71446 research outputs found
Sort by
Mechanistic studies of the Staphylococcus aureus cell division protein SmdA
No full textUnderstanding bacterial cell division is essential in the fight against bacterial infections and the
current antibiotic resistance crisis. In Staphylococcus aureus, cell division must be carefully coordinated
to successfully divide with consideration to its unique geometry. Many proteins in S. aureus have been
directly and indirectly implicated in promoting correct division. One protein, named SmdA, was
identified by our group and others to be essential for proper septal placement and resistance to
antibiotics. However, no mechanism regarding how SmdA is able to promote successful cell division
was proposed. In this work, I characterized SmdA both in vitro and in vivo. SmdA has a nuclease
related domain, or NERD, and I showed that this domain has exonuclease activity. It cleaves single
DNA in the 3’ to 5’ direction. To the best of my knowledge, no other NERD protein has been shown
to have nuclease activity. I then used TnSeq to define ΔsmdA’s genetic interaction network and
discovered that several proteins important for S. aureus cell division become essential when smdA is
lost. Three of these proteins, EzrA, PcdA, and FacZ, interact directly or indirectly with FtsZ, a tubulin
homolog that assembles in the presence of GTP to form the Z-ring that initiates cell division. I
therefore hypothesized that SmdA may also interact with FtsZ. Consistent with this, a suppressor
mutation in ftsZ, previously characterized in the Walker lab, corrects the cell division defects caused
by loss of smdA. In silico modeling also predicts with high confidence that that SmdA and FtsZ form
a complex. I conclude that SmdA directly regulates FtsZ dynamics to promote successful cell division.
Although further work needs to be conducted to validate the proposed model for how SmdA regulates
FtsZ dynamics, the findings in this work advance our understanding of SmdA and will ultimately lead
to a better understanding of cell division in this critical pathogen.Chemical Biolog
Enhancing the analysis of the large-scale structure of the Universe for cutting-edge cosmological surveys with two-point correlation function and beyond
Get PDFFor many years, we have known that the Universe is vast and expands with acceleration on the largest scales. However, dark energy, the substance supposedly driving this acceleration, might also weaken with time. Or at least the current standard model of cosmology has a problem explaining all the highest-quality data available.
We present one of the key technical ingredients that enabled the Dark Energy Spectroscopic Instrument (DESI) Baryon Acoustic Oscillation (BAO) distance measurements — the semi-analytic covariance matrices for the two-point correlation functions of point tracers. We then briefly discuss the cosmological implications of DESI BAO results, including the suggestion of dark energy. Then, we discuss a curious possibility of relieving the Hubble tension, the discrepancy in the expansion rate of the Universe today obtained directly from a Hubble diagram versus inferred indirectly from the CMB, without introducing fundamentally new physics. In the end, we explore a novel analysis technique combining galaxy redshift surveys with the data from the thermal Sunyaev-Zeldovich effect, a secondary anisotropy in the CMB. We aim to interpretably extract more valuable cosmological information from both than standard 2-point summary statistics allow. This would enable better consistency tests for the concordance model of cosmology and potentially new, exciting discoveries.Astronom
Manipulation of NK-cell immunity via metalloproteinases and chemokine receptor redirection in viral infection and cancer
No full textNatural killer cells perform their role of immunosurveillance and target killing by active engagement with ligands in the environment with their activating and inhibitory receptors, as well as their cytokine and chemokine receptors. It is well known that metalloproteinases can cleave ligands and chemokines to modify ligand-receptor interactions, chemotactic activity, and overall function. However, studies addressing how metalloproteinases manipulate these processes and how we can harness these to enhance natural killer cell activity in viral infection and cancer are lacking. Here, we discover the underlying mechanisms by which SARS-CoV-2 and related sarbecoviruses induce metalloproteinase-mediated cleavage of critical NK cells ligands, MIC-A/B, to evade natural killer cell immunity. We also probe the role of the metalloproteinase-cleaved chemokine CXCL16, which has demonstrated an increasingly important role in triple negative breast cancer. Lastly, as an extension of our work in probing chemokine-chemokine receptor axes, we set the groundwork to rewire NK-cell chemotaxis by introducing novel chemokine receptors as a therapeutic strategy to redirect them into tumor tissues and fully harness their cytotoxic potential.Graduate Educatio
Barriers and Bridges: Essays on Local-Migrant Dynamics and Refugee Integration
No full textThis dissertation investigates the conditions under which refugees are more welcomed by local communities. I focus on Turkey, the country that currently hosts the largest population of Syrian refugees in the world \citep{unhcrTurkey} and examine how host society attitudes, policy preferences and behaviors toward refugees are shaped by social interactions, aid framing and information-based interventions. Using a series of original survey experiments, I explore the underlying causes of exclusion and identify strategies that can foster social cohesion in contexts of forced displacement. Each chapter focuses on a distinct factor influencing refugee inclusion: residential preferences, the framing of international aid and intergroup contact followed by corrective information. These studies shed light on how local communities react to the arrival of refugees and offer useful lessons for shaping integration policies, especially in countries in the Global South.
