116964 research outputs found
Sort by
Adjoint-based data assimilation in a stochastically parametrized Lorenz 96 two-scale system
Full text linkThe success of stochastic parametrization in weather forecasting and subgrid-scale parametrization of ocean eddies shows great promise for its use in modeling the climate of the ocean and atmosphere. These kinds of parametrizations have yet to be employed, however, in full-scale ocean state estimation. I demonstrate the use of stochastic parametrization within a data assimilation context using adjoints by investigating these ideas in a simple toy model, the Lorenz 96 two-scale system (L96).
L96 is used as a proxy for highly resolved ocean dynamics, while a truncated version is used to compare the effectiveness of deterministic and stochastic parametrization schemes in analogy with subgrid-scale parametrization. I show that, in addition to deterministic models, stochastically parametrized models can be tuned via the adjoint method to more closely match the dynamics of a highly resolved model. The stochastic model shows additional promise by capturing a more realistic variability across frequencies. This plays an important role in uncertainty quantification, as shown by comparing an ensemble of stochastic model runs with a deterministic perturbed-parameter ensemble. This demonstrates significant promise for both the practicality and effectiveness of stochastic parametrization in ocean state estimation.Physic
Development of a radioisotope production facility using a cryogenic cooling system
Full text linkThis dissertation details the development of a Cryogenic Tracer Irradiation Facility (CTIF) at The University of Texas at Austin (UT). Radioisotopes are an important tool in the world of nuclear explosion monitoring, as they allow researchers to test detection equipment and simulate underground nuclear explosions before the real thing occurs. At UT, gaseous radioisotopes are typically produced using in-core irradiation procedures that take up time, are expensive, restrict research access, leave equipment contaminated and unable to be used for long periods of time, and their temporary nature poses an increased risk of accidentally releasing the radionuclides. One solution is a beam port based irradiation system, which is less intrusive on reactor operations but has a lower neutron flux than in-core based systems. To compensate for this, a liquid helium compressor is used to freeze a target gaseous isotope and increase the reaction rate to comparable levels with the in-core irradiation. Initial tests for this facility were promising, but equipment failure and incorrect flux measurements towards the end of the experiment resulted in only 0.043 Ci of ¹²⁷Xe being produced after a roughly one month irradiation despite the aim being 2 Ci. Issues with gas leaks and water pump failures were the expected culprit, and were fixed along with a change in methodology to measure short term irradiations using natural xenon. After a sample of natural xenon was irradiated for an hour at 50 kW of reactor power and counted using an high purity germanium (HPGe), a neutron flux and activation rate were determined. The neutron flux was calculated to be 5.13 ∗ 10¹¹ ± 3.94% neutrons/cm²/s for a reactor power of 900 kW. Assuming a one hour irradiation at the same reactor power with one liter of ¹²⁶Xe, the CTIF produced 3.62 ∗ 10⁷ ± 3.94% Bq of ¹²⁷Xe at the end of irradiation. In 2023, an irradiation of ¹²⁶Xe using the in-core facilities produced an activity of 3.81 ∗ 10⁸ ± 3.3% Bq under the same conditions and had an average flux of 1.57 ∗ 10¹³ ± 3.82% neutrons/cm²/s. The production rate for xenon isotopes in the cryogenic irradiation facility in the reactor beam port is found to be approximately 10% of the reaction rate in the previously demonstrated in-core irradiation facility. The cryogenic facility will be an excellent candidate for producing radioxenon isotopes on the order of 100 mCi (3.7x10⁹ Bq), but will not match the production rate of the in-core production method. Placing such a facility in a beam port with a higher flux provides a promising path forward.Mechanical Engineerin
Influence of adverse childhood experiences and protective and compensatory experiences on teacher absenteeism
Full text linkTeacher absenteeism continues to present a significant challenge for school leaders and policymakers, particularly in educational settings that serve high-need student populations. This study aimed to examine how adverse childhood experiences (ACEs), protective and compensatory experiences (PACEs), and resilience might influence patterns of teacher absenteeism in a public school district that serves economically disadvantaged communities. This research was grounded in Bolman and Deal’s four-frame model of organizational leadership as a multidimensional lens for analyzing complex issues within school systems.
