71446 research outputs found
Sort by
Estimating Curvature of Data Manifolds with Diffusion Models
No full textIn the quest to understand the geometry of data, curvature is a fundamental characterization. Differential geometry supplies many notions of curvature; this thesis presents exposition organizing a tree of curvature notions. Most importantly, we extend new research into a novel method harnessing diffusion models for estimating curvature from data. This new tool may aid in downstream applications such as shape analysis, learning theory, adversarial robustness, and more.
The key novel contributions of this thesis are (i) the introduction of a diffusion model to learn the data manifold and probe its latent representation rather than the raw data, (ii) a comparison of previous estimation methods using quadratic regression and diffusion maps investigating how they deteriorate with increased noise and dimension, and (iii) a new approach using geodesic interpolations generated by a diffusion model to estimate more precise directions of curvature.
Our findings on toy manifolds show that curvature estimation through a diffusion model proves more robust to noise, but this depends greatly on the fidelity of the diffusion model. Finally, we provide initial guesses at the curvature of small real world image datasets in the MNIST family, suggesting that these datasets might be relatively flat, which would be consistent with their empirical ease to learn.Computer Scienc
Uncovering Stakeholder Coalitions in the FCC’s Net Neutrality Rulemaking
No full textNet neutrality, the principle that all traffic on the internet should be treated the same, has been one of the most contentious issues in telecommunications policy. The objective of this thesis is to use network community detection methods to uncover stakeholder coalitions in the FCC’s net neutrality rulemaking. I used public comments and ex parte meetings data to construct various networks of stakeholders. I then applied the Spectral Clustering On Ratios-of-Eigenvectors (SCORE) algorithm to detect communities and Sankey diagrams to visualize community evolution over time. This analysis leads to several noteworthy findings. Two kinds of pivotal stakeholders, anti-net neutrality internet service providers (ISPs) and pro-net neutrality public interest organizations, were found to form their own communities or form coalitions with other communities. When the stakes of net neutrality rules were high in 2014, ISPs allied with the network equipment manufacturers community, while public interest organizations allied with the internet-based tech companies community. When the FCC turned against net neutrality in 2017, while ISPs dissolved their community, the public interest community remained solid, though internet-based tech companies left to form a community of their own. Astroturf organizations were also found to always form communities among themselves, detached from other anti-net neutrality communities.Computer Scienc
When Small Was Beautiful: Petty-Bourgeois Utopianism and the Sources of Neoliberalism, 1960 – 2000
No full textIn the decades that followed the Second World War, many commentators celebrated large-scale organizations – corporations, unions, welfare bureaucracies – as the foundation of the unprecedented economic growth that societies across the capitalist world enjoyed. At the same time, a loosely affiliated group of dissenters began to question the pretensions of “bigness” and to argue that “small is beautiful.” This dissertation tells the story of this intellectual current through a series of case studies, each of which considers the context, thought, and influence of a specific “partisan of smallness.” Although the protagonists of this story were largely anglophone, their ideas were inspired by their experiences in developing countries across the world – countries that also served as the main laboratories for their practical proposals. This story is thus a global one. The men and women studied here developed a sweeping and potent critique of midcentury capitalism, lamenting the impersonality of modern bureaucracies, the dreariness of work in large corporations, and the environmental harm caused by heavy industry. As a cure, these partisans of smallness prescribed a thorough shrinking of the basic institutions of government and economy, a retreat from expensive and complicated machinery, and an embrace of “appropriate technology” that could be purchased and used by individuals and families of modest means. In their most ambitious moods, they dreamed of establishing a decentralized political and economic order, a petty-bourgeois utopia.
Although the small-is-beautiful vogue is often seen as a short-lived trend that ran out of steam by the early 1980s, this dissertation argues that its significance and influence was deeper and more durable than is usually recognized. Beginning in the 1970s, some of the proposals of the partisans of smallness were taken up by major international development agencies, most notably the World Bank, where they helped inspire a turn away from earlier developmentalist ambitions and toward austerity. In an ironic and not entirely intentional manner, then, this heterodox critique of midcentury capitalism contributed to the rise of neoliberalism. Small-is-beautiful ideas also lived on in some of the most influential works of modern social science, especially in the thought of James C. Scott. Even today, prominent figures on both the left and right of the political spectrum can be seen as inheritors of the tradition studied here, whose story therefore constitutes an important chapter in the history of the origins of our times.Histor
Toward a Primordial Genetic Alphabet: Noncanonical Nucleotides in Nonenzymatic RNA Replication
No full textNonenzymatic RNA polymerization and copying are essential for the propagation of genetic information prior to the emergence of RNA polymerase. On the early Earth, in addition to the canonical nucleotides (A, U, C, G), alternative nucleotides composed of different sugars (arabino- and threo-nucleotides) and nucleobases (diaminopurine, 2-thiocytidine, 2-thiouridine, inosine) could have arisen through plausible prebiotic chemistry. Moreover, templated copying using the current genetic alphabet exhibits a strong bias toward G- and C-rich sequences. Given the potential diversity of nucleic acid building blocks and this inherent distribution bias, my research investigates how selection during both oligomerization and templated copying shapes the incorporation of noncanonical components. I study how variations in sugars and nucleobases influence RNA hybridization, structure, and sequence distribution.
