12023 research outputs found
Sort by
Exploiting Speckle to Image Deeper in Scattering Media
Get PDFOptical methods for imaging and focusing are advantageous in many scenarios as optics can provide exquisite spatial resolution, has multiple sources of contrast, and does not impart ionizing radiation. However, optical scattering remains a fundamental challenge which limits the depth at which we can perform imaging with good spatial resolution. This challenge motivated our investigations into methods that could make use of the scattered light in order to extend the depth of imaging through or within scattering media. In particular, we focus on answering: (1) Can one 'unscramble' the scattered light in order to recover information about the otherwise hidden object?; and (2) Can we preferentially detect the more forward scattered photons in an efficient manner in order to allow deeper penetration with modest resolution? These two questions are explored in the first two projects of the thesis:
1. The development of an imaging system that detects the scattered light and exploits correlations within the scattering process to enable imaging through scattering media at diffraction-limited resolution.
2. The introduction of a novel method, termed Speckle-Resolved Optical Coherence Tomography, that sensitively and preferentially detects the more forward scattered photons in a coherent, speckle-resolved fashion to allow deeper imaging at moderate resolution.
Optical methods offer the benefit of visualizing samples that would otherwise appear transparent. Using light, one is able to visualize and measure the thickness of transparent films and coatings in a non-contact manner. The third project in my thesis focuses on using light to non-destructively visualize and characterize the evenness of the silicone oil layer that typically coats the inner surface of prefilled syringes. Characterizing the evenness of this silicone oil layer is important as it impacts the functionality of the prefilled syringe and may correlate with particle formation, which is undesirable as the number of particles in a syringe is regulated due to potential health concerns. </p
Searches for Nonresonant Higgs Boson Pair Production and Long-Lived Particles at the LHC and Machine-Learning Solutions for the High-Luminosity LHC Era
Get PDFThis thesis presents two physics analyses using 137 fb−1 proton-proton collision data collected by the CMS experiment at √s = 13 TeV, along with a series of machine-learning solutions to extend the physics program at the LHC and to address the computational challenges in the High-Luminosity LHC era. The first analysis searches for nonresonant Higgs boson pair production in final states with two photons and two bottom quarks, with no significant deviation from the background-only hypothesis observed. The observed (expected) upper limit on the product of the Higgs boson pair production cross section and branching fraction into bb̅γγ is 0.67 (0.45) fb, corresponding to 7.7 (5.2) times the Standard Model prediction. The modifier of the Higgs trilinear self-coupling is constrained within the range -3.3 < κλ < 8.5. The modifier for coupling between a pair of Higgs bosons and a pair of vector bosons, along with the 2-dimensional constraint of the modifiers of Higgs self-coupling and Yukawa coupling, are also reported. A graph-based algorithm to identify boosted H → bb̅ jets to improve future Higgs search is presented. The second analysis searches for long-lived supersymmetry particles decaying to photons and gravitinos in the context of gauge-mediated supersymmetry breaking model. Results are presented in terms of 95% confidence level expected exclusion limits on the masses and proper decay lengths of the neutralino, which exceed the limits from the previous searches by up to 100 GeV for the neutralino mass and by five times for the neutralino proper decay length. A strategy for model-independent new physics searches is presented with an anomaly trigger based on unsupervised learning algorithms that can be deployed in both the high-level trigger and the Level-1 trigger in CMS. Three other machine-learning solutions are presented to address the computational challenges in the HL-LHC era: a layer based on multi-modal deep neural networks that can reduce the false-positive events selected by the trigger by over one order of magnitude while retaining 99% of signal events, a full-event simulation algorithm based on recurrent generative adversarial networks that has potential to replace traditional simulation method while being five orders of magnitude faster, and a fast simulation algorithm for specific analyses based on encoder-decoder architecture that would result in about an order-of-magnitude reduction in computing and storage requirements for the collision simulation workflow.</p
Antibody Targeting of HIV-1 Env: a Structural Perspective
