Title Information
Title
Nonparametric and Variable-Dimension Bayesian Mixture Models: Analysis, Comparison, and New Methods
Name: Personal
Name Part
Miller, Jeffrey W.
Role
Role Term: Text
creator
Origin Information
Copyright Date
2014
Physical Description
Extent
xiii, 205 p.
digitalOrigin
born digital
Note
Thesis (Ph.D. -- Brown University (2014)
Name: Personal
Name Part
Harrison, Matthew
Role
Role Term: Text
Director
Name: Personal
Name Part
Geman, Stuart
Role
Role Term: Text
Reader
Name: Personal
Name Part
MacEachern, Steven
Role
Role Term: Text
Reader
Name: Corporate
Name Part
Brown University. Applied Mathematics
Role
Role Term: Text
sponsor
Genre (aat)
theses
Abstract
Many nonparametric Bayesian models can be viewed as an infinite-dimensional limit of a family of finite-dimensional models. However, another way to construct a flexible Bayesian model is to take the same family and put a prior on the dimension --- that is, to use a variable-dimension model --- for example, putting a prior on the number of components in a finite mixture. Using theory and experiments, this thesis analyzes some of the differences and similarities between the nonparametric and variable-dimension approaches, develops new inference algorithms for these models, and explores new variable-dimension models. Primarily, we focus on the Dirichlet process mixture (DPM) and a variable-dimension alternative that we refer to as the mixture of finite mixtures (MFM) model. One of the main differences between DPMs and MFMs is the behavior of the posterior on the number of clusters. We show that for a large class of nonparametric mixtures, including DPMs and Pitman--Yor process mixtures over a wide range of families of component distributions, the posterior on the number of clusters does not concentrate at the true number of components when the data comes from a finite mixture. Meanwhile, it is known that the MFM posterior on the number of components concentrates at the true number, assuming the model is correctly specified. We explore the properties of the MFM, finding that it has many of the same attractive features as the DPM: a simple partition distribution, exchangeability properties, restaurant process, random discrete measure representation, and in certain special cases, a simple stick-breaking representation. As a result, many of the same approximate inference algorithms used for nonparametric mixtures can be easily adapted to the MFM. We also propose two new variable-dimension models: the hierarchical mixture of finite mixtures (HMFM) as an alternative to the hierarchical Dirichlet process (HDP), and the mixture of finite feature models (MFFM) as an alternative to the Indian buffet process (IBP). As with the MFM, these variable-dimension models exhibit some of the same appealing characteristics as their nonparametric counterparts.
Subject
Topic
Bayesian nonparametrics
Subject
Topic
variable-dimension models
Subject
Topic
mixture models
Subject (FAST) (authorityURI="http://id.worldcat.org/fast", valueURI="http://id.worldcat.org/fast/1423727")
Topic
Statistics
Record Information
Record Content Source (marcorg)
RPB
Record Creation Date (encoding="iso8601")
20141006
Language
Language Term: Code (ISO639-2B)
eng
Language Term: Text
English
Identifier: DOI
10.7301/Z0T1520K
Access Condition: rights statement (href="http://rightsstatements.org/vocab/InC/1.0/")
In Copyright
Access Condition: restriction on access
Collection is open for research.
Type of Resource (primo)
dissertations