Covariant matrix with zero eigenvalues

by Michael   Last Updated September 12, 2019 00:19 AM

For computing log-likelihood of points being in a certain multivariate distribution I am trying to invert the covariance matrix $\Sigma$. Unfortunately, $\Sigma$ turned out to be singular, with multiple zero eigenvalues. Using pseudo-inverse leads to very low log-likelihood, which is expected. How could I address this? Would it help to reduce the dimension of the problem?



Related Questions





Covariance matrix fit to the curvature

Updated March 26, 2017 05:19 AM

Correlation between two multivariate measures

Updated January 25, 2019 02:19 AM