Dissertations

Permanent URI for this collection

https://hdl.handle.net/20.500.14394/54496

Browse

Study of Charged Macromolecule Phase Behavior using Conventional and Modern Modeling Methods
(2025-02) Hoover, Samuel C.
Charged macromolecule systems serve as a wonderful basis for the study of the fundamental physics underlying biological phenomena. They provide relatively clean, simple systems, displaying rich physics due to their many degrees of freedom, short-ranged van der Waals interactions, long-ranged Coulomb interactions, entropic contributions arising from the mobile species, and chain connectivity. The self-assembly of charged macromolecules likely played a vital role in the emergence of life through equilibrium phase behavior like liquid-liquid phase separation and complex coacervation. In this work, we study the equilibrium phase behavior of charged macromolecule systems using conventional modeling techniques, like free energy minimization using numerical methods, and modern methods, like machine learning models, when computational barriers restrict more systematic analysis. We study the pH-responsive complex coacervation between polyzwitterions and polyelectrolytes — an interesting class of complex coacervates that are found to be stable at pH values considerably different from complex coacervates formed between polyelectrolytes. Developing new theory, we probe the many physicochemical parameters to explore the characteristics of their phase behavior. We then study the microphase separation transition behavior of sequence-defined polymers, those in which can only be distinguished by their unique monomer sequence and are analogous to proteins. The influence of monomer ordering plays a well known role in the assembly and conformations of charge macromolecules. Those sequence- dependent effects are only beginning to become unraveled and a systematic molecular dynamics study faces computational barriers. A machine learning-aided study of sequence-dependent effects on charged sequence-defined polymer phase behavior is conducted. First, we train a gradient boosted decision tree model to predict the microphase separation transition from the monomer sequence. From there, we study the learned patterns from the model to provide qualitative relationships between monomer order and microphase behavior.
Using bioinformatics to investigate the influence of sexual traits in health and disease
(2025-02) Edwards, Melise
The goal of this dissertation work is to investigate sex differences in the brain by evaluating sexual traits like hormones and gene expression and their contributions to disparate outcomes in health and disease. While numerous studies have looked at sex differences in the brain, few studies incorporate a multivariate model of sex into their work; this framework posits that sex is a constructed, nonbinary category that includes multiple, uncoupled sexual traits like chromosomal organization, physiology, behavior, hormone status and levels, and gene regulation. This expanded framework will better serve our inquiries into biological variation within and across sexes and species, as well as more accurately pinpoint causal mechanisms of disease. Using bioinformatics tools to study gene expression changes that occur along the aging trajectory, we investigate the role of sexual traits like hormones (e.g. estrogens) and hormonal targets in Alzheimer’s disease. In another approach, we examine the functional consequences of aromatase (estrogen synthase) inhibition in primate brain tissue to understand potential susceptibility to neurodegeneration and cognitive dysfunction. Lastly, we investigate sexual traits like gene expression and hormone status in a cohort of human Alzheimer’s patients to employ a multivariate model of sex in our analyses; this allows us to identify sex-biased cell populations uniquely vulnerable to disease. Together, these inquiries and bioinformatics analyses will allow us to more accurately define how sexual traits inform disease outcomes and contribute to a more inclusive scientific field.
EFFICIENT SOLUTIONS OF FIRST-KIND SURFACE INTEGRAL EQUATIONS
(2025-02) Lochner, Nash
The Method of Moments (MoM) solution of surface integral equations (SIEs) provides an accurate and flexible mathematical framework for solving general electromagnetic (EM) structures composed of various open/closed perfectly conducting, resistive, and/or penetrable bodies. However, the computational resources required to analyze modern EM devices and platforms have begun to outgrow industry-standard solver technologies. To answer the call for more throughput and higher fidelity in the engineering design cycle, the field of fast integral equation solvers has become an active area of research. While first-kind SIEs are known for their superior accuracy, flexibility, and efficiency, few, if any, fast solver frameworks fully leverage those intrinsic properties of first-kind SIEs for both iterative and direct solution strategies, or their combinations. This dissertation proposes a new algorithmic framework for the fast solution of first-kind SIEs. The proposed framework compresses dense MoM matrices in the form of $\mathcal{H}^2$-matrix representations and flat block-wise sparse low-rank matrix formats, which are suitable for constructing: (a) fast iterative solvers, (b) fast preconditioners, and (c) fast direct solvers, along with their combinations. At the core of all these developments is the idea of using a structure-specific set of (composite) basis functions that {\em radiate well}, and which is used to approximate far MoM matrix interactions in the case of fast iterative solvers, and {\em all} interactions in the case of fast direct solvers and preconditioners. The main novelty and efficiency of the method lie in the `inward-looking' approach of generating this orthogonal basis set and basis transformation. Much like rigorous approaches to computing optimal basis sets in electromagnetic systems, the proposed approach computes orthogonal group bases via an eigen decomposition of the self (group) resistance matrix, which is readily available from the near-field complex symmetric MoM (BEM) impedance matrix of the first-kind IE operator. The proposed first-kind SIE framework will first be formulated for the application of $\mathcal{H}^2$-representations of electrically large perfectly conducting and dielectric objects. With the core mathematical foundations established, a fast direct solver will be developed that sparsifies a `flat' $\mathcal{H}^2$-like representation into a blockwise sparse representation, which can efficiently be factorized using a custom block-sparse BLR LDL$^T$. With both the $\mathcal{H}^2$-matrix and direct solver frameworks in place, the proposed fast solver ecosystem will be deployed in the areas of robust preconditioning and the solution of composite conducting/dielectric objects. The asymptotic complexity of the proposed solvers is $\mathcal{O}(N^2)$ and $\mathcal{O}(N^3)$. However, for problems up to $32 \lambda$ in all three dimensions, the multilevel matrix representations and iterative solver were experimentally observed to have complexities of $\propto N^{1.2}$ for memory, $\propto N \log N$ for MxV times, and $\propto N^{1.5} \log N$ for setup times. The single-level sparsification-based fast solver was found to have $\propto N^{1.4}$ memory complexity and $\propto N^{2.12}$ factorization time. All methods were experimentally found to be error-controllable and resilient on a relatively small sample of realistic scattering problems.
