I wonder that what is the mu and sigma formula MLE(maximum likelihood estimates) for a 3 dimension guassian ? It is the same form as 1 and 2 dimension (+ 1 mu and sigma for the new vector) ?
- Anybody can ask a question
- Anybody can answer
- The best answers are voted up and rise to the top
As often happens with MLEs, it's just what you would expect.
The MLE for the matrix of covariances has the sample size $n$ in the denominator and the entries are the sample variances and sample covariances. This matrix has a Wishart distribution with $n-1$ degrees of freedom.
When I first saw the derivation of the MLE of the $3\times3$ or $p\times p$ variance, I was struck by the technique by which it turned out to be useful to view a scalar as the trace of a $1\times 1$ matrix. See this section: http://en.wikipedia.org/wiki/Estimation_of_covariance_matrices#Maximum-likelihood_estimation_for_the_multivariate_normal_distribution