Learning structurally consistent undirected probabilistic graphical models

In many real-world domains, undirected graphical models such as Markov random fields provide a more natural representation of the dependency structure than directed graphical models. For example, Bayesian networks cannot explicitly capture cyclic dependencies which occur commonly in real-world networks such as in biology. Unfortunately, structure learning of undirected graphs using likelihoodbased scores remains difficult because of the intractability of computing the partition function. We describe a new Markov random field structure learning algorithm which is motivated by canonical parameterization of Abbeel et al. We improve on their parameterization by learning per-variable canonical factors, which makes our algorithm suitable for domains with hundreds of nodes. Our algorithm is similar to learning dependency networks, but the learned structure is guaranteed to be consistent, and, therefore represents a consistent joint probability distribution. We compare our algorithm against s...

• Type: Conference
• Year: 2009
• Where: ICML
• Authors: Sushmita Roy, Terran Lane, Margaret Werner-Washburne
• Keywords: ICML 2009, Machine Learning, Structure Learning Algorithm, Undirected Graphical Models, Undirected Graphs