Finding the M Most Probable Configurations in Arbitrary Graphical Models

Abstract: Loopy belief propagation (BP) has been successfully used in a numberof di#cult graphical models to find the most probable configurationof the hidden variables. In applications ranging from proteinfolding to image analysis one would like to find not just the bestconfiguration but rather the top M . While this problem has beensolved using the junction tree formalism, in many real world problemsthe clique size in the junction tree is prohibitively large. Inthis work we address the...