Abstract
This article shows how the EM algorithm can be sped up by computing the gradient and Hessian of the observed log-likelihood and then using a Newton algorithm. The gradient and Hessian are computed using two identities associated with the expected log-likelihood. Illustrative examples are provided for co-variance estimation with missing observations and factor analysis.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have