Rice University Electronic Theses and Dissertations

Permanent URI for this collection

https://hdl.handle.net/1911/13110

Rice University makes all graduate theses and dissertations (1916-present) available online at no cost to end users. Occasionally a thesis or dissertation may be be missing from the repository. If you are unable to find a specific dissertation, please let us know and we will attempt to make it available through the repository, provided that the author has not elected for it to be embargoed.

News

Visit the web site for Rice University's Office of Graduate and Postdoctoral Studies for more information about Rice graduate student requirements for theses and dissertations.

Browse

Now showing 1 - 20 of 14122

Cluster-based methods for strongly-correlated systems
(2023-12-01) Papastathopoulos-Katsaros, Athanasios; Scuseria, Gustavo
We introduce three novel cluster-based methods to describe the ground states of strongly-correlated systems such as iron-sulfur clusters, conjugated hydrocarbons, and superconductors. These methods utilize a spatial tiling of sites as their core principle. The first approach employs unrestricted cluster mean-field theory (UcMF), with clusters of Sz eigenstates. Correlations between tiles are accounted for using perturbation theory (cPT2) and coupled-cluster (cCCSD). The second approach, generalized cluster mean-field theory (GcMF), allows Sz to break in each cluster, partially including missing intercluster correlations. A projection scheme, Sz GcMF, restores global Sz symmetry for further improvement. The third approach, a non-orthogonal configuration interaction-based theory (LC-cMF) which is still in development, is based on linear combinations of different system tilings. Various criteria, such as translational symmetry and spatial proximity, guide the selection of these tilings. Benchmark calculations on one- and two-dimensional spin models show the promise of these methods. GcMF and Sz GcMF provide a qualitative improvement over UcMF, while cPT2, cCCSD, and LC-cMF can quantitatively capture inter-cluster interactions in some systems. Overall, cluster-based methods offer valuable tools for investigating strongly-correlated spin systems with potential for further advancement.
Fluid-Mediated Slip of Shallow Subduction Thrust Faults
(2023-12-01) Belzer, Ben; French, Melodie
The shallow segment of subduction plate boundary faults (<10 km) hosts diverse slip modes including low-frequency earthquakes and slow slip. These slip behaviors are thought to be controlled by the mechanical properties of rocks and sediment present along the subduction interface and their thermal and hydrologic environment. The roles of fluids have particularly gained a lot of attention over the past 20 years as numerous geophysical studies have shown a correlation between slow earthquakes and evidence of high pore fluid pressure along subduction thrusts. However, the mechanics of slip in shallow, fluid-rich subduction shear zones are not well constrained. We present three studies addressing different roles that fluids play in affecting the strength and slip behavior of shallow subduction thrust faults. Using deformation experiments, we first quantify the frictional constitutive behavior of chlorite, which is a ubiquitous and important hydrous mineral in subduction zones. Rate-stepping shear experiments were performed under shallow hydrothermal conditions with varying temperature, pore fluid pressure, and slip rates from 10-9 to 10-5 m/s, representing slow slip speeds and the faster velocities previously used to study chlorite deformation. Our results show that chlorite strengthens with increasing temperature and transitions from stable to unstable frictional behavior with decreasing slip rate, indicating that slow slip can nucleate in chlorite-rich fault zones. Based on microstructural observations and micromechanical analyses, we interpret this transition is controlled by a competition between rate-strengthening mineral deformation at grain contacts, which promotes stable sliding, and the time-dependent strength of water films between grains, which promotes rate-weakening (unstable) behavior. Along shallow subduction faults, pore fluid pressure can become elevated due to disequilibrium compaction of sediments or fluid release due to dehydration reactions like the smectite to illite transition. In our second study, we investigate whether the source of fluid overpressure places a role on the mechanics of shallow subduction fault slip. We conducted hydrothermal shear experiments on chlorite gouge and natural cataclasite from the Rodeo Cove thrust (RCT) in California along stress paths that simulate disequilibrium compaction and dehydration reactions. Our results show that the effects of pore fluid pressurization on fault strength can generally be described using the critical state soil mechanics (CSSM) framework. The effects of path are more pronounced and persist to greater displacements in chlorite fault rock than in cataclasite, which we attribute to differences in microstructure. Because of the effects on microstructure, effective stress path also imparts a much more significant control on the stability of chlorite-rich faults, which is not predicted by CSSM. Our second study thus provides constraints on how both the source of pore fluids and fault rock composition should be considered in models of shallow subduction faulting. In our third study, we characterize the effects that fluid-mediated reactions have on the rheology of oceanic crust in a shallow subduction thrust environment (8-10 km). To do so, we present a geochemical and microstructural study of metabasaltic fault rock from the Rodeo Cove thrust zone. At the RCT, deformation is distributed along a dense network of reddish and greenish foliated cataclasites, which surround blocks of altered basalt that contain abundant calcite veins and cement. Our study indicates that faulting and cataclasis of the altered basalt followed by seawater-mediated K-metasomatism promoted extensive mineralization of celadonite within the RCT, which influenced its overall strength and deformation style. Transitions from spilitization to K-metasomatism of oceanic crust may therefore play an important role in the mechanics of shallow subduction fault slip, especially along active sediment-poor margins.