Three Essays on Shacklean Decision Theory
(2025-02) Kwak, Se Ho
G.L.S. Shackle was a major critic of probability based economic theory. During the 1980s and 1990s, his alternative framework of decision-making was mathematically reconstructed by Katzner. In this dissertation, I aim to advance Shacklean decision theory, as reformalized by Katzner, by exploring its mathematical foundation, methodological underpinnings, and applications to financial decision-making. In the first paper (chapter 1), we will construct an axiomatic system of ordering relation defined on the σ-field F* as the space of hypotheses about the future states in order to derive the functional representation of potential surprise formalized by Katzner. For this task, we introduced a base space, called monad F spanning all imaginable hypotheses by union operations in F. In addition, to investigate necessary topological conditions, we introduced the degree space F ̅, a quotient space of total preorder on monad F. Based on this, we will derive various continuous order-preserving functional representation of potential surprising ordering from F ̅ to [0, 1]. The second paper (chapter 2) presents a new interpretation of Shacklean theory by focusing on the common stage structure of decision-making in relation to expected utility theory. This paper shows that the characteristics of Shacklean framework can be explained as: (1) potential surprise, a non-distributive and non-additive, almost ordinal measure of subjective uncertainty defined on the incomplete list of imaginable future states, (2) attractiveness function, the valuation of “importance” reflecting various types of loss-psychology, and (3) decision-index, the final choice function of the action on the set of importance intervals. The third paper (chapter 3) explores individual portfolio adjustment behavior and potential scenarios that may trigger a panic response in asset portfolio management, using the Shackle-Katzner framework. It begins by examining a portfolio consisting of money and a single non-monetary asset, followed by an analysis of a portfolio containing two non-monetary assets. This study argues that if an investor relies solely on a dominant aspect—whether optimistic or pessimistic—in anticipating future payoffs shaped by market conditions, then not only abrupt and sharp updates of negative information but also a continuous and gradual increase in negative anticipation can lead to a sudden panic response upon crossing a critical tipping point.
Parameter Inference at the Large Hadron Collider using Neural Likelihood Ratios, and a Measurement of the Higgs Boson Decay Width using the ATLAS Experiment
(2025-02) Sandesara, Jay
A new statistical technique is developed for physics parameter estimation at the Large Hadron Collider (LHC) that uses modern deep-learning tools to realize a more general and fundamental approach to data analysis compared to the ad-hoc techniques commonly used. Coming under the general umbrella of Neural Simulation-Based Inference (NSBI) techniques, the new workflow uses a large number of Neural Networks (NNs) to directly learn event-by-event likelihood ratios and thus handles high-dimensional parameter estimation without the need to bin data into low-dimensional summary histograms. We developed novel techniques for parameterizing the likelihood ratios as a function of a large number of parameters common in LHC experiments and created modern computational workflows that make it possible to apply NSBI to a full-scale ATLAS experiment analysis. A measurement of the Higgs boson in the off-shell phase space is then performed using the new workflow, in the $H^*\to ZZ \to 4\ell$ channel. The evidence sensitivity is increased by a factor of $3.1$ using the new technique in the $H^*\to ZZ \to 4\ell$ channel, compared to the previous measurement published by ATLAS using the same Run-2 data with more standard techniques. A combination of this channel with the off-shell $H^*\to ZZ\to 2\ell2\nu$ channel is done finding evidence for the off-shell Higgs boson with $3.7 \sigma$ confidence level, superseding the last measurement. The off-shell measurement is then combined with the on-shell measurement for an indirect measurement of the Higgs boson decay width under a few assumptions. The observed (expected) value of the Higgs boson width at 68\% CL is $4.3^{+2.7}_{-1.9}$ ($4.1^{+3.5}_{-3.4}$)~MeV. This new method is promising for a wide range of measurements at the LHC, where no single observable may be optimal to scan over the entire theoretical phase space under consideration, or where binning data into histograms could result in a loss of sensitivity. It also offers easy re-interpretability and broader use.
