On Seven Fundamental Optimization Challenges in Machine Learning(2021-10-14) [Dissertation]
Advisor: Richtarik, Peter
Committee members: Carin, Lawrence; Ghanem, Bernard; Sra, Suvrit; Yin, WotaoMany recent successes of machine learning went hand in hand with advances in optimization. The exchange of ideas between these fields has worked both ways, with ' learning building on standard optimization procedures such as gradient descent, as well as with new directions in the optimization theory stemming from machine learning applications. In this thesis, we discuss new developments in optimization inspired by the needs and practice of machine learning, federated learning, and data science. In particular, we consider seven key challenges of mathematical optimization that are relevant to modern machine learning applications, and develop a solution to each. Our first contribution is the resolution of a key open problem in Federated Learning: we establish the first theoretical guarantees for the famous Local SGD algorithm in the crucially important heterogeneous data regime. As the second challenge, we close the gap between the upper and lower bounds for the theory of two incremental algorithms known as Random Reshuffling (RR) and Shuffle-Once that are widely used in practice, and in fact set as the default data selection strategies for SGD in modern machine learning software. Our third contribution can be seen as a combination of our new theory for proximal RR and Local SGD yielding a new algorithm, which we call FedRR. Unlike Local SGD, FedRR is the first local first-order method that can provably beat gradient descent in communication complexity in the heterogeneous data regime. The fourth challenge is related to the class of adaptive methods. In particular, we present the first parameter-free stepsize rule for gradient descent that provably works for any locally smooth convex objective. The fifth challenge we resolve in the affirmative is the development of an algorithm for distributed optimization with quantized updates that preserves global linear convergence of gradient descent. Finally, in our sixth and seventh challenges, we develop new VR mechanisms applicable to the non-smooth setting based on proximal operators and matrix splitting. In all cases, our theory is simpler, tighter and uses fewer assumptions than the prior literature. We accompany each chapter with numerical experiments to show the tightness of the proposed theoretical results.
Long Read Based Individual Molecule Sequencing and Real-time Pathogen Detection(2021-10) [Dissertation]
Advisor: Li, Mo
Committee members: Gao, Xin; Frokjaer-Jensen, Christian; Tang, FuchouWith the ability to produce reads with hundreds of kilobases in length, long-read sequencing technology is emerging as a powerful tool to decode complex genetic sequences that are previously inaccessible for short reads. Though the sequencing chemistry and base calling algorithm are being actively developed, the accuracy of the current long-read sequencing is still considerably low, thus limiting its applications. In this dissertation, I present three long read based DNA sequencing methods to overcome the limitation of read accuracy, contribute to a better understanding of Cas9 editing outcomes and mitochondrial DNA heterogeneity, and pave the way for real-time pathogen detection and mutation surveillance. The development of IDMseq enables the single-base-resolution haplotype-resolved quantitative characterization of diverse types of rare variants. IDMseq provides the first quantitative evidence of persistent nonrandom large structural variants following repair of double-strand breaks induced by CRISPR-Cas9 in human ESCs. The development of iMiGseq represents the first mitochondrial DNA sequencing method that provides ultra-sensitive variant detection, complete haplotyping, and unbiased evaluation of heteroplasmy levels, all at the individual mitochondrial DNA molecule level. iMiGseq uncovers unappreciated levels of heteroplasmic variants in single healthy human oocytes well below the current 1% detection limit, of which numerous variants are deleterious and associated with late-onset mitochondrial disease and cancer. It could comprehensively characterize and haplotype single-nucleotide and structural variants of mitochondrial DNA and their genetic linkage in NARP/Leigh syndrome patient-derived cells. The development of NIRVANA deals with the COVID-19 pandemic. NIRVANA can simultaneously detect SARS-CoV-2 and three co-infecting respiratory viruses, and monitor mutations for up to 96 samples in real time. It provides a promising solution for rapid field-deployable detection and mutation surveillance of pandemic viruses. Taken all together, IDMseq, iMiGseq and NIRVANA utilize the advantage of long reads, overcome the limitation of low accuracy, and facilitate the application of long-read sequencing technologies in multidisciplinary fields.
End-to-end Optics Design for Computational Cameras(2021-10) [Dissertation]
Advisor: Heidrich, Wolfgang
Committee members: Ghanem, Bernard; Michels, Dominik; Veeraraghavan, AshokImaging systems have long been designed in separated steps: the experience-driven optical design followed by sophisticated image processing. Such a general-propose approach achieves success in the past but left the question open for specific tasks and the best compromise between optics and post-processing, as well as minimizing costs. Driven from this, a series of works are proposed to bring the imaging system design into end-to-end fashion step by step, from joint optics design, point spread function (PSF) optimization, phase map optimization to a general end-to-end complex lens camera. To demonstrate the joint optics application with image recovery, we applied it to flat lens imaging with a large field of view (LFOV). In applying a super-resolution single-photon avalanche diode (SPAD) camera, the PSF encoded by diffractive op tical element (DOE) is optimized together with the post-processing, which brings the optics design into the end-to-end stage. Expanding to color imaging, optimizing PSF to achieve DOE fails to find the best compromise between different wavelengths. Snapshot HDR imaging is achieved by optimizing a phase map directly. All works are demonstrated with prototypes and experiments in the real world. To further compete for the blueprint of end-to-end camera design and break the limits of a simple wave optics model and a single lens surface. Finally, we propose a general end-to-end complex lens design framework enabled by a differentiable ray tracing image formation model. All works are demonstrated with prototypes and experiments in the real world. Our frameworks offer competitive alternatives for the design of modern imaging systems and several challenging imaging applications.
