Home

About

Advanced Search

Browse by Discipline

Scientific Societies

E-print Alerts

Add E-prints

E-print Network
FAQHELPSITE MAPCONTACT US


  Advanced Search  

 
Submitted for consideration for ISIT 2003, October 30, 2002. Poset Belief PropagationExperimental Results
 

Summary: Submitted for consideration for ISIT 2003, October 30, 2002.
Poset Belief Propagation­Experimental Results
Jonathan Harel, Robert J. McEliece, and Ravi Palanki
California Institute of Technology
Pasadena, California USA
Abstract: Poset belief propagation, or PBP, is a flexible generalization of ordinary belief
propagation which can be used to design algorithms for solving (approximately) many
probabilistic inference problems, including MAP decoding of binary linear codes. In this
paper, we will present some experimental results tha suggest that PBP can significantly
outperform conventional BP techniques.
1. Introduction.
In [4], building on the pioneering work of Yedidia, Freeman, and Weiss [10] on "generalized
belief propagation," McEliece and Yildirim introduced a class of algorithms called belief
propagation on partially ordered sets, or PBP. (Similar algorithms have recently been de-
veloped in [11] and [5, 6]). PBP includes as special cases ordinary belief propagation [7],
probability propagation [8], the generalized distributive law [1, 2], the sum-product algo-
rithm [3], generalized belief propagation [10], (all of these with and without loops), and
many other instances whose effectiveness has not yet been investigated in detail. In this
paper we summarize PBP (including a new "message-free" formulation of the algorithm)
and report the results of some experiments we have performed. Our tentative conclusion is

  

Source: Adolphs, Ralph - Psychology and Neuroscience, California Institute of Technology

 

Collections: Biology and Medicine