22691 research outputs found
Sort by
ML-Assisted Data Assimilation in Transitional High-Speed Boundary Layers
Full text linkNatural transition in high-speed boundary layers begins with the exponential amplification of infinitesimal instability waves. As these disturbance amplitudes grow, nonlinear interactions emerge, eventually leading to turbulent breakdown. Data assimilation provides a powerful framework to integrate sparse experimental measurements with Direct Numerical Simulations (DNS), enabling improved prediction of such complex, multiscale flows. In transitional high-speed boundary layers, one seeks to assimilate wall pressure measurements obtained using piezoelectric sensors (PCB) to infer the full flow field. This process requires solving an inverse problem and also demands efficient evaluation of the Navier–Stokes operator. This thesis develops a data assimilation framework for high-speed transitional flows using Deep Operator Neural Networks (DeepONet).
The first part of the thesis demonstrates how DeepONet can be used for forward modeling, predicting downstream flow in the nonlinear growth regime from upstream disturbances. Two cases are considered: a Mach 4.5 boundary layer with calorically perfect gas, and a Mach 10 flow with chemically reacting gas. The performance of DeepONet is evaluated using operator-specific metrics, and results show that for more complex flow physics, DeepONet offers greater benefit for fast, accurate prediction compared to DNS. The second part addresses the inverse problem: assimilating wall pressure measurements to estimate upstream instability characteristics. This problem is formulated as an optimization problem and solved using Bayesian Optimization (BO). An ensemble of DeepONets is used to construct a statistical surrogate model of the cost function. The reconstructed flow fields closely match reference DNS, and the method significantly outperforms conventional Kriging-based BO in computational efficiency. The final part extends the framework to predict transition location from inflow spectra using a Bayesian DeepONet. Because the transition Reynolds number is inherently stochastic, a latent-variable formulation is adopted to predict both the mean and uncertainty of the transition Reynolds number. An active learning strategy selects the most informative training samples, reducing computation cost by approximately 47\% compared to uniform sampling. By combining DNS, machine learning, and uncertainty quantification, this thesis advances data assimilation in high-speed boundary layers and provides an efficient tool for transition prediction and flow reconstruction
BEHAVIORS AND MINDSETS THAT LEAD TO FAMILY ENTREPRENEURSHIP OUTCOMES
No full textDespite business families creating more than 70% of global gross domestic product and driving 75% of start-ups, there is a paucity of research-backed approaches that consistently foster family entrepreneurship, creating economic value and strong family relationships simultaneously. This explanatory sequential mixed-methods study provides empirical evidence of a multidisciplinary, research-backed approach to developing family entrepreneurship capacity in the Healthy Family, Healthy Business model. By deploying family systems theory in addition to family business constructs, this model helps shift the focus from keeping a firm family owned to a universal approach to value creation and family connectivity occurring simultaneously. Healthy family relationships, characterized by non-reactive exploration, collaboration, and non-judgmental curiosity, form the foundation of family entrepreneurship. Conversely, families plagued by relationship conflict may see their businesses struggle to perform or even fail. Bowen family systems theory and its core construct differentiation of self explains the impacts of family functioning on individuals, families, communities, organizations, and societies. Thus, the behaviors and mindsets of individual family members who interact with the family and business system to foster positive outcomes emerge as an important, relatively unexplored area informing further research and practice. The research identified noteworthy quantitative findings enhanced by qualitative insights, demonstrating significant positive correlations between family member differentiation of self and their perception of the relative profitability of their business, among other findings. The Healthy Family, Healthy Business model’s interdisciplinary approach provides a new lens on the challenge of family business value creation, using solutions grounded in family therapy and tested across cultures to offer a potential platform for further research and practice. The Healthy Family, Healthy Business model holds the hope of amplifying the capacity of the vast majority of global gross domestic product to create growth and positive change. When entrepreneurial families get it right, the outcomes are incredible. However, all too often, family dysfunction leads to suboptimal, or worse, business and family outcomes. Until now, we have not known how to unlock this capacity. The Healthy Family, Healthy Business approach points to the answer to this important question
STRUCTURAL INTERSECTIONALITY: A POLICY-BASED MEASURE OF STATE-LEVEL DISCRIMINATION
No full textBackground: Identifying methods to measure structural determinants of health including structural discrimination is key to advancing health equity research. However, few structural measures of discrimination consider policy contexts nor the synergistic effects of racism, sexism, and classism on health inequities. The overarching goal of this dissertation is to apply an intersectional framework and state policy contexts to improve our understanding and measurement of structural discrimination.