Asymmetric Signaling: A New Dimension of Interference Management in Hardware Impaired Communication Systems(2021-10) [Dissertation]
Advisor: Alouini, Mohamed-Slim
Committee members: Shihada, Basem; Eltawil, Ahmed; Laleg-Kirati, Taous-Meriem; Dobre, Octavia A.Hardware impairments (HWIs) impose a huge challenge on modern wireless commu- nication systems owing to the characteristics like compactness, least complexity, cost ef- fectiveness and high energy efficiency. Numerous techniques are implemented to minimize the detrimental effects of these HWIs ,however, the residual HWIs may still appear as an additive distortion, multiplicative interference, or an aggregate of both. Numerous studies have commenced efforts to model one or the other forms of hardware impairments in the ra- dio frequency (RF) transceivers. Many presented the widely linear model for in-phase and quadrature imbalance (IQI) but failed to recognize the impropriety induced in the system because of the self-interfering signals. Therefore, we have presented not only a rigorous ag- gregate impairment model along with its complete impropriety statistical characterization but also the appropriate performance analysis to quantify their degradation effects. Lat- est advances have endorsed the superiority of incorporating more generalized impropriety phenomenon as opposed to conventional propriety. In this backdrop, we propose the improper Gaussian signaling (IGS) to mitigate the drastic impact of HWIs and improve the system performance in terms of achievable rate and outage probability. Recent contributions have advocated the employment of IGS over traditional proper Gaussian signaling (PGS) in various interference limited scenarios even in the absence of any improper noise/interference. It is pertaining to the additional degree of freedom (DoF) offered by IGS, which can be optimized to reap maximum benefits. This reduced-entropy signaling is the preferred choice to pose minimal interference to a legitimate network yielding another mechanism to tackle undesired interference. Evidently, the incorporation of both inherent and induced impropriety characteristics is critical for effective utilization. Most of the recent research revolves around the theoretical analysis and advantages of improper signaling with minimal focus on its practical realization. We bridge this gap by adopting and optimizing asymmetric signaling (AS) which is the finite discrete implemen- tation of the improper signaling. We propose the design of both structural and stochastic shaping to realize AS. Structural shaping involves geometric shaping (GS) of the symbol constellation using some rotation and translation matrices. Whereas, stochastic shaping as- signs non-uniform prior probabilities to the symbols. Furthermore, hybrid shaping (HS) is also proposed to reap the gains of both geometric and probabilistic shaping. AS is proven superior to the conventional M-ary symmetric signaling in all of its forms. To this end, probabilistic shaping (PS) demonstrates the best trade-off between the performance en- hancement and added complexity. This research motivates further investigation for the utilization of impropriety concepts in the upcoming generations of wireless communications. It opens new paradigms in inter- ference management and another dimension in the signal space. Besides communications, the impropriety characterization has also revealed numerous applications in the fields of medicine, acoustics, geology, oceanography, economics, bioinformatics, forensics, image processing, computer vision, and power grids.
Halide Perovskite-2D Material Optoelectronic Devices(2021-09-17) [Dissertation]
Advisor: Alshareef, Husam N.
Committee members: Schwingenschlögl, Udo; Mohammed, Omar F.; Quevedo-Lopez, ManuelMetal-halide perovskites have attracted intense research endeavors because of their excellent optical and electronic properties. Different kinds of electronic and optoelectronic devices have been fabricated using perovskites. A feasible approach to utilize these properties in real device applications with improved performance and new functionalities is by fabricating heterostructures with extraneous materials. We have developed mixed-dimensional heterostructure systems using three-dimensional (3D) metal-halide perovskites and different types of different two-dimensional (2D) materials, including semimetal graphene, semiconducting phosphorus-doped graphitic-C3N4 sheets (PCN-S), and plasmonic Nb2CTx MXenes. First, selective growth of single-crystalline MAPbBr3 platelets on monolayer graphene by chemical vapor deposition (CVD) is achieved to prepare the MAPbBr3/graphene heterostructures. P-type doping from MAPbBr3 is observed in the monolayer graphene with a decreased work function of 272 meV under illumination. The photoresponse of the fabricated phototransistor heterostructure verifies the enhanced p-type character in graphene. Such kind of charge transfer can be used to improve device performance. Then, bulk-heterojunctions made of MAPbI3-xClx and PCN-S are prepared in solution. The matched band diagram and the midgap states in PCN-S present a convenient and efficient approach to reduce the dark current and increase the photocurrent of the as-fabricated photodetectors. As a result, the on/off ratio increases from 103 to 105, and the detectivity is up to 1013 Jones with an order of magnitude enhancement compared to the perovskite-only device. Last, plasmonic Nb2CTx MXenes and MAPbI3 heterostructures are prepared for photodiodes to broaden the detection band to near-infrared (NIR) lights. The use of the perovskite layer expanded the operation of the diode to the visible range while suppressing the dark current of the NIR-absorbing Nb2CTx layer. The fabricated photodiode reveals a detectivity of 0.25 A/W with a linear dynamic range of 96 dB in the visible region. In the NIR region, the device demonstrates an increased on/off ratio from less than 2 to near 103 and much faster response times of less than 30 ms. The improved performance is attributed to the passivation of the MAPbI3/Nb2CTx interface.
