71446 research outputs found
Sort by
A Compact Piezoelectric Tilt Table for Cryogenic Electron Microscopy
Get PDFThis project designed, tested materials and machining techniques, and prototyped a piezoelectrically actuated motor to precisely orient a stage for a specimen within the compact space and restrictive environment of a transmission electron microscope (TEM). TEMs enable researchers to image and probe the structure of materials at the atomic scale, enabling advances in materials science, structural biology, and semiconductor scaling. However, many materials, such as superconductors, must be studied at cryogenic temperatures. Most materials and mechanisms that might be used to orient a specimen will not work at the very low temperatures, high vacuum, and magnetic fields within a TEM. Current specimen tilt stages typically use external DC motors which suffer from mechanical vibrations, instability, imprecision, and thermal drift affecting the specimen. There is a need to develop a tilting stage compact enough to fit within the TEM, precise enough to appropriately orient the specimen, and able to function within the cryogenic TEM environment.Mechanical Engineering S
Advancing Foundation Models for Medical Diagnosis and Biological Discovery
No full textThe open-ended and complex nature of medicine and biology poses fundamental challenges for building robust artificial intelligence (AI) and machine learning (ML) systems. This dissertation addresses key obstacles to the development of effective foundation models for medical diagnosis and biological discovery through three complementary contributions. First, I introduce frameworks designed to systematically decompose intricate and open-ended problems into tractable sub-components, facilitating the creation of meaningful benchmarks that accurately reflect scientific and clinical advancements. Chapter 1 introduces CRAFT-MD, an evaluation framework that simulates multi-turn doctor–patient conversations to assess diagnostic accuracy of Large Language Models (LLMs) in conversational settings. Chapter 4 presents a novel approach for evaluating the large-scale experimental viability of biological hypotheses generated by LLMs. Second, I develop scalable methods for identifying, curating, and harmonizing high-quality medical datasets crucial for training and evaluating foundation models. This is demonstrated in Chapter 3 through the construction of PanEndoAtlas, the largest endoscopic image dataset to date organized in a clinically meaningful hierarchy, and the accompanying PanEndoX benchmark. Third, I describe the development of specialized models tailored to medical and biological applications. Chapter 2 introduces BEANIE, a non-parametric statistical method for precise nomination of biological hypotheses from single-cell RNA-seq data in patient oncology cohorts. Chapter 3 presents PanEndoFM, a foundation model trained for endoscopic diagnosis on over 10 million images covering the entire GI tract. Together, these contributions advance the methodological, data-centric, and model-building foundations necessary for complex medical and biological domains.Biological and Biomedical Science
Understanding the factors controlling ozone pollution in East Asia
Get PDFEast Asia is one of the most severely polluted regions in the world. A key component of this pollution is tropospheric ozone, which poses a threat to both human and ecosystem health. Despite targeted efforts to reduce emissions, tropospheric ozone has continued to rise steadily in the region over the past twenty years. This dissertation investigates the factors controlling ozone pollution in East Asia using integrated data analysis from satellites, aircraft campaigns, and in-situ measurements, as well as chemical transport modeling. Specific topics addressed in my dissertation include the following:
Infer the spatial distribution of surface ozone concentrations in Asia using multispectral satellite ozone retrievals (Chapter 1):
Over the past two decades, satellite instruments have provided unprecedented information on global air quality, but direct inference of surface ozone from space remains challenging. Here, we develop a novel approach that combines multispectral ozone retrievals from the thermal infrared Tropospheric Emission Spectrometer (TES) and the ultraviolet-visible Ozone Monitoring Instrument (OMI) with a chemical reanalysis. Our results show the potential of combining satellite measurements and chemical reanalyses to augment air quality assessments in regions lacking robust surface monitoring networks.
Diagnose the causes of persistently high surface ozone in South Korea (Chapter 2):
Despite substantial efforts to reduce emissions, South Korea continues to experience widespread exceedances of their ozone standard. I examine trends in ozone and from 2015–2019 across South Korea, identifying volatile organic compounds (VOCs) as a dominant driver of ozone formation under current conditions. Simulations with anthropogenic emissions zeroed out reveal a significant external background contribution, implying that the air quality standard in South Korea is not practically achievable unless this background external to East Asia can be decreased.