New Designs in Perovskite-Organic Hybrid Materials with Enhanced Properties and Functionality
(2025-02) Cueto, Christopher
Lead halide perovskites (APbX3) are an unusual class of ionic semiconductors that first entered the scene as efficient solar absorbers in photovoltaics, and have since been adapted into colloidal nanocrystal (NC) form. The latter emit bright light and bear unique optoelectronic properties that set them apart from their metal chalcogenide quantum dot (QD) forerunners, including a bandgap that is tailorable across the visible range by halide selection (X = Cl, Br, I), a soft, dynamic lattice that undergoes facile ion-exchange and chemical rearrangement, extremely fast photoluminescence decay rates, and an innate defect tolerance the lifts the need for epitaxial passivation to achieve high luminescence. Unlike bulk polycrystalline films, reducing the ABX3 dimensions to the nanometer scale opens a large surface area and renders them dispersible in organic media, permitting intimate mixing with organic polymer and optoelectronic components than can lead to emergent new properties. Several volumes could be dedicated alone to the study of the inorganic chemistry, morphology and photophysics of nanocrystalline perovskites; in the first part of Chapter 1, a condensed, holistic description of their crystal structure, optical properties, ion-exchange behavior, shape-control mechanisms, chemical transformations, and unconventional surface chemistry is presented. In the second half, recent progress in the area of perovskite-organic hybrid materials is interpreted against this backdrop. In the ensuing chapters, efforts to design novel perovskite-organic designer materials are described. In Chapter 2, amphiphilic, zwitterion block copolymers are synthesized and used directly in the hot injection synthesis of CsPbBr3 NCs, permitting their direct attachment to the NC surface without intermediate ligand exchange steps. In Chapter 3, polystyrene derivatives with ammonium halide pendant groups are used to tune nanocomposite color in thin films and control the kinetics of inter-NC halide exchange. In Chapter 4, precise-length, π-conjugated zwitterions are introduced as replacements for the typical aliphatic ligands used on CsPbBr3 surfaces, demonstrating the impact of doing so on energy transfer (between ligand and perovskite) and NC-packing geometry in films. In Chapter 5, the photocatalytic activity of the perovskite NC is exploited in a new photolithography process for patterning fluorescence color (and shape) into nanocomposite films. Lastly, Chapter 6 entails a discussion of the future of this field, with commentary on lead-free perovskite NC compositions.
PDMS Brushes as Soft Materials in Elastomers and Photonic Crystals
(2025-02) Uchiyama, Takumi
Bottlebrush polymers exhibit ultra-softness and low viscosity compared to linear polymers due to minimal chain entanglement. These unique properties make them highly suitable for applications such as elastomers and photonic crystals, where their fast assembly and large domain spacing are advantageous. This dissertation focuses on three key developments: 1) PDMS brush polymers with hydrogen bonding for self-healing, 2) the development of polyrotaxane crosslinked bottlebrush polymers 3) the study of stretchable photonic crystals based on PDMS block bottlebrush copolymers are discussed. First, the self-healing properties were achieved in PDMS brush polymers by crosslinking via hydrogen bonding. Three different strengths of hydrogen bonding were evaluated, and their temperature response was compared as hydrogen bond association is a function of temperature. Additionally, PDMS brush polymers were crosslinked using polyrotaxane, a slide-ring crosslinker known to improve elongation and toughness. This novel combination of polyrotaxane and bottlebrush polymers was successfully synthesized. Finally, leveraging the softness and rapid assembly behavior of block bottlebrush copolymers, stretchable photonic crystals were developed. These photonic crystals were fabricated using two sets of PDMS-based block bottlebrush copolymers with different molecular weights and additives, forming spherical structures that exhibited red and blue colors, respectively. The reflection spectra of these crystals responded to applied strain, demonstrating color shifts (or no color shifts) that were influenced by the degree of crosslinking and polymer structure, highlighting the potential for switchable optical properties in flexible photonic materials.
Watershed ecohydrology and ecosystem services in the Ulua River watershed of Honduras
(2025-02) Quiñónez Camarillo, Ana Lorena
Watersheds are important for the sustenance of life since they provide ecosystem services to human communities and local biodiversity alike. The ongoing changes caused by the anthropogenic exploitation and use of the surrounding ecosystems have serious repercussions on the water quality and quantity of local water bodies. We have developed a Participatory Framework for analyzing threats, consequences, and solutions that will allow us to gain insight into the community-resource interactions. We evaluated socioeconomic and landscape factors that could be influencing the resilience of commons, including communities' perception of the threats to natural resources, the consequences of their loss, and the solutions they perceive as most effective to prevent this loss. The incentives required to promote widespread transition toward sustainable coffee production are poorly understood, leaving policy makers with insufficient information to design scalable forest conservation initiatives. We also developed a choice experiment to assess if Payment for ecosystem services as a viable option for improving the sustainability of coffee production in Honduras. Coffee farmers may be reluctant to set aside significant percentages of the landscape for forest restoration. Low-income coffee farmers appear to be more averse to forest conservation, suggesting the need for coupling monetary incentives with insurance against revenue losses. SWAT was used to develop a large-scale watershed model for the Ulua watershed. This model allowed the development of streamflow, sediment, and nutrient data to provide a baseline that can be used to define further work in the area and to consider the future evaluation of climate and land use changes and their effect on the watershed. Ulua is an important watershed in Honduras. It is the second biggest watershed in the country, has a high amount of agricultural production with economic relevance, and has extensive flooding problems in urban and rural areas. Developing a hydrologic model for this watershed could provide relevant information to improve its management.