Computational Imaging and Its Applications in Fluids(2021-09-13) [Dissertation]
Advisor: Heidrich, Wolfgang
Committee members: Ghanem, Bernard; Wonka, Peter; Schindler, KonradComputational imaging di↵ers from traditional imaging system by integrating an encoded measurement system and a tailored computational algorithm to extract interesting scene features. This dissertation demonstrates two approaches which apply computational imaging methods to the fluid domain. In the first approach, we study the problem of reconstructing time-varying 3D- 3C fluid velocity vector fields. We extend 2D Particle Imaging Velocimetry to three dimensions by encoding depth into color (a “rainbow”). For reconstruction, we derive an image formation model for recovering stationary 3D particle positions. 3D velocity estimation is achieved with a variant of 3D optical flow that accounts for both physical constraints as well as the rainbow image formation model. This velocity field can be used to refine the position estimate by adding physical priors that tie together all the time steps, forming a joint reconstruction scheme. In the second approach, we study the problem of reconstructing the 3D shape of underwater environments. The distortions from the moving water surface provide a changing parallax for each point on the underwater surface. We utilize this observation by jointly estimating both the underwater geometry and the dynamic shape of the water surface. To this end, we propose a novel di↵erentiable framework to tie together all parameters in an integrated image formation model. To our knowledge, this is the first solution that is capable to simultaneously retrieve the structure of dynamic water surfaces and static underwater scene geometry in the wild.
Applications of Magnetic Transition Metal Dichalcogenide Monolayers to the Field of Spin-orbitronics(2021-09) [Dissertation]
Advisor: Schwingenschlögl, Udo
Committee members: Manchon, Aurelien; El-Atab, Nazek; Laquai, Frédéric; Larsson, AndreasMagnetic randomaccess memory (MRAM) devices have been widely studied since the 1960s. During this time, the size of spintronic devices has continued to decrease. Conse quently, there is now an urgent need for new lowdimensional magnetic materials to mimic the traditional structures of spintronics at the nanoscale. We also require new effective mechanisms to conduct the main functions of memory devices, which are: reading, writ ing, and storing data. To date, most research efforts have focused on MRAM devices based on magnetic tun nel junction (MTJ), such as a conventional fielddriven MRAM and spintransfer torque (STT)MRAM devices. Consequently, many efforts are currently focusing on new alterna tives using different techniques, such as spinorbit torque (SOT) and magnetic skyrmions (a skyrmion is the smallest potential disruption to a uniform magnet required to obtain more effective memory devices). The most promising memory devices are SOTMRAMs and skyrmionbased memories. This study investigates the magnetic properties of 1Tphase vanadium dichalcogenide (VXY) Janus monolayers, where X, Y= S, Se, or Te (i.e., monolayers that exhibit inversion symme try breaking due to the presence of different chalcogen elements). This study is developed along four directions: (I) the nature of the magnetism and the SOT effect of Janus mono layers; (II) the Dzyaloshinskii Moriya interaction (DMI); (III) investigation of stability en hancement by adopting practical procedures for industry; and (IV) study of the effect of a hexagonal boron nitride (hBN) monolayer as an insulator on the magnetism of the VXY monolayer. This study provides a clear perspective for the next generation of memory de vices, such as SOTMRAMs based on transition metal dichalcogenide monolayers.
