Marlin B. Missing Data Problems in Machine Learning

Benjamin M. Marlin, 2008.A thesis submitted in conformity with the requirements for the degree of Doctor of Philosophy.Decision Theory, Inference, and Learning
Optimal Prediction and Minimizing Expected Loss
The Bayesian Framework
The Maximum a Posteriori Framework
The Direct Function Approximation Framework
Empirical Evaluation Procedures
A Theory of Missing Data
Categories of Missing Data
The Missing at Random Assumption and Multivariate Data
Impact of Incomplete Data on Inference
Missing Data, Inference, and Model Misspeci cation
Unsupervised Learning With Random Missing Data
Finite Mixture Models
Dirichlet Process Mixture Models
Factor Analysis and Probabilistic Principal Components Analysis
Mixtures of Factor Analyzers
Unsupervised Learning with Non-Random Missing Data
The Yahoo! Music Data Set
The Jester Data Set
Test Items and Additional Notation for Missing Data
The Finite Mixture/CPT-v Model
The Dirichlet Process Mixture/CPT-v Model
The Finite Mixture/Logit-vd Model
Restricted Boltzmann Machines
Comparison of Results and Discussion
Classi cation With Missing Data
Frameworks for Classi cation With Missing Features
Linear Discriminant Analysis
Logistic Regression
Perceptrons and Support Vector Machines
Basis Expansion and Kernel Methods
Neural Networks
Real Data Experiments and Results