71446 research outputs found
Sort by
From Crowded Classrooms to Empty Halls: How Enrollment Fluctuations Shape Students’ Academic Success
No full textThis thesis examines how enrollment fluctuations influence student academic performance through changes in school funding. Using data from Massachusetts and North Carolina, I leverage quasi-random variation in per-pupil funding generated by enrollment changes — a key determinant of district budgets under state funding formulas. Employing both Ordinary Least Squares (OLS) and instrumental variable (IV) approaches, with distance to Boston and local population growth as instruments, I estimate the causal effect of enrollment shifts on student outcomes in Math and English Language Arts (ELA). While OLS results suggest a positive relationship between enrollment growth and test scores, IV estimates reveal smaller and often insignificant effects, indicating that simple correlations may be biased by unobserved factors. Panel regressions show that in Massachusetts, higher enrollment levels are associated with declines in test performance, likely due to overcrowding and resource strain. In contrast, North Carolina schools with more stable or growing enrollment see modest improvements in ELA scores, highlighting different policy challenges in shrinking versus growing districts. These findings underscore the need for tailored policy interventions to manage the consequences of enrollment volatility and ensure sustained educational quality.Applied Mathematic
A New Game of Jenga: A Query into the Contours of U.S. Anti-Terror Law Enforcement Preparedness 1932-1972
No full textOn September 20, 2001 in order to reassure to the American public, President
George W. Bush proclaimed, “our war on terror begins with Al-Qaeda but it does not end
here.”1 This declaration became to be known as the Global War on Terror and invited an
avalanche of political scholars and historians to uncover a new understanding of the
Middle East. But the story of how the United States has responded to terrorism has an
untold story, going back further to the edges of the Cold War.
The purpose of my thesis is to show how the federal government implemented
anti-terrorism policies by combating the infiltration of Russian communist influence prior
to the Second World War. It was during the edge of the Cold War in the 1950s that the
federal government used the very policies that it adopted to address what it considered
domestic terrorism in the 1960s. I argue that presidential desperation to address the
communist threat, coupled with the growth of the FBI’s extension into local law
enforcement created a curved impromptu approach to terrorism, geared towards domestic
threats while removing the focus on threats that originated internationally. My contention
is that the collaboration between President Franklin D. Roosevelt and FBI Director J.
Edgar Hoover expanded the purpose of the FBI to address the Russian communist threat
in the 1930s. The next step came with the success of the war’s conclusion in 1945, where
President Harry S. Truman initiated the National Security Act of 1947.
The policy manufactured approaches to national defense and takes us through a maze of
juxtapositions to combating elevating crimes rates. Together with the expansion of
federal law enforcement responsibilities and with the threat of communism, socialism and
fascism coming into its own borders, the United States would begin to define what we
refer now to as “terrorism.” National Defense policies in the late 1940s would have to be
redefined and inversely applied in order to uproot this growing problem.
There is a litany scholarship that has been dedicated to the military and
intelligence response of the dilemma of Cold War internationally, but little has been
discussed of how the United States sought to protect itself once it had already arrived
inside of its own borders. If we analyze the federal government’s discourse of policies
and discussions that have gone largely awry, we can see that the American public was left
exposed and vulnerable through a misguided path of blinding ideologies from its elected
polity.
My approach is unconventional and establishes a series of patterns that converge
into single road of symmetrical evidence. I purposely avoid seismic arguments that
involve the Second World War, the Great Depression and US. Diplomacy in the Middle
East primarily because they would distract from the greater point at hand. My argument
rests on the idea that the ability to combat terrorism in the United States originates in its
need to combat domestic terrorism and then rotates precipitously to combat international
terrorism. Ultimately, we can see a cyclical pattern of the US government is in a constant
flux preventing two types of terrorism under various circumstances. By focusing on the
examination of personal letters and documents from J. Edgar Hoover, Franklin D.
