Learning Higher-Order Graph Structure with Features by Structure Penalty Shilin Ding 1 Department of {

In discrete undirected graphical models, the conditional independence of node labels Y is specified by the graph structure. We study the case where there is another input random vector X (e.g.