116964 research outputs found
Sort by
Graph-based methods for scalable and verifiable digital twins
Get PDFAs the digital twin paradigm increasingly gains traction across diverse domains, its adoption remains limited by three core challenges: scalability, verifiability, and predictive capability. To date, most work on digital twins has focused on specific implementations or case studies, often centered on assembling existing technologies. What remains largely absent is a generalized mathematical formulation that abstracts away from specific technologies and that can be applied consistently across applications. This thesis develops a cohesive theoretical formulation that addresses these challenges through formalized knowledge representation, verifiable system dynamics, and explainable generative modeling. The formulation defines a two-layer, graph-based representation of a digital twin. At the foundational layer, an ontological knowledge graph captures system structure by organizing entities, relationships, and semantic rules governing their interactions. At the adaptive layer, a probabilistic graphical model governs the temporal evolution of these elements, providing a consistent mathematical basis for updates. Building on this foundation, the thesis develops a rigorous method for the formal verification of digital twins using the Temporal Logic of Actions. We leverage the adaptive layer to derive formal specifications of a digital twin under asynchronous, bidirectional communication, conditions characteristic of real-world operation. Finally, the work addresses the challenge of predictive modeling in data-restricted settings, where developing accurate models often requires data that are high-dimensional, costly, or restricted. We develop an explainable approach for generating longitudinal synthetic data that preserves key dependencies, supporting the development of predictive models for digital twins. These methods are demonstrated through two applications: an Educational Digital Twin modeling statewide student pathways for large-scale analysis, and an unmanned aerial vehicle digital twin illustrating formal verification in safety-critical systems. Together, these contributions establish a mathematical foundation for digital twins that generalizes across domains, supports formal verification, and strengthens predictive capability through explainable synthesis.Computer Scienc
Engendering migration from the margins : rural ethnic women in migrant-sending communities in China
Get PDFThis dissertation is an ethnography of marginalized ethnic-minority Qiang women in Sichuan Province, China. These are mothers in their 40s-60s who live and work in highland Qiang villages, while their adult family members migrate to cities to find work in precarious labor markets. While much of the literature on labor migrations focuses on workers, and the wages they send “back home,” I examine how people “left behind” in migrant-sending communities facilitate and support out-migration. I draw on fifteen months of fieldwork, spent living and working in Changle, an impoverished Qiang migrant-sending community in the mountainous Himalayas of Sichuan Province, the province sending the highest number of migrant workers to urban China. I labored along with older Qiang women as they weeded, cultivated, and harvested crops to sell to urban markets. I also attended family gatherings and community events. I interviewed mothers, fathers, migrating sons and daughters, village and township officials, and teachers and health workers to map how communities grapple with immense socioeconomic precarity. My findings reveal that the “left behind” Qiang women have borne the economic and social costs of guaranteeing livelihoods, educating, and caring for the present and future migrant workers. I highlight the Qiang women’s strategies for fighting poverty and precarity to confront their structural exclusion. In the private sphere, older Qiang women take on back-breaking farm labor to eke out supplemental income to help support urban migrants in the low-wage informal urban economy. In an increasingly stratified and competitive marriage market, the minoritized and rural Qiang men have difficulty finding brides. These mothers also perform extraordinary relational work in order to arrange marriages and finances for their adult sons to help them establish stable lives. As grandmothers, these women face the daunting tasks of raising ethnic grandchildren to meet family and societal expectations of childrearing in late capitalist China. While trying to pass on the Qiang heritage to the children, these grandmothers also need to bear the “intensive childrearing” demands to ensure that the grandchildren are well assimilated into the Han-dominated society to have positive labor outcomes in the future. In the public sphere, rural Qiang women engage in political bargaining with the state to influence poverty alleviation and social provision programs.Sociolog
Elucidating properties of the plasma antibody repertoire response following erythrocytic-stage malaria vaccination and infection
Get PDFMalaria remains a leading cause of mortality among children in sub-Saharan Africa. Neutralizing antibodies against Plasmodium falciparum have long been associated with protection against severe disease, and next-generation vaccines hope to alleviate malaria burden. Although liver-stage vaccines exhibit partial efficacy, they fail to confer sterilizing immunity, emphasizing the need for blood-stage immunogens that induce a durable protective antibody response. In this dissertation, we delineate unknown features of the plasma antibody response to blood-stage infection and vaccination using various proteomics and next-generation sequencing (NGS) approaches.
