Alliance One Tobacco (Malawi)
Academic Research Repository at the Institute of Developing EconomiesNot a member yet
52305 research outputs found
Sort by
Measurements of Black Hole and Accretion Properties from Observations of Strong Field Gravitational Lensing
No full textAdvancements in Very Long Baseline Interferometry (VLBI) have ushered in a new era of high-resolution imaging that now allow us to probe electromagnetic signatures around black holes on unprecedentedly short gravitational length scales. These include the first images of black hole shadows produced from observations of the near-horizon environments of the supermassive black holes Sgr A∗ and M87∗ by the Event Horizon Telescope (EHT). Accessing this information, however, requires the development of novel techniques that can infer the properties of sources from the visibility domain products of VLBI. In this thesis, we present developments towards techniques for extracting information about black holes and their accretion environments from VLBI observations of low luminosity active galactic nuclei (LLAGN). We first measure the geometric features of the images of M87∗ from data acquired during the 2018 EHT campaign, where we show that the appearance of the source is consistent with the Kerr hypothesis. We then outline a novel, automatic differentiable, general relativistic ray tracing code which we use to define a dual-cone emission model for LLAGN. We show that the dual-cone model can accurately reproduce synthetic observations of the averaged accretion flows present in general relativistic magnetohydrodynamic (GRMHD) simulations of M87∗ and Sgr A∗. We then fit the dual-cone model to the EHT’s 2017 observations of M87∗ to measure properties of the system’s accretion flow and central supermassive black hole. We then study the effects of intrinsic source variability on the inference capabilities of the dual-cone model by performing a multi-epoch fit to the EHT’s 2017 and 2018 observations of M87∗, and present paths towards mitigating these effects. Finally, we discuss the development of a Gaussian Markov random field model as a potential solution for variability mitigation when performing snapshot and multi-epoch measurements from observations of M87∗ and Sgr A∗Physic
Engineering Vitrification Methods for Nanoscale Visualization of Dynamic Cell Processes
No full textBiological visualization is a critical tool for understanding and communicating biological processes, mechanisms, and interactions. A key gap in current visualization technology is the simultaneous imaging of processes that take place in nanoscale spatial regimes and millisecond temporal regimes. This work addresses these gaps through the development of a new tool for preparing biological samples, in addition to improvements in the sample preparation itself to allow access to a range of previously inaccessible thicker samples. Additionally, we address the development of biological cell lines expressing proteins crucial to human health and physiology, and the subsequent imaging of fluorescently tagged protein constructs we express in these cells. The data analysis to process these images is nontrivial, and we therefore develop new algorithmic techniques for analyzing the colocalization of two protein species in high density and noisy experimental contexts.Engineering and Applied Sciences - Applied Physic
Models of Human Decision-Making for Planning Digital Interventions Using Reinforcement Learning
No full textDigital interventions support people in sustaining effort toward their long-term goals, yet the user-level data available to personalize these interventions is scarce and noisy. This dissertation studies how reinforcement learning (RL) systems can make high-quality decisions under these data limitations by modeling the human user.
The first part of the thesis focuses on what human model to use. Drawing from behavioral science, I formalize Behavior Model RL (BMRL), a two-agent framework in which the human is represented as a sequential decision-maker with potentially maladapted Markov Decision Process (MDP) parameters, and the AI intervenes on these parameters to help reach their goal. BMRL requires us to specify how the AI agent models the other human agent. In practice, we must simplify our human models to support online learning for the AI, even knowing that these assumptions do not fully reflect real users. For instance, I introduce a simple, computationally tractable model of a user's goal-directed behavior in digital settings. I also point out that, as a field, RL simplifies its agents' discount models; we use exponential discounting to model agents (including human ones), despite evidence from psychology that humans discount hyperbolically. I examine how such modeling choices affect an AI’s ability to learn policies online, and I develop tools to understand whether these simplified models can generalize to more complex human behaviors.
The second part of the thesis examines how to learn human models online. I study the bias-variance trade-offs that arise when personalizing to individuals with limited data, and propose algorithms that manage model complexity over time. One method learns how model complexity should evolve by transferring "kernel evolution"' trajectories from prior users to new users, enabling rapid and stable online model selection in Gaussian Process regression. Another method increases the size of the AI's state space model of the human as more data is collected.
