116964 research outputs found
Sort by
The application of satellite radar interferometry to the study of future storm surge risks along the Gulf Coast
Get PDFCommunities in low-lying coastal regions are particularly vulnerable to hurricane-induced storm surges, and the overall flooding risk can be further elevated by climate changes and human activities. For example, land-cover changes, such as deforestation and wetland loss, can reduce surface roughness, which represents the ability of land surfaces to dissipate storm surge energy and slow down the flood water flow. Land subsidence, caused by groundwater pumping, hydrocarbon extraction, and wetland degradation, can accelerate relative sea level rise and further exacerbate coastal land loss. Mapping these surface property changes over the coastal areas is urgently needed to analyze future storm surge hazards, determine who lives in high-risk areas, and develop associated management plans. Recent advancements in satellite imaging radar techniques provide potential cost-effective solutions to these data collection needs. In particular, the goal of this Ph.D. work is to advance Synthetic Aperture Radar (SAR) and Interferometric Synthetic Aperture Radar (InSAR) techniques for the application of future storm risk analysis. In this dissertation, we show that a combined use of SAR amplitude, InSAR phase correlation, and InSAR phase decorrelation measurements can classify land-cover types with distinct surface roughness, a key input parameter required in storm surge models. The Houston and New Orleans surface roughness maps estimated from ALOS PALSAR data are consistent with surface roughness maps derived from optical sensors and in-situ measurements. Our algorithm only requires 0.3% of the radar pixels as training samples, and it can greatly increase the temporal resolution of the existing land-cover database. Furthermore, we present robust InSAR processing strategies that can effectively mitigate severe decorrelation noise in a large volume of InSAR data, which can achieve about 2 mm/yr accuracy in surface deformation estimation over traditionally challenging vegetated terrain. Our results reveal widespread subsidence that was previously undetected over coastal wetlands and less-populated communities. Using ADCIRC model predictions, we show that subtle subsidence features with a magnitude of millimeter-to-centimeter per year can substantially elevate future storm surge risks. In the New Orleans case, an average cumulative subsidence of 11 cm by 2100 could cause an additional 3047 km² of inundation, comparable to the impact of 91 cm absolute sea level rise. We demonstrate that the relationship between coastal subsidence and flood depth increase is complex, such that land subsidence (or land uplift) can reduce (or increase) the flood depth in certain areas. We found that white and high-income communities are at higher flooding risks if the observed land subsidence persists in the Ike case. By contrast, black and low-income communities are particularly vulnerable to flooding caused by future land subsidence in the Katrina case. Our flooding risk analysis framework can assimilate a broad range of remote sensing, field, and socio-economic data to inform decision-making.Aerospace Engineerin
Indigeneity, settler colonialism and public health : Xinka medicine as a practice of resurgence in southeast Guatemala
Get PDFThe Xinka medicine system is a fundamental aspect of Xinka identity. This dissertation centers Xinka onto-epistemologies of healing in examining everyday healing care as a form of resurgence that actively creates and maintains Indigenous health sovereignty. Healing care is enacted by a range of Xinka medicine practitioners (XMP), including midwives, acupressure specialists, herbalists, and communal feminists who view healing as a cosmo-political path. These onto-epistemologies of healing describe a profound and complex system of knowledge used to maintain the health and well-being of Xinka individuals, families, and communities. Furthermore, this care is fundamental to the broader Xinka cultural and identity revitalization movement and the current resistance against extractive industries in Xinka territory. This analysis draws from ethnographic and community research in southeast Guatemala between 2017 and 2020. Including interviews with Xinka and Maya elders, healers, and leaders. This work contributes to the visibility of the Xinka Pueblo’s cosmogonic and political thought and knowledge of medicine, health, and healing. This work centers aspects of the current and historic experience of the Xinka and, thinking about Indigenous futures, considers how Xinka onto-epistemologies can contribute to liberation and building a “good life” for Indigenous Pueblos in Iximulew.Latin American Studie
Modeling framework for systems analysis of the petrochemicals industry using network models