In Chapter 2, we demonstrate that malaria-naïve United Kingdom adults
immunized against P. falciparum reticulocyte-binding protein homolog 5 (PfRH5) elicit abundant, non-neutralizing plasma IgG lineages. A subset of these lineages, expressed as monoclonal antibodies (mAbs), synergize with both neutralizing and non-neutralizing mAbs to enhance in vitro parasite growth inhibition. These findings underscore that future PfRH5 immunogen design should minimize antibody responses to non-protective, linear epitopes while preserving epitopes recognized neutralizing and synergizing antibodies.
In Chapter 3, we characterize the plasma antibody repertoires to merozoite surface protein 1 (MSP1) and apical membrane antigen 1 (AMA1) in malaria-convalescent Ugandan adults. Because convalescent timepoints typically underrepresent B cells actively secreting antibodies, we employ a complementary proteomics technique to isolate immune-complexed IgG. While anti-MSP1 repertoires were highly polarized and exhibited shared V gene usage and higher SHM rates, anti-AMA1 repertoires were more diverse and did not share genetic features. Immune-complexed IgG overlapped top-ranking lineages
identified by traditional affinity chromatography, while also revealing additional lineages absent from those pulldowns. In one donor, we integrate Fab profiling and de novo bottom-up proteomics sequencing to enable the identification of the paired VH:VL sequence of a highly abundantly circulating plasma antibody. These results suggest that naturally acquired immunity to malaria may be driven by both polarized (MSP1) and diverse (AMA1) antibody repertoires, offering insights into why adults achieve partial protection that children lack.
Overall, the combination of proteomics approaches used here demonstrate that
plasma repertoires do not always align with circulating B cell repertoires. Further understanding of the serological response can elucidate mechanisms of blood-stage protection to inform future malaria vaccine and prophylactic strategies.Biomedical Engineerin
A New World Order? Careful What You Wish For
Get PDFThis paper examines the persistent attractions of the idea of a world order, and whether one may be said to exist today. It argues that we are now in a world adrift or, at best, between orders. It suggests that this may mark a return to the historical norm and represent an opportunity for ideas that can matter and shape political outcomes in this time of flux.LBJ School of Public Affair
Wars of the Greater Middle East, 1945–92
Get PDFThis article examines the history of war and society during the Cold War in the Middle East and parts of South Asia—two regions linked by geography, history, and culture. Few other regions have been so touched by war, or so fixed the attention of world leaders. Two themes run through the article. The first is how politics, technology, society, and culture changed the conduct of war. The second is how the conduct of war changed politics, society, and culture. The overarching argument is that a combination of pressures spread forms of war—namely, guerrilla warfare and terrorism—that put the use of force in the hands of the people. This democratization of violence complicated the consolidation of state authority and was intertwined with the return of Islam as a political force. If war after 1945 for the United States and Europe became, to quote Michael Howard, “an affair of states and no longer peoples,” then in much of the Middle East and South Asia, it became an affair of peoples as much as states.LBJ School of Public Affair
Investigation of supported ionic liquid membranes for olefin/paraffin gas separation
Get PDFMembrane-based separation processes offer significant energy savings compared to conventional separation technologies involving thermal phase changes. While membranes have found widespread application in various separation fields like desalination and air separation, they are not widely deployed in the industrial separation of olefin/paraffin gases. This is not just because olefins and their paraffin analogs have similar thermophysical properties but also because there are many technological challenges to overcome. This dissertation addresses some critical issues in the development of supported ionic liquid membranes (SILM) for olefin/paraffin gas separation. Our research focuses on three key areas: achieving carrier stability, enhancing transmembrane pressure stability, and exploring alternative carrier options for SILM systems. In our pursuit of carrier stability, we tackled the issue of silver ion carriers, widely employed for facilitating olefin transport, which tend to deactivate upon exposure to hydrogen, a common byproduct in olefin production. Leveraging anion selection within ionic liquids (ILs), we have achieved long-term carrier stability, exhibiting no loss of selectivity over more than a week, of pure hydrogen permeation. A comparison study of silver containing IL mixtures with two distinctive anions, bis(trifluoromethylsulfonyl)imide (Tf₂N⁻) and