{"id":{"repo_id":"msu","oai_identifier":"oai:d.lib.msu.edu:etd_53281"},"canonical_url":"https://search.ndltd.org/etd/msu/oai:d.lib.msu.edu:etd_53281","repository":{"repo_id":"msu","name":"Michigan State University","base_url":"https://d.lib.msu.edu/oai"},"display":{"title":"SCALABLE BAYESIAN INFERENCE FOR MODEL UNCERTAINTY WITH AUTOREGRESSIVE FLOWS","abstract":"Thesis (Ph.D.)--Michigan State University. Statistics - Doctor of Philosophy, 2026","abstract_html":"Thesis (Ph.D.)--Michigan State University. Statistics - Doctor of Philosophy, 2026","abstract_has_math":false,"creators":["Zhou, Jiefu"],"institution":null,"degree_name":null,"degree_level":null,"degree_discipline":null,"degree_department":null,"school":null,"contributors":["Bhattacharya, Shrijita SB","Maiti, Tapabrata TM","Boddeti, Vishnu VB","Xing, Yue YX"],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2026,"date_issued":"2026","date_published":"2026","updated_at":"2026-11-18T18:59:44Z","subjects":["Statistics"],"languages":["eng"],"rights":["Attribution 4.0 International"],"rights_urls":[],"identifier_entries":[{"key":"dc:identifier","label":"Identifier","values":["etd:53281","local:Zhou_grad.msu_0128D_22330"],"render_values":[{"text":"etd:53281","href":null,"code":true},{"text":"local:Zhou_grad.msu_0128D_22330","href":null,"code":true}]}],"person_identifier_entries":[]},"links":{"outbound_url":"https://doi.org/doi:10.25335/6pm4-za74","outbound_label":"DOI","outbound_source":"dc:identifier","record_url":"https://doi.org/doi:10.25335/6pm4-za74","record_label":"Repository record","record_kind":"doi","record_source":"dc:identifier","download_url":null,"download_label":null,"download_source":null,"access_status":null,"embargo_until":null,"access_note":null},"source_record":{"url":"https://d.lib.msu.edu/oai?verb=GetRecord&metadataPrefix=oai_dc&identifier=oai%3Ad.lib.msu.edu%3Aetd_53281","prefix":"oai_dc"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["Bhattacharya, Shrijita SB","Maiti, Tapabrata TM","Boddeti, Vishnu VB","Xing, Yue YX"]},{"key":"dc:creator","label":"Author","values":["Zhou, Jiefu"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date","label":"Dc Date","values":["2026"]},{"key":"dc:relation","label":"Dc Relation","values":["Electronic Theses & Dissertations"]},{"key":"dc:type","label":"Dc Type","values":["text","Electronic dissertations"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Statistics"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language","label":"Dc Language","values":["eng"]},{"key":"dc:rights","label":"Dc Rights","values":["Attribution 4.0 International"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["etd:53281","local:Zhou_grad.msu_0128D_22330","https://doi.org/doi:10.25335/6pm4-za74"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description","label":"Description","values":["Thesis (Ph.D.)--Michigan State University. Statistics - Doctor of Philosophy, 2026","Modern Bayesian analysis must often account for uncertainty at two interconnected levels: uncertainty about parameters within a model and uncertainty about the model itself. Although Markov chain Monte Carlo methods provide a general route to posterior inference, their computational burden can become substantial in high-dimensional or multimodal problems. Conventional variational methods are more scalable, but commonly used mean-field families can suppress posterior dependence and underestimate uncertainty. This dissertation develops autoregressive-flow variational methods that seek a more favorable balance among expressivity, computational scalability, and statistical reliability.The first part introduces \\texttt{FlowSelect}, a framework for high-dimensional Bayesian variable selection with point-mass spike-and-slab priors. Inverse autoregressive flows model dependence among regression coefficients and variable-inclusion indicators, while Binary Concrete relaxations permit gradient-based optimization over the discrete model space. The development begins with a Gaussian-slab formulation, whose conjugate quadratic structure provides a transparent setting in which to establish the core flow construction, posterior guarantees, and an updated variational-EM algorithm. It then advances to the principal Laplace-slab formulation, which introduces heavier tails and an $\\ell_1$-type shrinkage geometry. For this extension, strong selection consistency and posterior contraction are established for the exact posterior under growing sparsity, followed by a Kullback--Leibler approximation bound showing that an ideal flow-aided variational posterior inherits the same asymptotic selection and contraction properties. Extensive simulations and real-data analyses examine variable selection, coefficient recovery, uncertainty