Single-Step Conversion of Crude Oil to Petrochemicals using a Multi Zone Reactor: Reactor Design and Catalysts Assessment(2021-09) [Dissertation]
Advisor: Gascon, Jorge
Committee members: Rueping, Magnus; Castaño, Pedro; Yuriy, RomanOne hundred million barrels of oil are produced every day. Economic and population growth stirred a significant increase in the demand of oil over the last century. Today circa 75% of the crude oil barrel is dedicated to the manufacture of transportation fuels, while less than a 15 % is used for the production of chemicals. However, the demand for oil derived fuels is expected to first peak and then decrease. This is mostly due to environmental concerns related to CO2 emissions, the rapid development of green energy technology and improvements in efficiency. On the other hand, the demand for petrochemicals is still forecasted to continue growing in the foreseeable future. By 2040, the barrel split is indeed expected to reach a 34% for the production of petrochemicals. To fill the chemical demand gap, we need to maximize the production of light olefins and aromatics, preferably by developing direct conversion routes leading to chemicals yields in the order of 60-65 %. By converting crude oil directly to chemicals, several energy intensive refinery processes may be optimized and/or avoided, positively impacting the environment by reducing emissions. Equally important, this technology will be highly efficient for the production of highly valued chemicals, which will lead to cost-saving and, at the same time, double the profitability. This PhD Thesis describes a catalytic reactor concept consisting of a multi-zone fluidized bed (MZFB) able to perform several refining steps in one single reactor vessel along with a new catalyst formulation. The new configuration allows for in situ catalyst stripping and regeneration, while the incorporation of silicon carbide in the formulated catalyst confers it with improved physical, mechanical, and heat transport properties. As a result, this reactor has shown stable conversion of untreated Arabian Light crude to light olefins with yields per pass over 35 wt.% with a minimum production of dry gas on spray-dried catalysts containing 1:1 mixtures of ZSM-5 and FAU zeolites (alongside binder, clay, and Silicon carbide). Coke deposition and catalyst deactivation can be correlated to the nature and content of each zeolite component.
Laser investigations on a plasma assisted flame(2021-09) [Dissertation]
Advisor: Lacoste, Deanna
Committee members: Cha , Min Suk; Ruiz-Martinez, Javier; Starikovskaia, SvetlanaSustainable and low emission combustion requires new combustion paradigms and solutions to increase efficiency, comply with more stringent regulations on pollutants, and cope with the varying qualities of renewable fuels. Plasma Assisted Combustion (PAC) could be one of the tools to achieve these goals in practical combustion systems. Previous studies showed that PAC can be used in a variety of applications: to improve ignition in difficult environments, to extend the operating range of burners to leaner conditions, to contrast thermoacoustic instability, to allow flame-holding in extreme conditions, and more. While applications keep being proposed, there are efforts to model and understand the coupling between flames and plasma discharges. This work contributes to the unraveling of the action of plasma discharges on flames by performing a number of investigations on a simple PAC burner. Trends and temporal evolution of key chemical species and electric fields are measured during plasma actuation of the flame. Experimental datasets resulting from this work are meant to be used in cross-validating numerical simulations. The considered PAC burner generates a lean methane-air stagnation flame, across which discharges are applied, developing partially in the fresh and partially in the burned gases. Time-resolved 2D imaging of atomic hydrogen and oxygen is obtained by using two-photon absorption planar laser induced fluorescence (TALIF) while OH and CH radicals are measured by using planar laser induced fluorescence (PLIF). To measure the electric field, the Electric Field Induced Second Harmonic generation (EFISH) technique is used. A novel deconvolution-like post-processing procedure is proposed and used to calibrate the measurements and improve the spatial resolution, overcoming limitations and distortions typical of EFISH measurements. Presented results quantify the effect of the plasma actuation on the flame and lend themselves to the validation of numerical models.
Experimental and Computational Study on Pyrolysis and Combustion of Heavy Fuels and their Upgrading Technologies(2021-09) [Dissertation]
Advisor: Roberts, William L.
Committee members: Sarathy, Mani; Im, Hong G.; Cuoci, Alberto; Knio, OmarEngineering applications of unconventional fuels like HFOs require a detailed understanding of the physics associated with their evaporation. The processing of HFOs involves forming a spray; therefore, studying droplets is of particular interest. The work described in this dissertation tackles two of the most obscure aspects associated with HFOs modelling. The first aspect is the identification of a valid chemical description of the structure of the fuel. In particular, the author focused on finding a methodology that allows identifying a discrete surrogate to describe the complex pool of molecules of which the fuels are made. The second part of the work was devoted to understand and model thermally-induced secondary breakup, which is the primary cause of deviation from the "d2" that multi-component droplet experience. The formulation of a surrogate was successfully achieved by developing and implementing a new algorithm that allows building a surrogate from a set of easily accessible physical properties. A new methodology for the post-processing of experimental data was formulated. The methodology consists of studying the evolution of the normalized distance of the interface from the droplet’s centroid instead of its diameter. The new approach allowed the separation between interface deformation and expansion/shrinking. The information was then processed using the dynamic mode decomposition to separate the stochastic contribution associated with secondary atomization and the deterministic contribution of vaporization. Finally, thermally induced secondary atomization was studied using a CFD code appositely developed. The code is based on the geometric Volume of Fluid (VoF) method and consists of a compressible, multi-phase, multi-component solver in which phase change is considered. The novelty in the proposed approach is that the evaporation source term and the surface tension force are evaluated directly from the geometrically reconstructed interface. The code was validated against the exact solution of analytically solvable problems and experimental data. The solver was then used to study HFO secondary breakup and perform a parametric analysis that helped to understand the problem’s physics. A possible application of this framework is the formulation of sub-models to be applied in spray calculations.