Get PDFRecent technological advancements in hydraulic fracturing and horizontal drilling have lead to a substantial increase in the production of crude oil and natural gas in the United States. Unconventional oil and gas sources, such as shale formation are now economically favorable to exploit, and thus, shale gas has dominated natural gas production. In contrast to conventional sources, shale gas typically contains a high concentration of natural gas liquids (NGLs), which have to be removed to produce consumer-grade, dry natural gas. Therefore, the rapid increase in shale gas production has lead to an abundance of low-cost NGLs available for use, either directly as fuels or primarily as feedstock for petrochemicals manufacturing. The increasing production of shale gas, and consequently of natural gas liquids (NGLs), provides unique opportunities to expand the U.S. chemical industry, leading to questions about how to best use these resources. It is, thus, essential to assess any change before it is implemented in order to make prudent decisions, especially considering the magnitude of capital investment needed to make such changes or the potential disruptions that may occur. A common approach for determining the competitiveness of a new process relative to existing processes that generate the same product, or for comparing alternative new processes with one another, is to employ a techno-economic analysis. Despite being effective in comparing different processes on a consistent basis, this approach does not account for follow-on effects that may be caused by inserting a new process into the existing industrial ecosystem. These effects can often be substantial and could significantly alter the conclusions of simple techno-economic comparisons when taken into consideration. A systems approach that considers the entire petrochemicals industry is thus needed. For that reason, the overreaching goal of this dissertation is to solve the problem of optimally designing the petrochemical industrial network by developing and implementing a modeling framework that can determine the optimal industry configuration, i.e. the optimal technology selection and utilization as well as material flows, under one (or more) objective(s) and supply and demand limitations. In Part I, I present a new decision-making framework for analyzing the chemical manufacturing and refining industry in the United States by developing and implementing network models of the industry. First, I introduce the motivation for modeling the industry with network models and explain why they can effectively represent the system. Then, I provide an extensive literature review, followed by the presentation of linear network models, a relatively easy model to develop and solve that has been the go-to approach in literature. A comprehensive linear programming (LP) formulation of these models is shown, along with a discussion regarding their limitations. To deal with such limitations, I propose a new type of network models with variable-costs, along with an efficient way of solving such models, which are significantly more computationally complex. This renders the problem nonlinear and potentially discontinuous. I presented a nominal way to formulate the problem as a mixed-integer, nonlinear program (MINLP), which is significantly harder to solve. For this reason, I also propose a decomposition scheme as an efficient alternative to dealing with this problem, based on successive linear programming (SLP) and a cost-propagation algorithm. All the different models are compared with each other in terms of accuracy and computational efficiency, first on prototype networks, and then on industry-wide applications. In Part II, I implement the framework to analyze and assess competing pathways to utilize newly available NLG resources, particularly ethane and propane, the two biggest NGL components. First, I evaluate the impact of a potential, new catalytic dehydrogenation technology for converting ethane to ethylene, and then I evaluate potential, new catalytic oligomerization processes for converting ethylene to 1-butylene and to 1-octene. For each new technology evaluated, I determine the production level of the technology in the optimal industry network. By doing this over a wide range of net process cost points, a maximum adoption cost (the net process cost beyond which the technology would not be adopted into the optimal configuration) can be identified, and its sensitivity to the assumed product yield determined for each new technology studied. Scenarios in which the ethane supply is constrained to current values, and in which it is unconstrained, are considered. Similar studies are also performed for propane, where a new catalytic dehydrogenation process for converting propane to propylene was assessed and compared to propane exports. In Part III, I expand the framework to include more than one objectives, by developing and using multi-objective network models of the petrochemical industry that minimize both production cost and carbon loss across the industry. The model is first solved for each objective separately, and then the weighted sum and ϵ-constraint methods were employed to obtain the Pareto-optimal set of solutions. Moving along the Pareto front towards the minimum carbon loss point causes a number of structural changes in the industry network, which are presented and discussed.Chemical Engineerin
Does use of digital technologies help or harm cognition among older adults?