Environmentally friendly microcapsule based self-healing materials using castor oil in lieu of chlorobenzene
(2023-12-01) Nelson, Georgia; Getachew, Bezawit
Self-healing materials are a class of smart materials which can recover from physical or chemical damage autonomously. One type of self-healing materials involves the use of microcapsules which are embedded within a substrate. The microcapsules contain a reactive healing agent that is exposed to the outside environment when the substrate is damaged. Upon exposure to the environment, the healant inside the microcapsules reacts quickly and solidifies in place, repairing the damage at the interface. While microcapsule-embedded self-healing materials hold promise for improving the resilience of many systems including infrastructure and water treatment systems, current development has been limited to proof-of-concept work that uses carcinogenic chemicals such as chlorobenzene. Solvents such as chlorobenzene are used to dissolve the healing agent during the encapsulation process. Replacing the solvent used during microcapsule fabrication with more environmentally friendly solvents, such as castor-oil could improve the sustainability and practical application of self-healing materials. Practical applications include embedding these microcapsules to the surface of membranes for water treatment or applying them as a protective self-healing coating in pipes. In this work, microcapsules were fabricated with both chlorobenzene and castor oil, and the size distribution and the microcapsule core content of each solvent were compared using Fiji image analysis and Thermographic analysis (TGA) respectively. Using castor oil as a solvent resulted in microcapsules with an average diameter of 153um, in contrast to the chlorobenzene microcapsules that had an average diameter of 107um. Despite the larger average diameter of the castor oil microcapsules, TGA analysis suggested that microcapsules fabricated with chlorobenzene contained a higher percentage of healing agent. The successful fabrication of environmentally friendly castor oil-based self-healing microcapsules indicates that they can be used as an alternative to their toxic chlorobenzene counterparts, but future research is needed to determine if the lower percentage of healing agent will impact their performance.
Embargo
Accurate and Efficient Computational Approaches for Long-read Alignment and Genome Phasing of Human Genomes
(2023-12-01) Fu, Yilei; Treangen, Todd J
The arrival of long-read sequencing technologies has enabled analysis of human genomes at unprecedented resolution. Long-read technologies have facilitated telomere-to-telomere assembly of the human genome and shed light on difficult to resolve structural variations, single nucleotide variations and epigenetic modifications, which all play a critical role in disease etiology and individual genetic diversity. Despite the technological advancement, novel computational methods are still needed to fully leverage long reads. In this dissertation, I tackle three key computational questions by leveraging long-read sequences of human genomes: 1. I improve on the efficiency and precision of long-read alignment, 2. I develop a novel variant phasing techniques based on methylation signal, and 3. I provide a novel method for clinical analysis specific to cancer samples and tumor purity estimation. These accomplishments are represented by three software tools I have developed: Vulcan, MethPhaser and MethPhaser-Cancer, respectively. Vulcan is a read mapping pipeline that uses two distinct gap penalty modes, which is referred to as dual-mode alignment. Read aligners before Vulcan only use one type of scoring scheme during the pairwise alignment stage, which can struggle due to the variable diversity across the human genome. With Vulcan’s dual-mode alignment algorithm, the read-to-reference mapping quality and efficiency for Oxford Nanopore Technology (ONT) long-reads are improved for both simulated and real datasets. Notably, we also show Vulcan provides improvement in structural variation detection. Vulcan increased the SV detection F1 score of 30X human ONT reads from 82.66% (minimap2) to 84.94%. MethPhaser is the first method that utilizes methylation, an epigenetic marker, from Oxford Nanopore Technologies to extend SNV-based phasing. Long-read human genomic variant phasing is limited by read length and stretches of homozygosity along the genome. The key innovation of MethPhaser is the utilization of the haplotype-specific long-read methylation signals. In benchmarking against human samples, MethPhaser nearly triples the phase length N50 while incurring a minimal increase in switch error from 0.06% to 0.07% using ONT R10 reads at 60X coverage. As an extension method to existing long-read SNV-based phasing workflows, MethPhaser offers substantial enhancements with a negligible rise in switch error rates. Building upon MethPhaser, I have also innovated an algorithmic extension named MethPhaser-Cancer that uses methylation signals for the assessment of tumor purity and for categorizing reads. The tumor purity estimation is an important step in clinical treatment that is related to tailoring patient-specific therapeutic strategies and in the broader context of personalized medicine. MethPhaser-Cancer adeptly identifies hypomethylated areas within human tumor samples and utilizes the k-means algorithm to sort the reads into two distinct groups. This represents a pioneering approach in the long-read sequencing field to consider whole-genome methylation profiles in simulated clinical samples, capable of automatically estimating the tumor purity and distinguishing long-reads within specific regions between two samples. To conclude, this dissertation represents a set of novel and efficient approaches that enhances the long-read human genomic analysis. The real-life usage of Vulcan, MethPhaser and MethPhaser-Cancer includes long-read alignment, human genome variant phasing and tumor purity estimation.