A quantitative case study design was employed using a structured survey instrument distributed to teachers within the district. The survey gathered self-reported data from 340 teachers on exposure to ACEs, the presence of resilience-building experiences of PACEs, current levels of teacher resilience, and reported frequencies of professional absences. To strengthen the validity of the findings and reduce limitations associated with self-report bias, I incorporated archival absenteeism records provided by the district’s administrative database. This cross-verification allowed for a more accurate depiction of actual attendance patterns and increased the reliability of the analysis. Descriptive statistics were used to summarize the central tendencies within the data, and nonparametric correlational analyses were conducted to explore the relationships between variables measuring ACE, PACE, and resilience and teachers’ self-reported absenteeism. The results demonstrated statistically significant associations between higher ACE scores and increased absenteeism. In contrast, higher PACE scores were positively associated with greater resilience and lower reported absence rates. These findings suggest that a comprehensive understanding of teacher absenteeism requires attention not only to institutional or organizational factors but also to the developmental experiences that educators carry with them. The findings emphasize the importance of trauma-informed leadership approaches and the implementation of staff support systems that are both preventative and responsive in nature. The recommendations and implications might ultimately contribute to improved teacher retention and sustained engagement in classrooms that serve the most vulnerable student populations.Educational Leadership and Polic
On flexible and efficient new econometric approaches for choice data analysis
Full text linkThe theory and application of choice models have progressed by leaps and bounds in several directions over the last two to three decades. This dissertation adds to this expanding body of literature by contributing to, and elevating, the state-of-the-art in choice modeling through the introduction of more flexible model structures, newer dependent variable forms, efficient estimation procedures, and contextual empirical applications. Our first focus is on multivariate dependent outcome models. There is growing interest in multivariate mixed dependent outcome models that include a mixture of different kinds of discrete and continuous variables. Within the transportation literature, factorization-based approaches have received substantial attention in accounting for the jointness of such mixed dependent outcomes. These factorization approaches allow for a parsimonious estimation of the covariance matrix through the use of latent factors. However, there are two important assumptions in all such earlier mixed dependent outcome models: (i) marginal normality of unobserved factors (latent constructs) that generate jointness among the main outcome variables of interest, and (ii) independence between the unobserved factors (latent constructs) and the propensity equations underlying the main outcomes of interest. In our current effort, we simultaneously relax both these assumptions and develop a flexible Generalized Heterogeneous Data Model (GHDM) for mixed data modeling. To do so, we use the Yeo-Johnson (YJ) transformation (which allows for a single-parameter non-normal to normal transformation) for the unobserved factors (latent constructs) to relax the first assumption, while explicitly allowing these latent constructs to be correlated with the main outcomes (thus, relaxing the second assumption). We demonstrate an application of our proposed model in the context of individuals’ high-density residential neighborhood living choices and monthly bicycling frequency. Next, we individually extend the techniques behind each of the two relaxations undertaken in the previous study in unique directions. First, as an extension of the YJ transformation to multinomial discrete choice models, we propose the use of this transformation to account for non-normal random parameters in a multinomial probit (MNP) setting. Our proposed model, which we call the MNP-YJ model, provides a systematic implicit-Gaussian-copula-based framework to account for non-normality and dependencies across the random coefficients. We utilize this new multinomial model framework to investigate factors impacting commute mode choice in a developing economy. Second, following the flexible GHDM dependency approach, we extend the framework to the consideration of a rank-ordered variable among the main outcomes. Rank-based preference surveys can be exploited to achieve a certain desired precision in choice model estimation with a much smaller sample size, making ranked data surveys more cost-effective than traditional