No full textDoes use of technology affect cognition among older adults? This meta-analysis found use of digital technology was linked to reduced risk of cognitive impairment and rates of cognitive decline over time. Associations remained significant when accounting for other factors.Population Research Cente
Master's thesis recital (piano)
No full text4 unidentified works.MusicName of supervisor not provide
Identification of the effector phenotypes triggered by engagement of FcγRIIb in B cells
Get PDFThe interaction between Immunoglobulin G (IgG) and Fc gamma receptors (FcγRs) on immune cells plays a pivotal role in triggering a complex array of pro- and anti-inflammatory effects, which are central to the mechanisms of action of therapeutic antibodies. Despite the development of more than 30 Fc-engineered antibodies tailored for therapeutic purposes, our understanding of how engagement with FcRs influences the behavior of different immune cells remains far from complete. Notably, the effector phenotype resulting from the clustering of FcγRIIb, a receptor known for its distinct inhibitory role in the immune system, remains unclear. To investigate FcγRIIb-mediated effector functions, we employed a combination of methods, including extensive screening of large combinatorial libraries and protein engineering techniques, to create Fc domains (Fc2b variants) with selective binding to FcγRIIb. This variants exhibited no binding to FcγRI or FcγRIIIa and practically no interaction with FcγRIIa (K [subscript D] over 20 µM). Importantly, Fc2b variants did not exhibit activities such as ADCP, ADCC, and CDC under various conditions. We validated the functional binding of Fc to FcγRIIb by observing the phosphorylation of the Immunoreceptor Tyrosine-based Inhibitory Motif (ITIM) of FcγRIIb in B cells exposed to Anti-CD20 Fc2b variants. The capacity of Fc2b to selectively bind to FcγRIIb while avoiding activation of other FcγRs, particularly FcγRIIa, holds significant promise for diverse therapeutic applications. To gain insights into the effector phenotypes mediated by FcγRIIb in human B cells, we conducted a thorough investigation of B cell effector functions using human IgG immune complexes (ICs) of varying sizes. Interestingly, our results revealed that human IgG ICs did not induce FcγRIIb ITIM phosphorylation or trigger apoptosis in B cells. Additionally, we observed no noticeable inhibition of B cell proliferation and differentiation in response to ICs, which contrasts with previous publications. Furthermore, B cells did not exhibit trogocytosis when encountering opsonized target cells. Notably, our research uncovered the previously undocumented effector function of FcγRIIb-mediated internalization of small ICs. These results contribute to an enhanced comprehension of the functioning of the human immune system, providing a more precise perspective. Additionally, we offer insights into the potential applications of therapeutic antibodies.Biomedical Engineerin
Maternal alcohol use and stress experienced during the COVID-19 pandemic : effects on parenting and child outcomes
Get PDFDuring the early stages of the COVID-19 pandemic, increased alcohol use was observed among U.S. adults, especially women. However, little is known about alcohol use trends beyond the initial months. This study tracked alcohol consumption among mothers with young children over the first 10 months of the pandemic. It aimed to identify maternal alcohol use risk factors, changes in alcohol consumption during COVID-19, and the long-term effects of maternal alcohol use on maternal well-being and child development. A national sample of 298 mothers with young children aged 0-3 years was recruited in April 2020, during the lockdown period. They were followed for 20 months. Mothers reported alcohol frequency and quantity at three points during the pandemic. They also reported COVID-related stress and parenting stress initially, and parenting stress and child outcomes over the following 20 months. The study reveals a decline in maternal alcohol consumption during the initial ten months of the pandemic. Furthermore, it highlights the distinct influence of COVID-19-related stressors, such as financial concerns, isolation and partner conflict, uncertainty surrounding the pandemic, family health concerns, food insecurity and parenting stress on maternal drinking patterns. Each stressor uniquely impacted maternal alcohol use, underscoring the intricacies of pandemic-related stress. Additionally, it uncovered that a less favorable drinking pattern, characterized by a slower decline in alcohol consumption over time, was associated with higher parenting stress levels. This heightened stress, in turn, was significantly linked to increased internalizing and externalizing behavioral problems in young children.Human Development and Family Science
Data-driven analyses of kinematic patterns in speakers with amyotrophic lateral sclerosis
Get PDFAmyotrophic lateral sclerosis (ALS) is a neurodegenerative disease that rapidly impairs motor neurons in the cerebrum, brainstem, and spinal cord, leading to deterioration of voluntary motor functions. Despite increased disease awareness and advances in research, it remains a significant clinical challenge to detect bulbar (including speech) changes before perceptual symptoms manifest and to monitor bulbar degeneration, as current available clinical measures are highly subjective. The identification of objective markers of bulbar involvement will significantly strengthen disease management and treatments. Kinematic measures of articulatory movement are quantifiable and may be sensitive to subtle changes in speech-motor control. Therefore, the aim of this dissertation is to explore the use of speech bio-signals (i.e., tongue and lip movement), as an objective marker of subclinical changes in speech motor control due to ALS. This dissertation research includes three studies that comprehensively examine the effects of ALS on speech motor control and articulator signal variability. The first study analyzes statistical features of tongue and lip movement acceleration as potential predictors of speech performance decline. The second study investigates stimulus length and disease effects on articulatory stability, measured by the spatiotemporal index (STI). The third study examines a novel articulatory vowel/consonant distinctiveness space approach and two machine learning classification algorithms to detect subtle changes in articulatory behavior due to ALS. The scientific findings from these studies provide insights into the effectiveness of kinematic measures for detecting speech changes associated with ALS. Additionally, the results reveal potential methodological factors that can influence the measurement of tongue and lip stability.Communication Sciences and Disorder