quantification, and computation.The second part develops Neural Autoregressive Flow Bayesian Model Averaging (\\texttt{NAF-BMA}). By replacing affine autoregressive transformations with monotone neural transformers, the method can represent more complicated, including multimodal, within-model posteriors while jointly estimating model evidences and posterior model probabilities. Applications to linear and logistic regression, finite mixtures, crime data, nuclear-mass prediction, and heart-disease data illustrate the generality of the construction. Taken together, the results show how autoregressive flows can provide a scalable framework for Bayesian inference under model uncertainty, while clarifying the statistical and computational trade-offs that accompany increasingly expressive variational families.","Description based on online resource. Title from PDF t.p. (Michigan State University Fedora Repository, viewed ).","Includes bibliographical references."]},{"key":"dc:format","label":"Dc Format","values":["application/pdf","1 online resource (194 p.) : ill."]},{"key":"dc:title","label":"Title","values":["SCALABLE BAYESIAN INFERENCE FOR MODEL UNCERTAINTY WITH AUTOREGRESSIVE FLOWS"]}]}],"canonical_facts":{"dc:contributor":["Bhattacharya, Shrijita SB","Maiti, Tapabrata TM","Boddeti, Vishnu VB","Xing, Yue YX"],"dc:creator":["Zhou, Jiefu"],"dc:date":["2026"],"dc:description":["Thesis (Ph.D.)--Michigan State University. Statistics - Doctor of Philosophy, 2026","Modern Bayesian analysis must often account for uncertainty at two interconnected levels: uncertainty about parameters within a model and uncertainty about the model itself. Although Markov chain Monte Carlo methods provide a general route to posterior inference, their computational burden can become substantial in high-dimensional or multimodal problems. Conventional variational methods are more scalable, but commonly used mean-field families can suppress posterior dependence and underestimate uncertainty. This dissertation develops autoregressive-flow variational methods that seek a more favorable balance among expressivity, computational scalability, and statistical reliability.The first part introduces \\texttt{FlowSelect}, a framework for high-dimensional Bayesian variable selection with point-mass spike-and-slab priors. Inverse autoregressive flows model dependence among regression coefficients and variable-inclusion indicators, while Binary Concrete relaxations permit gradient-based optimization over the discrete model space. The development begins with a Gaussian-slab formulation, whose conjugate quadratic structure provides a transparent setting in which to establish the core flow construction, posterior guarantees, and an updated variational-EM algorithm. It then advances to the principal Laplace-slab formulation, which introduces heavier tails and an $\\ell_1$-type shrinkage geometry. For this extension, strong selection consistency and posterior contraction are established for the exact posterior under growing sparsity, followed by a Kullback--Leibler approximation bound showing that an ideal flow-aided variational posterior inherits the same asymptotic selection and contraction properties. Extensive simulations and real-data analyses examine variable selection, coefficient recovery, uncertainty quantification, and computation.The second part develops Neural Autoregressive Flow Bayesian Model Averaging (\\texttt{NAF-BMA}). By replacing affine autoregressive transformations with monotone neural transformers, the method can represent more complicated, including multimodal, within-model posteriors while jointly estimating model evidences and posterior model probabilities. Applications to linear and logistic regression, finite mixtures, crime data, nuclear-mass prediction, and heart-disease data illustrate the generality of the construction. Taken together, the results show how autoregressive flows can provide a scalable framework for Bayesian inference under model uncertainty, while clarifying the statistical and computational trade-offs that accompany increasingly expressive variational families.","Description based on online resource. Title from PDF t.p. (Michigan State University Fedora Repository, viewed ).","Includes bibliographical references."],"dc:format":["application/pdf","1 online resource (194 p.) : ill."],"dc:identifier":["etd:53281","local:Zhou_grad.msu_0128D_22330","https://doi.org/doi:10.25335/6pm4-za74"],"dc:language":["eng"],"dc:relation":["Electronic Theses & Dissertations"],"dc:rights":["Attribution 4.0 International"],"dc:subject":["Statistics"],"dc:title":["SCALABLE BAYESIAN INFERENCE FOR MODEL UNCERTAINTY WITH AUTOREGRESSIVE FLOWS"],"dc:type":["text","Electronic dissertations"]},"updated_at":"2026-11-18T18:59:44Z"}