71446 research outputs found
Sort by
Mate Choice in Phlox Wildflowers
No full textMate choice determines when, where, and how reproduction occurs. The summed effect of these decisions across generations drives evolutionary trajectories and patterns diversity across biological scales. Inspired by these far-reaching impacts, I focus on two primary arenas where plant mate choice decisions occur: 1) organismal interactions between plants and their pollinators and 2) cellular interactions between reproductive structures within the flower (pollen and pistil). In Chapter 1, I quantify the pollination environment in the Texas wildflower, Phlox drummondii, and identify high pollinator specialization towards a single butterfly species. Building on this empirical work in Chapter 2, I generate a novel theoretical framework for the role of pollinators as agents of dispersal. In Chapter 3, I test the hypothesis of increased self-fertilization in P. cuspidata as an adaptation to avoid costly hybridization with its related congener, P. drummondii. In Chapters 4 and 5, I use quantitative and functional genetics to investigate the genetic basis of the self-incompatibility mechanism active in P. drummondii. I map the genetic basis of intraspecific variation in the self-pollen rejection response and identify a novel gene causing self-pollen recognition in Phlox wildflowers. Taken together, my work integrates broad experimental approaches to explore how mate choice mechanisms function across biological scales.Biology, Organismic and Evolutionar
Molecular co-evolution between SARS-CoV-2 and human antibodies
Get PDFThe COVID-19 pandemic disrupted global health systems but simultaneously offered a rare
opportunity to observe the rapid co-evolution of a virus, SARS-CoV-2, with the human immune
system in real time. This evolutionary interplay is often thought of as a molecular arms race, in
which both pathogen and host must continually adapt to changes in each other’s phenotype. This
dynamic is particularly pronounced at the molecular recognition level. Human antibodies evolve
(via affinity maturation) to better neutralize the viral antigens (e.g., SARS-CoV-2 spike protein),
while the virus, in turn, accumulates mutations that confer immune escape. In this dissertation, I
present four projects that explore the co-evolutionary landscape between viral and immune
proteins. Chapters 1 and 2 focus on the combinatorial mutagenesis of the SARS-CoV-2 spike
receptor-binding domain (RBD) to measure the individual and combined effects of Omicron BA.1
mutations on ACE2 binding affinity (Chapter 1) and antibody evasion (Chapter 2). Chapter 3
extends this approach to later Omicron subvariants, using high-throughput genotype-to-phenotype
mapping to investigate how mutational effects depend on genetic background across divergent
lineages. Finally, Chapter 4 shifts focus to the host immune response, using deep sequencing of
human peripheral blood B cell repertoires and yeast display of candidate antibodies to measure
evolving antibody phenotypes. Together, these studies provide an integrated view of viral and
immune co-evolution, inferring the molecular principles that govern host-pathogen interactions
and the trajectory of viral adaptation.Biology, Organismic and Evolutionar
Ties that Bind: Semantic and Social Relationships in Inscribed Paintings of the “First Skipjack Tuna of the Season”
No full textThis dissertation is an exercise in comprehending the history and range of a single poetic topic/visual subject matter in inscribed paintings made in the context of early modern Japanese comic poetry circles. In doing so, it attempts to capture the full semantic reverberation of meaning evoked by these works. The first four chapters of the dissertation traces one subject matter, the skipjack tuna (katsuo, Katsuwonus pelamis), throughout poetry and art in the Edo period (1603–1868), beginning with its development as a new seasonal poetic topic in haikai comic linked verse in the mid-to late seventeenth and early eighteenth centuries. The poetic topic was later transformed into a visual motif of the skipjack tuna paired with bamboo grass. The iconography first developed in inscribed paintings in the “mad verse” (kyōka) community as a way to visually convey the essence of kyōka around the turn of the nineteenth century. It circulated in this community for about thirty to forty years before being introduced to the wider public in ukiyo-e prints.
