In the realm of discrete ill-posed problems, image deblurring is a challenging problem aimed at restoring clear and visually appealing images from their blurred counterparts. Over the years, various numerical techniques have been developed to solve this problem, each offering unique approaches to tackle blurring and noise.This thesis studies multilevel methods using Daubechies wavelets and Tikhonov regularization. The Daubechies wavelets are a family of orthogonal wavelets widely used in various fields because of their orthogonality and compact support. They have been widely applied in signal processing, image compression, and other applications. One key aspect of this investigation involves a comprehensive comparative analysis with Krylov methods, well-established iterative methods known for their efficiency and accuracy in solving large-scale inverse problems. The focus is on two well-known Krylov methods, namely hybrid LSQR and hybrid generalized minimal residual method \linebreak(GMRES). By contrasting the multilevel and Krylov methods, the aim is to discern their strengths and limitations, facilitating a deeper understanding of their applicability in diverse image-deblurring scenarios. Other critical comparison factors are the noise level adopted during the deblurring process and the amount of blur. To gauge their robustness and performance under different blurry and noisy conditions, this work explores how each method behaves with different noise levels from mild to severe and different amounts of blur from small to large. Moreover, this thesis combines multilevel and Krylov methods to test a new method for solving inverse problems. This work aims to provide valuable insights into the strengths and weaknesses of these multilevel Krylov methods by shedding light on their efficacy. Ultimately, the findings could have implications across diverse domains, including medical imaging, remote sensing, and multimedia applications, where high-quality and noise-free images are indispensable for accurate analysis and interpretation.

Gene expression models are key to understanding and predicting transcriptional dynamics. This thesis devises a computational method which can efficiently explore a large, highly correlated parameter space, ultimately allowing the author to accurately deduce the underlying gene network model using discrete, stochastic mRNA counts derived through the non-invasive imaging method of single molecule fluorescence in situ hybridization (smFISH). An underlying gene network model consists of the number of gene states (distinguished by distinct production rates) and all associated kinetic rate parameters. In this thesis, the author constructs an algorithm based on Bayesian parametric and nonparametric theory, expanding the traditional single gene network inference tools. This expansion starts by increasing the efficiency of classic Markov-Chain Monte Carlo (MCMC) sampling by combining three schemes known in the Bayesian statistical computing community: 1) Adaptive Metropolis-Hastings (AMH), 2) Hamiltonian Monte Carlo (HMC), and 3) Parallel Tempering (PT). The aggregation of these three methods decreases the autocorrelation between sequential MCMC samples, reducing the number of samples required to gain an accurate representation of the posterior probability distribution. Second, by employing Bayesian nonparametric methods, the author is able to simultaneously evaluate discrete and continuous parameters, enabling the method to devise the structure of the gene network and all kinetic parameters, respectively. Due to the nature of Bayesian theory, uncertainty is evaluated for the gene network model in combination with the kinetic parameters. Tools brought from Bayesian nonparametric theory equip the method with an ability to sample from the posterior distribution of all possible gene network models without pre-defining the gene network structure, i.e. the number of gene states. The author verifies the method’s robustness through the use of synthetic snapshot data, designed to closely represent experimental smFISH data sets, across a range of gene network model structures, parameters and experimental settings (number of probed cells and timepoints).

This thesis addresses the problem of approximating analytic functions over general and compact multidimensional domains. Although the methods we explore can be used in complex domains, most of the tests are performed on the interval $[-1,1]$ and the square $[-1,1]\times[-1,1]$. Using Fourier and polynomial frame approximations on an extended domain, well-conditioned methods can be formulated. In particular, these methods provide exponential decay of the error down to a finite but user-controlled tolerance $\epsilon>0$. Additionally, this thesis explores two implementations of the frame approximation: a singular value decomposition (SVD)-regularized least-squares fit as described by Adcock and Shadrin in 2022, and a column and row selection method that leverages QR factorizations to reduce the data needed in the approximation. Moreover, strategies to reduce the complexity of the approximation problem by exploiting randomized linear algebra in low-rank algorithms are also explored, including the AZ algorithm described by Coppe and Huybrechs in 2020.

