Convergence of the Belief Propagation Algorithm for RB Model Instances
Author:
Affiliation:

Clc Number:

Fund Project:

National Natural Science Foundation of China (61462001, 61262006)

  • Article
  • |
  • Figures
  • |
  • Metrics
  • |
  • Reference
  • |
  • Related
  • |
  • Cited by
  • |
  • Materials
  • |
  • Comments
    Abstract:

    Belief propagation algorithm is very effective in finding satisfying assignments for RB(k,n,α,rc,p) model instances where hard region becomes narrower. However, belief propagation algorithm does not always converge for factor graphs with cycles. Unfortunately, rigorous theoretical proof of this phenomenon is still lacking. Belief propagation algorithm is the most basic message passing algorithms. This study derives the convergence conditions of the belief propagation algorithm for solving RB(k,n,α,rc,p) model instances. In the RB(k,n,α,rc,p) model with k=2, α>(1/k), rc>0 and which proves that BP will be converged with high probability if p∈(0,n-2α). Experimental results show that such convergence conditions of belief propagation algorithm are very effective in two different group data from the random RB(k,n,α,rc,p) model instances. In the RB(k,n,α,rc,p) model, when n increases, the experimental convergence interval is fixed range, while the theory convergence interval become narrower. It is because the number of incompatible assignment are determined by n and p in the RB(k,n,α,rc,p).

    Reference
    Related
    Cited by
Get Citation

王晓峰,许道云. RB模型实例集上置信传播算法的收敛性.软件学报,2016,27(11):2712-2724

Copy
Share
Article Metrics
  • Abstract:
  • PDF:
  • HTML:
  • Cited by:
History
  • Received:March 09,2014
  • Revised:March 10,2015
  • Adopted:
  • Online: March 30,2016
  • Published:
You are the firstVisitors
Copyright: Institute of Software, Chinese Academy of Sciences Beijing ICP No. 05046678-4
Address:4# South Fourth Street, Zhong Guan Cun, Beijing 100190,Postal Code:100190
Phone:010-62562563 Fax:010-62562533 Email:jos@iscas.ac.cn
Technical Support:Beijing Qinyun Technology Development Co., Ltd.

Beijing Public Network Security No. 11040202500063