Engineering and Discovery of Novel Biocatalysts(2021-09) [Dissertation]
Advisor: Rueping, Magnus
Committee members: Arold, Stefan T.; Michels, Dominik; Stingl, UlrichBiocatalysis is considered a green and environmentally friendly technology. Therefore, novel enzymes and enzymatic systems, together with cascades and protein engineering approaches, are in high demand. Here, three very different biocatalytic approaches have been studied. First, the richness of enzymes in the Red Sea brine pools has been assessed, and the discovery and characterization of a novel halophilic γ-carbonic anhydrase is described, together with the protein engineering approach, which boosted the initial catalytic activity of the γ- carbonic anhydrase. The understanding of polyextremophilicity principles from enzymes from the Red Sea brine pool, contributes to the bioengineering effort of turning mesophilic enzymes into more stable variants. Next, focus is given to the use of amine-transaminases in cascades for chiral amine synthesis. This resulted in the development of a self-sufficient sustainable cascade for chiral and non-chiral amine synthesis. This cascade was achieved by combining a lysine decarboxylase with an amine-transaminase to generate a cheap amino donor source for a more sustainable reaction economy. Finally, gas vesicle nanoparticles are functionalized by various engineering principles to create floating platforms for the immobilization of enzymes. The proof-of-concept was achieved by anchoring a phytase via anchoring peptides on the gas vesicle nanoparticles surface. These bioengineering approaches contributed to the effort of generating first principles for protein engineering.
The Development of the Oceanic Mixed Layers and the Processes Governing the Evolution of their Properties in the Red Sea(2021-09) [Dissertation]
Advisor: Hoteit, Ibrahim
Committee members: Sofianos, Sarantis. S.; Knio, Omar; Duarte, Carlos M.; Sun, ShuyuThe importance of understanding the Red Sea (RS) circulation and its response to external forcing variability has become increasingly recognized in recent years. The RS circulation presents a complex behavior and is driven by both strong air-sea buoyancy fluxes and winds. The air-sea exchanges are mediated by the oceanic Mixed Layers (MLs), which constitute the active part of the ocean that interacts with the atmosphere and plays a critical role in the general circulation. The goal of this thesis is to build a deeper understanding of the processes that drive the evolution of the upper layer properties of the Red Sea, focusing on the development of the ML and its spatiotemporal variability. On account of the sparsity of observations, this is primarily achieved using state-of-the-art high resolution regional oceanic simulations, which are extensively validated against the available observations. We first analyze the model results to examine the relative contribution of the different components of the atmospheric forcing (buoyancy fluxes and momentum forcing) to the variability of the upper layer properties and the ML depth (MLD) distribution. Using closed and complete tracer (potential temperature and salinity) budgets, integrated over the MLD, we further investigate the role of internal oceanic processes in the evolution of the RS MLs. Our analysis separately considers the advective fluxes, diapycnal mixing, and entrainment of heat and salt that ultimately define the tracer concentration inside the ML. We further identify anomalous years in terms of their annual MLD and investigate the dependence of the ML development on the interannual accumulation of heat and salt in the water column. In light of recent reports of warming trends in the RS that raise concerns about the response of the RS under an increasingly warming climate, we extend our analysis on longer timescales than the model simulation period, using sea surface temperature (SST) as a proxy of the long-term RS upper layers’ variability. Increased understanding of the processes governing the evolution of the RS will not only serve to improve our knowledge on the basin’s dynamical functioning but also lead to greater understanding of the physical-biological interactions in the RS.
Stylistic and Spatial Disentanglement in GANs(2021-08-17) [Dissertation]
Advisor: Wonka, Peter
Committee members: Michels, Dominik; Ghanem, Bernard; Yang, Ming-HsuanThis dissertation tackles the problem of entanglement in Generative Adversarial Networks (GANs). The key insight is that disentanglement in GANs can be improved by differentiating between the content, and the operations performed on that content. For example, the identity of a generated face can be thought of as the content, while the lighting conditions can be thought of as the operations. We examine disentanglement in several kinds of deep networks. We examine image-to-image translation GANs, unconditional GANs, and sketch extraction networks. The task in image-to-image translation GANs is to translate images from one domain to another. It is immediately clear that disentanglement is necessary in this case. The network must maintain the core contents of the image while changing the stylistic appearance to match the target domain. We propose latent filter scaling to achieve multimodality and disentanglement. Previous methods require complicated network architectures to enforce that disentanglement. Our approach, on the other hand, maintains the traditional GAN loss with a minor change in architecture. Unlike image-to-image GANs, unconditional GANs are generally entangled. Unconditional GANs offer one method of changing the generated output which is changing the input noise code. Therefore, it is very difficult to resample only some parts of the generated images. We propose structured noise injection to achieve disentanglement in unconditional GANs. We propose using two input codes: one to specify spatially-variable details, and one to specify spatially-invariable details. In addition to the ability to change content and style independently, it also allows users to change the content only at certain locations. Combining our previous findings, we improve the performance of sketch-to-image translation networks. A crucial problem is how to correct input sketches before feeding them to the generator. By extracting sketches in an unsupervised way only from the spatially-variable branch of the image, we are able to produce sketches that show the content in many different styles. Those sketches can serve as a dataset to train a sketch-to-image translation GAN.