At the sugar level, arabino- and threo-nucleotides likely emerged alongside ribonucleotides, as they share a common synthetic pathway. I find that ribo- and arabino-nucleotides exhibit comparable incorporation efficiencies during non-templated primer extension, whereas threo-nucleotides are significantly less reactive. Moreover, the incorporation of an arabino-nucleotide at the end of a primer acts as a chain terminator. Competition experiments further reveal a bias against the incorporation of threo-nucleotides. These inherent biases, when considered alongside selective prebiotic synthesis and the known preference for ribonucleotides in templated copying, offer a plausible explanation for the exclusion of arabino- and threo-nucleotides from primordial oligonucleotides.
At the nucleobase level, I investigate the diaminopurine:uracil (D:U) base pair. Replacing adenine with diaminopurine (D) enhances pairing with uracil, but also introduces potential fidelity issues, as D can form a wobble-type mismatch with cytosine. Reassuringly, D:C mismatches exhibit high stalling factors, limiting erroneous extension. Deep sequencing of templated copying reactions further demonstrates that the noncanonical DUCG system yields more uniform product distributions and fewer mismatches than the canonical AUCG system. These findings position diaminopurine as a promising nucleobase for artificial nonenzymatic RNA replication systems.
Expanding beyond diaminopurine, I further explore a noncanonical genetic alphabet composed of 2-thiouridine (s2U), 2-thiocytidine (s2C), inosine (I), and adenine (A) to address the distribution bias present in the canonical alphabet. Thermodynamic and crystallographic studies show that the I:s2C and A:s2U base pairs are both isomorphic and isoenergetic. While I:s2C is slightly weaker than the canonical G:C pair, A:s2U is stronger than A:U, resulting in a balanced base-pairing landscape. Consistent with this, kinetic analyses of nonenzymatic templated primer extension reveal similar binding in the s2U/s2C/I/A system. Together, these results support the feasibility of a primordial genetic system based on s2U, s2C, I, and A, providing a potential solution to the challenge of biased nucleotide incorporation in early RNA replication. Overall, my work investigates the roles of noncanonical nucleotides in nonenzymatic RNA replication and provides new insights into constructing a primordial genetic alphabet.Chemistry and Chemical Biolog
Modeling Astronomical Data using Deep Learning by Integrating Embeddings
No full textAstronomical surveys produce time-series data by observing stellar objects across multiple wavelength bands. Foundational transformer-based models, such as Astromer, encode each time-series as a sequence of embeddings of uniform dimensions. However, such models operate independently on each band at a single time and do not natively leverage information across telescope filters. We extend this framework by introducing a fusion mechanism that maps the collection of single-band embeddings to a unified sequence representation, enabling multiband modeling for downstream tasks. The challenge lies in devising a mechanism within the encoder to coordinate between data from different wavelengths, which are often recorded at asynchronous times. We pre-train multiband models on a subset of 600 000 high signal-to-noise light curves from the MACHO survey and fine-tune them using the Alcock and ATLAS survey datasets. Experimental results show that both our proposed multiband architectures outperform the single-band models by approximately 10% in F1-score, with jointly pre-trained multiband encoders further improving performance over a collection of independently pre-trained single-band encoders. Furthermore, our experiments show that there are minimal differences in multiband performance when sampling individual band data asynchronously versus sampling all individual bands on the same set of time-steps. However, jointly pre-trained models can take more than twice the time to pre-train. These results demonstrate the trade-offs of the multiband approach where multivariate data are available.Extension Studie
Automated and Flexible Stress-Testing for the Robustification of Large Language Model Systems
No full textLarge language models are incredibly powerful but incredibly brittle and unreliable computing ob- jects. The same models that can convincingly generate rap lyrics in the style of Kanye West also struggle to consistently perform basic arithmetic accurately. The same models that can pass the interview bar at Amazon also struggle to ground their responses in facts. These models are a walking contradiction. As these systems become more embedded into our everyday lives, the key question becomes if we can properly trust and robustify these systems. The perspective that we adopt in this thesis is that, in or- der to prevent these systems from failing in high-stakes settings, we must preemptively discover all the ways in which they can fail. That is, we desire very powerful evaluation, red-teaming, and stress- testing technologies and tooling for language models. To this end, this thesis introduces the project of automated and flexible stress-testing and develops 1) REALM, a comprehensive robustness bench- mark and publicly hosted leaderboard with twenty-six hosted models in partnership with the biggest machine learning platform company in the world; 2) a suite of three novel and tailored red-teaming algorithms alongside three case studies of their successful application against in-the-wild LLM use cases; and 3) a multiagent stress-testing framework that universally jailbreaks five state of the art large language models that have been specifically finetuned to prevent jailbreaks. It is our hope that the technology developed here can enhance the conversation around safe, responsible, and secure AI de- velopment with regards to practical failure modes and correspondingly the methods to prevent them. In particular, with the technology developed here, all language model researchers, developers, users, businesses, and stakeholders will be able to discover failure cases before they arise in production. This brings us one step closer to the dream of robust language model systems.Computer Scienc
