DocumentCode :
2268324
Title :
The serial and parallel belief propagation algorithms
Author :
Tan, Peng Hui ; Rasmussen, Lars K.
Author_Institution :
Dept. of Comput. Eng., Chalmers Univ. of Technol., Gothenburg
fYear :
2005
fDate :
4-9 Sept. 2005
Firstpage :
729
Lastpage :
733
Abstract :
It has been shown that the stable fixed points of belief propagation (BP) algorithms correspond to extrema of the Bethe free energy. In this paper, we describe the dual problem for the minimization of the Bethe free energy and solve it using simple nonlinear block Gauss-Seidel and Jacobi algorithms. The use of the nonlinear block Gauss-Seidel algorithm corresponds to serial scheduling for the BP algorithm. In addition, it is shown that applying the nonlinear block Jacobi algorithm on the dual of the Bethe free energy corresponds to the parallel BP algorithm
Keywords :
graph theory; information theory; Bethe free energy; nonlinear block Gauss-Seidel algorithm; nonlinear block Jacobi algorithm; parallel belief propagation algorithm; serial belief propagation algorithm; Belief propagation; Convergence; Councils; Gaussian processes; Jacobian matrices; Lakes; Minimization methods; Random variables; Scheduling algorithm; Telecommunication computing;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Information Theory, 2005. ISIT 2005. Proceedings. International Symposium on
Conference_Location :
Adelaide, SA
Print_ISBN :
0-7803-9151-9
Type :
conf
DOI :
10.1109/ISIT.2005.1523432
Filename :
1523432
Link To Document :
بازگشت