Together, these contributions provide a computational foundation for embedding behavioral insights into human models so that we can plan digital interventions using RL. They illustrate that the right inductive biases can enable AI systems to plan effective, interpretable, and personalized support for behavior change.Engineering and Applied Sciences - Computer Scienc
Nature and Nation: Sociobiology and the Emergence of Feminist Science Critique in the Postwar United States
No full textThis dissertation expands and explores the archive of the 1970s-1980s “sociobiology debate” to redefine its political stakes, that far exceed academic impropriety and the emotions of two tenured Harvard professors, namely Edward O. Wilson and Richard C. Lewontin. It is a mistake to assume, this dissertation contends, that the sociobiology debate was merely an egomaniacal contest between two Ivy League personalities when in fact Wilson’s theories of human nature, that attracted opposition among political activists in the U.S. and abroad, were of consequence to the marginalization of women in the American workforce, the backlash against civil rights in the U.S. North, and the cultural power of the Left in Western higher education. An expanded archive reveals that this infamous scientific controversy was not only a significant iteration of the nature-nurture debate after the Second World War but also a conflict between Wilson and the radical science movement in the U.S. and elsewhere, whose Cold War valence cannot be ignored. We should, this dissertation suggests, understand the sociobiology debate as the story of the anticommunist repression – and subsequent deradicalization – of the radical science movement, that aimed to redirect science in the service of social benefit, during the Cold War. My study focuses on the American radical science movement in particular, and among sociobiology’s broad coalition of critics, I examine Science for the People (SftP), science feminists, and the International Committee Against Racism (InCAR). My work intervenes not only in the history of biology but also in the history of radical science. This dissertation makes the case that the archive of U.S. biology should include its suppression of science critique, and that the archive of radical science should, similarly, include its history of anticommunist repression. Ultimately, I pose the provocation that the real casualty of the sociobiology debate was not E.O. Wilson’s scientific reputation but the second wave of the radical science movement in the United States. Reckoning with this recent history, I conclude, is imperative to today’s efforts to correct the past wrongs of science, and reorient its course towards democracy and justice.History of Scienc
Learning, Optimization, and Control for Real-World Physical Systems
No full textRecent breakthroughs in machine learning and artificial intelligence have led to key advancements in many areas such as natural language understanding, game playing, and simulated locomotion. Two important algorithmic frameworks underpinning this success are reinforcement learning (RL) and data-driven, gradient-based optimization. Leveraging these RL and optimization frameworks for real-world physical systems, however, remains challenging: real systems yield limited data, rarely offer accurate analytical models, suffer from sim-to-real discrepancies, and often operate as large, interconnected networks. In this thesis, I seek to remedy these challenges in the following ways. First, I present a theoretically principled representation-based RL framework for stochastic nonlinear control which utilizes dynamics knowledge and structure to improve sample efficiency in RL. I will then discuss how this approach can be extended to the important challenge of scalable control in networked dynamical systems. Next, I will introduce efficient Bayesian and zeroth-order toolkits designed for sample-constrained black-box optimization, a pervasive challenge in many real-world engineering systems. Together, these contributions chart a path toward reliable, data-efficient learning-based optimization and control of complex real-world physical systems.Engineering and Applied Sciences - Applied Mat
Beyond Intention-to-Treat Analyses in Randomized Trials: Utilizing Per-Protocol Estimands and Observational Data to Investigate Strategies for Cardiovascular Disease Prevention
No full textRandomized trials are considered the gold standard study design for conducting comparative effectiveness research. Yet, evidence from randomized trials to inform cardiovascular disease (CVD) prevention efforts remains scarce for some patient populations, partly due to restrictive eligibility criteria and shorter follow-up of surrogate outcomes. Emulating target trials in observational data can provide insights into the effects of existing medications for CVD prevention in new target populations and on longer-term clinical endpoints. Regardless of study design, that is, whether we use data from randomized trials or emulate target trials in observational data, complementing intention-to-treat analyses, which estimate the effect of the assigned treatment strategies, with appropriate per-protocol analyses, which estimate effects that would have been observed had there been full adherence to the assigned treatment strategies, can be important to facilitate informed clinical decision-making. In this dissertation, I estimate both intention-to-treat and per-protocol effects using data from a large-scale, pragmatic randomized trial and from electronic health records and health insurance claims to elucidate the impact of medications important for CVD prevention in populations for whom evidence has been limited based on existing intention-to-treat analyses of randomized trials.