first-choice surveys. From a substantive standpoint, we contribute to the study of land use-transportation relationships by investigating residential location effects on auto-ownership levels and rank-based mode preferences of individuals, within a hypothetical futuristic autonomous vehicle landscape. Owing to the advantage in the efficiency of using a rank-based approach (as opposed to the first-choice approach) in a multinomial choice context, we further focus our efforts on the use of a rank-ordered dependent variable in a spatial model that accounts for the stochastic dependence in choice-making across individuals. Specifically, we propose a methodological framework for a spatial rank-ordered probit model (SROP) that accommodates both spatial lag effects as well as spatial drift effects. An application of the proposed model is demonstrated in a travel mode choice ranking experiment among seven alternatives, including autonomous vehicle (AV) private ride-hailing and AV pooled ride-hailing.Civil, Architectural, and Environmental Engineerin
Sources and fate of terrestrial organic matter in Arctic coastal waters
No full textLarge quantities of terrestrially-derived organic matter (OM) are delivered to the Arctic Ocean. This terrestrial OM is an energy subsidy for heterotrophs, supporting secondary marine production, and, if remineralized, releases inorganic nutrients that support primary production. Additionally, this OM may be decomposed to produce greenhouse gases like carbon dioxide and methane, impacting the global carbon cycle. While the land-to-ocean fluxes and bioavailability of riverine OM are relatively well constrained, much less is known about OM from Arctic coastal erosion or coastal groundwater. This work provides novel information about the composition and biodegradability of OM in eroding soils and coastal groundwater along the Alaska Beaufort Sea coast. We found that the geochemical composition of eroding permafrost varies from tundra surface to sea level, related to both the landscape geomorphology and processing within permafrost. Using experiments to measuring carbon dioxide production from bulk soils and permafrost in seawater, as well as biodegradability of dissolved organic carbon (DOC) leached from those soils in seawater, we found that marine-derived OM in late-Pleistocene age permafrost was more bioavailable compared to terrestrial and/or lacustrine OM in Holocene age active layer soils and permafrost. Pairing these experimental data with shoreline loss rates and soil carbon stocks, we show that coastal erosion supplies notable quantities of bioavailable organic carbon to the Alaska Beaufort Sea. We expect erosion to be relatively important during summer when inputs of bioavailable OM from rivers are low. We also compared the composition and biodegradability of DOC in supra-permafrost groundwater to surface waters. Despite being highly aromatic, groundwater DOC was more biodegradable than river DOC. In both groundwater and rivers, biodegradable DOC declined from spring break-up to fall freeze-up. Given the high concentrations of DOC in coastal groundwater, groundwater is potentially an important source of bioavailable OM to nearshore Arctic waters, fueling marine productivity and/or greenhouse gas emissions. These results are particularly relevant as the Arctic is rapidly warming, likely altering the mobilization of terrestrial OM to the Arctic Ocean.Marine Scienc
Artist Diploma Recital (soprano)
No full text10 son I'umile ancella from Adriana Lecouvreur / Cilea, Francesco -- 3 Sonetti del Petrarca, S.270 (Pace non trovo, Benedetto sia 'I giorno, I vidi in terra) / Liszt, Franz -- 3 Browning Songs, op.44 (The Year's at the spring, Ah, love, but a day! I send my heart up to thee!) / Beach, Amy -- 2 Chinese folk songs (Yearning and love, Swan goose) -- Good-Bye! / Tosti, PaoloMusicName of supervisor not provided
Evolution of the voice : vocal production in frogs, bats, and humans
Full text linkThe three chapters of this dissertation examine the evolution of the voice from the four perspectives originally laid out by Nikolaas Tinbergen: ontogeny, mechanism, phylogeny, and adaptive function. Chapter 1 focuses on the development of the larynx in the túngara frog (Physalaemus pustulosus), a neotropical amphibian. By tuning mathematical models with micro-CT-based measurements taken from an ontogenetic series of larynges, we show that male túngara frogs could likely vocalize during a “silent phase” in their development, suggesting that their muteness is facultative and not physically enforced. This result helps to illuminate morphological and behavioral consequences of an interplay of natural and sexual selection. Chapter 2 analyzes morphology, scaling, and constraint in the vocal apparatus of neotropical bats. We used micro-CT imaging and bioacoustics analysis to investigate larynx morphology and allometric scaling in eleven