Roosevelt, official Congressional records, CIA declassified materials and historical
polling data so that we can see that many of these greater events created a new
unforeseen path to the events leading up to the tragedy of 9-11.Extension Studie
Odors as ''natural language'': sparse neural networks in mammalian olfactory systems and large language models
No full textThe studies of physics, neuroscience, and artificial intelligence (AI) have a long intertwined history. Particularly, sparse connectivity is a common feature of the brain neural networks and a key focus in AI for efficient computation; notably, pruning trained networks for sparse connectivity has a long history, partially inspired by neuroscience. This thesis explores sparse neural networks through two linked research topics: one focused on the brain (bilateral alignment in olfactory systems), and the other on AI (pruning large language models for on-device AI assistants).
For the first topic, inspired by mammalian dual nostrils creating two cortical neural representations of odors, in Chapter 1, we studied how to construct the inter-hemispheric projections aligning these representations. We hypothesized that this construction originates from online learning since mammals are constantly breathing. With a local Hebbian rule, we found that sparse inter-hemispheric projections suffice for bilateral alignment and discovered an inverse scaling that more cortical neurons allow sparser projections. Also, the local Hebbian rule was found to approximate the global stochastic gradient descent (SGD) rule since their update vectors align, suggesting that biologically plausible learning rules can approximate global learning rules if they contain the gradient information of the latter.
The next chapter extends Chapter 1 from four perspectives: an analysis of the update vector alignment between Hebbian and SGD rules and how it depends on the network parameters; a simple theory that recurrent connections in olfactory cortex may improve the bilateral alignment, inspired by the Hopfield Networks (associative memory) and similar to the design of Google Titans model that combines recurrent neural networks with Transformers; the dynamical properties of Hebbian learning; and finally, the geometric landscape of Hebbian learning.
A similar inverse scaling has been discovered in the Transformer attention matrices used in large language models (LLMs), which motivated the second topic. Concretely, we pruned pretrained Meta Llama-2 and Llama-3 models to obtain models with fewer parameters and develop on-device AI assistants, explored their sparsity limits, and compared their performance at the limits. We found that more than 50% of the
parameters in both models could be pruned, and Llama-3 produced fewer factual errors at the sparsity limit but required more parameters presumably due to its training settings and dataset.
In summary, by studying sparsity in both biological and artificial neural networks, this thesis may provide valuable insights into the general bilateral alignment problem in neuroscience (across different modalities and brain regions such as the frontal cortex responsible for short-term and motor response and the medial entorhinal cortex for spatial memory), open the door to interesting theoretical questions, and inspire more efficient AI algorithms or applications.Biology, Molecular and Cellula
Revolution of the Heart: The Alternative New Woman in Early Soviet Media
No full textThe early Soviet government of the 1910s and 1920s was one of the first countries in the world to grant women equal rights with men. Among the Bolshevik Party members was a belief that so-called “new people” would come into being through the material and economic changes that the Revolution accomplished. For women and their representation in fictional narrative arts, this change in political and economic systems provided them with new opportunities for self-determination. State propaganda campaigns sought to bolster and harness women’s empowerment for political goals. Against this, a variety of authors of the 1920s instead explored a more individual-focused application of new opportunities for women. The authors analyzed in this project are Aleksandra Kollontai, the noted Bolshevik advocate for women’s rights, author Evgeni Zamyatin, avant-garde playwright Sergei Tret’yakov and Abram Room, filmmaker. I identify a fictional construct in works by these four artists which I call the “Alternative New Woman” through a feminist-informed close reading of key fictional works by these authors. Vasilisa Malygina by Kollontai, We by Zamyatin, I Want a Baby! by Tret’yakov, and Bed and Sofa by Room all feature a female protagonist who pursues her own desires (rather than conforming to societal expectations) throughout her narrative arc, which forms the basis for my definition of the Alternative Woman. The Alternative New Woman challenges both traditional and Bolshevik understandings of gender, femininity, and individuality. The Alternative New Woman of the 1920s also explores the changing social and material conditions of everyday life during the New Economic Policy (NEP), especially on topics that are traditionally understood as part of the feminine domain: family, marriage, sexuality, and reproduction.Slavic Languages and Literature