In Chapter 1, my co-authors, Karen Ferree and Kristen Kao and I examine whether residential segregation between migrants and hosts reflects symmetric preferences to live among co-ethnics or is primarily driven by host-side avoidance. This chapter draws on a door-to-door survey of 5,000 Turkish and Syrian residents in the city of Adana, a major hub of Syrian displacement. We embed a conjoint experiment in the survey to isolate how neighborhood demographic composition, crime, social capital and public services affect individuals’ willingness to move into hypothetical neighborhoods. The results reveal a striking asymmetry. Turkish citizens strongly prefer to live in neighborhoods with high concentrations of their ingroup and are particularly sensitive to Syrian outgroup size, often preferring higher-crime areas over neighborhoods with more than 30 percent Syrian residents. In contrast, Syrian respondents show no evidence of preferences for segregation. Refugees care more about feeling safe in their neighborhoods and don’t place much importance on the ethnic makeup of the area. This suggests that segregation in Turkey is not something both sides choose, it’s mainly driven by locals who prefer to live apart. That raises a bigger question: why are so many citizens hesitant to live next to refugees? The following chapters address this question by examining how the framing of refugee-related international policies shapes public responses and by exploring the role of intergroup contact and exposure to accurate information in reducing exclusionary attitudes.
In Chapter 2, co-authored with Melani Cammett, I investigate how the framing of international aid programs influence local support for refugee inclusion. Drawing on an original online panel survey with a randomized embedded experiment, we test how Turkish citizens respond to three types of aid directed at refugees: cash transfers, vocational training and social cohesion programs. We also vary whether the aid is described as benefiting refugees or the local economy and whether it is funded by international organizations or the Turkish government. We find that aid programs framed as promoting refugee self-sufficiency, particularly those described as also benefiting the host economy, generate significantly less exclusive policy preferences. The mediation analysis shows that people’s sense of economic burden plays a big role in how aid shapes their views on refugees. This chapter highlights why the way aid programs are framed matters. When done right, they can help build local support for including refugees. It also adds to broader conversations about how foreign aid works in countries that host large refugee populations.
In Chapter 3, I explore how online intergroup contact and fact-checking from academic sources jointly influence attitudes, policy preferences and behaviors toward refugees. Although research on the effects of contact and misinformation correction on intergroup relations has typically developed along separate lines, combining these approaches can produce more substantial improvements in public perceptions and responses to refugees. I design and implement a three-week randomized controlled trial using WhatsApp groups in Turkey. I find that online intergroup contact meaningfully improves attitudes, reduces support for exclusionary policy preferences and increases willingness to donate to refugee-related causes. Fact-checking on its own has no measurable effect. However, when fact-checking follows contact, it significantly reduces support for exclusionary policies. These results suggest that the effectiveness of information correction depends on prior relational engagement. Misinformation is not simply an informational deficit but a product of social distance. Efforts to correct it tend to be more effective when there’s already some level of personal connection or familiarity in place.
Across the three chapters, my dissertation shows that refugee inclusion depends on reducing economic anxieties but also on creating opportunities for meaningful social interaction and ensuring that corrective information is delivered in trusted and relationally grounded contexts. In doing so, it advances our understanding of the microfoundations of social cohesion in settings of mass displacement. These findings suggest that fostering inclusion requires a coordinated set of structural, interpersonal and informational interventions.Political Economy and Governmen
Worker-Centric AI for Decision Support
Get PDFJust as the Industrial Revolution reshaped manual labor, AI technologies are now transforming cognitive work—offering intelligent support and fundamentally changing the nature of workers' tasks and workflows. The design of AI-powered decision-support systems affects not only output-centric values of work --- such as quality, efficiency, and creativity ---but also human-centric values, including workers' skills in AI-supported tasks, their agency, collaboration, and the meaning they derive from their work. However, current AI decision-support paradigms typically focus only on output-centric metrics (e.g., decision accuracy) and overlook how people process information, their motivation to engage with AI recommendations, and their ability to critically assess AI outputs. As a result, these systems often lead to overreliance, fail to enable human-AI complementarity, and may even deskill workers.
