71446 research outputs found
Sort by
Sparse Representations in Artificial and Biological Neural Networks
Get PDFThis thesis explores how sparsity, the idea that only a small fraction of neurons are active at any time, is a common thread connecting biological brains and artificial intelligence. By combining theory, experiments, and real-world applications, we show how sparsity is a key ingredient underlying core cognitive abilities like attention, memory, and learning.
We start by uncovering a surprising link between the "attention" mechanism powering recent artificial intelligence (AI) breakthroughs and a classic theory of human memory called Sparse Distributed Memory (SDM). This suggests that brains and AI may leverage similar computational tricks.
Taking inspiration from the brain's cerebellum, we then use SDM to improve an AI's ability to learn continuously without forgetting previous knowledge. This showcases sparsity's ability to enable more flexible learning.
We also find that simply adding noise during training pushes AI to use sparse representations, causing it to develop more brain-like properties. This provides clues about why sparsity emerges in the brain while offering an easy way to encourage it in AI.
Finally, we use sparsity to peek inside the black box of large language models like ChatGPT and Claude. By pulling apart the tangled web of information these models use to think, we make progress towards more transparent and controllable AI.
Together, these findings paint sparsity as a unifying principle for intelligent systems, be they made of biological neurons or silicon chips. By connecting the dots between neuroscience and AI, this thesis advances our understanding of intelligence while charting a course towards more capable and interpretable AI systems.Systems Biolog
Ecology and evolution of a dominant skin commensal and its phage
Get PDFPoor predictability of stable microbial colonization undermines our ability to harness microbes in biotechnological and human health applications. How microbes colonize and evolve within ecosystems to create microbiomes with unique compositions, in which genetic variation occurs both within and across species boundaries is unclear. Bacteriophages are a likely determinant of colonization, as bacteria-phage dynamics are hypothesized to promote intraspecies diversity in microbiomes by generating strain-level population fluctuations. Human sebaceous skin offers a tractable model system to take a reductionist approach that combines metagenomics and culture-based whole genomics to extensively study the role of phage in microbial community colonization and intraspecies diversity in a natural ecosystem. This thesis studies the ecology and evolution of the highly abundant and ubiquitous skin commensal Cutibacterium acnes and its phage. The first chapter examines the role of phage-mediated selection in the assembly and structure of on-person C. acnes populations. We report findings that phage resistance is not a major determinant of C. acnes intraspecies diversity. As evidenced by widespread susceptibility to phage, resulting from weak selective pressure to maintain or diversify the limited pan-immune repertoire of C. acnes. Furthermore, despite the high prevalence of C. acnes phage in global facial skin metagenomes, the virus-to-microbe ratio is low, and phage-sensitive strains are the most prevalent and abundant members of on-person C. acnes populations. We therefore propose that the physiology and spatial structure of human skin buffers against strong phage-mediated selection and thus minimize the ecological relevance of encoding phage resistance. The second chapter examines the population structure and ecoevolutionary dynamics of C. acnes phage on human skin. We find strong evidence of one or more distinct phage lineages coexisting on individuals’ skin with person-specific genetic signatures that likely arose from independent colonization events. Within these lineages, closely related phage sublineages can coexist and diversify within the same individual, indicating stable phage engraftment and subsequent on-person evolution rather than transient colonization. Overall, this work enhances understanding of phage-mediated ecological and phylogenetic determinants of microbial colonization in human ecosystems, highlights the potential of viruses to colonize and adapt in individual microbiomes, and contributes to better design of microbial-based products with higher potential for durable colonization.Systems Biolog
En-algebras in m-categories
Get PDFThe study of EE_n-algebras in higher categories has attracted growing interests, both from various categorification programs in mathematics as well as the study of higher dimensional topological orders in physics. However, the complexity of these structures increases rapidly with the category level. In this thesis,
we prove a connectivity bound for maps of infinity-operads of the form AA_{k_1} otimes \cdots otimes AA_{k_n} -> EE_n, and as a consequence, give an inductive way to construct EE_n-algebras in m-categories. To prove this result, we first develop a theory of arity restricted unital infinity-operads.
Given k >= 1, we define
unital k-restricted infinity-operads, which are variants of infinity-operads which have only (= k)-arity morphisms, as complete Segal presheaves on closed k-dendroidal trees, which are closed trees built from corollas with valences = k. Furthermore, we prove that the restriction functors from unital infinity-operads to unital k-restricted infinity-operads admit fully faithful left and right adjoints by showing that the left and right Kan extensions preserve complete Segal objects. Varying k, the left and right adjoints give a filtration and a co-filtration for any unital infinity-operad by k-restricted infinity-operads, generalizing the AA_k filtration for EE_1.
Second, We prove a version of Eckmann-Hilton argument that takes into account both connectivity and arity of infinity-operads.