Passive seismic event locating with full waveform inversion and machine learning methods(2021-08) [Dissertation]
Advisor: Sun, Shuyu
Committee members: Alkhalifah, Tariq Ali; Peter, Daniel; Bin Waheed, Umair; Yao, GangOne of the key goals of microseismic monitoring is the accurate estimation of the source location. The accuracy of both P- and S-wave velocities strongly influences the estimation of source locations and, hence, the fracture detection’s reliability. I use advanced methodologies based on full waveform inversion methods to obtain accurate P- and S- wave velocities and locate the source and its characteristics. I first use an elastic FWI for passive source and velocity inversion, in which an equivalent source represents the conventional source term of the elastic wave equation. Thus, I update the source locations, source functions, and velocities simultaneously using a waveform inversion scheme. Waveform inversion of passive events has severe nonlinearity due to the unknown source locations in space and their functions in time. I, thus, use a source-independent objective function based on convolving reference traces with both modeled and observed data to avoid cycle skipping caused by the unknown sources. I test the method on real microseismic monitoring data. Then, I extend the method to a 3D acoustic orthorhombic case. I also analyze the relationship of the proposed equivalent source term and the conventional elastic wave equation’s seismic moment tensor source term. Besides, locating numerous microseismic events by solving wave equations is computationally expensive, and manually picking all the event arrivals is challenging. To address the issues without event picking or detection, I use a novel artificial neural network framework to directly map seismic data to their potential locations. I train two convolutional neural networks (CNN) on labeled synthetic 5 acoustic data containing simulated micro-seismic events to fulfill such requirements. At last, I use the developed convolutional neural network to predict the source location for field micro-seismic monitoring data. I, especially, train the CNN with a large amount of synthetically generated data and the extracted coherent noise from the field data. The synthetic training data allow us to control the corresponding labels, and the extracted noise from the field data and the pre-processing steps vastly reduce the di↵erence between the field and the synthetic data.
An iPS-Based Approach to Study the Transcriptional and Epigenetic Consequences of X-Chromosome Aneuploidies(2021-08) [Dissertation]
Advisor: Adamo, Antonio
Committee members: Li, Mo; Mahfouz, Magdy M.; Battaglioli, ElenaKlinefelter Syndrome (KS) is a multisystemic disorder associated with a plethora of phenotypic features including mental retardation, cardiac abnormalities, osteoporosis, infertility, gynecomastia, type two diabetes and increased cancer risk. KS is the most common aneuploidy in humans (with a prevalence of 1:500 to 1:1000 born males) and is characterized by one or more supernumerary X-chromosomes (47-XXY, 48-XXXY, and 49-XXXXY karyotypes). While X-chromosome inactivation (XCI) represses extra Xs, few genes called “escape genes” elude the XCI mechanism and are actively transcribed from X inactive. The overdosage of escape genes has been considered the molecular landscape that underlies KS clinical features. In this project, we exploit an integration-free reprogramming method to generate the largest described cohort of iPSCs from seven patients with KS and healthy donor fibroblasts from two relatives. The unicity of this cohort relies on the derivation of 47-XXY iPSCs and their isogenic 46-XY healthy counterparts, along with multiple rare 48-XXXY and 49-XXXXY iPSC lines. Through X chromosome inactivation (XCI) assessment, we show consistent retention of n-1 XCI in all derived KS-iPSCs. We identify the genes within the PAR1 region as the most susceptible to dosage-dependent transcriptional dysregulation and therefore putatively responsible for the progressively worsening phenotype in higher grade X aneuploidies. Moreover, we explore the transcriptional impact of X overdosage on autosomes and identify that the X-dosage-sensitive autosomal transcription factor NRF1 is a master regulator of the X-linked escape gene ZFX. Finally, we dissect the potential pathological impact of the escape gene KDM6A on low- and high-grade supernumerary X iPSCs and differentiated derivatives. We highlight a considerable proportion of KDM6A targets that could be responsible for paradigmatic clinical manifestations of KS.