Liquid-like condensates determine actin network organization
Get PDFThe actin cytoskeleton underlies numerous cellular processes such as cytokinesis, trafficking, endocytosis, and migration. To achieve this, cells utilize hundreds of actin-regulating proteins to organize filament remodeling into higher-order bundles and networks. Recently, several actin-regulating proteins were found to phase-separate. How does phase separation influence the higher-order filament network? Here, I discovered that the processive actin polymerase, VASP, phase-separates into liquid-like droplets. These droplets catalyze actin polymerization and bundling, causing droplet deformation into elongated rods. Bundling results from competition between the droplet surface tension and the rigidity of the actin filaments, in agreement with simulations. Specifically, elongating filaments minimize curvature by forming an actin-rich ring within the inner droplet interface. As actin polymerizes, the ring thickens, until its rigidity overcomes the surface tension of the droplet, causing the ring to straighten into a linear bundle. The result is an elongated droplet comprising parallel bundles of linear actin filaments. Importantly, the liquid-like nature of the droplets is necessary for robust bundling. Furthermore, simulations revealed that VASP-actin binding kinetics dictate the morphologies of actin filaments within the droplet. Together, this work reveals a novel mechanism by which liquid-like condensates can bundle cytoskeletal filaments. However, at the cellular leading edge, both branching and bundling proteins are simultaneously present, functioning to generate rigid filaments that push against the plasma membrane. Filament branching and bundling proteins have separately been found to phase-separate. How do condensates composed of both branching and bundling proteins regulate actin organization? To answer this, I formed droplets with the actin-branching protein, Arp2/3, and VASP. Using this system, I discovered that Arp2/3 and VASP compete for a limited supply of actin. Specifically, at low actin concentrations, actin branching robustly inhibited bundling by VASP droplets, in agreement with simulations. However, as the actin concentration increased, droplets deformed into aster-shaped structures with actin bundles radiating from a branched actin core. This work highlights that multi-component condensates can mediate competition between branched and bundled filamentous organizations, resulting in large-scale morphological changes. Collectively, my work reveals that protein condensates can recreate higher-order filament networks reminiscent of those found in cells, such as in lamellipodium and filopodia.Cellular and Molecular Biolog
Scaling robot learning with heterogeneous data from the real world, simulation, and the web
No full textModern machine learning methods for robotics promise to learn generalist robots that can be deployed to everyday environments. In order to generalize effectively, these methods rely on massive datasets, but collecting robot demonstration datasets in the real world poses significant scalability challenges. This dissertation advocates for scaling robot learning by harnessing heterogeneous data sources, including prior real-world data, simulation, and general web data. These diverse data sources vary in scale and offer complementary advantages. The aim of this dissertation is to demonstrate how these alternative data streams can be harnessed to train generalist robots that can efficiently acquire new skills and perform tasks in real-world settings.
Specifically, this dissertation makes three key contributions. First, I introduce methods that leverage existing diverse robot datasets to efficiently learn novel tasks. These methods create a library of reusable skills from prior data and learn to retrieve relevant past skills to learn new complex downstream tasks. Next, I explore the complimentary question of how to expand the scope of robot datasets beyond real robot data. I present RoboCasa, a large-scale simulation framework designed to train robots for everyday tasks. RoboCasa includes thousands of assets, hundreds of simulated environments, and numerous tasks, many of which are generated using AI-driven tools. I demonstrate how automated trajectory generation can produce extensive robot demonstration datasets with minimal additional human effort and present a complementary large-scale benchmark to evaluate progress systematically. Next, I show how to leverage simulation tools and synthetic datasets to perform real-world tasks through sim-to-real transfer. Lastly, I introduce methods that draw on even larger, non-robotics datasets from the internet to solve robotic tasks.
I demonstrate how to most effectively leverage internet data by bridging the gap between robotics and web domains through intermediate representations.
Taken together, these contributions demonstrate how to effectively integrate and leverage diverse data sources to learn diverse robot tasks at scale. This approach enables the training of versatile, generalist robotic agents capable of performing a broad spectrum of tasks across varied, real-world environments.Computer Scienc