Center for Machine Perception, Department of Cybernetics, FEL CVUT, Prague, Czech Republic
Inference tasks on graphical models (in particular, Markov random networks) include computing modes and marginals of Gibbs ditribution. Algorithmically, these tasks lead to labeling problems. I want to discuss some suboptimal approaches to these problems for large and sparse instances, occuring e.g. in machine vision and pattern recognition.