Building Stronger Nanofiber Scaffolds for Regenerative Aortic Valves
No full textFor the many patients in need of heart valve replacements, a lifetime of medical difficulties accompanies the available mechanical and bioprosthetic implant options. Scientists are working to address these comorbidities by developing regenerative heart valves made of biodegradable fibrous scaffolds that are immediately capable of controlling one-way blood flow while guiding the patient’s own cells to reconstruct the valve with native tissue. Our group’s state-of-the-art regenerative valve scaffolds are manufactured quickly and consistently with focused rotary jet spinning of a PLA/PCL copolymer (PLCL). However, this valve design has thus far only proven to be functional in the pulmonary position, and a more robust design is needed for a valve scaffold to operate in the higher-pressure aortic position. This project presents an upgraded regenerative valve design for functionality in the aortic position by engineering a new PLCL scaffold material suitable for more extreme pressure and flow conditions by altering the copolymer composition to improve mechanical strength without sacrificing regenerative capability. The experiments conducted to analyze the potential of this novel material demonstrate initial success in biocompatibility with 2D scaffold cell culture and valve functionality with in vitro flow testing in a simulated aortic environment.Engineering Sciences S
SpiroSniff: A Machine Learning Driven Breathalyzer for Lung Cancer Detection
No full textSpiroSniff is a portable, affordable breathalyzer designed to detect early-stage lung cancer by analyzing volatile organic compounds (VOCs) in exhaled breath. Leveraging metal oxide chemiresistive sensors and machine learning algorithms, the device aims to provide a non-invasive, rapid, and accurate alternative to traditional diagnostic methods like CT scans and X-rays, which are often inaccessible and cost-prohibitive, especially in low-resource settings. With a production cost target of under $150 and a desired specificity and sensitivity of at least 80%, SpiroSniff is engineered for widespread use, including in underserved populations. The project addresses a critical need for early detection tools in light of the rising prevalence of respiratory diseases and the high mortality rate associated with undiagnosed lung cancer. By focusing on robust sensor selection, in-lab validation, algorithm development, and ethical deployment, SpiroSniff has the potential to transform global lung cancer diagnostics and improve public health outcomes.Mechanical Engineering S
Essays in Behavioral and Experimental Economics
No full textThis dissertation consists of essays in behavioral and experimental economics, with a focus on how bounded rationality and information-processing constraints shape economic behavior.
The first chapter, coauthored with Cassidy Shubatt, develops a theory of how tradeoffs govern the difficulty of comparing choice options, and studies the implications of this theory. We propose measures of comparison complexity in three choice domains: multi-attribute, lottery, and intertemporal choice, which formalize the intuition that comparisons are more difficult when they involve pronounced tradeoffs, and provide axiomatic foundations for the theory in each domain. We experimentally validate our theory using large-scale choice data: in all three domains, our complexity measures are strongly predictive of choice errors and inconsistency. We then study the behavioral implications of our theory. First, we show how our theory rationalizes a range of documented biases and instabilities in choice, such as decoy/asymmetric dominance effects, preference reversals, and apparent probability weighting and hyperbolic discounting – and makes novel predictions on how they can be reversed by varying the nature of tradeoffs. We confirm these predictions experimentally, documenting that these disparate choice patterns are outgrowths of comparison complexity. Second, we apply our model to study obfuscation in markets, analyzing a pricing game in which firms can influence how comparable their products are to their competitors. We find that comparison complexity leads to spurious differentiation: firms design seemingly dissimilar products to increase the difficulty of price comparisons, which softens price competition and leads to higher markups.
