Increasingly correct message passing averaging algorithms

We study averaging algorithms, when implemented in large networks of wirelessly connected elements. We extend the notion of ¿increasing correctness¿ (IC) which was defined for cycle-free graphs, to general graphs. An averaging algorithm that is IC has meaningful outputs at each iteration. This makes it possible to stop the algorithm at any time, and use the output values computed up to that time. We prove that the class of IC averaging algorithms is nontrivial. We then present a simple IC averaging algorithm that is based on ideas from graphical models, and study its properties. Finally, we give example applications and simulations of IC averaging algorithms.