1
$\begingroup$

I am searching for the log normalizer based on the natural parameters for the multivariate gaussian in the exponential family representation. For the univariate gaussian, it is given by $$ a(\eta) = \frac{-\eta_1^2}{4 \eta_2} - \frac{1}{2} log(-2\eta_2) $$ (See e.g. https://www.cs.princeton.edu/courses/archive/fall11/cos597C/lectures/exponential-families.pdf)

For the multivariate gaussian, however, I am completly stuck and couldn't find any ressources stating it. Can someone help me out?

Thanks in advance!

$\endgroup$
1
$\begingroup$

The log normalizer for the multivariate Gaussian is given by \begin{equation} F(\Theta) = \frac{1}{4} tr(\Theta_2^{-1} \Theta_1 \Theta_1^T) - \frac{1}{2} log | \Theta_2 | + \frac{\mathcal{D}}{2} log \pi \end{equation} with $\Theta = (\Theta_1, \Theta_2)$ being the natural parameters and $tr(\cdot)$ denoting the trace of a matrix.

$\endgroup$

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.