For both moral and economic reasons, AI tools must empower workers, help them develop new skills, and truly complement human expertise. Technologies that support worker agency and skill development are more likely to lead to long-term organizational performance, job satisfaction, and economic resilience. To achieve this, I argue that AI systems must be designed from a worker-centric perspective --- one that optimizes both output- and human-centric outcomes and is grounded in human cognition: how people think, decide, learn, and apply expertise.
In the first part of this dissertation, I demonstrate that existing AI decision-support systems do not sufficiently account for human cognition. They are built on the implicit --- yet incorrect --- assumption that users consistently engage cognitively with AI support. Challenging this assumption within the field of human-AI decision-making, I show that cognitive engagement is an essential mechanism for critically evaluating and effectively incorporating AI advice into decision-making. Drawing on the dual-process theory of cognition, I demonstrate that current paradigms of AI support exacerbate heuristic (System 1) thinking by offering readily available decisions and explanations that users can adopt with minimal effort. To counteract this, I introduce cognitive forcing functions ---interaction interventions that elicit cognitive engagement by disrupting heuristic processing at decision time --- and show that these significantly reduce overreliance on AI.
Building on these findings, in the second part of this dissertation, I introduce a suite of novel systems that operationalize worker-centric AI --- systems that are grounded in human cognition when optimizing output- and human-centric outcomes of work. I present AI decision-support systems that: (1) complement human judgment through adaptive AI support policies, learned via reinforcement learning, that personalize assistance based on contextual and cognitive factors; (2) augment human skills through human-centered contrastive explanations that address knowledge gaps by contrasting AI decisions with likely human reasoning; and (3) extend human perspective-taking capabilities in decisions requiring cognitive empathy by simulating diverse viewpoints, as demonstrated with the AHA! system for AI deployment decisions.
Together, these contributions advance the field of human-AI decision-making and chart a path toward worker-centric AI systems --- designed not only to enhance productivity but also to sustain workforce development. Such systems ensure that AI complement human expertise, preserves pathways for skill acquisition, and ultimately strengthens the workforce.Engineering and Applied Sciences - Computer Scienc
Gender differences in labor market outcomes
No full textThis dissertation studies how gender mediates labor market outcomes, broadly defined. The first chapter asks: do the jobs best suited to women’s talents offer the flexibility that they need? I document that temporal flexibility and social skill intensity are inversely correlated across occupations. Guided by an occupational choice model, I show evidence that this correlation generates a trade-off between women’s preference for flexible work hours and their comparative advantage in social tasks. As a consequence, women’s relative labor market returns to both flexibility and social skills are attenuated, which in turn widens the gender wage gap. Event-studies around first births -- a shock to women's relative demand for flexibility -- help confirm that gender gaps in inflexible jobs are driven by women's time constraints rather than other factors. The second chapter documents how social intensive environments may lead to systemic disadvantage for labor market minorities. In work environments where people interact, differences in communication styles could increase employment segregation and labor market disparities for minority groups. Building on this intuition, we develop a simple model where jobs vary in the intensity of social interaction, and productivity is increasing in both social skills and cultural similarity. We find evidence in line with the model's predictions. The third chapter asks: what caused the rapid rise in human capital accumulation in the early 20th century? I examine the role of the birth control movement, led by Margaret Sanger, which facilitated the rollout of over 600 birth control clinics across the United States. Using a staggered difference-in-differences design, I find that children whose mothers gained access to birth control were more likely to attend school, more likely to be literate, and less likely to report working. These findings are consistent with a quantity-quality tradeoff, in which investments in education came at the high opportunity cost of forgone income via child labor.Public Polic
Learning in Large Neural Networks
Get PDFIn this thesis, I will summarize my recent works on theoretical frameworks for learning, generalization, scaling limits, and scaling laws of large neural networks. In the first part of the thesis, we will examine the kinds of limits attained by training randomly initialized networks. The two limits of primary interest are the infinite width and infinite depth \textbf{feature-learning} limits.