In Chapter 1, I estimated the effect of adhering to assigned treatment strategies of pravastatin or usual care on death and CVD in the Antihypertensive and Lipid-Lowering Treatment to Prevent Heart Attack – Lipid-Lowering Trial. The high initiation of lipid-lowering therapy in the usual care comparator group has been previously suggested to explain the null intention-to-treat findings. I demonstrate that deviations from the pravastatin treatment strategy may better explain the intention-to-treat results, showing that pravastatin reduced the risk of death and CVD even when allowing individuals in the usual care group to initiate lipid-lowering therapy. In Chapter 2, I estimated the effect of statin therapy versus usual care on the 5-year risk of CVD among women with breast cancer. I first specified the target trial and then emulated the trial using electronic health record data from a large cohort of patients at several Kaiser Permanente networks. While intention-to-treat findings showed no difference in risk of CVD over five years, the risk of CVD was lower for the statin therapy group compared to usual care in per-protocol analyses, though estimates were imprecise. In Chapter 3, I estimated the effects of adding GLP1-RA to SGLT2i and adding SGLT2i to GLP1-RA on glycemic control and intermediate cardiovascular risk factors at one-year, and risk of CVD over three-years among persons with inadequately controlled type 2 diabetes. I outlined the protocol of two target trials, then emulated them using electronic health records and claims data from 12 insurance/healthcare systems in the US. Regardless of initial background treatment, adding SGLT2i to GLP1-RA reduced HbA1c, systolic blood pressure, and body mass index after one-year. Estimates for the effect of combination therapy on cardiovascular events were imprecise, and additional data are needed to confirm whether adding SGLT2i or GLP1-RA leads to long-term protection against cardiovascular disease.
In conclusion, when evidence from intention-to-treat analyses of existing randomized trials is insufficient to guide clinical decision-making alone, using per-protocol effects and pragmatic trials as a framework for conducting comparative effectiveness research in observational data can help inform future CVD prevention efforts for populations at high risk, such as breast cancer survivors and patients with type 2 diabetes.Population Health Science
Beholding Greece: Viewing Panhellenic Sanctuaries in Late Antiquity
No full textPrevailing scholarly narratives have characterized Panhellenic sanctuaries in Late Antiquity as enclaves of dwindling pagan worship that were subsequently destroyed or appropriated by Christians. Contrary to these narratives, this dissertation argues that Panhellenic sanctuaries continued to function in Late Antiquity as sites where both Christian and non- Christian visitors could participate in cultural, civic, and intellectual traditions. This dissertation offers a new approach to Panhellenic sanctuaries by examining both archaeological and textual evidence for interactions with the monumental landscapes of the sanctuary of Apollo at Delphi and the sanctuary of Zeus at Olympia in the late third through the sixth centuries AD. The results of this study reveal that rather than attempting to obscure monumental testimonies of the ancient Greek pagan past, the stewards of sanctuaries in Late Antiquity maintained the monumental landscapes as distinct, recognizable spaces by selectively preserving monuments, by making new additions that respected existing topography, and by investing in architecture that facilitated visitors’ ability to view the landscape. By situating this pattern alongside practices associated with the intellectual culture of the Second Sophistic in the earlier Roman imperial period, this dissertation proposes that these acts of preserving and framing the monumental landscapes of the sanctuaries in Late Antiquity allowed visitors of different religious commitments to express a shared paideia and draw personal connections to the ancient past through theōria, the contemplative viewing of the mythological and historical landscapes of these renowned sites.Classic
Structural and Functional Studies of Transcription Through Chromatin
No full textTranscription elongation by RNA polymerase II (Pol II) through chromatin presents a formidable challenge, as nucleosomes impede Pol II progression. Despite the nucleosomal barrier, eukaryotic cells routinely achieve high-fidelity gene expression while preserving chromatin structure, ensuring both transcriptional fidelity and epigenetic continuity. This dissertation investigates the molecular mechanisms by which Pol II overcomes the nucleosomal barrier and maintains nucleosome integrity during elongation. Through cryo-electron microscopy and in vitro biochemical approaches, I characterize structural intermediates that reveal how Pol II and associated elongation factors coordinate histone displacement, retention, and reassembly. I describe a nucleosome retention transcription elongation complex in the absence of FACT and a distinct hexasome intermediate stabilized by multiple acidic blocks in the N-terminal domain of SPT6 during FACT-assisted transcription. Together, these structures visualize chaperone-dependent and -independent models of nucleosome retention. Complementary in vitro assays further demonstrate how transcription-generated nucleosomal states modulate the enzymatic activity of transcription-associated chromatin modifiers RNF20/40-UBE2A ubiquitin ligase and MLL1 complex. Together, these findings establish a structural and biochemical framework for how the elongation machinery preserves chromatin architecture while directly coupling nucleosome traversal to the installation of transcription-associated histone modifications.Biological and Biomedical Science
Neuroanatomical Asymmetry Across Species: From Mice to Macaques to Human Insights
No full textHow left-right anatomical asymmetry is established in the human brain is a major unanswered question in the field of laterality. Efforts to address this area of brain development are limited by the scarcity of animal models with such asymmetries, particularly for mammals whose brain structure and development are most similar to that of humans. Identifying patterns of anatomical brain asymmetry in other mammalian species would lay the groundwork for pursuing mechanistic studies of the molecular and cellular processes underlying its generation. Here, I led two studies using automated image analysis to perform extensive characterization of anatomical brain asymmetry in two key mammalian species of interest: mice (M. musculus) and rhesus macaques (M. mulatta).
