Problem
1. Consider a particular parameterization (θ, η) to Max-margin. Show how we can use second-best MAP inference to either find a violated constraint or guarantee that all constraints are satisfied.
2. Let H∗ be a Markov network where the maximum degree of a node is d∗. Show that if we have an infinitely large data set D generated from H∗ (so that independence tests are evaluated perfectly), then the Build-PMap-Skeleton procedure of algorithm reconstructs the correct Markov structure H∗.