The large width limit will take the form of a dynamical mean field theory (DMFT), where neurons asymptotically decouple and the macroscopic dynamics of the network are governed by population averages over the neurons in each hidden layer. This theory computes the dynamics of the learned representations of data in each hidden layer of the network throughout training as well as the dynamics of the network output predictions. Asymptotic corrections to this mean field limit will be computed from fluctuations around the DMFT saddle point. The mean field limit will then be stressed tested on several realistic networks such as convolutional networks and transformers on realistic computer vision and language modeling datasets. We next investigate the infinite depth limit of residual neural networks, where each hidden layer is a trainable perturbation of the identity map. When the residual branches are scaled correctly, these models admit infinite width and depth limits which are computable with a DMFT where intermediate layers approach a continuum limit described by the solution to a set of stochastic integral equations. We empirically show that scaling the depth and width correctly enable hyperparameter transfer, where optimal hyperparameters (learning rates, batch sizes, momentum values, etc) in small width and small depth models are the same in large width and large depth models, reducing the need for hyperparamter tuning. We will extend these mathematical techniques to analyze several distinct infinite-parameter limits for transformer models.
Next, I will describe simplified models of learning which enable a statistical analysis of generalization in a data-limited or parameter-limited regime. As in the previous section, we present both static and dynamic versions of these results. For data distributions and architectures that generate power law spectra for their limiting kernels, these theories provide predictions for the scaling laws with respect to the key computational and statistical resources: training time, model parameters, and total available data. I show that this setting provides a toy model of compute optimal scaling laws where model size and training time are traded off optimally. Lastly, we use similar mathematical techniques to analyze a toy model of hyperparameter transfer in randomly initialized deep linear networks.
Lastly, we move beyond randomly initialized deep networks and attempt to address how \textbf{structured neural representations}, such as the cortical representations of external stimuli in real brains, encode an implicit learning bias. We start with a simple neural circuit trained with the delta rule, finding that the spectral decomposition of the population code controls which learning tasks can be learned in a sample-efficient manner. Next, we examine how different learning rules alter feature learning dynamics and inductive bias in multilayer neural networks, extending the DMFT for multilayer networks to other biologically plausible learning rules. Lastly, we illustrate how the geometry of neural codes which respect symmetries in the data generating process controls the manifold capacity of linear readouts.Engineering and Applied Sciences - Applied Mat
Theory of Learning in Wide Deep Neural Networks
Get PDFIn recent years, significant breakthroughs in artificial intelligence have been largely driven by advances in Deep Neural Networks (DNNs). Inspired by the layered, modular organization of the human brain, these computational models have achieved unprecedented success in fields such as image recognition, structural biology, medicine, and language. Their impressive performance is largely due to their ability to learn to extract patterns from training data and generalize these insights to previously unseen scenarios. The study of how these networks acquire knowledge - commonly known as Deep Learning Theory (DLT) - provides valuable insights into how complex networks extract meaningful features from data. Furthermore, examining the mechanisms of generalization in DNNs may yield crucial insights into how biological neural systems perform similar tasks. The remarkable generalization abilities of DNNs raise several fundamental questions: (1) How do DNNs avoid overfitting despite being heavily overparameterized? (2) What is the relation between the strength of feature learning and the DNNs' generalization capabilities? (3) DNNs commonly struggle with flexibly and continuously learning new tasks in changing environments, tasks which the human brain handles routinely and effortlessly. What factors enable or hinder DNN performance in these scenarios? To explore these questions, this dissertation introduces a framework based on a Bayesian formulation of learning, which allows us to abstract away the complexities of detailed training procedures and focus instead on the structure of the solution space. We develop a theoretical framework for wide DNNs that makes Bayesian Learning analytically tractable, and derive its main properties in learning of a single task as well as a sequence of tasks.
We first analyze learning in a family of simplified, analytically tractable architectures, Deep Linear Neural Networks (DLNNs), in which each unit has a linear activation function. In the thermodynamic limit, where both the number of training examples and the width of the network become very large, yet maintain a fixed ratio, the statistics of the input-output mapping of the network, averaged throughout the solution space, can be solved exactly. Our analysis enables the evaluation of critical network properties, including generalization error, the effects of network width and depth, the size of the training set, as well as the roles of regularization and stochasticity during learning. Our theory allows for computation of both system performance and layer-wise data representations.
