “provide key forms of social and institutional support” (Stanton-Salazar, 2011, p. 1066) navigating the Borderlands (Anzaldúa, 1987) created by their role within the institution, support of student activists, and their own social identities. I frame this study through the following research question: How do institutional agents experience their commitment to the institution as an employee, while supporting student activists with whom they shared marginalized identities? This study also addresses these sub-questions: (a) How do student affairs educators navigate the political context of their university amid student activism? (b) With social identities in mind, how do student affairs educators manage their relationships with student activists and colleagues? Five student affairs educators participated in 60 to 75-minute interviews and produced images to describe their experiences when engaging with student activists. I present findings in three themes: (a) how participants negotiate being part of the institution; (b) how participants navigate the borderlands; and (c) how institutional agents are part of the community. Across the three themes, several key takeaways addressed how student affairs educators: (a) address power dynamics, (b) do not assume the politics of their colleagues, (c) define the labor done on behalf of the university, (d) are a bridge for student activists, (e) cultivate co-conspirators, (f) manage a sliding scale of transparency, (g) set boundaries when working with student activists, and (h) build trust with student activists. This study contributes to the body of literature on how student affairs educators with marginalized identities support student activists and manage their roles. This dissertation concludes with implications for research and practice, with key takeaways highlighted by participants’ own words
Taking Root: Balancing Socotra's Sustainable Development Through Agro-Tourism
No full textPost-pandemic tourism has surged, leading to over-tourism, where popular destinations areexperiencing social, economic, and infrastructural displacements. This resistance, as well as the
desire for more meaningful travel, has shifted tourists’ desired destinations to those that offer
unique cultural experiences and natural beauty, such as the relatively underexplored Socotra, Yemen. While this shift in tourism benefits developing destinations that rely on tourism for their economic generation, like Socotra, it also poses a risk to their social and natural habitats. Unregulated tourism growth in developing regions like the Middle East and North Africa (MENA) threatens natural resources, culture, and local communities. Socotra, Yemen, faces these challenges as its tourism interest grows, particularly due to its designation as a UNESCO World Heritage Site and its endemic biodiversity. A balanced approach is necessary to ensure the island’s sustainable development, addressing both pre-tourism efforts, such as enhancing Socotran hospitality skills, and post-tourism strategies, like preserving its natural conservation zones. Additionally, Socotra’s development model has the potential to set a global standard for sustainable tourism, not only for its provided solutions towards tourism’s current wicked problems but also for its transitional desert climate’s potential to incorporate climate-responsive design, ethical material sourcing, and thermal comfort strategies within its architectural design. This framework can develop a measurable system in which developing nations in similar cultural and natural climates can balance their tourism benefits with ecological, economic, and social
sustainability, in a replicable design that can be modified depending on each nation’s
programmatic demand
The Price of Fairness in Algorithmic Decision-Making
No full textAlgorithms play a fundamental role in modern decision-making, impacting economic structures, cultural landscapes, and resource allocation at unprecedented scales. While optimization techniques often prioritize efficiency—finding solutions quickly with minimal resources—many real-world applications require an additional consideration: fairness. This thesis explores the intricate trade-offs between efficiency and fairness in algorithm design, focusing on two key do-mains: fair division and fair hierarchical clustering.
In the first part, I examine the fair division problem, which seeks to distribute resources among stakeholders in an equitable and efficient manner. I introduce novel approaches for balancing fairness constraints with computational feasibility, investigating both offline and online settings. In offline fair division, I analyze classical notions such as envy-free allocations and extend them to more general settings, including weighted agent priorities and multigraph valuation models. In the online setting, where decisions must be made without full knowledge of future arrivals, I present new algorithms for fair resource allocation, including solutions for the online Santa Claus problem and class-fair bipartite matching.
The second part of this thesis focuses on fair hierarchical clustering, a problem central to machine learning and data analysis. Many clustering algorithms reinforce biases in data representation, disproportionately impacting marginalized groups. I develop novel fair clustering algorithms that balance demographic representation while preserving computational efficiency.My work introduces the first explainable algorithm for fair hierarchical clustering and provides near-optimal polylogarithmic approximation guarantees, significantly improving upon prior results.