The skipjack-tuna-on-bamboo-grass motif is a unique example of a painting subject deriving entirely from contemporary Japanese comic poetry. Unlike other piscine subjects in Japanese art, it lies outside the bird-and-flower (C: huaniaohua J: kachōga) painting tradition, which is a significant and important painting category in East Asia. Nor does the iconography stem from earlier poetic precedents in classic waka poetry, which were visualized and standardized in later artwork. The skipjack-tuna-on-bamboo grass iconography therefore provides an important window into understanding the deep and resounding influence of haikai and kyōka poetry. Tracking the development of the theme of the skipjack tuna, the dissertation sheds light on the sophisticated ways in which the new poetic and seasonal topics in the early to mid-Edo period were inherited and ingested, spanning generations, genres, and modalities. By taking a diachronic and subject-based approach, this investigation offers a new strategy for tackling the inherent difficulties of understanding highly intertextual and “trans-textual” works such as inscribed paintings.
The fifth and final chapter of the dissertation examines how meaning is built over time in inscribed paintings in a case study of three nearly identical works by the artist and poet Kubo Shunman (1757–1820). The pieces depict the skipjack tuna on bamboo grass; each have seven poetic inscriptions written around the painted motifs. Meaning is evoked through complex interactions between the poetic inscriptions and visual subject matter. The internal semantic frameworks of the paintings are determined both by the deliberate selection of poems to maximize the array of referents, and the poems’ relative position on the paintings, which forms a set reading sequence. The paintings’ inscriptions “activate” the visual motifs’ latent meanings, filling in the context left open by the unfilled substrate of the paintings. Readers’ understandings of the meaning of the painted motifs change as they read each poem in the sequence over the course of their interpretive experience. To map out the semantic networks formed by interactions between the different parts of the inscribed paintings, I employ a data visualization system I invented called the Polar Relational Data Graph (PRDG), which is rooted in concepts from data science, basic graph theory, and Japanese poetic practices. Employing the concept of the “motif” from musical semiology, I show how referents develop through repetition throughout the poem sequence into overarching themes. By showing how poetry is the fundamental determinative aspect of the paintings both historically and semantically, my dissertation challenges common assumptions about where the value of an artwork lies. In doing so, it forges a new path for conducting text-image analysis.East Asian Languages and Civilization
International Corporate Tax Reform: Efficacy and Acceptance Assessment of the OECD/G20 BEPS Project.
Get PDFThis research focuses on evaluating the efficacy and acceptance of the ongoing international corporate tax reform led by the Organisation for Economic Co-operation and Development (OECD)/Group of 20 (G20) in addressing multinational enterprises (MNEs)’ tax avoidance and promoting a fairer international tax system. The OECD G20 Base Erosion and Profit Shifting Project (BEPS Project) was initiated in 2012 to set up an international framework to combat tax avoidance by MNEs using base erosion and profit shifting (BEPS) practices. The BEPS Project deliverable was delivered in 2015 and comprises a 15-point Action plan (BEPS package). To further Action 1, the OECD proposed a Two-Pillar Solution which was agreed by the Group of 7 (G7). Pillar 1 allocates rights of taxation of residual profits to market countries for large profitable MNEs while eliminating digital services taxes. Pillar 2 imposes a global minimum tax of at least 15% to large profitable MNEs.
I assess the effectiveness of the BEPS Project, introduced by the OECD, through a multi-level analysis over the 2010-2024 period. The multi-level analysis comprises (1) a country-level study of the evolution of corporate income revenues in OECD countries and (2) a firm-level study of the evolution of global effective tax rates paid by the largest MNEs worldwide.
The findings of these analyses are threefold. First, the country-level analysis shows no significant increase in countries’ corporate tax revenues in the immediate aftermath of the implementation of the BEPS package, suggesting a limited impact of the ongoing international tax reform. Second, the firm-level analysis reveals that the largest MNEs had lower global effective tax rates (ETRs) in the period following the introduction of the BEPS package compared to the period before, suggesting that the ongoing international tax reform did not increase the tax burden of the largest MNEs. Third, the firm-level analysis shows that most of the largest MNEs have an ETR above the global minimum tax rate of 15% introduced by Pillar Two of the BEPS Project, which indicates that either changes in domestic laws to implement Pillar Two are not necessary or the global minimum tax rate under Pillar Two is not ambitious enough.
While acknowledging limitations in the data availability and reliability, the findings of my multi-level analysis provide insights into the effectiveness of the OECD/G20 international corporate tax reform and insights for international tax policymaking. The discussion of the results and their limitations highlight the importance of data coverage, availability and accuracy in monitoring international tax policies as well as the need for further research. My findings also suggest that more targeted policies may be sufficient to address corporate tax avoidance, which provide valuable insights for stakeholders and policymakers.