Get PDFA key component of contemporary efforts toward a human immunodeficiency virus 1 (HIV-1) vaccine is the use of structural biology to understand the structural characteristics of antibodies elicited both from human patients and animals immunized with engineered 'immunogens,' or early vaccine candidates. This thesis will report on projects characterizing both types of antibodies against HIV-1. Chapter 1 will introduce relevant topics, including the reasons HIV-1 is particularly capable of evading the immune system in natural infection and after vaccination, the 20+ year history of unsuccessful HIV-1 vaccine large-scale efficacy trials, an introduction to broadly neutralizing antibodies (bNAbs), and a review of common strategies utilized in HIV-1 immunogen design today. Chapter 2 describes the isolation, high-resolution structural characterization, and in vitro resistance profile of a new bNAb, 1-18, that is both very broad and potent, as well as able to restrict HIV-1 escape in vivo. Chapter 3 reports the results of an epitope-focusing immunogen design and immunization experiment carried out in wild type mice, rabbits, and non-human primates where it was shown that B cells targeting the desired epitope were expanded after a single prime immunization with immunogen RC1 or a variant, RC1-4fill. Chapter 4 describes Ab1245, an off-target non-neutralizing monoclonal antibody isolated in a macaque that had been immunized with a series of sequential immunogens after the prime immunization reported in Chapter 3. The antibody structure describes a specific type of distracting response as it binds in a way that causes a large structural change in Env, resulting in the destruction of the neutralizing fusion peptide epitope. Chapter 5 is adapted from a review about how antibodies differentially recognize the viruses HIV-1, SARS-CoV-2, and Zika virus. This review serves as an introduction to the virus SARS-CoV-2, which is the topic of the final chapter, Chapter 6. In this chapter, structures of many neutralizing antibodies isolated from SARS-CoV-2 patients were used to define potentially therapeutic classes of neutralizing receptor-binding domain (RBD) antibodies based on their epitopes and binding profiles
Slip Patterns on Heterogeneous Frictional Interfaces
Get PDFUnderstanding the implications of heterogeneity on frictional interfaces for the resulting slip patterns is a challenging, highly nonlinear, and dynamic problem with special relevance to earthquake source processes. Natural fault surfaces are rarely homogeneous and host a spectrum of slip behaviors in response to slow tectonic loading where slow steady slip and earthquake ruptures are just the end members. Understanding how heterogeneous frictional properties translate into different slip patterns would enable us to constrain the heterogeneity of natural faults and get an insight into processes that are difficult to observe in the field such as earthquake nucleation, with important implications for the assessment of seismic hazard.
In this thesis, we advance our understanding of fault heterogeneity and its effects by conducting numerical simulations of long-term slip histories on heterogeneous frictional interfaces. We first focus on how irregular fault geometry affects the variability in repeating sequences by investigating a specific example of the SF-LA repeaters in the Parkfield segment of the San Andreas Fault (SAF) in California. We then investigate the effect of increasing heterogeneity in the effective normal stress on earthquake nucleation processes, complexity of earthquake sequences, and features of larger-scale ruptures. In both cases, we incorporate the heterogeneity in physical properties into 2D planar faults governed by rate-and-state friction and embedded into 3D homogeneous elastic bulk. Fully dynamic simulations are used to numerically solve the resulting elastodynamic problems with friction as a nonlinear boundary condition.
Our models reproduce many observations about SF-LA repeating sequences, in- cluding their mean moment, mean recurrence times, stress drops, the observed non- trivial scaling between the seismic moment and recurrence times of the repeaters, the ranges of variability in moment and recurrence time, and the ranges of triggering times between the two sequences. Multiple models produce slip behaviors com- parable to observations, indicating that the models cannot be uniquely constrained based on available observations. We also study how small-scale features of hetero- geneity affect model response. We find that smoothing the distribution over scales smaller than governing length scales in the problem, such as the nucleation size in our case, changes the specific evolution of slip, but preserves its key characteristics, such as the range of event variability and triggering times between events. However, smoothing the distribution on larger scales modifies the response qualitatively.