Methods: In Aim 1, we used a theory-driven confirmatory factor analysis approach to develop the Policy-Based State Discrimination (PBSD) measure, using cross-sectional policy data from 2012. In Aim 2, we assessed the construct validity of the PBSD and conducted an empirical evaluation of existing state-level discrimination measures using a nomological net approach. In Aim 3, we used a cross-sectional ecological design and nested negative binomial regression models to evaluate the relationship between the PBSD and maternal mortality from 2015-2019.
Results: We identified two models of intersectional structural discrimination with theoretical and empirical support: a 3-factor model that specifies factors for structural racism, sexism, and classism, and a unidimensional model that specifies one intersectional factor with the same policy indicators as the 3-factor model. Top policy-drivers of intersectional structural discrimination were domains related to social safety net, criminal justice, and sexual and reproductive health. Our nomological net indicated that existing measures of structural discrimination cluster based on measurement conceptualizations (e.g., policy-based, inequity-based etc.). Additionally, we provide evidence for segregation and incarceration as two contemporary racializing regimes that operate differently in different state contexts. Finally, we found that the PBSD was associated with a 17% increase in maternal mortality. Although Black populations had more than double the risk of maternal mortality, the effect of PBSD-I on maternal mortality was significantly attenuated in Black populations compared to White populations.
Conclusions: Policies are modifiable drivers of health inequity and are an important tool to measuring structural discrimination at the state-level. An intersectional approach, a historical perspective, and a critical assessment of measures of structural discrimination are central to advancing health equity
CHARACTERIZING ALLERGEN PROCESSING AND MHCII EXPRESSION CAPABILITIES IN AIRWAY EPITHELIAL CELLS DURING AIRWAY ALLERGY
No full textThe increasing prevalence of allergic airway diseases highlights the need to understand their underlying mechanisms. These diseases are characterized by type 2 immune responses in the lungs, sustained by dysregulated innate immune mediators. Airway epithelial cells play a critical role in initiating and perpetuating the allergic response and are gaining recognition as active participants in immune responses through MHCII-mediated antigen presentation to T cells. Understanding the mechanisms regulating MHCII expression on airway epithelial cells and their capacity to process and present allergens is crucial for developing targeted therapies for allergic asthma.
To understand their contribution to the allergic cascade, we investigate lower airway epithelial cells' allergen processing and MHCII expression capabilities. The results demonstrate the ability of these cells to uptake and process allergens as well as express MHCII, highlighting their potential as active participants in the adaptive immune response within the airways
DISSECTING THE MECHANISMS OF TUMOR PROGRESSION AND RESISTANCE TO TARGETED THERAPY IN ACRAL MELANOMA
Full text linkAcral melanoma (AM), a rare and aggressive cutaneous melanoma subtype with
5-year survival rates below 16%, exhibits limited durable responses to CDK4/6
inhibitors (CDK4i/6i) despite frequent genetic alterations in this pathway. Here, we
identify rapid rewiring of AKT-mTOR signaling as a critical escape mechanism
following CDK4/6 inhibition. Using a genetically diverse panel of AM cell lines, we
demonstrate that CDK4i/6i induces Rb dephosphorylation within 1–3 hours,
concurrent with hyperactivation of AKT (pS473) and mTORC1 (pS6 S240/244).
Mechanistically, CDK4/6 inhibition disrupts cytoplasmic Rb interactions with the
mTORC2 subunit SIN1, as evidenced by proximity ligation assays showing
elevated Rb-SIN1 puncta at 3–6 hours post-treatment. This dissociation correlates
with loss of Rb-mediated suppression of mTORC2 activity, enabling AKT-driven
mTORC1 activation.
Pharmacological co-targeting of CDK4/6 and mTOR pathways synergistically
enhances anti-tumor efficacy, reducing clonogenic survival and increasing
annexin+ cytotoxicity relative to single-agent therapies. Dual inhibition ablates
mTORC1 activity and elevates cleaved PARP levels, as confirmed by immunoblot
and immunofluorescence assays. Long-term colony formation assays reveal
sustained suppression of proliferative recovery only under combination conditions.
In vivo, catalytic mTORC1/2-inhibition with TAK228 significantly reduces tumor
volume and multi-organ metastases in AM xenograft models.