Determine the origin of increasingly high background ozone in East Asia (Chapter 3):
From Chapter 2, we find that severe surface ozone pollution in East Asia is due in part to an elevated background subsiding from the free troposphere. We find that increasing background ozone is driven by enhanced stratospheric downwelling in recent years. This growing stratospheric contribution poses a substantial obstacle to achieving air quality standards, suggesting that mitigation strategies must account for both anthropogenic emissions and climate-driven changes in stratosphere–troposphere exchange.Earth and Planetary Science
IL-33 Amplifies Regulatory T Cell Reprogramming
No full textCD4+ Foxp3+ T regulatory (Treg) cells are essential to maintain immune homeostasis and immunological self-tolerance. Their immunosuppressive functions are beneficial in preventing autoimmune diseases but detrimental in supporting tumor growth. Tumor infiltrating Treg cells suppress anti-tumor immune responses and promote tumor immune escape. To unleash anti-tumor responses and improve prognosis in cancer patients, multiple therapeutic strategies have been developed to selectively deplete tumor infiltrating Treg cells. In order to more effectively regulate T cell responses, activated Treg cells can express the transcription factors that define the Th1, Th2 and Th17 effector cell lineages and adopt some of their functional attributes, a phenomenon described as Treg plasticity. However, Th1-, Th2-, or Th17-polarized Treg cells in most cases do not express the respective effector cytokines. Our laboratory has previously shown that tumor-infiltrating Treg cells not only adopt Th1 features, but can also secrete IFN at a low rate, that the rate of IFN-secretion can be amplified by targeting the CARMA1-BCL10-MALT1 (CBM) signalosome complex in Treg cells, and that Treg cell-derived IFN inflames the tumor microenvironment (TME) selectively to render immune checkpoint therapy (ICT)-resistant tumors sensitive to PD-1 pathway-targeted ICT without leading to autoimmune responses outside of tumor. This has led to the novel concept to harness Treg cells therapeutically to inflame the TME instead of depleting them. To investigate which TME factors promoted IFN production by Treg cells selectively in the TME, we screened for TME cytokines and observed that the IL-1 family cytokines IL-18 and IL-33 synergized with IL-12 to reprogram Treg cells to secrete IFN. It has been shown that IL-18 enhances IFN production by Th1 cells, while IL-33 enhances IL-5 and IL-13 production by Th2 cells. Therefore, we first examined the effects of IL-33 on Th1-, Th2-, and Th17-polarized Treg cells. We found that IL-33 enhanced not only T-bet and IFN expression in Th1-like Treg cells, but also GATA3 expression in Th2-like Treg cells, while we did not observe an effect on RORt and IL-17A expression in Th17-like Treg cells. This result indicates that IL-33 has pleiotropic roles in promoting Treg cell polarization. In many patients, cancer cells express IL-33, and we hypothesized that cancer cell-derived IL-33 could amplify Treg cell-secretion of IFN, supporting sensitivity to ICT in these patients. In contrast to this hypothesis, we observed reduced tumor growth in Treg cell-specific conditional ST2 (IL-33-receptor) knockout mice implanted with IL-33-expressing tumors, indicating that the immunosuppressive function of IL-33/ST2 signaling in tumor infiltrating Treg cells outweighs any potential pro-inflammatory effects.Graduate Educatio
Analysis of Carbon Reduction Potential in China’s Civil Aviation Industry (2027-2060)
Get PDFChina has committed to peak carbon emissions by 2030 and achieve carbon neutrality by 2060, placing considerable pressure on high-emission sectors to decarbonize. Yet the aviation sector currently lacks a national-level reduction target or roadmap. In contrast, the European Union and the United States have already set clear aviation decarbonization strategies, increasing pressure on China to respond.
This research evaluated future emission trends under a Business-as-Usual (BaU) model and assessed the effectiveness of key policy measures, particularly Sustainable Aviation Fuels (SAF) and structural changes in transportation such as High-Speed Rail (HSR) reforms. It aimed to answer three questions: (1) How will China’s air travel demand change under the BaU model? (2) What is the potential of structural adjustments, especially HSR substitution, to reduce emissions? (3) To what extent could policy incentives drive further reductions?