Along the way, we prove a technical Blakers-Massey type statement for algebras of coherent infinity-operads.Mathematic
Tilted Richardson Varieties
Get PDFThe flag variety and its subvarieties, such as Schubert and Richardson varieties, are central objects in algebraic geometry and algebraic combinatorics. In this thesis, we introduce and study tilted Richardson varieties , a new family of subvarieties of defined for all pairs of permutations and . This family generalizes classical Richardson varieties when in the Bruhat order and provides a geometric framework for the quantum Bruhat graph. We establish their fundamental geometric properties, including irreducibility, explicit dimension formulas, and a well-defined stratification indexed by tilted Bruhat intervals, a generalization of classical Bruhat intervals introduced by Brenti, Fomin, and Postnikov. Moreover, we introduce a tilted generalization of the classical Deodhar decomposition, which leads to a combinatorial formula for tilted Kazhdan--Lusztig R-polynomials, a notion that arises naturally in our framework.
We further develop a theory of total positivity for tilted Richardson varieties. In particular, we define and study their totally nonnegative parts and prove that they form CW-complexes. This generalizes previous work on the totally nonnegative flag variety and addresses Bj\"orner's questions regarding geometric realizations of tilted Bruhat intervals.
Finally, we establish explicit connections between tilted Richardson varieties and quantum Schubert calculus. We prove that coincides with minimal-degree two-point curve neighborhoods, allowing us to compute their cohomology classes and derive new relationships among Gromov--Witten invariants of the flag variety.Mathematic
Essays on Financial Accounting, Entrepreneurship, and Innovation
Get PDFEntrepreneurship and innovation are widely regarded as key engines of economic growth. This dissertation examines how information environments influence the decisions and outcomes of entrepreneurs and innovators across three studies grounded in financial accounting and law. Chapter 1 evaluates whether easier access to information fosters entrepreneurial success. Using the staggered launch of online repositories for Franchise Disclosure Documents, which sharply reduced information processing costs for would-be franchisees, I find a 3.1 percentage-point increase in early-stage survival among new businesses. Chapter 2 investigates the impact of judicial efficiency on corporate innovation and disclosure. Exploiting the 2011 Patent Pilot Program, which assigned patent cases to judges with specialized expertise, I show that firms headquartered in participating counties expand patent based innovative output by 6.9% relative to comparable firms in non-participating counties. Chapter 3 explores how capital markets value patent based innovation, with a particular focus on the recent surge in AI-related patents.Business Administratio
FtsZ phosphorylation modulates tail-core binding to tune cell division in Bacillus subtilis
Get PDFThe bacterial cytoskeletal protein FtsZ orchestrates cell division in nearly all bacterial species, yet the regulatory mechanisms governing its assembly dynamics remain incompletely understood. This thesis identifies a novel intramolecular interaction within Bacillus subtilis FtsZ between its intrinsically disordered C-terminal linker (CTL) and its globular core that directly modulates FtsZ function. Through complementary biophysical, biochemical, and computational approaches, I demonstrate that the CTL specifically binds to the core’s C-terminal polymerization surface with high affinity, with residues L330-H337 forming the critical binding interface. I further establish that S333 within this region is phosphorylated in a PrkC-dependent manner, suggesting post-translational regulation of this interaction.
Disruption of tail-core binding through S333 mutations produces effects across multiple biological scales: at the molecular level, it reduces FtsZ’s critical concentration and increases its GTPase activity; at the cellular level, it decreases cell length; and at the population level, it enhances growth under hypoxic and cell wall stress conditions in liquid culture while producing smaller colonies on solid media. Together, these findings suggest a previously uncharacterized regulatory mechanism where PrkC-mediated phosphorylation may dynamically tune FtsZ assembly, and consequently bacterial cell division, in response to environmental signals.
This work reframes our understanding of FtsZ’s intrinsically disordered linker from an inert mechanical tether to an active regulatory element. It further highlights how phosphorylation of disordered regions can provide functional plasticity to cytoskeletal proteins.Biology, Molecular and Cellula
Bayesian and Spatiotemporal Modeling of Infectious Diseases and Population Health
Get PDFInfectious disease outbreak detection is a critical component of public health surveillance. However,
the data and methods available for this task vary, and there is limited guidance on which method
to apply to a given dataset. Additionally, outbreak detection and public health rate estimation rely on
accurate population denominators, yet in the U.S., it is unclear which data sources provide the most
reliable estimates. This dissertation addresses both issues by evaluating outbreak detection methods
for syndromic and wastewater-based surveillance and by developing a model to estimate U.S. county
populations.
In Chapter 1, we present a simulation study evaluating spatio-temporal models for syndromic
surveillance in low-resource settings. Conventional syndromic surveillance methods face challenges in
handling missing data and often do not leverage spatio-temporal structure. We compare a baseline syndromic
surveillance model, a frequentist spatio-temporal model, and a Bayesian spatio-temporal conditional
autoregressive (CAR) model. The Bayesian CAR model consistently achieves high specificity
across simulations, underscoring the importance of spatio-temporal modeling in syndromic surveillance.