UNVEILING THE ROLE OF GUT MICROBIOTA IN CURCUMIN METABOLISM IN VITRO AND IN VIVO
(2025-02) Luo, Minna
Curcumin, derived from the rhizome of turmeric plant (Curcuma longa) rhizome, has been used in traditional Asian medicine for its therapeutic benefits for centuries. Extensive research has shown that curcumin acts on various cellular signaling pathways, and this has been related to its potential to treat a range of disorders. Early clinical trials have demonstrated its promising role in preventing colon, oral cavity, and liver cancers. However, the plasma concentrations of curcumin required for effectiveness are much higher than what can be achieved in current settings. This is due to the hydrophobic nature of curcumin, which results in low plasma levels and poor bioavailability. Therefore, there is a paradox between the high bioavailability and low bioavailability of curcumin. To explain, recent studies noticed that curcumin could affect gut microbiota composition, increasing gut microbiota diversity and beneficial commensal bacteria, and decreasing potentially pathogenic bacteria. Therefore, it was hypothesized that gut microbiota might be the potential target for curcumin’s therapeutic effects. However, the interaction between gut microbiota and curcumin is not sufficient and requires more studies. Therefore, the goal of this research is to enhance our understanding of the interactions between curcumin and gut microbiota and to discover and identify gut microbiota capable of metabolizing curcumin and gut microbiota-derived metabolites. First, the role of gut microbiota in curcumin metabolism in vivo remains poorly understood. To address this, we used antibiotics to deplete gut microbiota and compared curcumin metabolism in control and antibiotic-treated mice. Using Q-TOF and triple quadrupole mass spectrometry, we identified and quantified curcumin metabolites, revealing distinct metabolic pathways in these two mice groups. The novel metabolites, hexahydro-dimethyl-curcumin and hexahydro-didemethyl-curcumin were exclusively derived from gut microbiota. Additionally, gut bacteria deconjugated curcumin metabolites back into their bioactive forms. Moreover, control mice exhibited significantly lower curcumin degradation, suggesting a protective role of gut microbiota against degradation. In conclusion, these results indicated that gut microbiota might enhance the effectiveness of curcumin by deconjugation, production of active metabolites, and protection against degradation in the large intestine. Second, interindividual variation in this process remains unexplored. In this study, we anaerobically fermented curcumin with gut microbiota from seven donors in vitro. Curcumin metabolites were qualified and quantified using UPLC-orbitrap fusion tribrid mass spectrometer and LC-MS, respectively. Sixteen metabolites were identified, with methylated and acetylated metabolites being reported for the first time. Tautomers of five compounds were also identified, predominantly in enol forms. Quantification of these metabolites in different human fecal samples over time revealed significant variations in the types and levels of curcumin metabolites produced among individuals. These differences are attributed to variations in gut microbial profiles based on 16S rRNA sequencing results. Significant correlations were found between specific bacterial taxa and curcumin metabolites. In summary, our study identified new metabolites of curcumin and highlighted significant interindividual differences in the capacity of gut bacteria to metabolize curcumin. Lastly, although curcumin is widely recognized for its health benefits, the role of gut microbiota in its metabolic transformation was not well-studied. In this study, bacterial strains capable of metabolizing curcumin were isolated from human stool samples. Using 16S rRNA and whole-genome sequencing, two novel strains (Clostridium butyricum UMA_cur1 and Escherichia coli UMA_cur2) were identified. In addition, the metabolic products were analyzed using liquid chromatography-mass spectrometry (LC-MS). These strains efficiently converted curcumin into dihydro-curcumin (DHC) and tetrahydro-curcumin (THC). Notably, E. coli UMA_cur2 also produced hexahydro-curcumin (HHC) and octahydro-curcumin (OHC), marking the first identification of a strain capable of such transformations. The absence of the YncB gene (typically involved in curcumin conversion) in C. butyricum UMA_cur1 suggests an alternative metabolic pathway. Curcumin metabolism begins during the stationary growth phase, indicating that it is not crucial for primary growth functions. Furthermore, E. coli UMA_cur2 produced these metabolites sequentially, starting with DHC and THC and progressing to HHC and OHC. These findings identified two novel strains that can metabolize curcumin to hydrogenated metabolites. In conclusion, these studies improve our understanding of the interactions between curcumin and gut microbiota.