Gallium Oxide as Photonic Integrated Platforms in UV–Visible Spectrum
(2023-12-01) Zhou, Jingan; Zhao, Yuji
We report the gallium oxide (Ga2O3) as a photonic integrated platform and its nonlinear optical effects in the UV–visible spectra. The β-Ga2O3 as optical waveguides on sapphire substrates grown by metal-organic chemical vapor deposition (MOCVD). For linear properties, their propagation losses in the visible spectrum were comprehensively studied via experiments and simulations of the finite difference method by varying the dimensions of the waveguides. The fabrication process of the waveguides was shown and their propagation losses were measured by collecting the top scattered optical power along the propagation direction using a charge-coupled device (CCD) camera. The minimum measured loss was 3.7 dB/mm at a wavelength of 800 nm depending on the dimensions of the waveguides. For nonlinear properties, by focusing a pulsed laser beam onto a polished ε-Ga2O3 thin film, we collected the generated second harmonic photons with an ultra-sensitive femtowatt photodetector, obtaining effective second-order nonlinear optical susceptibility from visible to UV spectra. Two different measurement systems collecting reflected and transmitted photons were applied separately to make the result more convincing. The wavelength dependence from 790 nm to 900 nm and polarization dependence from TM mode to TE mode were measured as well. In addition, we compared the losses of waveguides with different widths, heights, wavelengths, and polarizations. It is revealed that the total propagation loss is mainly contributed by the bulk and sidewall scattering. Combined with theoretical simulations, various loss mechanisms from two-photon absorption, sidewall scattering, top surface scattering, and bulk scattering, were discussed for β-Ga2O3 waveguides, and their contributions to the total optical loss were estimated. After all these mechanism analyses on absorption and scattering, a large improvement and optimization were applied to the material quality and etching recipe, which improves the top surface and sidewall roughness respectively. This work provides valuable information for the fabrication of optical devices.
Embargo
The Applications of Multidomain Peptide Hydrogels as Scaffolds for 3D Cell Culture
(2023-12-01) Leyva Aranda, Claudia Viridiana; Hartgerink, Jeffrey
3D cell culture constructs have emerged as indispensable tools in the field of cell biology and tissue engineering. These innovative scaffold-based platforms offer a more physiologically relevant environment for studying cell behavior, tissue development, disease modeling, and drug testing. In this work, we explore the importance of 3D cell culture scaffold-based constructs, the common biomaterials used for their creation, and prospective research routes that promise to further advance this field. In addition to that, we evaluate the use of multidomain peptides (MDPs) as platforms for 3D cell culture, and their applications for stem cell preservation, T cell-based therapies, and growth factor and nucleobase-modified MDP scaffolds. By providing a soft three-dimensional substrate, neutral MDP hydrogels proved to be successful at inducing stem cell quiescence and at maintaining stem cell phenotype and immunomodulatory properties. Additionally, MDP-based hydrogels demonstrated promising properties to be used as a 3D platform for T cell culture, enabling T cell in vitro activation, expansion, and maintenance of antigen specificity. Because of their chemical tunability, ongoing work explores the development of more complex chemistries by introducing growth factor peptide mimics, as well nucleobases for the generation of higher order supramolecular structures within the MDP hydrogels. This thesis describes the use of multidomain peptides as chemically tunable, biocompatible, versatile biomaterials for the generation of 3D culture scaffolds, with potential applications in the fields of cell therapy, drug delivery, stem cell preservation, and translational medicine.
PIE: Perceptual Image Enhancement via Local Manifold Sampling on Pretrained Diffusion Models
(2023-12-01) Mayer, Paul Michael; Baraniuk, Richard G
PIE: Perceptual Image Enhancement via Local Manifold Sampling on Pretrained Diffusion Models
Embargo
Localization for Autonomous Underwater Vehicles inside Harsh and GPS-Denied Environments
(2023-12-01) Ben Moallem, Issam; Ghorbel, Fathi H.
The localization of Autonomous Underwater Vehicles (AUVs) deployed for integrity inspection of liquid storage facilities, to prevent failure of the process, is a critical and challenging task. This is primarily due to the harsh and GPS-denied work environment, as well as to the high degree of accuracy required by such confined-space activities to ensure accurate motion control and navigation, and generate rigorous inspection data associated with their true physical locations. Conventionally, an AUV performing general surveying operations is equipped with an Inertial Navigation System (INS) and/or a Doppler Velocity Log (DVL) for real-time state estimation and positioning, with respect to some inertial reference frame, while navigating its local environment. Due to inherently accumulating measurement errors over time, an INS/DVL device usually relies on the GPS for periodic recalibrations, which requires surfacing of the submersible robot. In deep waters, this strategy is energy resource and time inefficient, hence costly. Furthermore, for covered and underwater environments such as storage tanks, GPS signals are not even accessible at the liquid surface. Moreover, neither the INS/DVL-GPS system nor the traditional baseline acoustic positioning systems, based on trilateration techniques, provide a satisfactory solution accuracy as demanded by precision tasks such as pinpointing defects in steel storage and underwater structures. To overcome the shortcomings of the conventional underwater localization techniques, and achieve high-fidelity mapping between inspection data and real physical locations, we propose in this thesis a novel, accurate, and robust method to solve the robot localization problem inside confined, harsh, and GPS-denied environments. This method uses affordable sensors and fast algorithms to develop new techniques that provide accurate positions of the mobile agent. Given the geometry of the asset under investigation, a map representation for the robot's workspace is constructed based on range measurements over its boundaries. Then, the robot's position and orientation are accurately estimated relative to some defined reference landmarks (features) extracted from the map. In the event that the robot fails to recognize any landmark, a point-set registration technique is employed. In this case, the robot recursively matches map observations while in motion, which yields a relative position with respect to the most recently determined landmark-based position. The devised localization method will unleash fully autonomous robotic operations in confined, harsh, and GPS-denied environments. It will also facilitate Risk-Based Inspection (RBI) by employing predictive capabilities to optimize maintenance planning. This method can be applied in the oil and gas industry for inspecting liquid storage assets such as Aboveground Storage Tanks (ASTs) and Floating Production Storage and Offloading units (FPSOs).
