The role of normalization in the belief propagation algorithm

A number of problems in statistical physics and computer science can be expressed as the computation of marginal probabilities over a Markov random field. Belief propagation, an iterative message-passing algorithm, computes exactly such marginals when the underlying graph is a tree. But it has gained its popularity as an efficient way to approximate them in the more general case, even if it can exhibits multiple fixed points and is not guaranteed to converge. In this paper, we express a new sufficient condition for local stability of a belief propagation fixed point in terms of the graph structure and the beliefs values at the fixed point. This gives credence to the usual understanding that Belief Propagation performs better on sparse graphs.

2012 International Conference on Control, Automation and Information Sciences (ICCAIS), 2012

Numerous inference problems in statistical physics, computer vision or error-correcting coding theory consist in approximating the marginal probability distributions on Markov Random Fields (MRF). The Belief Propagation (BP) is an accurate solution that is optimal if the MRF is loop free and suboptimal otherwise. In the context of error-correcting coding theory, any Low-Density Parity-Check (LDPC) code has a graphical representation, the Tanner graph, which is a particular MRF. It is used as a media for the BP algortithm to correct the bits, damaged by a noisy channel, by estimating their probability distributions. Though loops and combination thereof in the Tanner graph prevent the BP from being optimal, especially harmful topological structures called the trappingsets. The BP has been extended to the Generalized Belief Propagation (GBP). This message-passing algorithm runs on a non unique mapping of the Tanner graph, namely the regiongraph, such that its nodes are gatherings of the Tanner graph nodes. Then it appears the possibility to decrease the loops effect, making the GBP more accurate than the BP. In this article, we expose a novel region graph construction suited to the Tanner code, an LDPC code whose Tanner graph is entirely covered by trapping-sets. Furthermore, we investigate the dynamic behavior of the GBP compared with that of the BP to understand its evolution in terms of the Signal-to-Noise Ratio (SNR). To this end we make use of classical estimators and we introduce a new one called the hyperspheres method.

2005

In (Wainwright et al., 2002) a new general class of upper bounds on the log partition function of arbitrary undirected graphical models has been developed. This bound is constructed by taking convex combinations of tractable distributions. The experimental results published so far concentrates on combinations of tree-structured distributions leading to a convexified Bethe free energy, which is minimized by the tree-reweighted belief propagation algorithm. One of the favorable properties of this class of approximations is that increasing the complexity of the approximation is guaranteed to increase the precision. The lack of this guarantee is notorious in standard generalized belief propagation. We increase the complexity of the approximating distributions by taking combinations of junction trees, leading to a convexified Kikuchi free energy, which is minimized by reweighted generalized belief propagation. Experimental results for Ising grids as well as for fully connected Ising mode...

Neural Computation, 2004

We derive sufficient conditions for the uniqueness of loopy belief propagation fixed points. These conditions depend on both the structure of the graph and the strength of the potentials and naturally extend those for convexity of the Bethe free energy. We compare them with (a strengthened version of) conditions derived elsewhere for pairwise potentials. We discuss possible implications for convergent algorithms, as well as for other approximate free energies.

2002

Bayesian belief propagation in graphical models has been recently shown to have very close ties to inference methods based in statistical physics. After Yedidia et al. demonstrated that belief propagation fixed points correspond to extrema of the so-called Bethe free energy, Yuille derived a double loop algorithm that is guaranteed to converge to a local minimum of the Bethe free energy. Yuille's algorithm is based on a certain decomposition of the Bethe free energy and he mentions that other decompositions are possible and may even be fruitful. In the present work, we begin with the Bethe free energy and show that it has a principled interpretation as pairwise mutual information minimization and marginal entropy maximization (MIME). Next, we construct a family of free energy functions from a spectrum of decompositions of the original Bethe free energy. For each free energy in this family, we develop a new algorithm that is guaranteed to converge to a local minimum. Preliminary computer simulations are in agreement with this theoretical development.