In Chapter 2, we introduce the Spatially-Weighted Ensemble for Estimation of Populations (SWEEP),
a Bayesian ensemble model that combines the American Community Survey (ACS), Population Estimates
Program (PEP), and WorldPop (WP) to generate intercensal population estimates. SWEEP uses
spatially varying weights that adapt to geographic patterns in product accuracy. Using 2019 product
estimates to predict 2020 census counts, SWEEP improves population estimates, particularly for the American Indian and Alaska Native (AIAN) population, and reveals systematic geographic variation
in data accuracy. These findings demonstrate the potential of spatially adaptive ensemble modeling
to improve population estimates and support more equitable disease and mortality rate estimation.
In Chapter 3, we develop a wastewater-based outbreak detection method using an exponential
growth model and evaluate its performance relative to clinically-defined outbreaks. Applied to countylevel
COVID-19 data, this method outperforms a reproductive number (Rt)-based approach. Detection
performance improves with spatial aggregation yet declines in extreme temperatures, high humidity,
and after 2021. These results suggest that wastewater surveillance can reliably detect outbreaks,
though its performance varies with environmental context and its evaluation depends on the quality
of reference clinical data.Biostatistic
Strong Coupling Topological Phases in Moiré Bands
Get PDFA central goal of quantum condensed matter physics is to understand, realize, and control phases of matter that exhibit macroscopic quantum phenomena. Moiré materials offer an unprecedented ability to do so through hosting strongly interacting electrons in topological bands. This setting was previously restricted to the FQHE, where electrons under massive magnetic fields split into new particles that carry a fraction of the electron's charge. This newly central experimental setting demands new theoretical tools that are applicable to strongly interacting topological bands. Existing theories are, naturally, specific to the only prior existing example, the lowest Landau level associated with the traditional fractional quantum Hall effect, and by their nature rule out several phases of matter including superconductivity.
In this thesis, we develop strong coupling theories in the topological setting and use them make predictions on the interacting physics of twisted graphene systems. In Chapter 1, we will show how to analytically predict fractionalization in topological bands without relying on mimicking the lowest Landau level. Chapter 2 will compare and contrast a class of twisted graphene systems using a variety of theoretical tools. In Chapter 3, we report on a theoretical framework that accesses Mott physics in the topological bands of TBG. Mott physics, a key ingredient of high temperature superconductors, is typically studied in bands without topology. We report on qualitatively new phenomena that emerge from the combination of Mott physics and band topology.Physic
How Do Children with Conduct Disorders Engage in Mentoring Relationships: A Qualitative Study
Get PDFEarly conduct issues or Conduct Disorder (CD) may serve as a predictor for
persistent conduct issues throughout the life course. These persistent issues are associated
with criminal behavior, intimate partner violence or discord, peer rejections, and
diagnosis of comorbid disorders such as antisocial personality disorder or depression later
in life (Fairchild et al., 2019). The presence of a mentor, either through a formal program
such as the Big Brothers and Big Sisters of America or through informal relationships,
has been shown to serve as a protective factor for children with conduct issues
(Caldarella et al., 2009; Owora et al., 2018). However, the mechanisms through which
mentorships are effective are still not fully understood. To explore the hypothesis that a
mentee’s perception of feeling seen and understood by their mentor has a protective
effect in reducing conduct problems, most likely by modeling prosocial behavior, a
mixed methods investigation using measures of social competence and self-perceived
social support will be combined with retrospective interviews.Extension Studie
Monetizing Coffee Leaves and Fruit to Fund Regenerative Agriculture: A Case Study in Puerto Rico
Get PDFCoffee cultivation sustains more than 100 million livelihoods but illustrates
agriculture’s environmental paradox: 70–80 % of smallholder households earn below
living-income thresholds while contributing disproportionately to deforestation,
biodiversity loss, and climate change. Regenerative agroforestry can restore ecosystems
through enhanced biodiversity, carbon sequestration, and soil health, yet adoption
remains limited by a “worse-before-better” gap of 10%–25 % yield penalties and
establishment costs of 3,000 per hectare. Existing instruments, including
certifications, carbon credits, and subsidies, have not closed the estimated 119,128/ha, reflecting Puerto Rico's high labor costs and input premiums. Regenerative
baseline (S₂) achieved positive viability in 100% of iterations with mean NPV of
141,944/ha with 100% positive
outcomes, representing 171% improvement over regenerative baseline and transforming
negative conventional systems into highly profitable operations. Counterintuitively, leaf-
only systems (S₄) achieved the highest performance at 53,363/ha due to superior per unit values and temporal
complementarity advantages. The cascara-only scenarios (S₃) remained negative at a
mean value of -$32,644/ha, confirming insufficient standalone viability.
The study contributes three advances: (1) a dual-metric viability framework
distinguishing economic versus financial barriers; (2) the theory of temporal
complementarity, explaining how off-season leaf processing converts coordination
problems into capacity-optimization opportunities; and (3) evidence that existing
cooperative infrastructure can transform biomass waste into regenerative finance.Extension Studie