On the uniqueness of loopy belief propagation fixed points

被引：70
作者：
Heskes, T [1 ]
机构：
[1] Univ Nijmegen, SNN, NL-6525 EZ Nijmegen, Netherlands
来源：
NEURAL COMPUTATION | 2004年 / 16卷 / 11期
关键词：
D O I：
10.1162/0899766041941943
中图分类号：
TP18 [人工智能理论];
学科分类号：
081104 ; 0812 ; 0835 ; 1405 ;
摘要：
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.
引用
收藏
页码：2379 / 2413
页数：35
相关论文
共 19 条
  • [1] Heskes T., 2003, ADV NEURAL INFORM PR, V15, P359
  • [2] HESKES T, 2003, UAI, P313
  • [3] Factor graphs and the sum-product algorithm
    Kschischang, FR
    Frey, BJ
    Loeliger, HA
    [J]. IEEE TRANSACTIONS ON INFORMATION THEORY, 2001, 47 (02) : 498 - 519
  • [4] LAURITZEN SL, 1988, J ROY STAT SOC B MET, V50, P157
  • [5] Luenberger DG., 2015, LINEAR NONLINEAR PRO
  • [6] McEliece RJ, 2003, IMA VOL MATH APPL, V134, P275
  • [7] Turbo decoding as an instance of Pearl's "belief propagation" algorithm
    McEliece, RJ
    MacKay, DJC
    Cheng, JF
    [J]. IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, 1998, 16 (02) : 140 - 152
  • [8] Minka T. P., 2001, P 17 C UNC ART INT U, P362
  • [9] Murphy KP, 1999, UNCERTAINTY IN ARTIFICIAL INTELLIGENCE, PROCEEDINGS, P467
  • [10] PAKZAD P, 2002, 2002 C INF SCI SYST
12