By leveraging techniques from combinatorial optimization, game theory, and online algorithms, this thesis provides theoretical insights and practical methodologies for designing algorithms that uphold fairness constraints without sacrificing efficiency. These contributions have broad implications for resource allocation, machine learning, and algorithmic fairness in large-scale systems
DEVELOPMENT OF MODELING AND OPTIMIZATION METHODOLOGIES TO FACILITATE TRANSITION OF HEAT PUMP AND REFRIGERATION SYSTEMS TO LOWER GWP REFRIGERANTS
No full textIn light of Kigali Amendment, Fluorinated Gas (F-Gas) regulations, and the United States Environmental Protection Agency (EPA) SNAP program calling for the phase out of hydrofluorocarbon (HFC) refrigerants which have an extremely high global warming potential (GWP), it is necessary to transition to lower GWP alternatives. For example, natural refrigerants such as hydrocarbons (HC, e.g., propane (R290), isobutane (R600a), etc.) and carbon dioxide (CO2) have gained traction as attractive low GWP alternatives for a wide range of residential and commercial heating, ventilation, air-conditioning and refrigeration (HVAC&R) applications which have historically depended on high-GWP refrigerants such as R134a, R410A, R22, and R404A. Moreover, researchers worldwide have also shown immense interest in utilizing refrigerant blends to achieve the required performance whilst minimizing environmental impact. One such area is supermarket refrigeration systems, which often employ multi-stage cascade cycle configurations due to large temperature lifts, making single-stage vapor compression (VC) systems impractical. A key challenge in this field is selecting environmentally-friendly refrigerants, as many common low-temperature refrigerants have high GWP, high ozone depletion potential (ODP), and/or are flammable. The use of HC refrigerants is also highly regulated due to their flammable nature, hence necessitating high-performance heat exchangers (HXs) with significant size, weight, cost, and refrigerant charge reductions compared to current state-of-the-art to comply with flammable refrigerant charge limits.Decarbonization of buildings is essential to reducing greenhouse gas emissions, as space and water heating account for a substantial share of energy-related emissions. Transition to heat pump technologies is central to this effort, offering a low-emission alternative to fossil fuel and electric resistance heating. They can substantially reduce greenhouse gas emissions from space and water heating when powered by renewable or low-carbon electricity, making them a key technology for building electrification and climate mitigation goals.
This thesis focuses on the development of modeling and optimization methodologies to facilitate the transition of heat pump and refrigeration systems to lower GWP refrigerants and is applied to three different case studies: (i) refrigerant blend optimization and selection for a supermarket refrigeration system, (ii) a water source heat pump system, and (iii) low-charge air-to-refrigerant HXs for a residential air-conditioning system.
In the first part of the thesis, an approach was developed to design optimal refrigerant blends to serve as lower-GWP alternatives to conventional high-GWP refrigerants while also enhancing system efficiency for a two-stage cascade refrigeration system. Significant improvements in COP by up to 49% and decreases in GWP (< 50) were observed when the optimal blends were incorporated into the system.
In the second part of the thesis, experimental validation of a water source heat pump system model utilizing R32 as the refrigerant and a brazed plate heat exchanger (BPHX) as the condenser was conducted for a steady-state VC system simulation platform to demonstrate the platform’s capability to accurately model and simulate VC systems with lower-GWP refrigerants and plate heat exchangers.