Investigating The Impact Of Computing-focused Hackathons on Self-efficacy Development Towards Computing, Computational Thinking And Innovation.
(2025-02) Kathala, Krishna Chaitanya Rao
This dissertation investigates the impact of computing-focused hackathons incorporating Wearable Learning (WL) technology on the development of self-efficacy in computing, computational thinking, and innovation among middle and high school students. Employing a sequential explanatory mixed-methods design, the study examines how immersive, hands-on learning environments foster critical competencies in today’s technology-driven society. The quantitative phase analyzed pre- and post-survey data from 54 participants across three program sites: Upward Bound, Massenberg STEM Institute, and the RIDE Center. Paired-sample t-tests revealed statistically significant improvements in computing self-efficacy (pre-test mean: 3.46, post-test mean: 3.62, p-value: 0.02859) and innovation self-efficacy (pre-test mean: 3.43, post-test mean: 3.61, p-value: 0.03706). A secondary analysis focusing on computing-related items and innovation-driven measures revealed amplified gains, emphasizing the WL curriculum’s dual impact on computational and creative problem-solving skills. Improvements were particularly evident in curricular stages that emphasized implementation, testing, and iterative refinement. The qualitative phase enriched these findings through exit tickets, semi-structured interviews, field notes, and student artifacts, providing nuanced insights into students’ learning experiences. The analysis underscored the transformative role of hackathons in enhancing students’ confidence, problem-solving abilities, and active engagement. Exit ticket responses and observational data captured students’ progression from initial apprehension to confidently presenting their completed games, illustrating growth in computational thinking, self-efficacy, and innovation skills. This research highlights hackathons as effective educational models for fostering self-efficacy and computational thinking, aligning with Bandura’s Social Cognitive Theory and Gee’s Sociocultural Perspective on Opportunity to Learn. By integrating quantitative and qualitative findings, the study supports the adoption of hackathon-based curricula in STEM education to enhance computing and innovation skills. It concludes with a proposal for a scalable framework to integrate WL interventions into mainstream education, advancing equity and inclusivity in access to computational competencies and innovation-driven learning opportunities.
MORAL INJURY AND SUICIDALITY IN POST-9/11 VETERANS: TRANSITION FROM MILITARY TO CIVILIAN LIFE
(2025-02) Garland, Eugene
In 2021, suicide in the veteran population accounted for 13.7 percent of all suicides among adults in the United States, with an average of 17.2 veteran suicides per day. Serving in the military potentially exposes veterans to unique risk factors that may contribute to suicidality. For example, physical injuries, mental health disorders, post-traumatic stress disorder (PTSD), traumatic brain injury (TBI), and moral and ethical challenges place veterans at higher risk for suicidality. The primary purpose of this study is to explore the relationships between moral injury and suicidality in post-9/11 military veterans who have recently separated from military service. Through exploring these relationships, the study aims to contribute to the existing body of knowledge on mental health challenges faced by veterans. Secondary data for this study was obtained from The Veterans Metrics Initiative (TVMI) to examine the potential contributing factors to suicidality in the veteran population. Moral injury was found to significantly predict suicidality. Combat exposure alone did not emerge as a significant predictor, nor did the interaction between moral injury and combat exposure significantly moderate the relationship with suicidality. Finally, examination of the role of PTSD as an effect modifier in the relationship between moral injury and suicidality revealed that PTSD does not significantly moderate the relationship between moral injury and suicidality. Understanding the association of moral injury, combat exposure, and PTSD to suicidality in this cohort provides a new pathway to build effective early interventions and treatment protocols targeted at multiple contributing factors that can lead to suicidality and suicide.
EXPLOITING PERVASIVE LEAKED EM SIGNALS FOR COMMUNICATION, CHARGING AND SENSING
(2025-02) Cui, Minhao
Wireless technologies are becoming increasingly important in our daily lives. As we use 4G and 5G services, researchers are also working on the development of future 6G networks. In addition to traditional communication functions like Wi-Fi, wireless signals are now being used for localization, sensing, and even charging. However, one significant challenge with these new applications is that they require dedicated signal transmissions, which can interfere with the original communication functions. In my Ph.D. thesis, I aim to leverage the pervasive ambient leakage signals, which are typically seen as detrimental, to enhance wireless communication performance and enable new functions like sensing and charging. My approach is based on the observation that there is a significant amount of leakage RF signals in our environment. For instance, powerlines continuously emit 50/60 Hz electromagnetic (EM) signals due to the alternating current flowing through them. In my thesis, I implement innovative designs in both hardware and software to transform these ambient leakages from adversaries into valuable assets. We first explored that during the transmission of VLC, the transmitter not only emits out visible light signals but also leaks out RF signals due to the intensity modulation scheme. What's more interesting is that the leakage RF signal contains a copy of the data transmitted in the light signals and this finding renders VLC--the generally believed most secure wireless technology--not secure any more. Building upon this finding, we further demonstrate a novel utilization of these leakage signals for carrying extra data to double the data rate of existing VLC systems. Following the successful utilization of VLC leakage signals for communication, we further view the leakage signals as a form of wasted energy and devote our effort to harvesting it with the help of wearable bracelet antenna. Last but not least, we are going to utilize the leakage signals for sensing purposes. This electromagnetic leakage, stemming from alternating current in the electric appliance, is governed by Maxwell's Equations. We can infer the body motions by analyzing such leakage signals received by the human body. Therefore, we decide to leverage the EM leakage from electric vehicles (EVs) to enable in-vehicle sensing. We observe that numerous components within the EVs including battery, powerline, and power inverter, emit EM signals during their operation. And such leakage can be utilized to sense the body motions of the driver/passenger without any dedicated signal transmitters.