Response evaluation of nonlinear dynamic systems endowed with fractional-order derivatives under evolutionary stochastic excitation
(2023-12-01) Zhang, Wei; Spanos, Pol D
Natural hazards and excitations often exhibit stochastic characteristics, such as winds, earthquakes, and ocean waves. These load scenarios deserve extensive attention and investigation that account for the uncertain characteristics; otherwise, it may lead to unpredictable damage. In addition, the viscoelasticity phenomenon is prevalent in a variety of engineering materials. When resisted by dynamic loads, the viscoelastic materials exhibit viscous, smoothly varying, and time-dependent deformation, associated with energy dissipation. It can be essential to understand the viscoelastic behavior and its potential influence on structural response, particularly when structural design and response analysis are considered. In this regard, during recent decades it has been shown that the implementation of fractional derivatives allows a more descent description of the viscoelasticity phenomenon. Therefore, in this thesis, the challenge of viscoelastic oscillators subjected to evolutionary stochastic loads is addressed. More specifically, the fractional-order derivative element is introduced to effectively represent the viscoelastic nature of the materials. Further, several analytical and numerical methods are examined. To start, the statistical linearization method is extended for oscillators with fractional derivative elements, where a quite versatile discretization approach is introduced that makes the proposed method applicable to any kind of nonlinearity. Next, the stochastic averaging method is applied on fractional oscillators reducing the dimensionality of the system, and thus accelerates the following computation. Thirdly, the wavelets-galerkin method is adopted to address the evolutionary response statistics of either linear or nonlinear systems. Note that, by accounting unnoticeable overlapping of the basis wavelets functions, the method predicts accurately responses of relatively flexible and/or lightly damped systems. Results in juxtaposition with the pertinent Monte Carlo simulation data demonstrate the reliability and accuracy of the proposed methods of analysis.
Embargo
Catalyst design for water treatment using ab initio simulation
(2023-11-30) Chen, Yu; Senftle, Thomas P
Shortage of clean water sources due to climate change, development of industrialization, and population growth is a concerning problem worldwide. Heterogenous catalysis is a promising strategy to reduce the concentration of undesirable substances during water treatment. In this thesis, I apply ab initio simulation to identify key material properties and fundamental reaction mechanisms that dictate catalyst performance for the treatment of two important water contaminants: nitrate and per-fluoroalkyl substances (PFAS). This insight in turn informs design strategies for designing better catalysts for these applications. Perfluorooctanoic acid (PFOA) is one of the most prevalent PFAS contaminants in surface and ground water. Working with experimentalist collaborators, we reported that hexagonal boron nitride (hBN) is a promising photocatalyst for PFOA degradation under UVC illumination, with an activity ~2x higher than TiO2. In my thesis, I applied density functional theory (DFT) in a grand canonical (GC) formalism (Bhati and Chen et al., J. Phys. Chem. C, 2020, 124, 49, 26625–26639) to determine the photo-catalytic mechanism responsible for PFOA degradation on hBN. (Chen et al. Environ. Sci. Technol., 2022, 56, 12, 8942–8952) I confirmed the favorability of the proposed photo-oxidation step of PFOA on the hBN surface: CnF2n+1COO− + h+ → CnF2n+1ꞏ + CO2. Furthermore, by investigating the electronic properties of hBN, I found that NB substitutional point defect introduces mid-gap states that enable the UVC light absorption and enhance charge carrier separation. Therefore, introducing more NB defects is a promising strategy to enhance the photocatalytic degradation performance of hBN. My work also helped to determine the role of surface hydrophobicity in promoting PFOA degradation, which is attributed both to stronger adsorption of the hydrophobic fluorinated tale of PFOA and the exclusion of water molecules that can scavenge photo-excited holes. (Wang and Chen et al., submitted) Thus, increasing surface hydrophobicity is another strategy for enhancing catalyst performance during PFOA degradation. Using this insight, we are now developing covalent organic framework (COF) catalysts with tunable functionality to tailor hydrophobic and electrostatic interactions, thus maximizing PFOA adsorption. Besides PFOA, nitrate is another pervasive surface and groundwater contaminant found worldwide. Nitrate anions are highly soluble and mobile, and can cause harmful health effects in humans, including diseases such as blue baby syndrome, cancer, etc. Investigating the reaction network of electrocatalytic nitrate reduction, we found Cu and Pd catalysts can play a synergistic role in nitrate removal. (Lim and Chen et al., ACS Catal. 2023, 13, 1, 87–98) Using DFT, we discerned how the electronic properties of the metal catalyst affect the nitrate reduction reaction mechanism, steering the product selectivity to either N2 or NH3. (Chen and Senftle, submitted) We propose that metals like Pd, with less-occupied and more-delocalized d orbital exhibit higher N2 selectivity due to adsorbate-adsorbate interactions that promote N–N bond formation over N–H bond formation. This insight sets the theoretical basis for the design of better Pd/Cu bimetallic catalysts for the selective disposal of nitrate from water.