The findings have significant implications. For the BEPS Project to achieve real efficiency and acceptance, it must address the challenges of data quality and reporting harmonization to allow reliable monitoring by stakeholders. For MNEs to comply with new data and reporting requirements, the BEPS Project’s implementation needs to provide further tax certainty and administrative simplification. More importantly and urgently, the BEPS Project will have to resist fragmentation of the Inclusive Framework and ensure effective inclusivity, particularly for developing countries.Extension Studie
Learning structured representations in neural networks
Get PDFDeep learning has revolutionized engineering fields where feature extraction and model-based approaches were traditionally used, yielding groundbreaking results in computer vision, natural language processing, and their intersection. These advances were made possible by the widespread use of GPUs for training deep networks, and these models frequently have billions (if not trillions) of parameters and require the equivalent of years to train. However, a multitude of learning problems in science and engineering are bounded by constraints: either user-defined (such as cost, compute, or memory constraints) or problem-defined (such as data or task symmetries, or physical constraints).
In this dissertation we translate such constraints into structured priors; and present how that structure can be embedded in neural networks by construction. In particular, in the first part we propose architectures that follow optimization-based priors (which we term user-defined) with provable guarantees, and in the second part we propose architectures that have algebraic structure (which we term problem-defined). Both of these frameworks lead to neural networks with provable priors, significantly reduced parameter counts, while maintaining (or improving) the performance of downstream tasks.Engineering and Applied Sciences - Computer Scienc
Scripts of Shame in Old Spanish Verse: Libro de Apolonio, Vida de Santa María Egipçiaca and Libro de buen amor
No full textThis dissertation traces “scripts of shame”—or guidelines for performing this emotion that readers can glean from engaging with literature—in a series of medieval Iberian narrative poems: Libro de Apolonio [Book of Apollonius], Vida de Santa María Egipçiaca [Life of Mary of Egypt, and Libro de buen amor (Book of Good Love). My main argument is that while these poems generate shame scripts, they address questions of female holiness, poetics, courtship, and, most notably, sex and sexual sin.
Chapter One situates this dissertation’s scholarly intervention within two related fields: affect studies and shame studies, laying the groundwork for the project’s historical, theoretical, and methodological frameworks. Chapter Two focuses on how the 13th-century Libro de Apolonio enacts discursive shame, reconceptualizing this poem as a sustained riddle about sex and sexual sin. Chapter Three explores the 13th-century Vida de Santa María Egipçiaca through the lens of medieval confession, arguing that the text both follows and deviates from the affective scripts of this sacrament. Finally, Chapter Four argues that the 14th-century Libro de buen amor engages with affective scripts that conceive of shame as both a deterrent to love and an integral part of love. By uncovering these scripts, my research provides new insights into the study of affect and medieval Iberian narrative poetry, demonstrating that while this poetry often produces and reifies emotional norms based on social and cultural hierarchies, it also allows for the subversion of these norms.Romance Languages and Literature
Essays on the Supplemental Nutrition Assistance Program
No full textThe Supplemental Nutrition Assistance Program (SNAP) is the U.S.’ largest nutrition assistance program, providing food vouchers to approximately 1 out of 8 Americans every month. This dissertation, comprised of three chapters, uses economics to examine how policies and social norms affect the costs of participating in SNAP and related programs for eligible, low-income households. My work evaluates how innovative policy changes can address the social and administrative challenges of participating in safety net programs.
Following the COVID-19 pandemic, the national budget of SNAP doubled, its caseload increased by 10 percent, and its application denial rate increased by nearly 50 percent. In Chapter 1, I investigate the factors behind these persistent enrollment changes, including economic conditions and policy changes. I compile a new dataset on state policy waivers during the COVID-19 public health emergency, including unprecedented flexibilities in enrollment procedures and increases in benefit amounts, and I use state-level policy variation to understand the effects of each policy on SNAP caseloads. I find that emergency supplemental benefits and recertification waivers drove enrollment increases. I estimate an elasticity of SNAP enrollment with respect to benefit size of 0.09-0.18. These results suggest that government policies can be more influential than economic conditions in determining transfer program caseload patterns.