These findings establish the Rb-SIN1-mTORC2 axis as a novel mediator of
intrinsic CDK4i/6i resistance and provide preclinical rationale for co-targeting
CDK4/6 and mTORC1/2 to improve AM outcomes
INVESTIGATING PRECISE MANAGEMENT AND ARRHYTHMOGENIC MECHANISMS IN GENETIC HEART DISEASE USING CLINICAL DATA AND HEART DIGITAL TWINS
No full textArrhythmogenic right ventricular cardiomyopathy (ARVC) is a genetic heart disorder that predisposes individuals, particularly young and athletic populations, to ventricular tachycardia (VT) and sudden cardiac death. Despite advancements in medical technologies, managing ARVC remains a significant challenge in clinical practice. Catheter ablation, a primary treatment for VT in ARVC, is associated with high recurrence rates. Although ARVC encompasses various genetic subtypes, each with distinct disease mechanisms, current management strategies largely adopt a one-size-fits-all approach, reflecting an incomplete understanding of genotype-specific differences. This thesis addresses these challenges by investigating personalized management strategies for ARVC patients and elucidating the underlying arrhythmogenic mechanisms. Through the development of a genotype-specific digital platform that integrates genetic information, clinical data, and biophysical heart models, we successfully predicted VT circuits and optimal ablation sites. Furthermore, this platform revealed critical genotype-specific differences in arrhythmogenic mechanisms. The innovations described in this thesis lay the foundation for advancing precision medicine and enabling more personalized management of cardiac diseases, ultimately improving patient outcomes
There and Back Again: Top-Down and Bottom-Up Perspectives on Generalized Symmetries
Full text linkQuantum field theory is a wild and ferocious beast in the field of physics. However, it also has proven time and time again to be one of our best ways to probe the universe in which we live. One tool that has helped tame this ferocity is symmetry. In this dissertation, we will explore many facets of (generalized) symmetries in quantum field theory, including examples from string theory as well as the mathematical structure that symmetries form.
This dissertation can be sorted into four main parts. In the first, we present and study a family of strongly coupled 4d \cN=2 SCFTs, dual to a new family of M-theory solutions. We use various wrapped M-branes to find field-theoretic quantities of the 4d theories, including the holographic central charge and operator scaling dimensions.
In the second part, we further analyze the generalized symmetries of different holographic field theories. We will see how brane configurations yield topological symmetry generators, and moreover how tachyon condensation in string theory gives rise to non-invertible symmetries. This includes both discrete and continuous symmetries alike, and concerns families of constructions dual to 4d \cN=1 and 3d \cN=2 theories.
In the third part, we discuss the fate of symmetry operators in gravitational theories. We first argue that a symmetry operator insertion in a theory requires a regulator, understood as the thickness for the operator. The topological nature is recovered as this thickness is taken to zero. We then argue that in the presence of gravity, the regulator cannot be taken to zero and hence the operators cannot be made topological. Thus, there is a natural and universal obstruction to the presence of symmetry operators when a theory is coupled to gravity.
In the fourth and final part, we turn to the mathematical structure of generalized symmetry operators. We show that, from purely physical arguments, the symmetry operators of a -dimensional quantum field theory naturally satisfy a proposed definition for a fusion -category. We focus particularly on Karoubi completeness, and study how higher condensation operators factorize into gauging interfaces
ESSAYS ON MICROECONOMIC THEORY
No full textThis dissertation presents three essays on consumer and firm behavior, with a particular focus on integrating bounded rationality. The first paper develops a list-based random attention model to study how stochastic limited attention affects decision-making when an agent selects one item from an ordered list of alternatives. The model attributes stochastic choice to variations in the number of items considered by the agent, which may differ across lists. I show that a finite dataset of lists’ choice distributions can be rationalized by the model if and only if these distributions satisfy two testable conditions. When the choice distributions are rationalizable, both conditions convey information on the agent’s unobserved attention behavior and preference. As an application, I explore how a list designer can construct a list of alternatives that maximizes worst-case profit, given partial information (extracted from the data set) on the agent’s attention behavior and preference. The second paper introduces revealed preference tests for the profit-maximizing behavior of a monopolist producing either a single good or multiple goods. The tests impose no parametric assumptions on the firm’s production or demand functions. I characterize when observed market outcomes are consistent with profit maximization, allowing for both time-invariant and time-varying demand functions, whether known or unknown to the modeler. I then propose a revealed preference test for Bertrand-Nash equilibriua among firms producing differentiated goods. The third paper presents a model where some consumers pay only limited attention to the good on offer and may choose not to consider a product for purchase simply because of its price, without considering its net utility. We study the implications of this type of bounded rationality on firm behavior in both a monopoly market and a duopoly market