Under the BaU model, I estimated China’s future air travel demand using authoritative projections of economic growth and population structure. With 2019 as the baseline, demand was projected to reach 1.15 times the 2019 level by 2035 and more than double by 2060, underscoring the need for additional measures to achieve net-zero.
Building on this baseline, the analysis examined HSR’s substitution effect by evaluating China’s major 2019 civil aviation routes and calculating replacement potential based on travel times between origin–destination pairs. Future scenarios considered possible HSR speed increases: by 2035, no infrastructure upgrades but operational adjustments such as timetable changes; and by 2060, infrastructure improvements. Results projected that, with 2019 as the baseline year, approximately 23.1% of air travel would shift to HSR by 2035, increasing to 37.6% by 2060.
When GDP growth, demographic change, and HSR substitution were jointly considered—while excluding aircraft efficiency improvements—aviation kerosene demand was estimated at 31.17Mt by 2035 and 44.88 million tonnes by 2060. Rising fuel demand highlighted the urgent need to curb aircraft carbon emissions while accommodating mobility growth.
Currently, SAF is internationally recognized as the primary technological pathway for aviation decarbonization, with hydroprocessed esters and fatty acids (HEFA) as the only commercially viable route. As used cooking oil (UCO) remains the primary SAF feedstock but is in limited supply, even efficient nationwide collection would not provide sufficient capacity to achieve a 20% reduction in aviation emissions from 2019 levels by 2035. Therefore, China must accelerate commercialization of alternative SAF pathways through R&D support, while also adopting policies such as mandatory blending targets and carbon credit schemes to stimulate emissions reduction in the aviation sector.Extension Studie
THREE ESSAYS ON HEALTH SYSTEM BARRIERS TO WOMEN’S ACCESS TO HIGH‑QUALITY HEALTH CARE IN SUB‑SAHARAN AFRICA
No full textDespite remarkable progress in reducing maternal and newborn mortality over the past 20 years, 800 women and 6,500 newborns still die every day globally during delivery or in the days and weeks afterward. These deaths are preventable with high-quality care provided throughout the lifecycle from preconception to childhood. Yet, women-newborn dyads often fall through the cracks of a weak continuum of care, especially after birth. Various factors contribute to these challenges, ranging from individual-level determinants to health-system constraints, such as limited health financing and low quality of care. In particular, postnatal period has received less focus than other periods, and its quality of care remains under-measured. More evidence is needed to identify the key factors that prevent women from accessing life-saving care during this critical time.
The following three chapters investigate how different components of health system influence variations in postnatal care quality and access to reproductive health services. By employing both quantitative and qualitative methodologies, I provide a comprehensive understanding of the health system barriers that impede women’s access to high-quality care. Following the introduction, Chapter 2 uses data from direct delivery observations in public facilities in Dire Dawa Administration, Ethiopia to examine health system competency on risk detection and management for newborns. Results find that both at-risk and healthy newborns receive similarly low quality immediate postnatal care, with non-clinical factors like mother’s education contributing to variation in care quality. Chapter 3 further investigates barriers to care quality for women and newborns after discharge from delivery facilities in Kakamega, Kenya. Employing an explanatory sequential mixed method, the study shows that the content of care received before discharge influences mothers’ decisions to seek routine postnatal care. Key drivers include trust in the provider’s supervision, counseling on the importance of postnatal care, and the formal scheduling for the next postnatal visits. Chapter 4 used Demographic Health Surveys to investigate the impact of disruptions in development assistance for health in social marketing programs, a key health financing scheme in low resource settings, on women’s reproductive health behaviors in Zambia. Analysis adopting difference-in-difference method supplemented by a synthetic control approach shows that funding discontinuation do not affect overall modern contraceptive use or pregnancy rate but do lead to decline in using condoms and oral contraceptives.