In the final part, a comprehensive system-level optimization methodology was developed for air-to-refrigerant tube-fin and finless non-round tube HXs which is capable of designing optimal HX pairs (i.e., condenser and evaporator) with minimal refrigerant charge to replace the baseline HXs in an air-to-R410A air-conditioning system to facilitate the transition to lower GWP refrigerants R32 and R290. System-level simulations with the optimal HX pairs suggest that the optimal HX pairs have comparable thermal-hydraulic performance to the baseline system with significant reduction in HX-level charge (> 70%) , thereby supporting the adoption of lower-GWP natural refrigerants such as R290
Study of Atmospheric Polarization with Ground-Based Robotic Hyperspectral Measurements and HARP2 Data Analysis
No full textReport completed at the end of the 2024 CISESS Summer Internship Program.Atmospheric polarization offers insight into air quality monitoring and satellite image quality, as it can impact polarization sensitivity and lead to image striping. This study evaluates polarization patterns produced by atmospheric scattering by enhancing and deploying a ground-based Robotic Hyperspectral Polarization Measurement System (RHPMS) in preparation for the calibration and validation of satellite observations. The measurement system, designed by NOAA/STAR and integrated by the University of Maryland's Cooperative Institute for Satellite Earth System Studies (CISESS), is composed of a telescope, polarimeter, spectrometer, optical fiber, and Raspberry Pi controller. Using Python to visualize and analyze hyperspectral data, the study examined polarization across wavelengths under clear-sky conditions. These ground-based findings were compared with satellite observations from the HARP2 polarimeter and Ocean Color Instrument (OCI) aboard NASA's PACE mission. The findings support the improvement of aerosol characterization and quantification for accurate atmospheric and climate research.This study was supported by NOAA grant NA19NES4320002 (Cooperative Institute for Satellite Earth System Studies-CISESS) at the University of Maryland/ESSIC
PROTECTOR OF THE REFORMED: OLIVER CROMWELL’S RELIGIOPOLITICAL MOTIVES AND THE DEVELOPMENT OF PROTECTORATE IDEOLOGY
No full textAligning AI with Human Values: A Path Towards Trustworthy Machine Learning Systems
No full textMachine learning has become a powerful tool for harnessing vast amounts of data across diverse applications. However, as artificial intelligence (AI) technologies advance and become more deeply integrated into daily life, they also introduce risks such as malicious exploitation, misinformation, and unfair decision-making, which can undermine their reliability and ethical integrity. Given AI’s growing influence, ensuring that these systems are trustworthy and aligned with human values is essential for their responsible and safe deployment. To address these challenges, this dissertation investigates trustworthiness across the AI pipeline, focusing on training-time vulnerabilities, inference-time robustness and alignment, and the long-term impacts of decision-making models. At the training stage, it examines how manipulated training data can compromise vision-language models, facilitating the spread of coherent misinformation. At the inference stage, it develops methods to enhance adversarial robustness in image classifiers and align frozen language models with human values at test time through reward guidance. For the long-term impact, it formulates fairness in sequential decision-making and proposes strategies to mitigate bias accumulation over time. Together, this dissertation aims to provide a holistic framework for improving AI reliability, safety, and fairness, fostering more trustworthy and responsible AI deployment
The Art of Invention: Antonio Canova's Bozzetti
No full textThe Italian sculptor Antonio Canova (1757-1822) modeled sketches in terracotta and wax, known as bozzetti, that serve as unique and invaluable witnesses to his creative process. In stark contrast to his exquisite marble sculptures, those palpitating sketches reveal Canova’s complex method of invention, which can be understood as both a mental and physical activity. Long dismissed as mere studio props, the bozzetti are, in fact, crucial to understanding Canova’s significance at the intersection of classical ideals and the pursuit of modern inventiveness. Challenging the myth of spontaneous creation, my dissertation argues that Canova’s sculptural sketches are legitimate works of art in their own right. To achieve this, it is crucial to examine the artistic purpose of the bozzetti and reveal the method or rather the methodical research that Canova conducted in clay. This process entails deconstructing the phases of invention into a succession of deliberate choices operated directly in the material. Additionally, my dissertation establishes distinct philosophical frameworks for evaluating their aesthetic value—frameworks that diverge from the Kantian model applied to the marble sculptures, which proves inadequate for the bozzetti. Drawing on the works of Christian Wolff (1679-1754), Alexander Gottlieb Baumgarten (1714-1762), François Hemsterhuis (1721-1790), and Johann Gottfried Herder (1744-1803), I examine their relevance to a revised aesthetics of the bozzetti. Finally, tracing the singular historiography of the bozzetti, distinct from that of the marble sculptures, will reveal how they played a pivotal role in revitalizing the modernity of the neoclassical sculptor in the mid-twentieth century. Instead of being viewed as ancillary products of the finished marble sculptures and spontaneously created forms, the bozzetti deserve to be examined for their distinctive artistic, aesthetic and historiographic significance