Conjugate Analysis for the Multivariate Gaussian

I work through Bayesian parameter estimation of the mean for the multivariate Gaussian.

The goal of this post is to derive the likelihood, posterior, and posterior predictive for a multivariate Gaussian model with an unknown mean parameter. Consider the -variate Gaussian,

with density function :

The likelihood is over i.i.d. observations, denoted with the design matrix , is

Assume is known. A common prior for the mean parameter is another Gaussian,

Our goal is to show that is a conjugate prior, meaning the posterior is also Gaussian. The posterior is

which we can write out explicitly as

We can write the terms inside the exponents as

where . Since the posterior is w.r.t. to , we can drop terms that do not depend on —we’ll still retain the Gaussian kernel and can just properly normalize it—and combine like terms:

Now we just complete the square. First, let

Then Eq. is equivalent to

This is the posterior for a multivariate Gaussian with unknown mean. We can write it a standard form, e.g. see (Murphy, 2007), as:

To compute the posterior predictive,

we observe the following:

Notice that and are independent. Intuitively, if I tell you the value of , that tells you nothing about the value of because the distribution on does not contain the parameter . Formally,

Since and are independent, we can simply add the means and covariances of the two distributions when adding the random variables, implying:

  1. Murphy, K. P. (2007). Conjugate Bayesian analysis of the Gaussian distribution. Def, 1(2\sigma2), 16.