Matching Items (14)
Description
The introduction of parameterized loss functions for robustness in machine learning has led to questions as to how hyperparameter(s) of the loss functions can be tuned. This thesis explores how Bayesian methods can be leveraged to tune such hyperparameters. Specifically, a modified Gibbs sampling scheme is used to generate a distribution of loss parameters of tunable loss functions. The modified Gibbs sampler is a two-block sampler that alternates between sampling the loss parameter and optimizing the other model parameters. The sampling step is performed using slice sampling, while the optimization step is performed using gradient descent. This thesis explores the application of the modified Gibbs sampler to alpha-loss, a tunable loss function with a single parameter $\alpha \in (0,\infty]$, that is designed for the classification setting. Theoretically, it is shown that the Markov chain generated by a modified Gibbs sampling scheme is ergodic; that is, the chain has, and converges to, a unique stationary (posterior) distribution. Further, the modified Gibbs sampler is implemented in two experiments: a synthetic dataset and a canonical image dataset. The results show that the modified Gibbs sampler performs well under label noise, generating a distribution indicating preference for larger values of alpha, matching the outcomes of previous experiments.
ContributorsCole, Erika Lingo (Author) / Sankar, Lalitha (Thesis advisor) / Lan, Shiwei (Thesis advisor) / Pedrielli, Giulia (Committee member) / Hahn, Paul (Committee member) / Arizona State University (Publisher)
Created2022
Description
Uncertainty Quantification (UQ) is crucial in assessing the reliability of predictivemodels that make decisions for human experts in a data-rich world. The Bayesian approach to UQ for inverse problems has gained popularity. However, addressing UQ in high-dimensional inverse problems is challenging due to the intensity and inefficiency of Markov Chain Monte Carlo (MCMC) based Bayesian inference methods. Consequently, the first primary focus of this thesis is enhancing efficiency and scalability for UQ in inverse problems.
On the other hand, the omnipresence of spatiotemporal data, particularly in areas like traffic analysis, underscores the need for effectively addressing inverse problems with spatiotemporal observations. Conventional solutions often overlook spatial or temporal correlations, resulting in underutilization of spatiotemporal interactions for parameter learning. Appropriately modeling spatiotemporal observations in inverse problems thus forms another pivotal research avenue.
In terms of UQ methodologies, the calibration-emulation-sampling (CES) scheme has emerged as effective for large-dimensional problems. I introduce a novel CES approach by employing deep neural network (DNN) models during the emulation and sampling phase. This approach not only enhances computational efficiency but also diminishes sensitivity to training set variations. The newly devised “Dimension- Reduced Emulative Autoencoder Monte Carlo (DREAM)” algorithm scales Bayesian UQ up to thousands of dimensions in physics-constrained inverse problems. The algorithm’s effectiveness is exemplified through elliptic and advection-diffusion inverse problems.
In the realm of spatiotemporal modeling, I propose to use Spatiotemporal Gaussian processes (STGP) in likelihood modeling and Spatiotemporal Besov processes (STBP) in prior modeling separately. These approaches highlight the efficacy of incorporat- ing spatial and temporal information for enhanced parameter estimation and UQ. Additionally, the superiority of STGP is demonstrated compared to static and time- averaged methods in time-dependent advection-diffusion partial differential equation
(PDE) and three chaotic ordinary differential equations (ODE). Expanding upon Besov Process (BP), a method known for sparsity-promotion and edge-preservation, STBP is introduced to capture spatial data features and model temporal correlations by replacing the random coefficients in the series expansion with stochastic time functions following Q-exponential process(Q-EP). This advantage is showcased in dynamic computerized tomography (CT) reconstructions through comparison with classic STGP and a time-uncorrelated approach.