Workload-Based Task Selection for Automation
(2023-12-01) Taffese, Tewodros Belete; Byrne, Michael
Workload is a construct used to understand and predict human-automation performance. For an effective human-autonomy collaboration, it is important to maintain workload at an optimum level. Excessive workload impedes performance, while low workload might lead to boredom, lack of vigilance, and lack of situation awareness. While there is existing research on workload and its impact on automation, there seems to be lack of guidelines regarding which components of a task should be subject to automation. The purpose of this dissertation was to address this gap by providing a framework for task selection in automation. Three experiment was conducted to investigate the efficacy of workload-based task selection for automation. Participants performed the Theatre Defense Task, which is a simulated military operation task. The first experiment compared subtask workloads while participants performed the task without automation, identifying high and low workload subtasks. Workload differences between subtasks were measured using subjective workload ratings and performance-based metrics. In the second experiment, these subtasks were automated and compared in different conditions. The third experiment introduced a design intervention to address observed strategies from the second experiment and to optimize the automation. Participants exhibited higher performance scores in high workload subtask automation compared to low workload subtask automation conditions, especially when the task difficulty was high. The workload rating of different subtasks appeared to be influenced by the automation. The results showed that automation alters human-system interactions. Workload-based task selection demonstrated performance improvement and reduced workload. More importantly, automating high workload subtasks significantly improved and reduced overall workload, while automating low workload components of a task had no impact on performance. This study underscores the significance of exploring the impact of subtask automation on both the overall task and individual subtask performances.
Embargo
How Do Skills Shape Trainee Perceptions of Effort and Interest in Training? A Lifespan Development Perspective
(2023-11-28) Davenport, Meghan; Beier, Margaret E
Due to a confluence of an aging workforce and technology change, individuals must engage in skill learning throughout the lifespan, often outside of the bounds of an employer. Based on lifespan development theories, people choose goals in response to the growth, maintenance, and decline inherent in aging (P. B. Baltes & Baltes, 1990; Carstensen et al., 1999). Therefore, a person’s existing skills’ alignment to training should impact how they judge the effortfulness and desirability of engaging in the training (Kanfer & Ackerman, 2004). Age-related changes in motivational processes should impact how a person’s skill profile influences their effort perceptions and interest, which may differ between types of training (people-focused vs. things-focused skill training). I conducted a pilot study (N = 70, Mage = 40.1) and a focal study (N = 493, Mage = 39.1) of job-seeking adults ages 18-60. Participants self-reported their skill level across twenty skills, then reported their perceptions of how interesting and effortful four trainings seemed based on descriptions. A separate sample of 10 subject matter experts’ ratings of each of twenty skill’s relevance to each of the trainings were used to create four separate keys (one for each training). Each key was used to score participants’ skill profiles’ alignment to each training. I used multilevel modeling to analyze the data, which included responses related to each of the four trainings clustered within-person. A participant’s level of skill alignment was significantly negatively related to perceptions of effort required to learn from a training (γ = -0.171, p < .001), and positively related to their level of interest in the training (γ = 0.349, p < .001). Age interacted significantly with skill alignment to impact effort perceptions, such that older participants were more sensitive to their skill alignment to a training when judging its effortfulness (γ = -0.133, p < .05). Surprisingly, age did not significantly interact with skill alignment to affect interest in training and training type did not change the relationship. By integrating the job search, lifespan development, and motivation literatures, this work can inform potential interventions to improve career outcomes for jobseekers.
A General Method for Efficient Distributed Training and Federated Learning in Synchronous and Asynchronous Scenarios
(2023-11-30) Dun, Chen; Kyrillidis, Anastasios
In the past decades of development of machine learning systems, there is an eternal conflict: model performance versus model scale versus computation resources. The never ended desire to improve model performance significantly increases the size of machine learning model, the size of training dataset and the training time, while the available computation resources are generally limited due to limited memory size and computation power of computation devices and limited data usage (due to data storage or user privacy). In general, there are two main research attempts to solve such eternal conflict. The first attempt focuses on decreasing the needed computation resources. Accordingly, synchronous distributed training systems (such as data parallelism and model parallelism) and asynchronous distributed training system have been widely studied. Further, federated learning system has been researched to address the additional restriction of data usage due to data privacy or data storage. The second attempt to solve the eternal conflict instead focuses on improving model performance with limited model scale with Mixture of Expert (MoE) system. As we find there is hidden shared essence between these two directions, we aim to create a general methodology that can solve the problems met in both directions mentioned above. We propose a novel methodology that partitions, randomly or by a controlled method, the large neural network model into smaller subnetworks, each of which is distributed to local workers, trained independently and synchronized periodically. For the first direction, we demonstrate, with theoretical guarantee and empirical experiments, that such methodology can be applied in both synchronous and asynchronous systems, in different model architectures, and in both distributed training and federated learning, in most cases significantly reducing communication, memory and computation cost. For the second direction, we demonstrate that such methodology can significantly improve the model performance in MoE system without increasing model scale, by guiding the training of specialized experts. We also demonstrate our methodology can be applied to MoE systems on both traditional deep learning model and recent Large Language Model (LLM).
