Introduce random variables
1. Define random variable ***z*** in lower dimension with Guassian distribution
$
p(\mathbf{z}) = \mathcal{N}(\mathbf{z}|0,\mathbf{I})
$
2. Observed random variable ***x*** has a conditional distribution given by:
$
p(\mathbf{x|z}) = \mathcal{N}(\mathbf{x}|\mathbf{Wz}+\boldsymbol{\mu}, \sigma^2\mathbf{I})$where $\mathbf{x} = \mathbf{Wz}+\boldsymbol{\mu}+\boldsymbol{\epsilon},\quad\boldsymbol{\epsilon}~\mathcal{N}(\mathbf{z}|0,\sigma^2\mathbf{I})
$
4. Marginal distribution of data:
$
p(\mathbf{x}) = \int p(\mathbf{x|z})p(\mathbf{z})d\mathbf{z}=\mathcal{N}(\mathbf{x}|\boldsymbol{\mu},\mathbf{C})
$
where
$
\mathbf{C} = \mathbf{W}\mathbf{W}^T+\sigma^2\mathbf{I}
$
5. Posterior distribution is:
$
p(\mathbf{z|x})=\mathcal{N}(\mathbf{z}|\mathbf{M}^{-1}\mathbf{W}^T(\mathbf{x}-\boldsymbol{\mu}),\sigma^2\mathbf{M})
$
Estimate parameters
- $\boldsymbol{\mu}_{ML}=\bar{\boldsymbol{x}}$
- sample mean
- $\mathbf{W}_{ML} = \mathbf{U}_M(\mathbf{L}_M-\sigma^2\mathbf{I})^{\frac{1}{2}}\mathbf{R}$
- $\mathbf{U}_M$ - eigenvector matrix
- $\mathbf{L}_M$ - matrix of eigenvalues along diagonal
- $\mathbf{R}$ - arbitrary rotation matrix
- $\sigma^2_{ML} = \frac{1}{D-M}\sum_{i=M+1}^D\lambda{i}$
- average of eigenvalues we don't use
Can use [[Expectation Maximization (EM) Algorithm|EM Algorithm]]
- computational advantages in high dimensions
- can handle missing data
- required for factor analysis