1
$\begingroup$

Given that we partition a Gaussian random vector $\textbf{x}$ into three groups, $\textbf{x}_a$, $\textbf{x}_b$, and $\textbf{x}_c$, with a corresponding partitioning of the mean vector $\mu$ and of the covariance matrix $\Sigma$ in the form

$\mu = \begin{pmatrix} \mu_a\\ \mu_b\\ \mu_c\\ \end{pmatrix}, \enspace \Sigma=\begin{pmatrix} \Sigma_{aa} & \Sigma_{ab} & \Sigma_{ac} \\ \Sigma_{ba} & \Sigma_{bb} & \Sigma_{bc} \\ \Sigma_{ca} & \Sigma_{cb} & \Sigma_{cc} \\ \end{pmatrix} $

What is the expression for the conditional distribution $P(x_a|x_b)$ in which $x_c$ has been marginalized out? (Note the 3 groups are also multivariate)

$\endgroup$
1
$\begingroup$

Marginalising out $x_c$ means that you can forget about the last row and column of the covariance and $\mu_c$ in the mean vector.

http://en.wikipedia.org/wiki/Multivariate_normal_distribution#Conditional_distributions $$p(x_a|x_b)=\mathcal{N}\left(\mu_{a|b},\Sigma_{a|b}\right)$$

Where $\mu_{a|b},\Sigma_{a|b}$ is given in the above link.

$\endgroup$
  • $\begingroup$ +1 for good answer, but it's essentially just the first sentence plus a link ... even though it's a standard result I think it would be better to give the values in an answer, perhaps with a link to a question which gives more details, like this one $\endgroup$ – Glen_b Mar 19 '14 at 7:26

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.