Towards reliable black-box variational inference
(2025-02) Agrawal, Abhinav
Probabilistic models are essential for understanding complex systems across various scientific fields. Inferring the posterior distribution over the unknown quantities is central to generating insights from these models, but the posterior is often analytically intractable and must be approximated. Black-box variational inference (BBVI) is a leading approximate inference framework that uses optimization to find a tractable posterior approximation. BBVI methods apply to a wide variety of models and offer practitioners a faster alternative to expensive Markov chain Monte Carlo methods. While BBVI is promising, it suffers from several issues that hinder its widespread adoption. We identify four such challenges: robustness, scalability, evaluation, and accuracy, and address them in this thesis to make BBVI a reliable inference tool. The first chapter addresses the issue of robustness. Naive BBVI approaches are often not robust enough to work out of the box and fail to converge without model-specific tuning. We improve this by integrating key algorithmic components like initialization strategies, gradient estimators, variational objectives, and advanced variational families. Extensive empirical studies demonstrate that our proposed scheme outperforms competing baselines without requiring model-specific tuning. The second chapter improves scalability. When models are large, BBVI methods fail to scale and suffer from slow convergence. We address this by proposing a structured and amortized BBVI scheme that maintains accuracy while offering faster and more scalable inference than existing approaches. The third chapter improves evaluation. The posterior predictive distribution is used to make predictions on unseen data; however, sometimes posterior predictive evaluations can be extremely noisy. We identify the conditions under which the simple Monte Carlo estimator of the posterior predictive distribution can exhibit an extremely low signal-to-noise ratio. Based on this analysis, we introduce an adaptive importance sampling approach that significantly improves the evaluation accuracy. The fourth chapter improves accuracy. Normalizing flow-based variational inference (flow VI) is a promising class of BBVI methods, but its performance is mixed, with some works reporting success and others reporting optimization challenges. We conduct an empirical analysis to disentangle the impact of various factors like representational capacity, objectives, gradient estimators, batch sizes, and step sizes. Our analysis leads to practical recommendations for each factor, culminating in a flow VI approach that matches or surpasses leading turnkey Hamiltonian Monte Carlo methods on a wide variety of targets. Collectively, the advances in this thesis make BBVI a reliable inference method.
“Prime Harvest”: The Bioarchaeology of Body Acquisition for Iceland's Early Medical Training
(2025-02) Netzer Zimmer, Adam
Dissection of the human body has long been a key part of Western medical education. To acquire bodies, anatomists often resorted to exploitative methods, including grave robbing and trading bodies through colonial networks targeting marginalized populations. Bioarchaeological research, particularly by scholars of color, has shown how such practices were central to colonial European nations’ constructions of race, identity, and difference. However, less attention has been paid to cadaver acquisition and medical teaching in marginal European nations. This dissertation examines the history of Icelandic cadaver collection through the Læknagarður Anatomical Skeletal Collection at the University of Iceland. Using queer archaeology and Black feminist theory, I explore how Iceland navigated its dual position as both a colonial dependency and part of European racial hierarchies. By analyzing the collection alongside historical and ethnographic records, I investigate the intersections of colonialism, medical education, and race in Icelandic history. The collection includes human remains retained after surgeries, excavated from local archaeological sites, purchased through international trade networks, and donated by physicians. These findings reveal how Icelandic medical education relied on global networks while reflecting local constraints. I argue that the creation of the Læknagarður Collection was not merely a medical teaching tool but also a nationalist project, reflecting global trends in using anatomical knowledge to construct identity. Drawing on Sylvia Wynter’s critique of human hierarchies and queer archaeology’s focus on marginalized narratives, I show how the collection embodies broader colonial histories and power dynamics in the circulation of human remains. Ultimately, this research seeks to make legible the forgotten histories of the individuals whose remains are now housed in Læknagarður and reveals how both local and global dynamics shaped Iceland’s early anatomical practices.