In the second chapter, I show how information-processing constraints can make sense of an empirical puzzle that has been documented in multiple domains: that the relationship between individuals’ beliefs over economic quantities and their behavior is often quantitatively attenuated relative to theoretical benchmarks. The idea is that due to the complexity of many decisions, individuals find it difficult to translate their beliefs into optimal decisions – they face uncertainty over the belief-action map. I develop a model of how this uncertainty affects beliefs and behavior, and empirically test its predictions by measuring and manipulating uncertainty over the belief-action map. In a portfolio allocation experiment, I find that higher uncertainty over the mapping predicts a more attenuated relationship between subjects’ return expectations and investment decisions, weakens behavioral responses to information regarding returns, and reduces information acquisition. One implication of these results is that information provision interventions, which aim to improve behavior by correcting beliefs, are unlikely to be successful if individuals face frictions in translating beliefs into behavior. On the other hand, the existence of this friction points to the usefulness of interventions that reduce uncertainty over the belief-action map. I demonstrate the effectiveness of one such intervention in increasing subjects’ responsiveness to their beliefs.
In the third chapter, I develop a theory of how boundedly-rational agents learn from data. I model a decision-maker who observes data and is exposed to a multiplicity of models, or accounts of how information should be interpreted. The decision-maker does not average across these models as a Bayesian would, but instead adopts a single model through which to interpret the data. I propose a theory of model selection based on the insight that individuals seek decisive models that reduce residual uncertainty over the optimal course of action. I show how the decisiveness criterion is characterized by a demand for extreme models, which generates documented inferential biases such as overprecision and confirmation bias. The dependence of the decisiveness criterion on the decision-maker’s objectives rationalizes a range of documented patterns in inference and choice, and generates novel predictions as to how belief polarization can arise along heterogeneity in decision-makers’ objectives. Finally, I apply the model to study the provision of expert advice, as well as social learning through the exchange of models; the theory predicts the supply of overly confident advice in the former setting and predicts group polarization in the latter.Economic
The Economics of Urban Mobility
No full textCities offer immense potential for economic growth and improved livelihoods; however, the realization of these benefits hinges critically on efficient transportation and mobility. This dissertation includes three chapters that use a combination of natural experiments and large-scale randomized controlled trials to study how barriers to physical mobility in urban settings impact the economic outcomes of low-income populations.
The first chapter studies how the spatial market segmentation of minibus associations in Johannesburg, South Africa, impacts service provision and commuter mobility. The need for coordination between associations results in inefficiently low and rigid supply of minibuses, leading to large increases in wait times for commuters. The second chapter uses a field experiment to study whether reducing public transit fares improves mobility and socioeconomic outcomes for low-income residents in Allegheny County, Pennsylvania. Completely subsidizing transit fares leads to mode substitution away from private vehicles and into public transportation, and increases employment for participants that were unemployed at baseline. The third chapter documents a new barrier to urban mobility: lack of exposure. Using a field experiment, we find that workers in Nairobi, Kenya, require a wage premium to work in neighborhoods they are not already familiar with, an effect that disappears after a single visit to the neighborhood. This familiarity premium implies that lack of exposure to parts of a city restricts residents’ access to economic opportunities.