Using seven separate mouse datasets totaling over 3500 animals, I identified a global anterior-posterior asymmetry pattern in the mouse brain but the absence of regional asymmetries at individual structures. Anterior regions in the mouse brain are greater in volume and surface area on the right compared to left and are shifted anteriorly on the right compared to left. Posterior regions are greater in volume and surface area on the left compared to right and are shifted anteriorly on the left compared to right. In macaques, using seven datasets totaling nearly 700 animals, I identified 80 regional asymmetries, including a right hemisphere frontal lobe expansion and right-larger auditory regions, which are present as early as 1 month after birth and persist into adulthood. Yet, in contrast to mice, I observe no global asymmetry pattern in macaques. Extensive use of cross-validation demonstrates the results to be reproducible across independent datasets, across two different image modalities in mice, and across two different image registration software.
Together, my studies establish high-confidence, atlas-based patterns for studying neuroanatomical asymmetry in both mice and macaques, robust to numerous biological and technical variables. These atlas-based patterns can serve as a foundation for future studies examining the action of genetic factors in neuroanatomical asymmetry, characterizing the underlying cellular architecture, or further probing the relationship between anatomical and functional brain asymmetries. Yet, interrogating these patterns also reveals three significant observations that advance our understanding of brain laterality. 1) The presence of human-like regional asymmetries in macaques but not mice suggests this aspect of laterality may have emerged within the primate lineage. 2) The mouse results show no relationship between anatomical asymmetries and known mouse functional asymmetries, indicating that functional asymmetry can be decoupled from macro-anatomical asymmetry. 3) The regional pattern of anatomical asymmetry in macaques is not correlated at all with that seen in humans, suggesting that, although some instances of consistent asymmetries may exist across species, the regional patterns are generally not evolutionarily conserved. Finally, my findings in mice reconcile two previously published studies on the question of neuroanatomical asymmetry in mice which on their surface, appeared to show major differences in their findings but which co-exist within the results of my analysis. Overall, my thesis research has provided a foundation for future cellular-level studies and advanced evolutionary understanding of anatomical left-right asymmetry.Medical Science
Epistemic Limits of Trustworthy Machine Learning
No full textTheoretical understanding of a system’s limits has long driven technological breakthroughs. Carnot delineated the fundamental limits of heat engine efficiency, paving the way for the design of modern state-of-the-art engines. More than a century later, Claude Shannon unraveled the fundamental limit of communication, known as channel capacity. This insight revolutionized communication systems, enabling continual improvements that ultimately led to wireless communication as we know it today.
This thesis discusses the epistemic limits of machine learning (ML) and leverages them to improve the trustworthiness of ML systems. ML models have an epistemic limit when proving one of their properties is impossible. Epistemic refers to the impossibility of providing theoretical guarantees (knowledge) about a model's property. Epistemic limits are information-theoretic converse results on the hypothesis test that checks a model's property.
First, we prove a limit on how much information personalized models can use while ensuring reliable test for performance gains across all users -- epistemic limits of personalization. We leverage this limit to develop a tool to help with feature selection. Second, we show a limit for reliably testing if model performance is equitable across multiple demographic groups --epistemic limit of fairness testing. We exploit this limit to design a metric for efficient algorithmic bias detection.
Third, we prove a limit for testing if one model outperforms another on average -- epistemic limit of model selection.
We use this result to delineate the set of indistinguishably good models --Rashomon set. Finally, we argue that the epistemic limits in model selection imply that explaining the predictions of ML models is necessary. Then, we develop efficient methods for explaining the content produced by large language models.Engineering and Applied Sciences - Applied Mat