Embargo
Predicting Post-surgery Walking Function of Pelvis Sarcoma Patients using Personalized Neuromusculoskeletal Models
(2023-12-01) Li, Geng; Fregly, Benjamin J
Surgical treatments for pelvic sarcomas have been historically challenging due to the complex anatomy of the pelvic region and large heterogeneity in tumor conditions across patients. Orthopedic oncologists and implant companies must therefore make many decisions in the design of the surgical treatments and the prostheses to address these challenges. However, most of these decisions were still made based on subjective judgments, which raised the question of whether a more objective decision-making approach can be used to further improve the patients’ walking function post-surgery. Predictive simulations of human movement using personalized neuromusculoskeletal models is one approach to design interventions. Surgeons and biomedical engineers can use this approach to generate exhaustive set of treatment designs and objectively evaluate all of them before deciding on the best option. Before the great potential of clinical translation of this technology in treating pelvic sarcomas can be fully realized, it is important to make sure the predictive simulations can reproduce reality at first. This dissertation seeks to use the three technical chapters to methodically address the technical challenges associated with generating realistic model-based predictive simulations. In the first technical chapter (Chapter 2), a personalized musculoskeletal model was created to simulate the gait movement of a pelvic sarcoma patient. The model met our need for being capable of independently simulating movement actuated by both their trunk muscles and lower extremity muscles, since most existing models were usually specialized for musculoskeletal system in only one region of the body e.g. trunk only or leg only. In addition to detailing model development, this technical chapter also addressed the practical problem where insufficient electromyography channels were available to record muscle activities of both trunk and leg muscles. The proposed computational approach would use muscle synergies extracted from the measured activity of leg muscles to provide realistic estimates of trunk muscle activities. In the second technical chapter (Chapter 3), experimental gait data of a pelvic sarcoma patient pre- and post-surgery were collected, analyzed, and compared to provide an understanding on how neural controls of the lower extremity muscles changed after the surgery. Three hypotheses about how partial information of pre-surgery neural controls could be used in predict post-surgery muscle activities (1. Fixed SynVec, pre-surgery synergy vectors were retained, 2. Fixed SynCmd, pre-surgery synergy commands were retained, and 3. Shifted SynCmd, pre-surgery synergy commands were retained but allowed to shift in time) were evaluated. The Fixed SynCmd and Shifted SynCmd hypotheses accurately reconstructed post-surgery muscle activities. In the third technical chapter (Chapter 4), optimizations based on four different assumptions about optimality principles for predicting post-surgery walking were formulated and evaluated. The first proposed minimization of changes in muscle synergies from pre-surgery models. The second proposed minimization of deviations in synergy vector weights from set values. The third proposed minimization of changes in muscle activations from pre-surgery activations, and the fourth proposed minimization of muscle activations during post-surgery walking. The optimization based on minimization of changes in muscle synergies most accurately predicted the muscle activations during post-surgery walking. The research reported in this dissertation establishes the foundation for future works in model-based prediction of post-surgery gait of pelvic sarcoma patients. As more experimental and clinical data are collected, more analyses can be performed and more predictive simulations can be generated using the methods develop in this dissertation, to shed more lights on the appropriate methods to use for predicting post-surgery walking.
Embargo
Resilience Planning for Water Distribution Systems
(2023-11-30) Zhou, Xiangnan; Duenas-Osorio, Leonardo
Urban water distribution systems are lifeline infrastructure, playing a crucial role on community resilience, safety, and governance. They are vulnerable to various threats, including extreme hazard events, inevitable infrastructure deterioration, disruptions from upstream dependencies, and climate change. Facing these evolving challenges, it is urgent for water utilities to incorporate resilience management into their daily operations and long-term strategic planning. In the past two decades, existing research on the resilience management of water distribution systems primarily focuses on resilience assessment. Besides quantifying resilience, an important question for water utilities is what proactive measures can be taken before disruptions to improve system resilience. This dissertation aims to develop algorithms and tools to guide water utilities’ long-term resilience planning. First, we identify and establish useful performance measures for the resilience planning of water utilities. Second, we develop novel approaches for guaranteed deterministic and probabilistic vulnerability analysis of water distribution to inform mitigation strategy development. For deterministic vulnerability diagnosis, we develop a guaranteed N-k contingency analysis method, which builds on a novel search algorithm that integrates enumeration and a guaranteed-error sampling scheme. For probabilistic vulnerability quantification, we establish a probabilistic functionality analysis scheme that incorporates the guaranteed-error sampling scheme for principled estimations. Third, we introduce advanced measures and tools to facilitate heuristic and optimal mitigation strategy development. Regarding decision modeling, we suggest using a superquantile measure as a probabilistic decision criterion to account for risk averseness and deep uncertainty. We propose network augmentation as an alternative mitigation strategy that goes beyond traditional component hardening. For optimal mitigation, we introduce the Cross Entropy method, a promising optimization analysis tool for combinatorial stochastic optimization problems. Finally, we develop generic system modeling and policy search methodologies for water distribution reconfiguration, exploring opportunities to alter the layout of water distribution systems. Specifically, we develop a meso-scale water distribution system representation model that approximates large-scale water systems using reduced backbone networks to enable policy search. We establish a multi-objective optimization model that explores trade-offs across alternative distributed water system configurations. We demonstrate and test our approaches and tools on hypothetical and practical water systems. We find that the internal vulnerabilities within a water distribution system are determined by the complex interactions among its network topology, physical characteristics, and demand patterns. There are usually a few critical components whose failure can cause significant performance losses to water distribution systems, while the rest of the system components have a uniformly small impact on system functioning. Informed hardening and network augmentation strategies can yield efficient vulnerability reduction, improving distribution resilience. Additionally, reconfiguring centralized water systems to distributed alternatives presents an opportunity to enhance supply resilience. This dissertation aims to provide water utilities with the knowledge and tools they need to navigate alternative resilience-improving strategies, ranging from component hardening to network augmentation to system reconfiguration. The algorithms and tools we develop can be implemented on resilience-oriented planning platforms, such as the Interdependent Networked Community Resilience Modeling Environment (IN-CORE) and the Computational Modeling and Simulation Center (SimCenter) to guide practical resilience planning for water distribution systems.
