Problem
Consider the problem of applying EM to parameter estimation for a variable X whose local probabilistic model is a tree-CPD. We assume that the network structure G includes the structure of the tree-CPDs in it, so that we have a structure T for X. We are given a data set D with some missing values, and we want to run EM to estimate the parameters of T. Explain how we can adapt the EM algorithm in order to accomplish this task. Describe what expected sufficient statistics are computed in the E-step, and how parameters are updated in the M-step.