PN3P Rhodium Pincer Complexes: Coordination Chemistry and Reactivity(2021-08) [Dissertation]
Advisor: Huang, Kuo-Wei
Committee members: Yu, Han; Lai, Zhiping; Wang, ShaowuAbstract: The choice ofsuitable ligand platforms is crucial to organometallic coordination chemistry and homogeneous catalysis. Among the various ligand platforms available, pincer ligands offer a convenient route to manipulate the properties of the resulting complexes. The pincer chemistry of rhodium has attracted attention for over 40 years, and Rh complexes are dominated by Rh(I) and Rh(III) low-spin states, thus they are more predictable than other paramagnetic species. Compared to other pincer ligand platforms, pyridine-based pincer complexes are particularly attractive as they exhibit diverse reactivities. Our group realized a new class of the PN3 (P) pincer system, with altered the unique catalytic performances, thermodynamic and kinetic properties due to their pseudo-dearomatized nature. In Chapter 2, selective carbonylation of benzene to benzaldehyde using a phosphorus nitrogen PN3P Rh(I) complex was realized. The PN3P Rh pincer chloride complex cPePN3PRhCl was capable of activating C−H bond of benzene to give the phenyl complex cPePN3PRh(C6H5) using KN(SiMe3)2 as a base. Furthermore, the benzoyl complex cPePN3PRh(CO)(C6H5) was obtained by treating a benzene solution of cPePN3PRh(C6H5) with CO gas. In dilute HCl, a high yield of 90% benzaldehyde was formed with regeneration of the cPePN3PRhCl. This is the first example of selective carbonylation of benzene into benzaldehyde accomplished by directly inserting CO without irradiation. In Chapter 3, the ligand-centered reactivity of a pseudo-dearomatized PN3P *rhodium complex towards molecular oxygen wasrealized. For the dearomatized rhodium carbonyl complex (tBuPN3P*RhCO), one of the C−H bonds of the pseudo-dearomatized pyridine ring was oxidized by O2 to create an α, β-unsaturated carbonyl functionality. Moreover, the resulting metal complex with the post-modified PN3P ligand could react with thiophenol and 4-methylaniline to afford the corresponding oxidative Michael addition products. In Chapter 4, to further explore the ligand-centered reactivity of tBuPN3P *RhCO, a series of second-generation diimine-amido PN3P-pincer carbonyl complexes were synthesized by reaction of tBuPN3P*RhCO and various alkyl/benzyl halides via a post-modification strategy, and these complexes were well characterized by NMR, HRMS, FT-IR, and single crystal diffraction. Moreover, a plausible mechanism for the formation of 2nd -generation PN3P complexes was proposed
Stochastic Optimization in Target Positioning and Location-based Applications(2021-08) [Dissertation]
Advisor: Al-Naffouri, Tareq Y.
Committee members: Al-Naffouri, Tareq Y.; Zhang, Xiangliang; Park, Shinkyu; Ballal, Tarig; Swindlehurst, LeePosition information is important for various applications, including location-aware communications, autonomous driving, industrial internet of things (IoT). Geometry based techniques such as time-of-arrival (TOA), time-difference-of-arrival (TDOA), and angle-of-arrival (AOA) are widely used and can be formed as optimization prob lems. In order to solve these optimization problems efficiently, stochastic optimization methods are discussed in this work in solving target positioning problems and tackling key issues in location-based applications. Firstly, the direction of arrival (DOA) estimation problem is studied in this work. Grid search is useful in the algorithms such as maximum likelihood estimator (MLE), MUltiple SIgnal Classification (MUSIC), etc. However, the computational cost is the main drawback. To speed up the search procedure, we implement random ferns to extract the features from the beampatterns of different DOAs and use these features to identify potential angle candidates. Then, we propose an ultrasonic air-writing system based on DOA estimation. In this application, stochastic optimization methods are implemented to solve gesture classification problems. This work shows that stochastic optimization methods are effective tools to address and benchmark practical positioning-related problems. Next, we discuss how to select antennas properly to reduce the expectation of DOA estimation error in a switch-based multiple-input-multiple-output (MIMO) system. Cram`er Rao lower bound (CRLB) expresses a lower bound on the variance of an unbiased estimator, but it does not work well for low SNR scenarios. We use DOA threshold-region approximation as an indicator and propose a greedy algorithm and a neural network-based algorithm. Finally, we propose a joint time difference of arrival (TDOA) and phase difference of arrival (PDOA) localization method. It is shown that the phase difference, which is also widely used in DOA estimation, can improve the performance of the well established TDOA technique. Although the joint TDOA/PDOA cost function has a lot of local minima, accurate estimates can be obtained effectively by choosing an appropriate initial estimation and using particle swarm optimization (PSO).
Spatial and Temporal Biodiversity Patterns of Coral Reef Cryptofauna on the Arabian Peninsula(2021-08) [Dissertation]
Advisor: Berumen, Michael L.