nitrate (NO₃⁻), reveals that the Tf₂N anion, which tends to form ion aggregates with silver ions, plays a pivotal role in preventing silver ion nucleation. In our pursuit of pressure stability, structural modification of the porous support overcame the inherent low transmembrane pressure limit of SILMs due to IL blow-out. By incorporating a mesoporous silica layer onto the base pore structure, we have successfully reduced the effective pore size, enabling solution-diffusion transport through the IL even under high transmembrane pressure. Increasing transmembrane pressure across the modified SILMs resulted in the reduction of the IL thickness in the SILM, opening up the possibility of producing ultra-thin IL layers with enhanced transport properties. Lastly, we focused on the development of alternative non-silver olefin carriers. We identified one iron-hexafluoroacetylacetonate (hfac) based complex exhibiting reversible olefin binding ability. While promising, we concluded that effectively taking advantage of this complex's functionality will require a separation system with longer diffusion lengths than afforded by a SILM.Chemical Engineerin
Fabricating large scale graphene oxide films for structural applications
Get PDFGraphene’s ultra-high specific surface area, conformational complexity, and capacity to be functionalized give it unparalleled capability to dissipate mechanical energy through interfacial friction. Previous work has examined the damping and stiffness of graphene oxide films from an experimental perspective but has hardly scrutinized the mechanisms responsible for such behavior. In this thesis, the morphological mechanisms responsible for high stiffness and damping are connected with the processing conditions used to create graphene oxide films. Parameters such as pH, initial concentration, particle size and oxidation, and substrate material are connected to the structure and mechanical properties of films placed in tension. Particle size, oxidation, and pH are found to affect the stress transfer of GO sheets; this in turn affects the stiffness, strain at maximum stress, and maximum stress. Initial concentration, oxidation, and substrate affect the damping factor. Films which display high damping factor are flat overall but are curled internally. This internal curling, caused by gradients in swelling and delamination, places lamellae in bending when the overall film is in tension. While in bending, GO sheets can shear against each other and dissipate large amounts of energy through interfacial friction. This research provides an interesting case study on the interplay between capillary action, evaporation, and jamming in a charged, anisotropic colloidal suspension. It also addresses a few barriers to the scale up of production for graphene oxide films. GO films with high damping could be used to mitigate vibrations and dampen oscillations in critical components, enhancing structural integrity and reducing fatigue.Mechanical Engineerin
Evaluating the role of mechanical and hydrologic properties of input sediment to fluid pressure development in the shallow subduction zone
Get PDFOceanic sediments subducted with the downgoing oceanic plate play a critical role in controlling the physical state and mechanical behavior of subduction zones. Seismic velocities provide valuable insights into the physical state of subducting margins, yet interpreting these data requires additional constraints. Laboratory experiments offer a means to directly link seismic observations to rock properties and pore pressure, enabling quantitative assessment of the mechanical state at depth. In this dissertation, I investigate the mechanical behavior of subduction zone sediments from the Alaskan and Northern Hikurangi subduction margins. I focus on the evolution of porosity, permeability, P-wave (Vp) velocities to better understand the processes governing fluid release and to refine interpretations of seismic velocity data in terms of physical and hydrologic conditions.
First, I present laboratory measurements of hydrologic and mechanical properties in pelagic sediments entering the underthrusting section of the Alaska margin, cored at Site U1417 during IODP Expedition 341. I use these data to inform a simple loading model which I compared to field-based observation and show that sediment thickness, initial burial depth, and loading rate are key factors in controlling drainage of the underthrusting sediment. Secondly, I investigate the origins of low seismic velocities observed in underthrusting sediments, which are often interpreted as indicators of high fluid pressure. I report laboratory measurements of Vp and Vs in exhumed, relic accretionary wedge metasediments from Kodiak Island, Alaska, which exhibit strong velocity anisotropy. Comparison of these results with physics-based models for transversely isotropic media suggests that combination of fabric orientation and optimally oriented cracks can contribute to lower observed velocities, without requiring near lithostatic pore pressure.