Slippages of the Collective Present: The Aesthetic and Political Frictions of Oil in the Twenty-First Century
(2025-02) Vazquez Sanchez, Eric
This dissertation examines literary, cinematic and collective responses to oil from the first two decades of the twenty-first century as they attempt to historicize presents and conceive of futures. This project posits that disparate geographic zones and modes of life are engaged in an act of mutual recognition through oil. Therefore, this project is about the slipperiness and friction caused by oil and the ways in which oil insinuates itself into every aspect of modern life. Situating oil as the medium through which global processes become legible, the works under examination respond to a variety of ecological, historical and narrative imperatives as they enact sites of remembrance that expose the conditions for social exclusion, environmental degradation, and shared complicity. Consequently, this study employs specific case studies that reproduce and disrupt established oil depictions, thus examining the cracks in petroculture. The first chapter analyzes the political unproductivity of oil through an examination of the overlapping of historical memory and a dissociated witnessing in the waning days of the petrostate as presented in two novels from Nigeria and Venezuela. In the second chapter, I focus on two novels from Argentina and Gabon that choose not to revel in the moment of witnessing but in what comes after as they depict the emergence of toxic communities located at the erratic resource frontiers of oil. Oil as a political actor in a network of labor and consumption is at the crux of chapter three, which examines two films from China and Latvia that contrast an ethics of slowness with petrocapitalism. The final chapter meditates on the political potential of protest as it reconsiders oil as a site of resistance and reimagining in the collective work of Indigenous activists in North America. Using friction as a mode of engagement sets the scope and limits of my project, as friction arises from the assemblage of the selected works, the dialogues each pair represents, their treatment of genre, and the ways they contest established claims surrounding oil culture.
Investigating Contributors to Pancreatic Toxicity Following Developmental Perfluorooctanesulfonic Acid (PFOS) Exposure in Zebrafish (Danio rerio)
(2025-02) Tompach, Madeline
Perfluorooctanesulfonic acid (PFOS), a legacy member of the per- and polyfluoroalkyl (PFAS) chemical class, has been shown to broadly affect organ systems using unknown or multifaceted mechanisms. Epidemiological studies demonstrate that PFOS crosses the placenta and is found in breast milk, underscoring the importance of studying developmental exposure. Developmental PFOS exposure in zebrafish has been shown to affect both the endocrine and exocrine pancreas. This dissertation identifies and investigates factors that contribute to pancreatic toxicity with developmental PFOS exposure. The endocrine pancreas is composed of islets of Langerhans containing insulin producing β-cells that regulate blood glucose while the exocrine pancreas secretes digestive enzymes to break down nutrients for intestinal absorption. Due to its structure, PFOS has been identified as a fatty acid mimic, aiding in its ability to enter cells through nutrient transporters. Here we compare the effects of PFOS and a non-fluorinated structural analog alpha lipoic acid (ALA). We demonstrate that, like PFOS, ALA reduce islet area and alters lipid parameters, indicating that that fatty acid mimicry plays a role in islet toxicity with exposure to ALA and PFOS during development. We also investigate how PFOS exposure affects the development and function of the exocrine pancreas across multiple larval stages. We show that increased yolk utilization can lead to an insurmountable nutrient deficit, leading to reduced exocrine pancreas size in post-yolk feeding larvae. Additionally, we show that even at a timepoint and PFOS concentration where there are no detectable changes to the size of the exocrine pancreas, there is reduced exocrine pancreas function. Finally, we study how PFOS exposure affects islet vascularization where we saw a decrease in contact between the vasculature and the β-cells, likely driven by changes to cell-cell adhesion, namely reduced integrin gene expression with PFOS exposure. This dissertation advances the understanding of contributors to pancreatic toxicity with PFOS exposure. Future studies should be focused on studying the mechanisms underlying the processes described in this work to better understand the contribution of PFOS exposure during development on the risk of chronic diseases later in life.
Impacts of Microbial Diversity on Soil Carbon and Ecosystem Function
(2025-02) Shinfuku, Melissa
Climate change induced biodiversity loss threatens to diminish the ability of diverse microbial communities to promote the stability and productivity of soils. These communities not only sequester carbon by producing microbial necromass, but also serve as the underpinning for the critical functions that drive terrestrial carbon cycling. Understanding how ecosystems and soil carbon stocks respond to shifts in microbial community composition and diversity will influence the role of soils as a carbon sink or source on a warming planet. Using both laboratory incubations and observational field samples from the Harvard Forest Long-Term Ecological Research (LTER) site, I investigated how microbial diversity affected the persistence of microbially-derived soil organic matter and ecosystem function. Overall, taxonomic diversity appears to have little effect on either necromass persistence or ecosystem function. The diversity of the microbial community producing soil organic matter is not a driver of the carbon use efficiency of microbial necromass, suggesting that the microbially-transformed soil organic matter bears similar susceptibility to decomposition, regardless of the community composition that produces the necromass. Additionally, bacterial taxonomic diversity was not a driver of ecosystem function at the Harvard Forest LTER. However, bacterial functional diversity, in particular acquisition trait richness, had a negative relationship with ecosystem function in soils exposed to chronic warming. Altogether, these results indicate that the microbial taxonomic diversity has little impact on the persistence of microbial necromass in the short-term or on ecosystem function at the Harvard Forest LTER. However, microbial acquisition trait richness did have a negative relationship with ecosystem function in soils exposed to chronic warming. Together these results highlight that microbial functional diversity, not microbial taxonomic diversity, may be a better approach to understanding how microbial diversity drives ecosystem function in a changing climate.
