33760 research outputs found
Sort by
Constructions of Gender, Race, and Sexuality in Nineteenth-Century Bound Sheet Music Collections
No full textThis thesis examines constructions of gender, race, and sexuality in sixteen bound sheet music collections (binder’s volumes) held in the Jacklin Bolton Stopp Collection on Nineteenth-Century Music at the University of Maryland’s Special Collections in the Performing Arts. I provide a literature review and overview of relevant theoretical and methodological frameworks; address the importance of extrafamilial and intergenerational relationships in binder’s volumes, as well as memorialization and playfulness, all under the auspices of gender; examine constructions of race as exemplified through these books’ inclusion of blackface minstrelsy and songs linked to abolitionism; and detail constructions of both homosexuality and heterosexuality through music selection and biographical details. Drawing on feminist, speculative historical, microhistorical, and archival methodologies, I argue that bound sheet music collections provided opportunities for girls and young women both to subvert and affirm the white supremacist heteropatriarchal expectations of the nineteenth-century United States. This thesis contributes to literature on amateur music-making and, more specifically, binder’s volumes, expanding current scholarship by introducing speculative history and queer theory to this subfield
A Tunable Optical Centrifuge for Spectroscopy and Collisional Dynamics of Superrotor Molecules
No full textThe work presented in this dissertation utilizes a tunable optical centrifuge to study the spectroscopy and collisional dynamics of N2O and CO2 superrotor molecules. High-resolution transient IR absorption spectroscopy is used for time-dependent, polarization-sensitive, and rotational state-resolved measurements of superrotor molecules. An optical centrifuge combines ultrafast laser pulses with opposite chirp and opposite circular polarization to create a linear optical field polarization that angularly accelerates thermal molecules into extreme rotational states. The maximum rotational state populated by the optical centrifuge is related to the maximum angular frequency of the optical trap. We present a tunable optical centrifuge, which controls the final rotational state population distribution of the optical centrifuge via a systematic reduction in the spectral bandwidth of the optical pulse. In this way, the rotational population can be controlled down to a specific maximum rotational quantum state J. The studies presented here are a mixture of high-resolution spectroscopy measurements and polarization-sensitive collisional dynamics measurements of high-J N2O and CO2. First, we measure new N2O IR transition frequencies with J≤ 200 in the R-branch and J≤ 155 in the P-branch of the (0001) ← (0000) vibrational transition. New, more accurate spectral fitting parameters are reported and a combination difference analysis reveals the effects of high angular momentum on bond stretching. Next, the new N2O spectroscopy is used to measure the collisional dynamics of J= 67-180 with time-dependent and polarization-sensitive IR absorption signals. Here time-dependent measurements were performed to determine relative populations, molecular alignments and rotation-to-translation energy transfer pathways of high-J N2O. Next, we discuss measurements of new IR transitions of CO2 with J= 186-282. A spectral perturbation was discovered near J= 220. Collisional dynamics studies were then completed in the region of J= 244-282. Time-dependent and polarization-sensitive measurements of CO2 molecules in this region were performed to determine relative populations, molecular alignments and rotation-to-translation energy transfer pathways. Finally, two optical centrifuge traps were characterized to test the effects of the optical trap spectral bandwidth on the collisional relaxation dynamics of CO2 superrotors
HOLOGRAM, HEADING INTO THE LIGHT: EXPLORING EMBODIMENT, IDENTITY, AND DIGITAL MOTION CAPTURE
No full textAs digital technologies increasingly shape perception, interaction, and representation, dance artists are uniquely positioned to interrogate their impact on embodied experience. Hologram, my MFA thesis project, investigates how motion capture tools such as Rokoko and Unity mediate the expressive capacity of the body and influence the construction of identity in performance. This interdisciplinary research draws on rehearsal-based inquiry, digital experimentation, and somatic practice to examine the feedback system I call the Body-Machine Loop: a dynamic in which digital tools both discipline and are shaped by the bodies that use them.The process began with the collection of motion capture data from 35 dancers across various geographies, forming a digital archive of improvisatory material. Collaborating with a cast of four performers, we mined this archive using miscalibration, improvisation, and kinesthetic empathy to generate both live and digital choreography. The resulting performance explored glitch aesthetics, avatar logic, labor and reliability within digital systems, and the porous boundary between physical and virtual selves.
