Optimizing Reweighted Belief Propagation for Distributed Likelihood Fusion Problems
نویسندگان
چکیده
Belief propagation (BP) is a powerful tool to solve distributed inference problems, though it is limited by short cycles in the corresponding factor graph. Such cycles may lead to incorrect solutions or oscillatory behavior. Only for certain types of problems are convergence properties understood. We extend this knowledge by investigating the use of reweighted BP for likelihood consensus problems, which are characterized by equality constraints along possibly short cycles. Through a linear formulation of BP, we are able to analytically derive convergence conditions for certain types of graphs and optimize the convergence speed. We compare with standard belief consensus and observe significantly faster convergence.
منابع مشابه
Information Fusion with Distributed Probabilistic Networks
We introduce a robust approach to diagnostic information fusion within a network of probabilistic models distributed throughout a system of agents. In particular, we introduce an algorithm for inter-agent belief propagation, which combines partial fusion results from local probabilistic models in a consistent manner. This approach allows sequential fusion of large amounts of heterogeneous infor...
متن کاملIn-Network Belief Propagation on Sensor Networks
Sensor Networks provide a cheap, unobtrusive, and easy to deploy method for gathering large amounts of data from an environment. However, this data is often significantly noisy. One way to improve the confidence in the data is to build a model of the correlations among nearby sensor readings and do inference on this model to derive data with less error. Previous methods of inference on sensor n...
متن کاملTree-reweighted belief propagation algorithms and approximate ML estimation by pseudo-moment matching
In previous work [10], we presented a class of upper bounds on the log partition function of an arbitrary undirected graphical model based on solving a convex variational problem. Here we develop a class of local message-passing algorithms, which we call tree-reweighted belief propagation, for efficiently computing the value of these upper bounds, as well as the associated pseudomarginals. We a...
متن کاملLinear Programming Relaxations and Belief Propagation - An Empirical Study
The problem of finding the most probable (MAP) configuration in graphical models comes up in a wide range of applications. In a general graphical model this problem is NP hard, but various approximate algorithms have been developed. Linear programming (LP) relaxations are a standard method in computer science for approximating combinatorial problems and have been used for finding the most proba...
متن کامل