IEEE Transactions on Information Theory, 2005

Note: This technical report is superseded by MERL TR2004-040, available at http://www.merl.com/papers/TR2004-040/.The region graph method is the most general of these methods, and it subsumes all the other methods. Region graphs also provide the natural graphical setting for GBP algorithms. We explain how to obtain three different versions of GBP algorithms and show that their fixed points will always correspond to stationary points of the region graph approximation to the free energy. We also show that the region graph approximation is exact when the region graph has no cycles. This work may not be copied or reproduced in whole or in part for any commercial purpose. Permission to copy in whole or in part without payment of fee is granted for nonprofit educational and research purposes provided that all such whole or partial copies include the following: a notice that such copying is by permission of Mitsubishi Electric Research Laboratories, Inc.; an acknowledgment of the authors and individual contributions to the work; and all applicable portions of the copyright notice. Copying, reproduction, or republishing for any other purpose shall require a license with payment of fee to Mitsubishi Electric Research Laboratories, Inc. All rights reserved.

Proceedings of the 18th IFAC World Congress, 2011

In this paper, we study the convergence of belief propagation algorithms (BPAs) on binary pairwise Gibbs random fields (BP-GRFs). Exploiting the equivalence of BPA on the graph associated with BP-GRF and the corresponding computation tree, we first express each BPA message from a node to its parent (on the computation tree) as a function of the messages from its descendants (on the computation tree) at a certain distance. Then, we introduce the notion of a message range by restricting message functions to appropriate domains that correspond to arbitrary initialization of the BPA. We show that message ranges contract at each BPA iteration and characterize the asymptotic behavior of BPA messages on BP-GRF graphs. For a special but important class of BP-GRFs, we are able to derive necessary and sufficient conditions for the BPAs to converge. This should be contrasted with existing methods which typically provide sufficient conditions for convergence (but may not impose any restrictions on the BP-GRFs). We have experimentally shown that these necessary and sufficient conditions can be checked with low computational complexity and, hence, can be applied to large graphs.

2010 IEEE International Conference on Acoustics, Speech and Signal Processing, 2010

Belief propagation is known to perform extremely well in many practical statistical inference and learning problems using graphical models, even in the presence of multiple loops. The iterative use of belief propagation algorithm on loopy graphs is referred to as Loopy Belief Propagation (LBP). Various sufficient conditions for convergence of LBP have been presented; however, general necessary conditions for its convergence to a unique fixed point remain unknown. Because the approximation of beliefs to true marginal probabilities has been shown to relate to the convergence of LBP, several methods have been explored whose aim is to obtain distance bounds on beliefs when LBP fails to converge. In this paper, we derive uniform and non-uniform error bounds on messages, which are tighter than existing ones in literature, and use these bounds to derive sufficient conditions for the convergence of LBP in terms of the sum-product algorithm. We subsequently use these bounds to study the dynamic behavior of the sum-product algorithm, and analyze the relation between convergence of LBP and sparsity and walk-summability of graphical models. We finally use the bounds derived to investigate the accuracy of LBP, as well as the scheduling priority in asynchronous LBP.

Computing Research Repository, 2006

Recently, Chertkov and Chernyak derived an exact expression for the partition sum (normalization constant) corresponding to a graphical model, which is an expansion around the belief propagation (BP) solution. By adding correction terms to the BP free energy, one for each "generalized loop" in the factor graph, the exact partition sum is obtained. However, the usually enormous number of generalized loops generally prohibits summation over all correction terms. In this article we introduce truncated loop series BP (TLSBP), a particular way of truncating the loop series of Chertkov & Chernyak by considering generalized loops as compositions of simple loops. We analyze the performance of TLSBP in different scenarios, including the Ising model on square grids and regular random graphs, and on Promedas, a large probabilistic medical diagnostic system. We show that TLSBP often improves upon the accuracy of the BP solution, at the expense of increased computation time. We also show that the performance of TLSBP strongly depends on the degree of interaction between the variables. For weak interactions, truncating the series leads to significant improvements, whereas for strong interactions it can be ineffective, even if a high number of terms is considered.

IEEE/CAA Journal of Automatica Sinica, 2020

Gaussian belief propagation algorithm (GaBP) is one of the most important distributed algorithms in signal processing and statistical learning involving Markov networks. It is well known that the algorithm correctly computes marginal density functions from a high dimensional joint density function over a Markov network in a finite number of iterations when the underlying Gaussian graph is acyclic. It is also known more recently that the algorithm produces correct marginal means asymptotically for cyclic Gaussian graphs under the condition of walk summability (or generalised diagonal dominance). This paper extends this convergence result further by showing that the convergence is exponential under the generalised diagonal dominance condition, and provides a simple bound for the convergence rate. Our results are derived by combining the known walk summability approach for asymptotic convergence analysis with the control systems approach for stability analysis.