Finally, I assess the partitioning of fluids in input sediments to the northern Hikurangi subduction margin. I show experimentally that Vp is only sensitive to the total water content of the sediments, which has important implications for interpretation of porosity transformed from velocity at margins with high hydrous mineral content. I then use X-ray diffraction (XRD) measurements of hydrous mineral content to correct for mineral bound water in previously reported estimates of total water content. Overall, this dissertation advances the understanding of where, and how the mechanical and hydrologic properties of underthrusting sediments influence the mechanical behavior of the shallow subduction zone.Earth and Planetary Science
On optimizing data locality for data-intensive parallel workloads: a compile-time framework for machine learning models and a run-time abstraction for task-based systems
Get PDFModern data-intensive workloads, from high-performance computing (HPC) to machine learning (ML), are increasingly bottlenecked by data movement rather than computation. This "memory wall" is especially pronounced in heterogeneous systems combining multi-core CPUs and many-core GPU accelerators. Optimizing data locality -- ensuring data is at the right place, at the right time -- is therefore a critical and universal challenge. This dissertation presents a comprehensive, two-part solution that addresses this challenge at two different granularities: a compile-time framework to maximize intra-kernel locality for ML models, and a run-time abstraction to manage inter-task locality for task-based parallel systems. The primary contribution of this dissertation is FlashLight, a compiler-native framework within the PyTorch ecosystem that accelerates emerging variants of the Attention mechanism. Attention is a fundamental component of modern Large Language Models (LLMs). While I/O-efficient implementations like FlashAttention leverage tiling and kernel fusion to optimize standard Attention, these hand-tuned kernels are not trivially adaptable to new, complex Attention variants designed to enhance model quality or efficiency. Existing template-based solutions like FlexAttention offer a partial remedy but lack composability and cannot express more general, data-dependent formulations. FlashLight bridges this gap by automatically generating FlashAttention-style fused kernels from idiomatic PyTorch programs, without relying on static templates. It extends PyTorch's compilation backend, TorchInductor, with novel compiler optimization techniques. These include semantic fusion via algebraic transformations that generalize the "online softmax" algorithm to fuse state-dependent reductions, and structural fusion passes, like dimension demotion, that leverage a unified reduction IR to tile and fuse operator subgraphs. By fusing the entire attention subgraph (e.g., matmul-softmax-matmul) into a single, tiled kernel, FlashLight optimizes intra-kernel data locality by eliminating the materialization of large intermediate tensors to global memory. Our results show that FlashLight generates kernels with performance competitive to FlexAttention on static workloads and superior on dynamic workloads, while supporting a more general class of models, including those beyond FlexAttention's expressiveness. While FlashLight optimizes locality within a kernel at compile-time, a complete solution necessitates the management of locality across devices at run-time. To address this, we also present CrossPy, a heterogeneous array abstraction designed and contributed as the data management layer for the Parla task-based runtime. In Python's ecosystem, developers face the significant challenge of managing heterogeneous data across the distinct memory spaces of CPUs (using e.g. NumPy) and GPUs (using e.g. CuPy). CrossPy solves this by providing a unified, shared index space over physically partitioned arrays. It integrates directly with the Parla scheduler to enable inter-task data locality by facilitating data-aware task placement and automating data movement. Critically, its optimized, cacheable alltoall interface moves communication orchestration out of the Python interpreter, bypassing the Global Interpreter Lock (GIL) bottleneck to provide efficient, parallel inter-device communication. We demonstrate its efficacy with a multi-GPU K-Means implementation that achieves strong and weak-scaling efficiency, proving that CrossPy provides programming convenience without sacrificing performance. Ultimately, this dissertation demonstrates a multi-level strategy for optimizing data locality. By contributing both a compile-time kernel fusion framework for ML models and a run-time data management abstraction for task-parallel systems, this dissertation provides a comprehensive and practical path to unlocking performance for data-intensive parallel workloads on modern heterogeneous hardware.Computer Scienc
The development of a wearable sensing device for swallow monitoring and classification
Get PDFThis study presents the development and evaluation of a non-invasive, wireless wearable sensing device for continuous monitoring and classification of swallowing and non-swallowing behaviors in healthy individuals. The device employs knitted fabric sensors for enhanced comfort and seamless integration into daily life. Using machine learning algorithms, swallowing patterns were identified to provide insights into the behaviors’ features and characteristics. Data collected from three strategically placed sensors were processed and analyzed, with feature extraction and selection methods employed to optimize input for the machine learning models. The fine k-Nearest Neighbors (KNN) model demonstrated superior performance, achieving an accuracy of 75.6%, precision of 72.4%, sensitivity of 81.5%, specificity of 74.4%, and F1-score of 76.7%. Results indicated that incorporating multiple sensors improved classification accuracy, with the highest accuracy achieved using three sensors (75.6% with 51 features). The wearable sensing device and accompanying algorithms show potential in dysphagia diagnosis, treatment, and monitoring, in hopes of paving the way for further refinement and potential applications in other aspects of dysphagia care, such as swallow therapy and telemedicine integration.Mechanical Engineerin