 |
Center for Nonlinear Studies |
Graphical models are widely used to represent statistical dependence among a set of random variables. Learning the graphical model structure given the (empirical) distribution of random variables is a well known and well studied problem. Though this is hard in general, there are certain cases where this is tractable. For instance, the Chow-Liu algorithm tells us how to efficiently compute the best tree graphical model for a given distribution. In this talk, we propose a greedy heuristic for obtaining the best zero external magnetic field planar Ising model and best outer planar Ising model with external magnetic field. Our algorithm exploits the fact that inference on such models is tractable. We also present simulation results for the same.