Together, this dissertation demonstrates that health systems underperform by providing low levels of routine care quality throughout the postnatal period and by reflecting weakness in health financing that hinder the consistent medical supplies. Findings can be used to design future interventions to improve service quality during the postnatal period and to develop strategies that enhance financing resilience in low-resource health systems.Population Health Science
Addressing the Extremes of Reactivity in Small-Molecule-Catalyzed Stereoselective Glycosylation
Get PDFGlycosides play disparate yet essential roles in biology, and as such their study is of great
importance to human health. To perform such studies, the stereocontrolled laboratory synthesis of
glycosides remains a necessary endeavor. Despite this, the development of general methodology
for stereoselective glycosylation is an unmet challenge in organic reaction development. This lack
of generality can be attributed, among other reasons, to the incredible range of reactivities present
across the various glycosyl donors of biological relevance, owing in turn to their diversity of
structures. This is especially problematic for the stereoselective glycosylation of donors that sit at
the extremes of reactivity, both high and low, where methodology for glycosylation with biomass
sugars is typically not applicable. We aim to address these shortcomings through the development
of general catalytic glycosylation protocols that are applicable across donor classes.
In Chapter 1, we discuss the relationship between structure and reactivity in chemical
glycosylation, highlighting key studies that probe this relationship. We will discuss the
stereochemical implications of these structure-reactivity effects, with an emphasis on
glycosylation with highly reactive 2-deoxyglycosyl donors and highly unreactive glucuronyl
donors. We will then discuss current strategies for the stereoselective preparation of 2-
deoxyglycosides and glucuronides, as well as the limitations of these strategies. Finally, we will
summarize our research group’s development of bis-thiourea-catalyzed stereoselective
glycosylation, including key findings, mechanistic studies, and limitations that studies documented
in this dissertation aim to address.
In Chapter 2, we document the development of bis-thiourea-catalyzed methods for βselective 2-deoxy- and 2,6-dideoxyglucosylations of natural products, carbohydrates, and amino
acids. Disarming ester protecting groups were necessary to counter the high reactivity of 2-
deoxyglycosyl electrophiles toward non-stereospecific SN1 pathways. Differing catalyst structures
were found to be optimal for use of 2-deoxy- and 2,6-dideoxyglycosyl donors. Alcohol and phenol
nucleophiles with both base- and acid-sensitive functionalities were compatible with the catalytic
protocol, enabling access to a wide array of 2-deoxy-β-O-glucosides.
In Chapter 3, we document the successful application of metal–salen catalysis towards β-glucuronidation of alcohols, phenols, and anilines via stereospecific opening of 1,2-
anhydroglucuronate electrophiles. The optimized protocol is mild and pH-neutral, enabling β-glucuronidation of complex pharmaceuticals and natural products bearing acid-sensitive and
Lewis-basic functionality. Kinetic studies are consistent with a transformation that is overall first order in catalyst, which contrasts previous metal–salen-catalyzed epoxide opening reactions where
second-order rate dependence on catalyst is observed.Chemistry and Chemical Biolog
Essays on Global Bond Market Integration: The Roles of Foreign Ownership, Auction Signals, and Currency Risk
Get PDFThis dissertation investigates how global investors shape bond market integration through three interconnected mechanisms: cross-border asset ownership, informational dynamics in bond auctions, and currency-related risk preferences. The first chapter demonstrates that increased foreign ownership of sovereign bonds significantly strengthens the co-movement of bond yields across countries, using China's 2017 bond market liberalization as a case study. The second chapter identifies U.S. Treasury auctions as uniquely informative events that reveal global investor demand for long-term bonds, resulting in systematic declines in global yields. The third chapter explores how currency risk influences investor behavior, documenting systematic differences in return sensitivity and capital flow volatility driven by currency denomination and hedging practices.Economic
New knock in mouse lines Dmp1em1(CreERT2) and Dmp1em2(ZsGreen) enable precise osteocyte-specific targeting and visualization
