Bayesian methods for system reliability and community detection

Thumbnail Image
Date
2011-01-01
Authors
Guo, Jiqiang
Major Professor
Advisor
Alyson G. Wilson
Committee Member
Journal Title
Journal ISSN
Volume Title
Publisher
Altmetrics
Authors
Research Projects
Organizational Units
Organizational Unit
Statistics
As leaders in statistical research, collaboration, and education, the Department of Statistics at Iowa State University offers students an education like no other. We are committed to our mission of developing and applying statistical methods, and proud of our award-winning students and faculty.
Journal Issue
Is Version Of
Versions
Series
Department
Abstract

Bayesian methods are valuable for their natural incorporation of prior information and their practical convenience for modeling and estimation. This dissertation develops flexible Bayesian parametric methods for system reliability and Bayesian nonparametric models for community detection.

The Bayesian parametric models proposed allow the assessment of system reliability for multi-component systems simultaneously. We start with a model that considers lifetime data at every component. Then we generalize to a unified framework with heterogeneous information. We demonstrate this unified methodology with pass/fail, lifetime, and degradation data at both the system level and the component level. Further, we propose a Bayesian melding approach to combine prior information from multiple levels.

For community detection, we propose a series of statistical models based on Bayesian nonparametric techniques. These statistical models provide a natural approach for identifying communities in networks using only data on edges. We take advantage of the Bayesian nonparametric approach to include an important feature in our models: the number of communities is an implied parameter of the model, which is therefore inferred during estimation. We also introduce an “Erdős Rényi” group for those nodes that do not belong to communities. Other important aspects of this series of models include increasing flexibility of modeling probabilities for edge presence, linking these probabilities to community sizes, and obtaining communities from posterior samples under a decision theory framework. When presenting our models, we discuss model selection and model checking, which are necessary considerations when applying statistical approaches to real problems.

Comments
Description
Keywords
Citation
Source
Subject Categories
Copyright
Sat Jan 01 00:00:00 UTC 2011