Together, these three chapters offer new insights into how the design of urban transport systems shapes economic behavior and outcomes.Economic
Harmony and Rupture in Desdemona: Toni Morrison's Remaking of Shakespeare's Othello
No full textIn Toni Morrison’s Desdemona, a theatrical remaking of Shakespeare’s Othello, two characters are wholly reconfigured: the white Venetian Desdemona, and a barely mentioned figure in the Shakespearean text, Barbary, Desdemona’s Black maid of African descent. In Morrison’s portrayal, Barbary, through deft storytelling, becomes a conduit for Desdemona to access other contexts, other worlds and other multivoiced discourses. In the play’s broad strokes imagery, the tension between the two characters becomes deeply symbolic of conflictual white/Black relations in America as Morrison sees them, a state of affairs she dramatizes as allegorically existing in a Dantean purgatorial state, or midway station, or state of undecidability. This ambiguity arises from the continued presence in the play of master/slave histories. But Morrison’s Desdemona goes beyond the context of America and slave histories of the American South, extending her interrogation, widening her angle to the Caribbean, but also to England, France, Spain and other European contexts, as well as to West African rulers and slave traders associated with the four-hundred-year history of transatlantic slaveries. That the character of Barbary has always been performed by an African tale-teller/singer in African narrative style is the central dramatic device giving the play global aesthetic and historical scope. The multi-locale setting extends Morrison’s interrogation to a more global level. This immense expansion of scope making Desdemona an intertextually rich work influenced by prior works, may also –indeed, almost certainly will— influence future texts, performances and theater criticism of the canonical Othello.Extension Studie
High-resolution dissection of cis-regulatory elements
No full textGene expression is a highly regulated process that governs almost all aspects of life. Precise gene regulation guarantees that every cell selectively expresses a subset of all genes to perform specialized cellular functions. Therefore, dysregulated gene expression often leads to disease and is implicated in aging. A key step of eukaryotic gene regulation is transcriptional regulation, where protein factors such as transcription factors (TFs) and nucleosomes bind to cis-regulatory elements (CREs) in the genome to determine the transcription of target genes. Our understanding of gene regulation therefore heavily depends on our ability to observe the action of regulatory proteins on genomic DNA, and scientists have worked for decades building ever better technologies to measure protein-DNA interactions at regulatory elements. This dissertation presents technological advances that allows measurement of protein-DNA interaction at cis-regulatory elements with significantly improved cellular (cell-type/state- resolved), genomic (single base-pair), and molecular (single molecule) resolution.
First, we describe PRINT and seq2PRINT, which are computational methods for tracking transcription factor and nucleosome binding within cis-regulatory regions using single-cell ATAC-seq (scATAC-seq) data. PRINT detects footprints of DNA-binding proteins across spatial scales by accurately modeling enzymatic sequence bias and signal dispersion using machine learning. Using pseudo-bulked scATAC-seq, PRINT achieves cell-type- and cell-state-resolved TF and nucleosome footprint landscapes in systems with complex cell type composition and across hundreds of thousands of CREs genome-wide. Building upon PRINT, we describe seq2PRINT, which is a deep-learning model that uses local DNA sequence as the sole input to predict footprint patterns in the same locus. By extracting sequence features learned during training, seq2PRINT can accurately predict TF binding events with single base-pair resolution.
Aided by low-rank approximation (LoRA), we can scale up seq2PRINT to hundreds of samples or cell-type/states. With PRINT and seq2PRINT, we reveal complex dynamics of TF and nucleosome binding within CREs in human hematopoiesis with unprecedented cell state and genomic resolution. We show that many CREs display distinct combinations of TF binding across cell types undetectable in traditional accessibility-based analyses. We further use PRINT and seq2PRINT to characterize murine hematopoietic stem cell (HSC) aging and show widespread reorganization of CREs and identify age-associate TF cooperations.
Second, we describe TDAC-seq, which is a technology that achieves single-molecule long-read profiling of chromatin accessibility and protein footprints using a double-strand DNA deaminase DddA11. We show that DddA11 can introduce C-to-T mutations in DNA regions that are accessible and unprotected from proteins, allowing accessibility and footprint measurements with nanopore sequencing. We show that TDAC-seq allows simultaneous read out of chromatin organization and genetic perturbations such as deletions introduced by CRISPR Cas9 or A-to-G mutations introduced by an adenine base editor (ABE). This allowed high throughput pooled CRISPR screen or ABE screen where the effect of each deletion/editing outcome on local chromatin organization can be individually assessed.
In summary, the body of work presented in this dissertation resolved several long-standing technological challenges in measuring chromatin-level biological processes in gene-regulation. The technologies described enable cell-type/state-resolved single base-pair tracking of regulatory factor binding, as well as single-molecule long-read measurement of chromatin organization, providing powerful new tools for gene regulation studies.Biological and Biomedical Science