Negative social judgments or “stigma” about the receipt of government benefits may discourage participation and impose costs on those already receiving benefits. But who experiences stigma in SNAP, how it affects participant decisions, and whether it can be reduced remain unclear. In Chapter 2, co-authored with Alice Heath and Michael Holcomb, we use a nationally representative survey experiment to measure levels of stigma across social groups, and we test whether low-touch interventions intended to alleviate stigma may affect SNAP participation. We find that stigma varies by political affiliation and SNAP participation status: Democrats report lower levels of stigmatizing beliefs than Republicans, and SNAP participants report lower levels of stigma than non-participants. Our three randomized interventions have heterogenous effects: they decrease stigma among Democrats and those with low expectations of judgment from others, increase stigma among Republicans, and have no effect on those with high expectations of judgment. One intervention that addresses a common zero-sum concern—that enrolling in SNAP prevents others from receiving benefits—increases interest in take-up among eligible non-participants while decreasing support for SNAP spending among the general population. Our findings indicate the importance of both social norms and political orientation for influencing public support for SNAP and willingness to participate. Reducing stigma may require targeted messaging towards different demographic groups and program design that reduces potential for stigma in settings where benefits use is observable.
In Chapter 3, I investigate the growth of online grocery purchasing with food vouchers across retailer markets and its effects on low-income households who use benefits. I collect new data on initial authorizations of Electronic Benefit Transfer (EBT) transactions online across more than 400 food retailers and all fifty states. Although the policy initially affected only SNAP, online purchases with new voucher programs for school children—Pandemic EBT and Summer EBT—were later covered under the authorizations. I find that stores authorized for online benefit purchases were more likely to be large retailers located in urban and higher-income neighborhoods compared to those authorized for in-store purchases only. Then, I use quasi-experimental spatial variation in EBT online purchasing availability to estimate effects on benefit redemption patterns and SNAP enrollment. The proliferation of online shopping for benefit users was expected to improve participant experiences by reducing stigma, improving convenience and access to food stores, and streamlining voucher purchases—despite facing service fees and potentially higher product prices online. I estimate that online EBT exposure led to a $16 increase in monthly online grocery spending per EBT-using household. Households substitute away from in-store spending at large food retailers. Finally, I find that online grocery purchasing availability increases local SNAP participation by 4 percent, primarily by increasing retention of existing participants. These results suggest that policies which streamline participants’ benefit redemption experiences can improve the effectiveness of in-kind benefit programs.Public Polic
Crime and Democracy: Central America in Comparative Perspective
No full textThis dissertation explores the relationship between organized crime and liberal democracy. Across three papers, I draw on evidence from Central America to generate new theoretical and empirical insights about how crime (re)shapes—and often undermines—democratic institutions. The first paper examines the relationship between organized crime and the durability of democracy. I show that organized crime can facilitate democratic reversals through both "supply-side" and "demand-side" effects. The second paper introduces the concept of criminal electioneering: deliberate efforts by criminal groups to influence elections. I provide a typology of criminal electioneering strategies and argue that criminals are more likely to engage in electioneering (a) when they are locked in direct competition with rival criminal groups and (b) when candidates are in close "resource races" with their electoral opponents. The final paper examines the determinants of support for "authoritarian bargains" related to public security. Using survey evidence from Guatemala, I show that information about the costs of harsh anti-crime policies for democracy tempers support for those policies among voters—even in contexts of high crime and insecurity.Governmen
Semi-supervised and Representation Learning for Improved Classification and Stratification in EHR Data
Get PDFThe rapid digitization of healthcare has given rise to vast repositories of electronic health record (EHR) data, offering unprecedented opportunities for data-driven advancements in disease prediction, patient stratification, and clinical decision-making. However, the high dimensionality, sparsity, and heterogeneity of EHR data present unique statistical and computational challenges. Moreover, the scarcity of high-quality labels—due to the cost and complexity of manual annotation—further complicates supervised modeling efforts. This dissertation addresses these challenges through a unified framework of semi-supervised learning and representation learning for improved classification and stratification in EHR data, with applications to phenotyping, disability prediction, and patient subgroup discovery.