EXPLORING REPRESENTATIONS FOR 3D RECONSTRUCTION FROM IMPAIRED REAL-WORLD DATA
(2025-02) Selvaraju, Pratheba
3D reconstruction from real-world data is essential in applications like augmented reality, robotics, medical imaging, and autonomous navigation. However, this data is often noisy, incomplete, occluded, or corrupted. Despite these imperfections, utilizing this data is necessary to develop reconstruction methods that can be applied in real-world scenarios. Each application comes with unique requirements and constraints, making it important to select representations tailored to the specific characteristics. Recognizing that our world is primarily composed of two types of objects, static (rigid) and dynamic (non-rigid) body structures, this thesis focuses on reconstruction tasks by exploring representations best suited to each type, ensuring adaptability to applications with similar characteristics, rather than reinventing wheels for each case. We focus on static structure reconstruction in fabrication industries that produce real-world products. It often deals with non-malleable materials which has zero-gaussian curvature property. To address reconstruction with this property constraint, we introduce Developability Approximation for Neural Implicits through Rank Minimization, a neural network model that represents surfaces as piecewise zero-gaussian curvature patches. The model encodes data implicitly, offering an advantage over prior explicit methods that struggle with high tessellation and shape fidelity. Applying this method to large-scale urban planning requires understanding building structures which is made of several different components with different non-malleable materials. Thus, automatically identifying these components becomes essential. To this end, we created a large-scale dataset of 2,000 diverse building exteriors (e.g., residential, commercial, stadium) named BuildingNet. Using this dataset, we developed a Graph Neural Network (GNN) model to automatically label building components. Next, we explore dynamic object reconstruction, focusing on human faces, by introducing OFER: Occluded Face Expression Reconstruction. OFER reconstructs expressive human faces from occluded images. Occlusion introduce new sources of ambiguity in hidden regions, requiring multi-hypotheses solution. Toward this, OFER employs a parametric face model and trains hybrid UNet-Attention diffusion models to generate diverse expression coefficients. This representation ensures smooth, plausible reconstructions with integrity to the visible parts and ease of animatability through simple parameter adjustments. In facial animation, real-time performance is crucial for applications like gaming and augmented reality, which require computational efficiency while preserving high quality. Traditional UNet-based diffusion models often suffer from slower temporal coherence and long range sequence, while attention computation results in computational overhead and slower inference time. To tackle this, we explore efficient computational representations and introduce FORA: Fast-Forward Caching for Diffusion Transformer Acceleration. FORA employs a caching mechanism that reuses intermediate outputs, thereby minimizing computational overhead without requiring model retraining, enabling faster processing with minimal trade-offs in quality.
EFFECTS OF OXYGEN ON THE SURFACE TENSION OF AN AEROSPACE ALLOY
(2025-02) SanSoucie, Michael
Containerless processing allows deep undercooling of liquid metals, alloys, glasses, and ceramics. Thermophysical properties, such as viscosity, surface tension, density, and specific heat, may be measured using non-contact methods. While electrostatic levitation is done under high vacuum, surface tension and viscosity results may be affected by oxide films or dissolved oxygen. Oxygen control is desirable for investigation of novel high-temp materials, e.g. aerospace alloys, where precise knowledge of transport phenomena or oxygen diffusion defines the potential application limits. The surface tension of molten metals is often affected by even a small amount of adsorption of surface active elements such as oxygen. Models were used to predict the surface tension of Inconel 718, and measurements were taken to compare with the models.
Ambient Air Pollution During Spermatogenesis and Impact on Semen Quality and Infertility Treatment Outcomes
(2025-02) Russo, Lindsey
While research suggests a detrimental association between criteria air pollutants and semen quality in countries with high levels of air pollution, a data gap exists in exploring this association in U.S. cities which tend to be characterized by low to moderate levels of air pollution. Additionally, few studies have characterized the impact of exposure to ambient air pollution during spermatogenesis with downstream couple-level infertility treatment outcomes. Both particulate and gaseous pollutants can cause an inflammatory response in the lungs, leading to an increase in inflammatory markers and an overproduction of reactive oxygen species which can damage the blood-testis barrier, leading to impaired spermatogenesis. Exposure to ambient air pollution may also result in DNA strand breaks and epigenetic changes which may lead to poor infertility treatment outcomes. Leveraging data from the Folic Acid and Zinc Supplementation Trial (FAZST) (2013-2018) for male partners of couples seeking infertility treatment in the Salt Lake City, Utah region, this dissertation aimed to answer critical questions surrounding the impact of ambient air pollution on semen quality (n=1,965) and couple-level infertility treatment outcomes in an IUI cohort (n=505 couples and 1,223 cycles) and an IVF cohort (n=221 couples and 280 cycles). This dissertation sheds new insight on the potential role of male preconception exposure to ambient air pollution during spermatogenesis on semen quality and infertility treatment outcomes in Salt Lake City, which may improve family planning.

Browse

Recent Submissions