Our study of the earthquake initiation processes on interfaces with normal stress heterogeneity reveals that systematic increase in heterogeneity induces a continuum of behaviors, ranging from purely fault-spanning events to persistent foreshock-like events interspersed between fault-spanning mainshocks. In models with strong heterogeneity, most smaller-scale and larger-scale events initiate from scales much smaller than the nucleation size estimates calculated for uniform interfaces with equivalent average properties. While the variations in normal stress induce inversely proportional variations in the instability length scale often called nucleation size, we find that the nucleation-size variations by themselves are insufficient to cause such behavior, and that the associated strong heterogeneity in frictional strength is also required. In models with uniform friction strength but the same nucleation-size variation, the nucleation processes of larger-scale events are similar to those on uniform interfaces, with an addition of multiple triggered small-scale earthquakes. Our simulations show that several hypothesized scenarios of earthquake nucleation and foreshocks on natural faults may be viable and reflect different types and levels of heterogeneity on different faults the effects of which, in addition, vary as fault conditions evolve. For example, even with strong fault heterogeneity, some large- scale events have foreshocks and some do not, in the same simulation.
The increasing fault heterogeneity generally leads to increasing complexity of the resulting earthquake sequences and moment-rate release (also called source-time function) of large-scale, fault-spanning events, as intuitively expected, although with some saturation at the higher heterogeneity levels. We find that, in the presence of significant normal-stress heterogeneity, source-time functions of many larger-scale events exhibit prolonged seismic initiation phases, similar to some observations, as the events nucleate from the heterogeneity scale and re-rupture the areas pres-lipped quasi-statically and in foreshocks. The source-time functions also reveal that larger-scale events in our models -- that are arrested by velocity-strengthening barriers -- have a more abrupt arrest phase than natural earthquakes, which places constraints on rupture-arresting mechanisms that should be used in modeling. The initial moment rates are similar for events of different eventual sizes on interfaces with strong heterogeneity, implying that, in those cases, large events are just small events that ran away.</p
Non-Native Chemistry of Metalloenzymes
Get PDFMetalloenzymes are important catalysts in biochemistry, but the scope of their naturally occurring activities is dwarfed by the range of chemistry achieved by synthetic transition-metal catalysts. To date, efforts to expand the catalytic repertoire of metalloproteins beyond their native activities have focused almost exclusively on heme-binding proteins, which have been engineered to catalyze a wide variety of carbene- and nitrene-transfer chemistry. Heme-binding proteins represent only a limited subset of the vast diversity of metalloproteins that exists in Nature, and the non-native chemistry of the rest of the metalloproteome remains largely unexplored. This thesis details the discovery and engineering of non-native catalytic abilities of non-heme metalloproteins. Chapter 1 introduces metalloproteins as biocatalysts in synthetic chemistry, and various approaches to expand their catalytic activities. Chapter 2 describes efforts towards enzyme-catalyzed hydrosilylation, including the curation and development of a diverse library of non-heme metalloproteins. In Chapter 3, a non-heme iron-dependent dioxygenase (Pseudomonas savastanoi ethylene-forming enzyme, PsEFE) is found to catalyze nitrene-transfer chemistry, and is engineered by directed evolution to improve this non-native activity. The nitrene transfer activity and selectivity of PsEFE can be modulated by small-molecule metal-coordinating ligands. Chapter 4 describes the discovery and development of a PsEFE-catalyzed olefin aminoarylation reaction, a previously unknown reaction of sulfonyl azides and olefins. This reaction is unprecedented in the existing chemical literature, and displays a number of unusual mechanistic features. Together, the work described here represents the expansion of non-native chemistry to a new class of metalloenzymes, enabling the discovery of previously unknown catalytic activities.</p
Geometry Synthesis and Multi-Configuration Rigidity of Reconfigurable Structures
Get PDFReconfigurable structures are structures that can change their shapes to change their functionalities. Origami-inspired folding offers a path to achieving shape changes that enables multi-functional structures in electronics, robotics, architecture and beyond. Folding structures with many kinematic degrees of freedom are appealing because they are capable of achieving drastic shape changes, but are consequently highly flexible and therefore challenging to implement as load-bearing engineering structures. This thesis presents two contributions with the aim of enabling folding structures with many degrees of freedom to be load-bearing engineering structures.
The first contribution is the synthesis of kirigami patterns capable of achieving multiple target surfaces. The inverse design problem of generating origami or kirigami patterns to achieve a single target shape has been extensively studied. However, the problem of designing a single fold pattern capable of achieving multiple target surfaces has received little attention. In this work, a constrained optimization framework is presented to generate kirigami fold patterns that can transform between several target surfaces with varying Gaussian curvature. The resulting fold patterns have many kinematic degrees of freedom to achieve these drastic geometric changes, complicating their use in the design of practical load-bearing structures.