The written thesis contextualizes the performance across five chapters: Chapter 1 introduces key definitions and methods; Chapter 2 frames the theoretical groundwork for digital embodiment and the Body-Machine Loop; Chapter 3 analyzes the construction and performance of Hologram; Chapter 4 documents technical strategies, failures, and workarounds; and Chapter 5 extends the research to issues of surveillance, labor, and digital materiality. Ultimately, this project argues for the choreographic and critical potential of glitch, the liveness of digital bodies, and dance as a methodology for investigating the sociopolitical logics embedded in contemporary technologies
Automated metrology for additively manufactured parts using deep learning and computer vision
No full textThis paper presents a comprehensive, fully automated pipeline for dimension analysis of machine parts from 2D image data. The pipeline is designed to process high-resolution grayscale scans, leveraging advanced computer vision and deep learning techniques to extract and summarize part dimensions. The first stage employs the Line Segment Detector (LSD) to identify linear features, filter irrelevant segments, and calculate distances between parallel lines. The second stage utilizes a CNN-based Ellipse Detector (ElDet) to detect and measure elliptical and circular features, including concentric ellipses, through a recursive detection strategy for improved accuracy. The proposed method significantly enhances measurement speed and automation over traditional computer vision-based techniques while maintaining accuracy, efficiently handling both simple and complex geometries while reducing the need for manual analysis. Extensive evaluation on real and synthetic datasets highlights the pipeline's robustness and effectiveness in industrial applications.https://doi.org/10.1117/12.306250
Towards Reliable Agentic LLMs
No full textIn recent years, rapid advancements in large language models (LLMs) have steadily shifted their applications from simple chatbots to increasingly complex, autonomous agents. Agentic applications require LLMs to interact with a broad range of external information sources, tools, and environments to solve intricate tasks with minimal human oversight—posing significant challenges to their reliability. This dissertation presents a series of contributions toward (more) reliable agentic LLMs.
Firstly, we explore how LLMs can be made more robust when incorporating external references—an essential capability for many agentic applications. We introduce chain-of-defensive-thought, a simple yet effective technique that instructs LLMs to generate a chain of thought mimicking a structured reasoning process of cross-checking. This highly accessible approach significantly improves the robustness of a wide range of LLMs against reference corruption. Importantly, it highlights a promising direction: exploiting the reasoning abilities of LLMs for robustness on tasks that are not necessarily reasoning-centric, which is a timely insight given the growing interest in LLM reasoning and the increasing reliability demands of agentic applications.
Secondly, we examine the reliability of tool use in agentic LLMs. While external tools can dramatically extend the capabilities of LLMs, the current paradigm—where models choose tools based solely on text descriptions—proves fragile. We demonstrate how strategic edits to tool descriptions can substantially bias tool usage, revealing a vulnerability in standard tool/function-calling protocols. These findings underscore the need for a grounded mechanism for agentic LLMs to select and utilize tools and resources.
Finally, we address the reliability of LLM evaluations, particularly in the presence of test set contamination, where models may (knowingly or not) train on test data prior to evaluation. We propose DyePack, a novel framework that repurposes backdoor techniques into a principled mechanism for identifying such contamination. DyePack operates without requiring access to model internals and supports both multiple-choice and open-ended tasks. More importantly, it provides provable guarantees by enabling exact false positive rate (FPR) computation before flagging any model as contaminated—effectively preventing false accusations while offering strong evidence for every case detected. This positions DyePack as a powerful tool for maintaining the integrity of open benchmarks and safeguarding our pathway toward reliable agentic LLMs
Differential Diagnosis of Disordered Language in African American English-Speaking Children: A Comparison Between Computerized Black English Sentence Scoring and Developmental Sentence Scoring
No full textBackground: Children who speak African American English (AAE) are disproportionately likely to be diagnosed with language disorders. Traditional language sample analysis (LSA) metrics, such as Developmental Sentence Scoring (DSS), are based on the morphosyntactic structures of Mainstream American English (MAE) and may not accurately reflect the language abilities of AAE-speaking children. This study examined the effectiveness of computerized Black English Sentence Scoring (BESS) system in distinguishing between TD and LI groups in AAE-speaking children compared to use of DSS. Method: Language samples from 88 children (22 LI, 66 TD) between the ages of 5;0 and 7;02, including 44 AAE-speaking children and 44 MAE-speaking children as the control group, were analyzed using Computerized Language Analysis (CLAN) DSS and BESS options. Results: Findings suggested BESS is a culturally responsive LSA metric for AAE-speaking children, minimizing linguistic bias seen in DSS. Results revealed neither DSS or BESS are statistically reliable ways to identify LI in children who speak either AAE or MAE. Discussion: Clinical ramifications and future directions are discussed