This dissertation explores applications of machine learning methods in service of the design of screening tests, which are ubiquitous in applications from social work, to criminology, to healthcare. In the first part, a novel Bayesian decision theory framework is presented for designing tree-based adaptive tests. On an application to youth delinquency in Honduras, the method produces a 15-item instrument that is almost as accurate as a full-length 150+ item test. The framework includes specific considerations for the context in which the test will be administered, and provides uncertainty quantification around the trade-offs of shortening lengthy tests. In the second part, classification complexity is explored via theoretical and empirical results from statistical learning theory, information theory, and empirical data complexity measures. A simulation study that explicitly controls two key aspects of classification complexity is performed to relate the theoretical and empirical approaches. Throughout, a unified language and notation that formalizes classification complexity is developed; this same notation is used in subsequent chapters to discuss classification complexity in the context of a speech-based screening test. In the final part, the relative merits of task and feature engineering when designing a speech-based cognitive screening test are explored. Through an extensive classification analysis on a clinical speech dataset from patients with normal cognition and Alzheimer’s disease, the speech elicitation task is shown to have a large impact on test accuracy; carefully performed task and feature engineering are required for best results. A new framework for objectively quantifying speech elicitation tasks is introduced, and two methods are proposed for automatically extracting insights into the aspects of the speech elicitation task that are driving classification performance. The dissertation closes with recommendations for how to evaluate the obtained insights and use them to guide future design of speech-based screening tests.

During the inversion of discrete linear systems, noise in data can be amplified and result in meaningless solutions. To combat this effect, characteristics of solutions that are considered desirable are mathematically implemented during inversion. This is a process called regularization. The influence of the provided prior information is controlled by the introduction of non-negative regularization parameter(s). Many methods are available for both the selection of appropriate regularization parame- ters and the inversion of the discrete linear system. Generally, for a single problem there is just one regularization parameter. Here, a learning approach is considered to identify a single regularization parameter based on the use of multiple data sets de- scribed by a linear system with a common model matrix. The situation with multiple regularization parameters that weight different spectral components of the solution is considered as well. To obtain these multiple parameters, standard methods are modified for identifying the optimal regularization parameters. Modifications of the unbiased predictive risk estimation, generalized cross validation, and the discrepancy principle are derived for finding spectral windowing regularization parameters. These estimators are extended for finding the regularization parameters when multiple data sets with common system matrices are available. Statistical analysis of these estima- tors is conducted for real and complex transformations of data. It is demonstrated that spectral windowing regularization parameters can be learned from these new esti- mators applied for multiple data and with multiple windows. Numerical experiments evaluating these new methods demonstrate that these modified methods, which do not require the use of true data for learning regularization parameters, are effective and efficient, and perform comparably to a supervised learning method based on es- timating the parameters using true data. The theoretical developments are validated for one and two dimensional image deblurring. It is verified that the obtained estimates of spectral windowing regularization parameters can be used effectively on validation data sets that are separate from the training data, and do not require known data.

Solving partial differential equations on surfaces has many applications including modeling chemical diffusion, pattern formation, geophysics and texture mapping. This dissertation presents two techniques for solving time dependent partial differential equations on various surfaces using the partition of unity method. A novel spectral cubed sphere method that utilizes the windowed Fourier technique is presented and used for both approximating functions on spherical domains and solving partial differential equations. The spectral cubed sphere method is applied to solve the transport equation as well as the diffusion equation on the unit sphere. The second approach is a partition of unity method with local radial basis function approximations. This technique is also used to explore the effect of the node distribution as it is well known that node choice plays an important role in the accuracy and stability of an approximation. A greedy algorithm is implemented to generate good interpolation nodes using the column pivoting QR factorization. The partition of unity radial basis function method is applied to solve the diffusion equation on the sphere as well as a system of reaction-diffusion equations on multiple surfaces including the surface of a red blood cell, a torus, and the Stanford bunny. Accuracy and stability of both methods are investigated.

Global optimization (programming) has been attracting the attention of researchers for almost a century. Since linear programming (LP) and mixed integer linear programming (MILP) had been well studied in early stages, MILP methods and software tools had improved in their efficiency in the past few years. They are now fast and robust even for problems with millions of variables. Therefore, it is desirable to use MILP software to solve mixed integer nonlinear programming (MINLP) problems. For an MINLP problem to be solved by an MILP solver, its nonlinear functions must be transformed to linear ones. The most common method to do the transformation is the piecewise linear approximation (PLA). This dissertation will summarize the types of optimization and the most important tools and methods, and will discuss in depth the PLA tool. PLA will be done using nonuniform partitioning of the domain of the variables involved in the function that will be approximated. Also partial PLA models that approximate only parts of a complicated optimization problem will be introduced. Computational experiments will be done and the results will show that nonuniform partitioning and partial PLA can be beneficial.