species of neotropical bats across six families. Within the sampled taxa, call acoustics and larynx morphology were strongly constrained by body size. Laryngeal cartilage mineralization patterns appeared to be more determined by phylogenetic structure than taxon-specific differences in echolocation call structure. We also failed to find a systematic relationship between larynx morphology, call structure, and the acquisition of vocal production learning. Overall, our results suggested that strong physical and phylogenetic constraints act on the form and acoustic output of the chiropteran larynx, despite its specializations. In Chapter 3, we measured the acoustic space occupied by human speech, non-linguistic, and musical vocalizations along with the calls of chimpanzees, bonobos, and chacma baboons. We use Mel-frequency cepstral coefficients to create an acoustic space depicting the spectro-temporal features of over 750,000 brief vocal segments. Speech and song occupied significantly less acoustic-feature volume than human non-linguistic vocalizations. In addition, the volumes of speech and song were not statistically different from those of non-human primates. These results suggest that speech was not enabled by an expansion of human vocal range. Anatomical shifts unique to humans may have led to an elaboration of non-linguistic utterances, but learned vocalizations use a surprisingly small fraction of this space.Ecology, Evolution and Behavio
Visions of migration : news images and the production of visual knowledge on the U.S. southern border
Full text linkMigration is a natural and ever-present human process that affects nearly every single country in the contemporary globalized world. Migration is also the most consistently politicized, denigrated and demonized social process in many societies. In the United States, migration exists as a political and social conflict that produces prejudices, stokes nationalistic fervor and influences acts of violence. Without direct experience, much of the knowledge of migration in American life is constructed by news media and especially news images, which are positioned as documentary evidence of these social processes. But news images are not natural, disinterested reflections of reality, they are motivated symbolic constructions that have the capacity to influence material conditions for migrating people. Photographs influence subsequent social practices through their ability to reflect discourses while obscuring their origins; they have the ability to harm and the ability to heal. Through poststructural theories of representation, this dissertation explores the nature of the visual construction of migration in the U.S. news industry. This dissertation employs visual analyses, interviews and ethnographic observation on the U.S.-Mexico border to illuminate the complex social practice of visual news work, the conventions of representation and their knowledge potential. The visual construction of migration is a product of competing interests that influence how photographers work in the field. Multiple stakeholders apply physical, social and temporal pressures, which act as discursive and embodied barriers that divide photographers and migrants during the photographic encounter. This ultimately results in visual conventions such as a priority on moments of enforcement, asymmetrical power relations between symbols of authority and migrating people, and a separation between the viewer and depicted migrants. These conventionalized images reflect beneficial discourses for powerful entities’ distinct agendas—an example of hegemony in action. Though photographers are often bound by these influences, they still retain the ability to change these representations. When conflicts fracture this hegemonic influence, photographers can break conventions and make images that don’t benefit the most powerful stakeholders, but benefit the migrants themselves.Journalism and Medi
Towards robust deep neural networks via synthesizing task-driven adversarial scenarios
Full text linkDeep neural networks (DNNs) have become indispensable in autonomous systems such as robotics and self-driving vehicles, enabling advanced perception, forecasting, and control. Yet their reliability is often undermined by vulnerabilities to out-of-distribution (OoD) inputs and adversarial perturbations. Current robustness methods are predominantly task-agnostic, applying random augmentations or adversarial examples that improve vision models in isolation but may not affect downstream decision-making. As a result, autonomous systems remain susceptible to control-critical failures. This dissertation develops task-specific and game-theoretic approaches to enhance robustness and generalization across three key domains: perception, forecasting, and black-box control. First, I propose a framework for synthesizing adversarial visual scenarios explicitly