ContributorsLi, Shuyi (Author) / Lan, Shiwei (Thesis advisor) / Hahn, Paul (Committee member) / McCulloch, Robert (Committee member) / Dan, Cheng (Committee member) / Lopes, Hedibert (Committee member) / Arizona State University (Publisher)
Created2023
Description
When most people think of Phoenix, Arizona, they think of sprawling cityscapesand hot desert mountains full of saguaros and other cacti. They rarely think of water and
fish, and yet, the Arizona landscape is home to many lakes, ponds, rivers and streams,
full of both native fish and sportfish, including in the urban areas. According to the report
by DeSemple in 2006, between the years 2001 and 2006, the Rio Salado Environmental
Restoration Project worked to revitalize the dry river bed that runs through Phoenix, that
included the construction of two urban ponds, the Demonstration Pond and the Reservoir
Pond. At the start of this study, it was unknown what vertebrate species inhabited these
ponds, but it was known that these urban ponds have been used to dump unwanted
aquatic pets. The bluegill Lepomis macrochirus was found to reside in both ponds, and as
it is such an important sportfish species, it was chosen as the focal species for these
studies, which took place over periods in March, May, July, and September of 2021.
Single-season occupancy models were used to attempt to determine how L. macrochirus,
use the microhabitats within the system, and a multi-season model was used to estimate
their recruitment, and seasonal changes in occupancy. In addition, this study also
attempts to understand the size structures of the L. macrochirus population in the
Reservoir Pond and the population in the Demonstration Pond, and if that size structure
varies from March to September. As the populations of these ponds are physically
isolated from one another, statistical tests were also done to determine if the size
structures of the two populations of L. macrochirus differ from one another and found
that the two populations do indeed differ from one another, but only during two of the
sampling periods.
ContributorsKeister, Emily Jan (Author) / Saul, Steven (Thesis advisor) / Bateman, Heather (Committee member) / Suzart de Albuquerque, Fabio (Committee member) / Arizona State University (Publisher)
Created2022
Description
Farmers markets (FMs) serve an important role in local food economies. FMs are multi-scalar operations that involve a number of decision makers: farmers, market managers, and local residents. FMs provide economic benefits to individual farmers, as they serve as a marketplace where local and regional growers and producers can sell products to customers, yet, unlike traditional retailers who have devoted merchandising managers, FMs are constrained by a lack of operational efficiencies that would allow FMs to effectively mimic this marketing strategy to increase profitability. The purpose of this study is assess how FM managers can optimize sales revenue at their markets and expand market reach to increase traffic to their markets. We assemble a revenue history from market vendors for the years 2016-2019 and perform a portfolio optimization problem. This approach assumes that a FM’s decision of which vendors to allow to sell at the market is akin to an investor’s problem of deciding which assets to hold in an investment portfolio. In a case study of a farmers market in the Southwest, we find that the current vendor mix is sub optimal and lies much below the efficient frontier. The implications of these results for FM managers is improvements can be made by changing the vendor mix to match one of the portfolios that lie along the efficient frontier.
ContributorsDavid, Raphael (Author) / Chenarides, Lauren (Thesis director) / Hahn, Paul (Committee member) / Mallory, Mindy (Committee member) / Barrett, The Honors College (Contributor) / Industrial Engineering (Contributor)
Created2022-05
Description
Corynorhinus townsendii, a bat species residing in north-central Arizona, has historically been observed hibernating in highly ventilated areas within caves and abandoned mines, but there is little to no specific data regarding this tendency. Understanding how air movement may influence hibernacula selection is critical in bettering conservation efforts for Arizona bats, especially with white-nose syndrome continuing to devastate bat species populations throughout the United States. My study aimed to begin filling in this knowledge gap. I measured wind speed in three known Arizona hibernacula during the winter hibernation season and combined this data with the locations of bats observed throughout each of the three survey locations. I modeled our findings using a generalized linear model, which confirmed that wind speed is indeed a predictor of C. townsendii roost selection.