The overarching goal of this work is to develop scalable, robust, and interpretable methods that leverage both labeled and unlabeled EHR data, improve generalizability across populations, and uncover clinically meaningful structure in complex disease settings. The dissertation is composed of three interrelated papers, each tackling a key methodological bottleneck in modern EHR-based machine learning: (1) evaluating model performance under distributional shift, (2) learning rich patient representations in the presence of limited labels, and (3) stratifying heterogeneous patient populations using outcome-informed embeddings.
In Chapter 1, we consider the problem of evaluating the performance of binary classifiers when labeled data are unavailable in a target population. This setting is common in clinical phenotyping tasks, where models are trained using limited chart-reviewed labels in one cohort and then applied to other cohorts with potentially different covariate distributions. We propose STEAM Semi-supervised Transfer lEarning of Accuracy Measures), a doubly robust estimation procedure for receiver operating characteristic (ROC) parameters under covariate shift. STEAM combines calibrated density ratio weighting with robust outcome imputation, using both unlabeled source and target data to improve efficiency while protecting against model misspecification. Through theoretical guarantees and empirical results, we demonstrate that STEAM enables accurate performance assessment in unlabeled target populations, with applications to phenotyping models in rheumatoid arthritis on temporally evolving EHR cohort.
Building on the challenge of label scarcity, Chapter 2 shifts focus to semi-supervised representation learning for predictive modeling. We propose SCORE (Semi-supervised Clustering thrOugh REp-
resentation learning), a generative embedding framework that models the joint distribution of high-dimensional EHR features using a multivariate Poisson-LogNormal distribution, with pretrained code embeddings capturing semantic relationships between clinical concepts. SCORE integrates limited labeled data via a hybrid Expectation-Maximization and Gaussian Variational Approximation algorithm, enabling efficient and theoretically sound inference in large-scale, partially labeled cohorts. We show that SCORE produces informative and transferable patient embeddings, improving prediction of disability status in multiple sclerosis (MS) and outperforming conventional supervised and unsupervised methods.
Finally, Chapter 3 addresses the critical task of patient stratification in heterogeneous diseases. We focus on Alzheimer’s disease (AD), where progression and prognosis vary substantially with age. We propose SOLAR (age-Specific Outcome-guided representation Learning for pAtient clusteRing), a novel clustering framework that incorporates time-to-event outcomes and explicitly models age-group structure using a multitask learning paradigm. SOLAR jointly learns low-dimensional patient representations across age groups, encouraging shared structure while allowing age-specific flexibility. By integrating survival information and modeling age-related heterogeneity, SOLAR identifies clinically meaningful AD subtypes with distinct prognostic profiles, improving both interpretability and clinical utility over existing age-unaware or outcome-agnostic methods.
Together, these three works present a cohesive framework for semi-supervised and representation learning in EHR analysis. The methods developed here contribute new strategies for evaluating, predicting, and stratifying patient outcomes in data-scarce, high-dimensional clinical settings. In doing so, they aim to advance the broader goals of personalized medicine and evidence-based healthcare by making machine learning more robust, scalable, and clinically relevant.Biostatistic
Peering from the Parapet of Perturbative QFT
Get PDFModern experiments such as the Large Hadron Collider probe new physics through precise measurements that over-constrain Standard Model parameters to sub-percent accuracy, alongside broad searches for beyond-Standard-Model signals. Central to these efforts is perturbation theory, which approximates quantum field theory computations as series expansions in small parameters. Although the complete result lies in the full series, only the first few terms are computationally accessible, raising the question: when can perturbative predictions be trusted, and what alternatives exist when they fail? In this dissertation, we demonstrate that effective field theories (EFTs) can restore convergence by resumming perturbative series to all orders in the coupling. We illustrate this approach with examples from backward scattering in gauge theories, and the extraction of the strong coupling constant using heavy jet mass distribution from electron-positron colliders. Furthermore, by leveraging precision experimental data, we refine power counting in bottom-up flavor EFTs, imposing strong constraints on UV models of the SM flavor hierarchy. Finally, we address the breakdown of perturbation theory--manifested in non-convergent asymptotic series—-by interpreting diagrammatic renormalons as saddle points of the effective action, thereby developing a new probe into their existence via the path integral.Physic