Privy Pits and Lost Toys: Hauntology and Contemporary Archaeology in Bassett, Virginia
No full textMany archaeologists experience a certain level of connection with the past when interacting with sites and artifact assemblages. Personal artifacts such as toys can be particularly captivating. This thesis is interested in the quotidian, unrecorded aspects of a twentieth century domestic site with multiple dwellings occupied by textile factory employees and their families and a hauntological interpretation of that site. How do this site, its artifacts, and its past haunt us?Site 44HR0219 was excavated in two survey efforts during 2021 and 2022. The focus of this survey was the early twentieth century housing constructed by the Bassett-Walker textile factory for its employees and their families. Each house sat on piers and had a private privy. Some blocks of houses shared communal yard space. Sanborn insurance maps indicate that the houses encompassed by the site occupied a single street block, with dwellings facing the streets on all side and with communal yard space at the center of the ring of structures. Privies were located at the rear of each house, near the edges of the communal yard space. The houses represented in this site were occupied between the 1920s and 1970s, after which they were demolished to make space for additional parking for the textile factory. The major feature type observed during Phase II-level survey was privy pits, containing primarily personal materials. Features like this – intact, sealed, and dense in artifacts – can teach us much about the private lives of the employees and their families, particularly the parts that are not commonly recorded by companies or government censuses.
Records from the early twentieth century – for example, insurance maps, company records, census records, and oral histories – are often available to a certain extent. There is a local historical society that keeps extensive archives. Many houses from the same construction event are still extant and occupied across the street from the site. The data and dwellings impart a certain sense that we already know what there is to know about the people who occupied this site. Despite this knowledge, sites like this one can be interpreted as haunted (metaphorically). The following thesis uses hauntology as a theoretical framework for discussion of 44HR0219, in order to fully understand how the Smith River Site and the intimate artifacts recovered there impact us in the present
EFFECTS OF RED CLOVER LIVING MULCH ON WEEDS, INSECT PESTS, BENEFICIALS, AND CANTALOUPE YIELD
No full textVegetative diversity in agroecosystems can reduce weed and insect pests, enhance beneficial arthropods, and improve biological control. Field studies in 2023 and 2024 investigated the synergistic use of a living mulch with pesticide inputs compared to plasticulture cantaloupe. Whole plot treatments included a living mulch or plasticulture, while subplots included standard or reduced pesticides. Plasticulture had more weeds than living mulch each growing season respectively. Standard pesticides improved weed suppression for living mulches in 2023. Striped cucumber beetles and aphids were suppressed in living mulch compared to plasticulture. Natural enemies were more abundant in living mulch than plasticulture, and some were greater in reduced pesticide subplots. Predation was higher in living mulch than plasticulture in 2023. However, yield was reduced in living mulch compared to plasticulture. Living mulches manage weed and insect pests but may require mowing or chemical suppression to reduce competition with cantaloupe
Essays on Health, Environment, and Public Policy
No full textIn this dissertation, I examine the impact of public policies on health and well-being across three different contexts. My research explores how elements of the physical and policy environment influence health and educational outcomes.
In Chapter 1, I study a policy that reduced air pollution and its effects on cognitive performance among schoolchildren. Using detailed administrative data covering the universe of public school students in Texas and North Carolina, I examine if reducing pollution from diesel-powered school buses affects student academic outcomes. To do so, I leverage variation from the installation of emissions control devices called retrofits, of which over 9,000 were installed between 2004-2014. I find that installing retrofits increased math and reading test scores by 0.02 standard deviations for the average retrofitting district. By scaling these estimates by the fraction of the bus fleet retrofitted, I estimate that improving air quality by this margin for all school bus riders would raise district-average test scores by 0.05 standard deviations, which is equivalent to an improvement in teacher quality of half a standard deviation. Treatment effects are stronger for students who live farther from school and grow with years of exposure. I also compare installing retrofits with other investments studied in the education literature, such as reduced class sizes or intensive tutoring programs. I show that retrofitting buses was a very cost-effective investment in terms of gains in academic achievement per dollar spent. These results suggest that school bus retrofits were an effective intervention that had substantive benefits beyond improvements in health. I close with a discussion on how these results can be extended to the transition to electric school buses.