Boron Nitride Nanomaterials: Dispersions in Surfactants and Encapsulation of Photoluminescent Metal Complexes
(2023-12-01) Martinez Jimenez, Cecilia; Marti-Arbona, Angel
Boron nitride nanomaterials consist of a hexagonal lattice of sp2 hybridized alternating boron and nitrogen atoms. They can exist in one-dimensional nanotube form and as two-dimensional nanosheets. Their high thermal stability, and optical transparency in the visible range, among other properties, make them compelling materials for a wide range of applications. This thesis discusses the study of aqueous dispersions of hexagonal boron nitride with surfactants, encapsulating a Ru (II) photoluminescent metal complex inside boron nitride nanotubes, and dispersing carbon and boron nitride nanotubes in a photoluminescent surfactant. Chapter 1 introduces the nanomaterials discussed, while Chapter 2 reviews the different methods reported for the exfoliation and dispersion of hexagonal boron nitride. Chapter 3 describes the study of the exfoliation of hexagonal boron nitride using nine different surfactants and the assessment of the dispersion yield, exfoliation quality, and stability over time of the dispersions. Chapter 4 explores using boron nitride nanotubes to encapsulate a Ru (II) photoluminescent metal complex. We produced a series of samples with increasing loading of the metal complex and studied the changes in their photophysical properties. Chapter 5 examines using a photoluminescent surfactant derived from a Ru (II) metal complex to disperse either carbon or boron nitride nanotubes and perform spectroscopic studies to elucidate their interactions. Finally, Chapter 6 presents a conclusion for the projects discussed throughout the work.
Probing ultralight dark fields in cosmological and astrophysical systems
(2023-11-30) Zhang, Hong-Yi; Amin, Mustafa
Dark matter constitutes $26\%$ of the total energy in our universe, but its nature remains elusive. Among the assortment of viable dark matter candidates, particles and fields with masses lighter than $40 \mathrm{eV}$, called ultralight dark matter, stand out as particularly promising thanks to their feasible production mechanisms, consistency with current observations, and diverse and testable predictions. In light of ongoing and forthcoming experimental and observational efforts, it is important to advance the understanding of ultralight dark matter from theoretical and phenomenological perspectives: How does it interact with itself, ordinary matter, and gravity? What are some promising ways to detect it? In this thesis, we aim to explore the dynamics and interaction of ultralight dark matter and other astrophysically accessible hypothetical fields in a relatively model-independent way. Without making specific assumptions about their ultraviolet physics, we first demonstrate a systematic approach for constructing a classical effective field theory for both scalar and vector dark fields and discuss conditions for its validity. Then, we explore the interaction of ultralight dark fields, both gravitational and otherwise, within various contexts such as nontopological solitons, neutron stars, and gravitational waves.
Embargo
Modeling and quantitative analysis to understand evolution, prognosis, and drug delivery in complex diseases
(2023-11-27) Peláez Soní, María José; Cristini, Vittorio; Kono, Junichiro
Modeling and quantitative analysis stand as indispensable tools in medicine, offering invaluable insights, predictive capabilities, and actionable solutions to address a myriad of healthcare challenges. Their instrumental role in advancing our understanding of disease dynamics has paved the way for enhanced diagnosis, prognosis, and treatment capabilities. This dissertation showcases the pivotal role of modeling methods and quantitative analysis in medicine, presented through three distinct applications that span over advanced drug delivery systems, cancer prognosis and the evolution of chemoresistance. In the first part of this work, a comprehensive mathematical model of transdermal drug delivery via microneedle-based patches, integrated with a pharmacokinetics model, is introduced. Model-based simulations were conducted to pinpoint the key parameters governing systemic delivery, enabling the optimization of patch designs to improve drug pharmacokinetics. In the second part, survival analysis is employed to identify biomechanical and immune biomarkers, enabling the prospective prediction of tumor aggressiveness, invasiveness, treatment outcomes, and survival probability in breast cancer. Lastly, a novel hypothesis is presented, proposing that water exclusion zones within cells may act as insulation barriers, safeguarding the delicate quantum nature of specific biochemical reactions against environmental influences. This hypothesis gains additional support through a review regarding the role that interfacial water plays in several biological processes and a proof of concept example to illustrate the application of quantum mechanics models for understanding the evolution of chemoresistance. Through these multifaceted investigations, this dissertation underscores the vital role that modeling and quantitative analysis plays in investigating the complexities of diseases, promising new horizons in medicine.