ContributorsKitchel, Heidi (Author) / Moore, Marianne (Thesis director) / Saul, Steven (Committee member) / Barrett, The Honors College (Contributor) / College of Integrative Sciences and Arts (Contributor)
Created2022-05
Description
Humans cooperate at levels unseen in other species. Identifying the adaptive mechanisms driving this unusual behavior, as well as how these mechanisms interact to create complex cooperative patterns, remains an open question in anthropology. One impediment to such investigations is that complete, long-term datasets of human cooperative behaviors in small-scale societies are hard to come by; such field research is often hindered both by humans' long lifespans and by the difficulties of collecting data in remote societies. In this study, I attempted to overcome these methodological challenges by simulating individual human cooperative behaviors in a small-scale population. Using an agent-based model tuned to population-level measurements from a real-life marine subsistence population in the southern Philippines, I generated dynamic daily cooperative behaviors in a hypothetical subsistence population over a period of 1500 years and 42 overlapping generations. Preliminary findings from the model suggest that, while the agent-based model broadly captured a number of characteristic population-level patterns in the subsistence population, it did not fully replicate nuances of the population's observed cooperative behaviors. In particular, statistical models of the simulated data identified reciprocity-based and need-based cooperative behaviors but did not detect kinship-motivated cooperation, despite the fact that kin cooperation traits evolved positively and reciprocity cooperation traits evolved negatively over time in the agent population. It is possible that this discrepancy reflects a complex interaction between kinship and reciprocity in the agent-based model. On the other hand, it may also suggest that these types of statistical models, which are frequently utilized in human cooperation studies in the anthropological literature, do not reliably discriminate between kin-based and reciprocity-based cooperation mechanisms when both exist in a population. Even so, the completeness of the simulated data enabled use of more complex statistical methodologies which were able to disentangle the relative effects of cooperative mechanisms operating at different decision levels. By addressing remaining pattern-matching issues, future iterations of the agent-based model may prove to be a useful tool for validating empirical research and investigating novel hypotheses about the evolution and maintenance of cooperative behaviors in human populations.
ContributorsPhelps, Julia R. (Author) / Reiser, Mark (Thesis advisor) / Saul, Steven (Thesis advisor) / Morgan, Thomas (Committee member) / Arizona State University (Publisher)
Created2023
Description
As the future for our planet it is our job to understand what is happening in our world especially in an age of technology. We have the world at our fingertips yet many of us do not know what is happening around the world. Anthropogenic and natural threats are wreaking havoc on the sea turtle population from coral bleaching to bycatch(Shaver et al., 2020). We have come together as a population to reduce the amount of plastic straws in the oceans, but many have stopped there. Not realizing that sea turtles are keystone species that keep the oceans and the wildlife within it healthy (Why do sea turtles matter? 2020). All are listed under the Endangered Species Act but some of those most threatened species are the Kemp’s Ridley sea turtles, which are critically endangered and Leatherbacks which are endangered (Bandimere, 2020). Sea Life Survival is a board game that creates an interactive learning experience based on many different childhood favorite games. Learning takes on many forms and fun is definitely the best way to learn (de Freitas, 2018). This game provides a way to gain information while also experiencing an engaging and entertaining adventure. Understanding why people play games helped create a game that met the components of intention and enjoyment in order to produce a game that people would want to play (Hamari & Keronen, 2017). The purpose of the game is to spread information on sea turtles in a way that presents them in a light hearted way while still touching on the tragic life that some sea turtles succumb to. Future improvements to the game would include party packs which would showcase the new knowledge that has been discovered.
ContributorsMapp, Quiarrah (Author) / Saul, Steven (Thesis director) / Bateman, Heather (Committee member) / Barrett, The Honors College (Contributor) / College of Integrative Sciences and Arts (Contributor)
Created2022-05
Description
The Salt River wild horses are a historic population of unbranded, unclaimed, wild and free-roaming horses, that were born in the wild and merit protection within our National Forest and protection of the Wild Horse and Burro act of 1970. Terms like undomesticated or feral are thrown around in place of “wild”. The past couple of decades or so, there has been an ongoing debate about the current state of the horses on the range. The horses that are along the Salt River, are considered to be state protected and not federally protected, which has sparked a vast discussion on the social, ethical and moral aspects. There has been an overabundance of horses on the range and are causing potential issues to the environment and other farmland. According to the BLM, wild horse and burro populations have a demonstrated ability to grow at 18-20 percent per year. With the widespread and overabundance that is occurring with the horses and burros, it has been said to have a great ecological cost on the rangeland ecosystem by overgrazing native plants, exacerbating invasive establishment and out-competing other ungulates like cattle. Overabundant free-roaming horse and burro populations have large and growing economic and ecological costs for the American public. Without effective management actions, horse and burro populations will double within the next 4-5 years. In this project, with the help of Dr. Julie Murphree, the Salt River Horse Management group and Arizona’s State Liaison for the Department of Agriculture, I conducted various ride-a-longs and conducted my own literature study to further solidify the knowledge I gained when navigating through the Salt River Wild Horse Management group. I can use their data as well as my own observations in the field to catalog their behaviors and look for any signs that would give reason to why this method of population control may or may not be used. I was able to note the horses in their “natural state” which would give me the opportunity to see any behavior changes in various population groups (or otherwise known as Bands). The main objective of this paper is to understand PZP as a population control tool and the effect it has on the Salt River Horses in Arizona.