In Chapter 2, I study the relationship between income and health in the context of tax policy targeted at low-income mothers. While the correlation between income and health has been well established, the causal relationship is not well understood. In this chapter, I estimate the long-run effects of the Earned Income Tax Credit (EITC) on outcomes for low-income mothers in the United States. Using variation from federal and state expansions over three decades, I examine a range of health and economic outcomes observed up to 15 years after becoming a parent. In general, I find little evidence of long-run broad-based effects of EITC benefits for adults, in contrast to previously documented positive long-run effects for children. I find suggestive evidence of long-run mental health improvements, but estimates are imprecise. I also find some evidence of long-run improvements in physical health and economic outcomes for Black mothers. Despite fade-out of short-run effects of the EITC, improvements in maternal health may still be an important mechanism for how safety net programs create long-lasting benefits for children.
In Chapter 3, Michel Boudreaux and I study a statewide health policy on initiation of contraceptive methods. Delaware Contraceptive Access Now (DelCAN) was a statewide initiative implemented in 2015-2019, which executed a coordinated bundle of reforms to improve access to all types of contraception, particularly long-acting reversible contraceptives (LARCs). While most program components targeted publicly subsidized patients such as Medicaid and Title X, some components were applicable to all patients, such as clinical training and a public awareness campaign. In this study, which was published in PLOS ONE, we examine the impact of DelCAN on LARC initiation among privately insured women using a national commercial claims database and a difference-in-differences design. We estimate that DelCAN increased the probability of a LARC insertion among privately insured women in Delaware age 15-44 by 0.4 percentage points in an average year, a 13% increase from baseline. We also estimate that DelCAN increased LARC adoption in the immediate postpartum setting by 0.3 percentage points per year, which is a large increase relative to baseline. Treatment effects were stronger among individuals under age 30. Our study population had free access to all types of contraceptives via the ACA mandate, but our results suggest that clinical training and public awareness play important roles in ensuring access to the full range of contraceptives. These results help inform current and future policymakers in states engaged in similar contraceptive access policies
THE IMPACT OF STATE ALCOHOL EXCLUSION LAWS ON ALCOHOL-RELATED MORTALITY AND TREATMENT UTILIZATION
No full textAlcohol is one of the leading risk factors for preventable death and disability in the United States. It is a toxic substance that affects behavioral, physical, and cognitive functions of the body directly or indirectly. One of the federal and state efforts to intervene to reduce alcohol-related injuries and accidents includes the Alcohol Exclusion Law (AEL). Under AEL, insurance companies may refuse to pay healthcare providers for treating the injured while impaired by alcohol or any other drug that was not prescribed by a physician at the time of injury.
My dissertation investigated the changes in alcohol-induced mortality and alcohol treatment utilization in the states that repealed AELs, compared with the changes for the same time in the states that did not repeal. I collected data from the Centers for Disease Control and Prevention Wide-ranging Online Data for Epidemiologic Research (CDC WONDER), Treatment Episodes Data Set-Admissions (TEDS-A) from the Substance Abuse and Mental Health Services Administration, American Community Survey data from the U.S. Census Bureau, and the National Institute on Alcohol Abuse and Alcoholism. Combining these datasets, I conducted the panel data analysis using Two-Way Fixed Effects to examine the impact of AEL repeal.
The first aim examined the impact of the repeal of AEL on alcohol-induced mortality rates, and the death rate decreased after the ban of AEL. The second aim evaluated the effect of the repeal of AEL on the admission rates to treatment facilities for those with alcohol problems, especially for racial/ethnic minority groups across different referral sources. The results suggest that the alcohol treatment admission rate among Hispanic individuals referred by criminal justice systems might have increased following the ban of AEL. Also, the repeal increased the alcohol admission rate among Hispanic individuals with individual referrals. The third aim examined the impact of the repeal on the treatment admission rates among males, females, and pregnant women. The evidence does not suggest that the repeal affected the alcohol admission rates of any of these subgroups. The findings provide policymakers with insights into how racial/ethnic minoritized populations were affected by the AELs, who are vulnerable to alcohol use