No full textIn adult mammals, bone is a dynamic organ with both structural and endocrine functions, mainly mediated by osteocytes, the most abundant bone cells, and the principal source of bone-derived hormones such as fibroblast growth factor 23 (FGF23) and sclerostin (SOST). The widely used 10-kb promoter-driven transgenic Tg(Dmp1-Cre) and Tg(Dmp1-CreERT2) lines, while widely used and helpful, utilize transgenic approaches with transgene expression driven by the Dmp1 promoter. The transgene expression suffers from positional effects with off-target expression in muscle, brain, and other tissues, limiting their applications, causing a major technical gap in the investigation of osteocyte-specific functions. Here, we generated two novel mouse lines, Dmp1em1(CreERT2) and Dmp1em2(ZsGreen), via CRISPR-Cas9-mediated knock-in approaches that allow CreERT2 or ZsGreen expression under the genetic control of the endogenous Dmp1 locus by inserting an IRES–CreERT2 or IRES–ZsGreen cassette into the Dmp1 locus, preserving the Dmp1 gene expression. We showed that the Dmp1em1(CreERT2) enabled highly efficient tamoxifen-inducible recombination (>90% in cortical and trabecular osteocytes) with minimal off-target labeling in non-bone tissues. The Dmp1em2(ZsGreen) line showed robust, fixation- and decalcification-resistant green fluorescence in osteocytes from birth through adulthood, faithfully reflecting endogenous Dmp1 expression. Lineage tracing using the Dmp1em1(CreERT2) line further revealed that the cortical osteocytes are long-lived, whereas trabecular osteocytes undergo continuous renewal, uncovering compartment-specific differences in osteocyte lifespan. By eliminating off-target recombination or gene expression in skeletal muscle, brain, gastrointestinal tract, and marrow compartments, the two mouse lines offer powerful tools for rigorous studies in osteocyte biology, mechanotransduction, and bone regulation of systemic physiology that impact health, aging, and diseases, while minimizing complications due to off-target transgene expression.Graduate Educatio
Everything Is a Matrix: Minimizing Data Movement and Parameter Count Across the Machine Learning Stack
Get PDFMachine learning has revolutionized natural language processing, computer vision, and beyond. Yet as machine learning models scale in size and capability, the demand for computational resources likewise grows, exposing new challenges in efficient and scalable deployment. Extracting maximal performance from existing hardware is therefore vital to unlocking the next wave of progress in artificial intelligence.
In many modern workloads, matrix operations dominate resource consumption, sometimes accounting for more than 99% of the workload [1]. Thus, we will focus on matrices as the central unit of optimization. This thesis presents an array of novel techniques to reduce memory footprint, accelerate computation, and improve overall hardware utilization. We demonstrate substantial efficiency gains are achievable by rethinking how data is computed, stored, and compressed, with a special focus on matrices, the core computational structure underpinning both scientific computing and neural networks.
First, we address dense matrix multiplication by introducing CAKE, a method that partitions computation into optimally shaped blocks to minimize memory bandwidth bottlenecks (Chapter 2). We extend this method to tensor contractions with any number of loops with mCAKE (Chapter 3). Then, for neural networks exhibiting moderate sparsity, the Rosko framework (Chapter 4) exploits outer-product structure to efficiently skip zero-valued computations and enables the creation of hardware-compatible sparsity patterns through structured pruning.
Next, we investigate efficient representations of weight matrices of neural networks using Singular Value Decomposition (SVD) (Chapter 5), enabling both memory savings and accelerated inference. Building on this, we explore low-rank model compression, where the compact forms of decomposed weight matrices facilitate efficient training and adaptive fine-tuning (Chapter 6). We then introduce blockwise knowledge distillation techniques (Chapter 7) that allow highly compressed, SVD-based student models to learn directly from their full-rank teacher counterparts, preserving both efficiency and model accuracy. Lastly, we demonstrate a privacy-preserving framework for distributed inference that splits computation between local devices and cloud servers, ensuring user data labels remain on-device while leveraging powerful cloud-based feature extractors (Chapter 8).
Together, these contributions meaningfully advance the efficiency and scalability of both conventional scientific workloads and the latest state-of-the-art AI models.
Reference: [1] A. Ivanov, N. Dryden, T. Ben-Nun, S. Li, and T. Hoefler, “Data movement is all you need: A case study on optimizing transformers,” 2020.Engineering and Applied Sciences - Computer Scienc