Uncertainty is inherent in predictive decision-making, both with respect to forecasting plausible future conditions based on a historic record, and with respect to backcasting likely upstream states from downstream observations. In the first chapter, I evaluated the status of current water resources management policy in the United States (U.S.) with respect to its integration of projective uncertainty into state-level flooding, drought, supply and demand, and climate guidance. I found uncertainty largely absent and discussed only qualitatively rather than quantitatively. In the second chapter, I turned to uncertainty in the interpretation of downstream observations as indicators of upstream behaviors in the field of Wastewater-Based Epidemiology (WBE), which has made possible the near real-time, yet anonymous, monitoring of public health via measurements of biomarkers excreted to wastewater. I found globally, seasonality of air and soil temperature causes biomarker degradation to vary up to 13-fold over the course of a year, constituting part of the background processes WBE must address, or detrend, prior to decision-making. To determine whether the seasonal change in degradation rates was introducing previously unaccounted for uncertainty with respect to differences in observed summertime and winter-time populations, I evaluated demographic indicators recorded by the Census Bureau for correlation with their distance from all major wastewater treatment plants across the U.S. The analysis identified statistically significant correlation for household income, education attainment, unemployment, military service, and the absence of health insurance. Finally, the model was applied to a city-wide case study to test whether temperature could explain some of the trends observed in monthly observations of two opiate compounds. Modeling suggests some of the monthly changes were attributed to natural temperature fluctuation rather than to trends in the substances’ consumption, and that uncertainty regarding discharge location can dominate even relative observed differences in opiate detections. In summary, my work has found temperature an important modulator of WBE results, influencing both the type of populations observed and the likelihood of upstream behaviors disproportionally magnified or obscured, particularly for the more labile biomarkers. There exists significant potential for improving the understanding of empirical observations via numerical modeling and the application of spatial analysis tools.

I focus on algorithms that generate good sampling points for function approximation. In 1D, it is well known that polynomial interpolation using equispaced points is unstable. On the other hand, using Chebyshev nodes provides both stable and highly accurate points for polynomial interpolation. In higher dimensional complex regions, optimal sampling points are not known explicitly. This work presents robust algorithms that find good sampling points in complex regions for polynomial interpolation, least-squares, and radial basis function (RBF) methods. The quality of these nodes is measured using the Lebesgue constant. I will also consider optimal sampling for constrained optimization, used to solve PDEs, where boundary conditions must be imposed. Furthermore, I extend the scope of the problem to include finding near-optimal sampling points for high-order finite difference methods. These high-order finite difference methods can be implemented using either piecewise polynomials or RBFs.

A specific species of the genus Geobacter exhibits useful electrical properties when processing a molecule often found in waste water. A team at ASU including Dr Cèsar Torres and Dr Sudeep Popat used that species to create a special type of solid oxide fuel cell we refer to as a microbial fuel cell. Identification of possible chemical processes and properties of the reactions used by the Geobacter are investigated indirectly by taking measurements using Electrochemical Impedance Spectroscopy of the electrode-electrolyte interface of the microbial fuel cell to obtain the value of the fuel cell's complex impedance at specific frequencies. Investigation of the multiple polarization processes which give rise to measured impedance values is difficult to do directly and so examination of the distribution function of relaxation times (DRT) is considered instead. The DRT is related to the measured complex impedance values using a general, non-physical equivalent circuit model. That model is originally given in terms of a Fredholm integral equation with a non-square integrable kernel which makes the inverse problem of determining the DRT given the impedance measurements an ill-posed problem. The original integral equation is rewritten in terms of new variables into an equation relating the complex impedance to the convolution of a function based upon the original integral kernel and a related but separate distribution function which we call the convolutional distribution function. This new convolutional equation is solved by reducing the convolution to a pointwise product using the Fourier transform and then solving the inverse problem by pointwise division and application of a filter function (equivalent to regularization). The inverse Fourier transform is then taken to get the convolutional distribution function. In the literature the convolutional distribution function is then examined and certain values of a specific, less general equivalent circuit model are calculated from which aspects of the original chemical processes are derived. We attempted to instead directly determine the original DRT from the calculated convolutional distribution function. This method proved to be practically less useful due to certain values determined at the time of experiment which meant the original DRT could only be recovered in a window which would not normally contain the desired information for the original DRT. This limits any attempt to extend the solution for the convolutional distribution function to the original DRT. Further research may determine a method for interpreting the convolutional distribution function without an equivalent circuit model as is done with the regularization method used to solve directly for the original DRT.