ContributorsRendon, Chyna (Author) / Murphree, Julie (Thesis director) / Saul, Steven (Committee member) / Barrett, The Honors College (Contributor) / College of Integrative Sciences and Arts (Contributor)
Created2022-05
Description
This dissertation develops versatile modeling tools to estimate causal effects when conditional unconfoundedness is not immediately satisfied. Chapter 2 provides a brief overview ofcommon techniques in causal inference, with a focus on models relevant to the data explored
in later chapters. The rest of the dissertation focuses on the development of novel “reduced
form” models which are designed to assess the particular challenges of different datasets.
Chapter 3 explores the question of whether or not forecasts of bankruptcy cause bankruptcy.
The question arises from the observation that companies issued going concern opinions were
more likely to go bankrupt in the following year, leading people to speculate that the opinions themselves caused the bankruptcy via a “self-fulfilling prophecy”. A Bayesian machine
learning sensitivity analysis is developed to answer this question. In exchange for additional
flexibility and fewer assumptions, this approach loses point identification of causal effects
and thus a sensitivity analysis is developed to study a wide range of plausible scenarios of
the causal effect of going concern opinions on bankruptcy. Reported in the simulations are
different performance metrics of the model in comparison with other popular methods and a
robust analysis of the sensitivity of the model to mis-specification. Results on empirical data
indicate that forecasts of bankruptcies likely do have a small causal effect.
Chapter 4 studies the effects of vaccination on COVID-19 mortality at the state level in
the United States. The dynamic nature of the pandemic complicates more straightforward
regression adjustments and invalidates many alternative models. The chapter comments on
the limitations of mechanistic approaches as well as traditional statistical methods to epidemiological data. Instead, a state space model is developed that allows the study of the
ever-changing dynamics of the pandemic’s progression. In the first stage, the model decomposes the observed mortality data into component surges, and later uses this information in
a semi-parametric regression model for causal analysis. Results are investigated thoroughly
for empirical justification and stress-tested in simulated settings.
ContributorsPapakostas, Demetrios (Author) / Hahn, Paul (Thesis advisor) / McCulloch, Robert (Committee member) / Zhou, Shuang (Committee member) / Kao, Ming-Hung (Committee member) / Lan, Shiwei (Committee member) / Arizona State University (Publisher)
Created2023
Description
This dissertation centers on Bayesian Additive Regression Trees (BART) and Accelerated BART (XBART) and presents a series of models that tackle extrapolation, classification, and causal inference challenges. To improve extrapolation in tree-based models, I propose a method called local Gaussian Process (GP) that combines Gaussian process regression with trained BART trees. This allows for extrapolation based on the most relevant data points and covariate variables determined by the trees' structure. The local GP technique is extended to the Bayesian causal forest (BCF) models to address the positivity violation issue in causal inference. Additionally, I introduce the LongBet model to estimate time-varying, heterogeneous treatment effects in panel data. Furthermore, I present a Poisson-based model, with a modified likelihood for XBART for the multi-class classification problem.
ContributorsWang, Meijia (Author) / Hahn, Paul (Thesis advisor) / He, Jingyu (Committee member) / Lan, Shiwei (Committee member) / McCulloch, Robert (Committee member) / Zhou, Shuang (Committee member) / Arizona State University (Publisher)
Created2024