tailored to multi-step, model-based robotic control. Leveraging the end-to-end differentiability of perception–planning–MPC pipelines and generative models, the method perturbs latent scene parameters in ways that directly degrade control cost. Retraining on these adversarial scenarios improves control robustness on OoD data by up to 36.2%, significantly outperforming conventional task-agnostic data augmentation. This result highlights the importance of aligning robustness interventions with the ultimate control objective. Second, I introduce a game-theoretic framework for robust time-series forecasting in dynamic environments. In this two-player, zero-sum formulation, a forecaster predicts future trajectories while an adversary strategically perturbs historical inputs to maximize control cost. Training against such adversaries compels the forecaster to learn policies that minimize cost under worst-case perturbations. This approach, which converges to local Nash equilibria, yields a 30.1% improvement on OoD real-world lane-change data relative to baseline forecasters. The framework demonstrates how game theory can be used to systematically model and counteract distributional shifts in safety-critical decision-making. Finally, to address settings where gradient-based methods are infeasible, I develop a Bayesian optimization framework for finding local saddle points in two-player zero-sum black-box games. By constructing Gaussian process surrogates and embedding them within a bilevel optimization loop, the method identifies local Nash points using only zeroth-order samples. Strategic sampling of the true objective refines the surrogate and enables efficient discovery of equilibria in nonconvex–nonconcave landscapes. Applied to robust MPC, this approach achieves a 27.6% reduction in mean control cost on OoD data compared to nominal controllers. Together, these contributions form a comprehensive framework for task-aware robustness in machine learning systems. By bridging adversarial scenario synthesis, game-theoretic forecasting, and black-box optimization, this work advances both theoretical understanding and practical tools for ensuring safe, reliable operation of AI in uncertain, real-world environments.Electrical and Computer Engineerin
The influence of social organization on sperm competition in Verreaux's sifaka (Propithecus verreauxi) and the ontogeny of sexual size monomorphism in Verreaux's sifaka (Propithecus verreauxi)
Full text linkThis master’s report is a compilation of two papers. Each abstract is presented below. Chapter 1: Sexual selection predicts that males invest in sperm competition when females mate with multiple males. Because group composition facilitates access to mates, I hypothesized that variation in within-group mate competition influences investment in sperm competition. In a population of wild Verreaux’s sifaka (Propithecus verreauxi), which live in both single and multimale groups, I tested the predictions that (1) dominant males living in multimale groups have larger testes than males in single-male groups, and (2) within multimale groups, dominant males have larger testes than subordinates. Consistent with expectations, I found that dominant males in multimale groups had significantly larger testes mass than males in single-male groups. Within multimale groups, dominant males had significantly larger testes mass than subordinate males. These results demonstrate that (1) males flexibly respond to the current level of within-group sperm competition, and (2) through dominance, within-group contest competition impacts male investment in sperm competition. Chapter 2: Due to variation in life history trajectories, adult body size and age at physical maturation have different determinants and implications for male and female primates. Unlike many other primates, Verreaux’s sifaka (Propithecus verreauxi) exhibit sexual size monomorphism at adulthood, but it is unknown if male and female sifaka experience the same developmental trajectory to reach this same adult body size. I examined the growth rate and duration of 14 morphometric variables collected from a population of wild Verreaux’s sifaka at Kirindy Mitea National Park in Madagascar between 2006 and 2019. Females exhibited longer duration and rate of growth in several skeletal measurements, while males exhibited longer duration and rate of growth in muscular measurements. Adult females had significantly longer hindlimb length than adult males, while adult males had significantly larger thigh circumference, bicep circumference, and a higher intermembral index than adult females. Because most of the differences are only apparent and significant in adulthood, I suggest that these adaptations are tailored for adult behaviors such as infant carrying and intrasexual competition, and that the path to adulthood varies greatly depending on sex and life history.Anthropolog