Committee members: Carvalho, Susana; Jones, Burton; Wing, Rod Anthony; Brainard, RustyCoral reef cryptobenthic communities are largely understudied yet they contribute to the large majority of coral reef biodiversity. The main aim of this dissertation was to understand the effects of the organic C, temperature, surrounding benthic communities, salinity, catastrophic events, time, and limitations to dispersal of the cryptobenthic communities. Using 54 ARMS along the Saudi Arabian Red Sea coast, we found that temperature, chlorophyll-a concentration, and photosynthetic active radiation affected the number of OTUs of the cryptobiome, i.e., its biodiversity. We found temperature, energy available, and benthic structure to be associated with distinct cryptobenthic communities and to influence its diversity patterns. These environmental conditions affected differentially the abundance of specific organisms. We also investigated the inter-annual patterns of variability of this biological component in the central Red Sea. We deployed and collected ARMS in four reefs along a cross shelf gradient in three sampling periods spanning 6 years (2013-2019). This period included the 2015/2016 mass bleaching event. We observed cross shelf differences in community composition to be consistent over time and maintained after the bleaching event. However, turnover was significantly higher between prebleaching and post bleaching sampling years than between post bleaching comparisons. Cryptobenthic communities of 2019 presented a slight return to prebleaching composition. In light of predictions of returning bleaching events every 6 years, the observed return might not be sufficient for reaching a full recovery. We investigated the relative contribution of two ecological theories: the neutral theory (associated with the limitations to dispersal and therefore geographic distance) and the niche filtering (associated with environmental conditions that limit colonization). We used 50 ARMS collected from the north, central, and south Red Sea, the Arabian Gulf, and Oman Gulf. We found that limitation to dispersal and environmental filtering to influence beta diversity. However, the geographic distance had a better fit with the beta diversity patterns observed, suggesting a preponderance of the neutral theory of ecology explaining the community patterns. This dissertation provides fundamental information on characterization of the cryptobiome in the Arabian Peninsula.
Printable 3D MoS2 Architected Foam with Multiscale Structural Hierarchies for High-rate, High-capacity and High-mass-loading Energy Storage(2021-08-01) [Dissertation]
Advisor: Tung, Vincent
Committee members: Anthopoulos, Thomas D.; Huang, Kuo-Wei; Chen, Han-YiMaterials with three-dimensional (3D) hierarchical architectures exhibit attractive mechanical, energy conversion and thermal radiative cooling properties not found in their bulk counterparts. However, implementation of hierarchically structured 3D transition metal dichalcogenides (TMDs) is widely deemed not possible, by the lack of manufacturing solutions that overcome the hierarchy, quality, and scalability dilemma. Here we report dewetting-driven destabilization (DDD) process that enables simple, template-free, high throughput printing of 3D architected MoS2 Foam with hierarchy spanning seven orders of magnitude — from angstroms to centimeters. Although extremely simple, our manufacturing process combines electrohydrodynamic printing with dewetting-induced-patterning. This technique can be applied to a range of dissimilar twodimensional (2D) layered materials, including Ti3C2Tx MXene and reduced graphene oxide (rGO). The deposited MoS2 Foam achieves amplification of resilience and conductivity. It constructs hierarchically porous and spatially interconnected networks for both ions and electrons transfer. We further demonstrate the 3D MoS2 architected foam as high-performance anodes with an otherwise unachievable combination of a 99% battery yield, a dynamic recovery (up to 85%) to withstand excessive volume expansion, a strain-induced reduction in diffusion barrier (0.2 eV), and improved electron transport abilities across the entire structure. The result is the high Li-ion charge storage capacity with robust cycling stability at a bulk scale (~3.5 mg/cm2) and under a high current density (10,000 mA/g). The outstanding electrochemical performance arises from the architected structure-induced pseudocapacitive energy storage mechanism based on the redox reaction of Mo, rather than the traditional conversion reaction. Notably, the performance achieved is on par with or surpasses state-of-the-art anodes made of black phosphorus composites, Si-graphene and mesoporous graphene particle anodes, while the technique offers an evaporation-like simplicity for industrial scalability. This work is foundational, and the developed DDD process opens a new sight to manufacture structurally robust, multifunctional hierarchical structures from 2D materials. Given the high adjustability of synthesis conditions and a wide variety of 2D materials, we anticipate previously unattainable possibilities in the energy storage, flexible electronics, catalysis, separation and drug delivery.
Shape Matching and Map Space Exploration via Functional Maps(2021-07-29) [Dissertation]
Advisor: Wonka, Peter
Committee members: Ovsjanikov, Maks; Ghanem, Bernard; Pottmann, Helmut; Solomon, JustinComputing correspondences or maps between shapes is one of the oldest problems in Computer Graphics and Geometry Processing with a wide range of applications from deformation transfer, statistical shape analysis, to co-segmentation and exploration among a myriad others. A good map is supposed to be continuous, as-bijective-as-possible, accurate if there are ground-truth corresponding landmarks given, and lowdistortionw.r.t. different measures, for example as-conformal-as-possible to preserve the angles. This thesis contributes to the area of non-rigid shape matching and map space exploration in Geometry Processing. Specifically, we consider the discrete setting, where the shapes are discretized as amesh structure consisting of vertices, edges, and polygonal faces. In the simplest case, we only consider the graph structure with vertices and edges only. In this thesis, we design algorithms to compute soft correspondences between discrete shapes. Specifically, (1)we propose different regularizers, including orientation-preserving operator and the Resolvent Laplacian Commutativity operator, to promote the shape correspondences in the functional map framework. (2) We propose two refinement methods, namely BCICP and ZoomOut, to improve the accuracy, continuity, bijectivity and the coverage of given point-wisemaps. (3)We propose a tree structure and an enumeration algorithm to explore the map space between a pair of shapes that can update multiple high-quality dense correspondences.