Essays on Natural Gas and Electricity Markets
(2023-12-01) Min, Luke; Hartley, Peter; Medlock, Ken
This dissertation explores the dynamic and complex nature of energy markets, focusing on the impact of deregulation and market transitions in various geopolitical contexts. The first chapter examines the impact of partial deregulation on South Korea's electricity and natural gas markets, with a specific focus on the wholesale electricity market. In the early 2000s, reforms transitioned the vertically integrated market, previously dominated by KEPCO, into a competitive landscape. Concurrently, the natural gas market was liberalized, breaking KOGAS' monopoly and permitting private firms to import LNG. Utilizing a structural model, this study estimates hidden market variables, including the fuel costs for independent firms, and conducts counterfactual simulations based on extensive data from the power generation and gas markets. The findings indicate that partial deregulation resulted in a slight increase in wholesale electricity prices, contradicting the expected decrease commonly associated with market liberalization. This deregulation significantly enhanced the market share and profits of independent gas power plants, often at the expense of KEPCO and other entities. While there was a marginal overall increase in total market surplus, mainly benefiting independent firms, this was countered by a reduction in consumer surplus and an uneven distribution of benefits. These outcomes underscore the complexities of partial deregulation, suggesting that while it can promote competition and diversify LNG sources, its advantages and impacts are unevenly distributed across different market players. The study emphasizes the importance of nuanced policy approaches in the deregulation of energy markets, especially in scenarios where changes are partial or constrained. The second chapter of this dissertation focuses on the repercussions of Russia's invasion of Ukraine on February 24, 2022, particularly its impact on the security of natural gas supply in Europe. The unfolding events of 2022 have put the continent on alert, preparing for a winter that could pose significant challenges in terms of high prices and uncertainty. Despite a less severe winter than anticipated, the situation remains complex and unresolved. The drastic reduction in Russian gas imports casts a shadow of uncertainty over the future supply of natural gas in Europe. Germany, as the European Union’s largest economy, serves as a pivotal example in understanding the broader dynamics of the European natural gas market and the looming challenges. The reliance of manufacturing on natural gas underscores the potential far-reaching impacts on the availability of gas and the overall economic performance of the EU. To analyze the potential scenarios for the natural gas market balances in Germany in the upcoming winter season and beyond, we developed three demand-oriented scenarios: 1) a cold winter in 2022-23, 2) a mild winter in 2022-23, and 3) an extreme case scenario. This chapter presents the critical findings and implications derived from these scenarios, offering insight into the evolving landscape of energy security and economic stability in Europe. The third chapter of this dissertation delves into the burgeoning influence of renewable energy generation in Texas and its significant impact on the electricity market. This segment of the study investigates the effects of increased renewable energy penetration on the pricing strategies of firms in the wholesale electricity market. Leveraging data from the day-ahead market within Texas' wholesale electricity framework, the research examines the shift in market power across varying hours of the day. It is discovered that during peak hours, market power diminishes considerably in the wake of the planned expansion of solar generation. This finding highlights the transformative role of renewable energy sources in reshaping traditional market dynamics and influencing competitive behavior in the energy sector.
Identification of a Novel Role of BMP Signaling in Enteric Neural Crest Colonization of the Zebrafish Enteric Nervous System
(2023-11-29) Moore, Joshua Alan; Uribe, Rosa
The vertebrate enteric nervous system (ENS) consists of a series of interconnected ganglia within the gastrointestinal (GI) tract, formed during development following the migration of enteric neural crest cells (ENCCs) into the primitive gut tube. Much work has been done to unravel the complex nature of extrinsic and intrinsic factors that regulate the processes that direct the migration, proliferation, and differentiation of the ENCCs. However, ENS development is a complex process, and we still have much to learn regarding the signaling factors that regulate ENCC development. Here in zebrafish, through transcriptomic, in situ transcript expression, immunohistochemical analysis, and chemical attenuation, I identified a time-dependent role for bone morphogenetic protein (BMP) in the maintenance of Phox2bb+ enteric progenitor numbers and/or time of differentiation of the progenitor pool. In support of our in silico transcriptomic analysis, I identified the expression of a novel ENS ligand-encoding transcript, bmp5, within the developmental regions of the ENCCs. Through the generation of a novel mutant bmp5wmr2, I identified a functional role for BMP5 in the proper colonization of the GI tract, whereby phox2bb+ enteric progenitor numbers were reduced. Altogether, this work has identified time-dependent roles for BMP signaling and identification of a novel extrinsic factor, BMP5, that is necessary for the proper formation of the vertebrate ENS.

Browse

Recent Submissions