Midterm Review The Midterm • Everything we have talked about so far • Stuff from HW • I won’t ask you to do as complicated calculations as the HW • Don’t need a calculator • No books / notes Maximum Likelihood Estimation • How to apply the maximum likelihood principle – log likelihood + derivative + solve for 0 – You should know how to do this for Bernoulli trials and 1-D Gaussian • Conjugate distributions – Dirichlet, Beta Mixture Models and EM • What does the EM algorithm do? – Understand the E-step and M-step • Log-exp-sum trick – You should be able to derive this – You should understand why we need to use it Hidden Markov Models • Viterbi – What does it do? – What is the running time? • Forward-backward – What does it do? • Be able to compute the probability of a “parse” – Joint probability of a sequence of observed and hidden states Bayesian Networks • Understand d-separation criteria • Be able to answer simple questions about whether variables are independent given some evidence • Markov Blanket Markov Networks / Belief Propagation • Moralizing a graph (convert Bayesian network into Markov Network) • Belief propagation – What does it do, when is it guaranteed to converge to the correct posterior distribution.