Geometric Arthur Parameters
No full textWe study unramified principal Eisenstein series from the perspective of the geometric Lang-
lands equivalence. We prove a geometrization of the Langlands constant term formula and
show that it recovers the classical result after applying the categorical trace of Frobenius. By
analyzing the singular support filtration on the category of ind-coherent sheaves on the stack of
local systems with restricted variation, we produce a generalization of a formula of D. Kazhdan
and A. Okounkov. As a result we produce an upper bound on the spectrum of the Hecke
algebra associated to a rational point of a curve, relative to its action on unramified principal
Eisenstein series.Mathematic
Application of Hansen Solubility Parameters to Improve Oral Absorption of Nintedanib by a Self- Micro-Emulsifying Drug Delivery System
No full textNintedanib is a drug used to treat Idiopathic Pulmonary Fibrosis, but it has poor solubility and low oral bioavailability. This study explored how a Self-Microemulsifying Drug Delivery System (SMEDDS) could improve these issues. By using Hansen Solubility Parameters, two solvents, Benzyl Alcohol and Eugenol, were found to dissolve Nintedanib well and used in the oil phase. HSP provides a systematic way to predict solvent compatibility, making it a valuable tool for addressing insolubility challenges in poorly soluble drugs. By ensuring compatibility between Nintedanib and the chosen excipients, HSP helped optimize drug loading within the formulation, ultimately contributing to improved bioavailability. HSP enhances bioavailability by identifying excipients that not only dissolve the drug effectively but also facilitate its incorporation into a stable microemulsion, improving its dispersion and absorption in the gastrointestinal tract. The Eugenol: Benzyl Alcohol [1:2] blend created the largest microemulsion region, showing the best self-emulsification potential. Tween 20, chosen for its good surfactant properties, was tested at higher concentrations (50%), resulting in smaller droplet sizes (5 nm) and faster emulsification.
The permeability study showed that SMEDDS significantly improved Nintedanib’s ability to cross Caco-2 cell layers, with the Eugenol: Benzyl Alcohol [1:3] mix reaching a high permeability level (Papp = 1.08 × 1 0⁻⁶ cm·s⁻¹). Efflux ratio tests also suggested that Nintedanib is less likely to be affected by P-glycoprotein, which is important for improving absorption. However, there were some limitations, like the lack of stability testing, the
potential for phase separation, and the need to confirm these findings in animal models. Despite these, the results show SMEDDS could be a promising strategy to boost Nintedanib’s bioavailability, and further testing is needed.Extension Studie
Consequences of Unpaid Labor: Evidence from the U.S.
No full textUnpaid labor—such as caregiving and housework—remains foundational to the functioning of families and society, yet continues to be undervalued, invisible, and unequally distributed. This dissertation examines the lived experiences and consequences of unpaid labor focusing on three groups—grandparents, parents of minor children, and working parents—using nationally representative data. It asks: How do caregiving expectations shape labor force participation in later life? How do parenting practices aligned with intensive parenting norms relate to stress and fatigue? And how is unpaid labor timed across the workday in relation to paid work and subjective well-being?
The first chapter examines how gendered dynamics unfold across generations to shape older adults’ labor force participation. The second chapter introduces “backdrop parenting”—moments of child presence without active care—and suggests that mothers experience greater fatigue than fathers from these experiences due to overlapping unpaid tasks. The third chapter explores how working parents schedule unpaid labor around paid work, revealing that stress and fatigue vary with its timing across the day.
Together, these chapters offer a more contextualized understanding of unpaid labor, emphasizing not only how much time is spent, but also when care occurs, who is present, and how it is experienced. By highlighting less visible contours of unpaid labor—such as intergenerational interlocking of gender norms, backdrop parenting, and the temporal layering of tasks—this dissertation challenges conventional ideas about unpaid labor and underscores the need for intersectional, life course–oriented approaches to understanding unpaid labor.Sociolog
After Snow: The Case for an Alpine Public
No full textIn an era of climate transition, alpine skiing demands reexamination: What happens after snow? This thesis envisions the ski resort as a public terrain for contested remediation rather than a privatized, extractive enclave. Focused on the Telluride Ski Resort in Colorado, the project examines the ski mountain as a designed landscape—clear-cut, regraded, and reshaped—while tracing the industry’s corporate consolidation and shift from natural to artificial snow.
Telluride has taken drastic measures to sustain skiing’s illusion in a warming world, expanding its snowmaking network and turning water into frozen infrastructure. To counter this artifice, the thesis proposes new alpine infrastructures: solar screens to shield glare, revegetated clearings to act as windbreaks, and regraded micro-topography to retain natural snow. Through land stewardship, privatized operations are transformed into a public resort. Rather than projecting skiing’s end, After Snow proposes a designed transition, tying the sport’s long-term survival to the creation of an alpine public.Department of Landscape Architectur