We then heuristically extend our theory to fully connected nonlinear DNNs and validate it numerically. For a more rigorous extension to nonlinear DNNs, we propose a tractable nonlinear architecture, Globally Gated Deep Linear Networks (GGDLNs), which preserve key qualitative features of nonlinear networks while remaining analytically tractable. Compared to DLNNs, GGDLNs exhibit richer and more complex dependencies on network depth, width, and regularization. The gating operation enhances network capabilities by allowing for flexible ways to encode context. In particular, we show that GGDLNs are able to learn simultaneously multiple tasks with contradicting labels, by explicitly incorporating task-relevant information into their gating units.
Finally, we extend our theoretical framework to investigate continual learning (CL) in wide DNNs, where networks sequentially learn new tasks without losing previously acquired knowledge. We first consider the single-head scenario, where a single neural network is used to perform both training and inference on all tasks. For tasks with contradicting labels which the single-head architecture struggles with, we consider a multi-head architecture with task-specific readouts. In the multi-head scenario, learning a new task involves modifying the shared hidden-layer weights while adding a new task-specific readout, leaving previous readouts untouched. This architecture can be interpreted as a gated network similar to the GGDLN, where the task identity information is incorporated into non-overlapping sets of gating units. These units then activate the corresponding output pathways for each task. Building upon the previously developed Bayesian framework, we introduce order parameters (OPs) that quantify task similarity and accurately predict the degree of forgetting and anterograde interference. Our findings emphasize that task similarity and network depth significantly impact interference in both single-head and multi-head CL setups, highlighting conditions leading to catastrophic interference and suggesting effective strategies for reducing forgetting.
In summary, this dissertation presents a comprehensive theoretical analysis of learning in wide DNNs for both single and sequential tasks, demonstrating how generalization and internal representations critically depend on network architecture, hyperparameters, and task structures. These insights lay the groundwork for future exploration into generalization within more complex and practically relevant DNN architectures, and offer potential pathways toward understanding the neural mechanisms underlying representation learning and generalization in biological neural systems.Biophysic
Ultramodernism in Global Musical Thought, 1900–1950
Get PDFI cast a new critical perspective on music theory’s pre-World War II avant-garde, revealing an integrated intellectual network spread across Russia, Central and Western Europe, North America, and Latin America. In the early twentieth century, I show, teleological music theories divergent from that of Arnold Schoenberg proliferated in experimental and atonal music, only to be suppressed in later reception, largely because of Schoenberg’s perceived continuity with music’s “Great German Tradition.” I recover these alternative music theories through a mixture of global historiography and archival research: In the US, Henry Cowell forged an international coalition with his theories of dissonant counterpoint and acoustics; in Paris as a Russian émigré, Ivan Wyschnegradsky developed alternative axioms for atonal music; and in Mexico, Carlos Chávez and Julián Carrillo negotiated discourses surrounding modernism and Indigeneity, both with each other, and in dialogue with the dissertation’s other protagonists. By taking a transnational approach, I destabilize conceptions of the pre-war avant-garde’s intellectual locus in the Second Viennese School—conceptions ascendant at least since Theodor Adorno’s mid-century writings—in favor of a rich and heterogenous global network.Musi
Identification of growth cone molecular machinery controlling subtype-specific circuit construction during mouse cortical development
No full textPrecise construction of specific neural circuitry enables exquisite control over diverse nervous system functions. Circuit construction is primarily regulated by growth cones, subcellular compartments at tips of extending axons that rapidly integrate extracellular signals to control axon pathfinding, then mature into synapses. Although elegant prior work has identified molecules in neuronal cell bodies that regulate subtype-specific wiring, much less is known about downstream molecular machinery in growth cones, and how dysregulation of these intricate pathways disrupts circuit construction, leading to a range of neurodevelopmental and neuropsychiatric disorders.
Here, we identify RNAs and proteins in subtype-specific growth cones from developing mouse cortex that control distinct aspects of circuit construction by engineering new approaches for in vivo subcellular investigation. We elucidate non-canonical mechanisms underlying functions of these molecules in regulating axon guidance, and highlight the power of growth cone proteomics to discover new controls over subtype-specific circuit formation. Finally, we develop a synthetic system for manipulation of molecular abundances with subtype, stage, and subcellular specificity in vivo, enhancing future functional investigations of growth cone RNAs and proteins. Together, this work demonstrates how appropriate subcellular localization of diverse molecules is essential for specific neuron subtypes to form precise circuits in vivo, and how even subtle dysregulation of growth cone molecular machinery can result in aberrant circuitry that is implicated in a wide array of neurodevelopmental and neuropsychiatric disorders.Biology, Molecular and Cellula