EARLY LIFE DRIVERS AND DEVELOPMENT OF ADHD IN THE BOSTON BIRTH COHORT: APPLICATION OF MULTIDIMENSIONAL RISK FACTOR ANALYSIS, LIPIDOMICS, AND MEDIATION ANALYSIS
No full textAttention-Deficit Hyperactivity Disorder (ADHD) is a common neurodevelopmental disorder affecting approximately 7 million children and adolescents in the U.S., according to the 2022 National Survey of Child Health. Despite its prevalence, the etiology of ADHD remains incompletely understood, underscoring the urgent need to identify early-life determinants and inform primary prevention strategies. This dissertation examined the additive and interactive effects of multi-dimensional early life risk factors on childhood ADHD, as well as potential underlying biological pathways, using data from the Boston Birth Cohort (BBC) - a racially, ethnically, and socioeconomically disadvantaged population at elevated risk for ADHD. Specifically, this dissertation addressed three aims:
Aim 1 examined the combined impact of multilevel, across-domain risk factors on childhood ADHD risk in 2,265 mother-child dyads (591 ADHD, 1674 neurotypical) using a novel multilevel analytical approach. Six risk factors - maternal nativity, race/ethnicity, smoking, pregnancy stress, preterm birth and/or intrauterine inflammation, and child’s sex assigned at birth - were identified as key contributors, collectively explaining a large variation in ADHD risk ranging from 8% to 66% through their additive and interactive effects.
Aim 2 assessed the association between maternal and cord plasma lipid metabolites and childhood ADHD risk in 701 mother-child pairs (378 ADHD, 323 neurotypical) using lipidome-wide association analysis with adjustment for multiple tests. While no significant associations were observed for maternal lipids, 104 cord plasma lipid metabolites across 15 classes were significantly associated with ADHD risk, with most showing positive associations with ADHD.
Aim 3 explored potential biological pathways linking maternal nativity (U.S.-born vs. non-U.S.-born) to offspring ADHD risk via mediation analysis in 690 mother-child dyads (317 ADHD, 373 neurotypical). Two cord plasma fatty acids - myristoleate and myristoleic acid - each mediated approximately 20% of the association between maternal nativity and ADHD risk.
Collectively, these findings highlight the additive and interactive influence of multiple early-life risk factors on ADHD development and suggest the cord plasma lipidome as a potential early biomarker of ADHD risk and a window into possible biological pathways. If validated in future studies, these results may inform innovative strategies for early risk assessment, targeted screening, and preventive interventions
EDUCATION KNOWS NO BOUNDARIES: TOWARD EQUITY FOR TRANSNATIONAL FAMILIES IN AMERICAN EDUCATION
No full textIn the United States, a growing number of transnational individuals maintain simultaneous cultural, economic, and social ties across two or more countries. For educational settings, this shift has led to a rising population of transnational families and teachers who mobilize multiple linguistic and cultural repertoires to navigate and communicate across national and institutional boundaries (Nash & Skerrett, 2025). Yet, monolingual educational policies continue to dominate American public schooling, even in diverse metropolitan areas, making both teaching and learning misaligned with the lived realities of students, families, and educators.
While existing research has explored how transnational students negotiate their identities in school, little attention has been paid to how other key stakeholders–families and teachers–experience and respond to monolingual schooling norms. This qualitative study investigates how these transnational stakeholders (re)construct their identities in relation to language ideologies and cultural expectations in K-12 education. Guided by Bronfenbrenner’s (1994) Ecological Systems Theory and Spencer’s (2008) Phenomenological Variant of Ecological Systems Theory, this study explores how system-level factors, from ideological frameworks and institutional policies to classroom practices, intersect to influence transnational school communities’ identity development, sense of belonging, and coping responses.
Data were collected through open-ended questionnaires and follow-up interviews with families and teachers across three metropolitan cities: Los Angeles, Miami, and New York City. Findings reveal both shared challenges and moments of resistance in navigating monolingual language policies. Ultimately, by centering the voices of transnational educators and families, this study aims to contribute to the development of more sustainable and equitable educational practices and policies that affirm transnational identities and help shift the culture of power in schools (Villenas, 2019)