To address this challenge, the second part of this thesis introduces the concept of multi-configuration rigidity as a means of achieving load-bearing capabilities in structures with multiple degrees of freedom. By embedding springs and unilateral constraints, multiple configurations are rigidly held due to the prestress between the springs and unilateral constraints. This results in a structure capable of rigidly supporting finite loads in multiple configurations so long as the loads do not exceed some threshold magnitude. A theoretical framework for rigidity due to embedded springs and unilateral constraints is developed, followed by a systematic method for designing springs to maximize the load-bearing capacity in a set of target configurations. An experimental study then validates theoretical predictions for a linkage structure. Together, the application of geometry synthesis and multi-configuration rigidity constitute a path towards engineering reconfigurable load-bearing structures.</p
A Technical and Systems Analysis of Hydrogen Fuel in Renewable Energy Systems
Get PDFWithin the next century, we must tackle the dual challenges of continuing to meet the increasing global demand for energy services while stabilizing global temperatures to mitigate the effects of anthropogenic climate change. Doing so will require a major restructuring of all energy services on a global scale. Here, we contribute to the understanding of the role of hydrogen fuel in net-zero emissions systems from both a technical and systems perspective.
From the technical perspective, we evaluate the activation mechanism of an electrodeposited cobalt selenide hydrogen evolution reaction (HER) catalyst using operando Raman spectroscopy. During this activation process these films, which originally show no catalytic activity toward HER, undergo a compositional change in which selenium in the form of loose, polymeric chains is electrochemically reduced from the material. This work provides a facile method towards investigating catalytic materials under operando conditions, elucidates the changes that occur in this cobalt selenide material during the activation step, and offers potential paths toward the improvement of the cobalt selenide catalyst.
At the systems level, we use hourly weather data over multiple decades and historical electricity demand data to analyze the gaps between wind and solar supply and electricity demand for California (CA) and the Western Interconnect (WECC). We quantify the occurrence of resource droughts when the daily power from each resource was less than half of the 39-year daily mean for that day of the year. Using a macro-scale electricity model, we then evaluate the potential for both long-term storage (in the form of power-to-gas-to-power) and more geographically diverse generation resources to minimize system costs. For wind-solar-battery electricity systems, meeting California demand with WECC generation resources reduces the cost by 9% compared to constraining resources entirely to California. Adding long-duration storage lowers system costs by 21% when treating California as an island. This data-driven analysis quantifies rare weather-related events and provides an understanding that can be used to inform stakeholders in future electricity systems.</p
Experimental Studies on the Thermodynamics and Kinetics of Coexisting Olivine, Silicate Melt, and Vapor
Get PDFThis thesis focuses on experiments run in 1 atm gas-mixing furnaces exploring the thermodynamics and kinetics of coexisting olivine, silicate melt, and vapor. Chapter 1 provides a high-level introduction and summary of the results for each of the following chapters. Chapter 2 and Chapter 3 both involve experiments run on natural olivines containing melt inclusions. Chapter 2 describes a set of homogenization and cooling rate experiments designed to characterize chemical zonation that develops across melt inclusions during cooling. A diffusion model for MgO in the inclusion liquid was calibrated based on these experiments and then used to calculate the syneruptive cooling rates of lavas on Earth and on Mars based on comparison of the model to experimental and natural diffusion profiles in melt inclusions. Chapter 3 presents the first co-determined measurements of S and Fe oxidation state in experimental silicate melts that were equilibrated with the oxygen fugacity of a gas-mixing furnace. The use of melt inclusions as sulfur-bearing experimental vessels is explored, as are implications for interpreting room temperature measurements of the oxidation state of multivalent elements. A set of natural melt inclusions are used as a case study to demonstrate that the temperature-dependence of sulfur-iron electron exchange in basaltic liquids is either weak or leads to the conversion of ferric iron to ferrous iron during cooling. Chapter 4 presents a new parameterization of the composition-dependence of the olivine-liquid Fe-Mg exchange coefficient, Kᴅol/liq,Fe2+-Mg, based on experiments at low oxygen fugacity where corrections for Fe3+ are minor. A quantitative thermodynamic model is fit to the data, showing that the Kᴅ is a function of the Si, Al, Ti, Na+K contents of the liquid as well as olivine composition. Models of Kᴅol/liq,Fe2+-Mg that do not incorporate liquid compositional variables cannot account for the variability of Kᴅ (~0.22-0.38) observed at low oxygen fugacity in a compilation of high-quality literature experiments. Lastly, in Appendix 1, the published version of Richter, Saper, et al. (2021), GCA 295 is included. For this chapter, I contributed MELTS calculations (Ghiorso and Sack 1995; Smith and Asimow 2005) which were used to model crystallization processes and to set boundary conditions for models elemental and isotopic diffusion of Mg and Li in lunar olivines and martian olivines and augites. The combined elemental and isotopic diffusion profiles were used to discriminate between zoning formed due to crystallization from that due to diffusion.</p
Wearable Inductive Damping Sensors for Skin Edema Quantification
Get PDFThe electrical conductivity of human organs is closely related to the physiological or pathological changes occurring within the organ. For example, metastatic liver tumors significantly increase electrical conductivity compared to healthy liver tissues over a wide frequency range. Therefore, knowing when and where these conductivity changes happen within an organ is highly valuable for disease monitoring.
Skin is the largest human organ by surface area, and under its large surface, there are numerous tiny blood and lymphatic vessels that circulate body fluid and dissipate heat. Therefore, it contains critical information about systemic circulation. Diseases such as congestive heart failure, acute renal injury, and liver failure disturb the systemic circulation and allow extra interstitial fluid to accumulate in the form of peripheral skin edema. As the interstitial fluid is highly conductive, the overall skin conductivity significantly increases when edema occurs.
Consequently, quantification of skin edema allows us to track the progression of these diseases and is the main goal to pursue in this study. The current clinical standard uses a 0-to-4 grade system to quantify the severity of edema based on how the skin responds to a pressing force. However, it requires in-person examination and has relatively large inter-examiner variations, making it less suitable for real-time edema monitoring.
To solve the unmet need to quantify edema in real-time, I present a skin edema model that relates skin conductivity to the interstitial fluid volume fraction. The latter is used to quantify the severity of edema. Furthermore, I developed a wearable coil sensor that provides accurate real-time conductivity measurements on subcutis, a significant portion of the skin where edema typically occurs. The coil sensor uses alternating magnetic fields to induce eddy currents in the skin and measures the skin conductivity as a function of coil resistance change. The experimental results suggested that when grade-1 edema occurs, the subcutis conductivity increases from the average value of 0.09 S/m to 0.25 S/m. This change corresponds to an increase of interstitial volume fraction from 10% to 20% in the subcutis. These quantitative results are consistent with finite element simulations and allow direct comparison with ultrasonography measurements. Due to its high accuracy and portability, the proposed wearable sensor opens a new possibility for continuous monitoring of skin edema.</p
New Algorithms for Programmatic Deep Learning with Applications to Behavior Modeling
Get PDFRaw behavioral data is becoming increasingly more abundant and more easily obtainable in spatiotemporal domains such as sports, video games, navigation & driving, motion capture, and animal science. How can we best use this data to advance their respective domains forward? For instance, researchers for self-driving vehicles would like to identify the key features of the environment state that impact decision-making the most; game developers would like to populate their games with characters that have unique and diverse behaviors to create a more immersive gaming experience; and behavioral neuroscientists would like to uncover the underlying mechanisms that drive learning in animals. Machine learning, the science of developing models and algorithms to identify and leverage patterns in data, is well-equipped to aid in these endeavors. But how do we integrate machine learning with these spatiotemporal domains in a principled way? In this dissertation, we develop and introduce new algorithms in programmatic deep learning that tackle some of the new challenges encountered in behavior modeling.
Our work in programmatic deep learning comprises two main themes: in the first, we show how to use expert-written programs as sources of weak labels in domains where manually-annotated expert labels are scarce; in the second, we explore programs as a flexible function class with human-interpretable structure and show how to learn them via neurosymbolic program learning. Augmenting deep learning with programmatic structure allows domain experts to easily incorporate domain knowledge into machine learning models; we show that this results in significant improvements in many behavior modeling applications like imitation learning, controllable generation, counterfactual analysis, and unsupervised clustering.</p