Self-Guided Belief Propagation - A Homotopy Continuation Method

被引:0
作者
Knoll, Christian [1 ]
Weller, Adrian [2 ,3 ]
Pernkopf, Franz [1 ]
机构
[1] Graz Univ Technol, Signal Proc & Speech Commun Lab, A-8010 Graz, Austria
[2] Univ Cambridge, Cambridge CB2 1TN, England
[3] Alan Turing Inst, London NW1 2DB, England
关键词
Computational modeling; Belief propagation; Probabilistic logic; Convergence; Graphical models; Couplings; Random variables; belief propagation; probabilistic inference; sum-product algorithm; partition function; inference algorithms; COMPUTATIONAL-COMPLEXITY; MIMO DETECTION; INFERENCE; CONVERGENCE; MODELS;
D O I
10.1109/TPAMI.2022.3196140
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Belief propagation (BP) is a popular method for performing probabilistic inference on graphical models. In this work, we enhance BP and propose self-guided belief propagation (SBP) that incorporates the pairwise potentials only gradually. This homotopy continuation method converges to a unique solution and increases the accuracy without increasing the computational burden. We provide a formal analysis to demonstrate that SBP finds the global optimum of the Bethe approximation for attractive models where all variables favor the same state. Moreover, we apply SBP to various graphs with random potentials and empirically show that: (i) SBP is superior in terms of accuracy whenever BP converges, and (ii) SBP obtains a unique, stable, and accurate solution whenever BP does not converge.
引用
收藏
页码:5139 / 5157
页数:19
相关论文
共 71 条
  • [1] Massive MIMO Detection Techniques: A Survey
    Albreem, Mahmoud A.
    Juntti, Markku
    Shahabuddin, Shahriar
    [J]. IEEE COMMUNICATIONS SURVEYS AND TUTORIALS, 2019, 21 (04): : 3109 - 3132
  • [2] Allgower E. L., 2003, INTRO NUMERICAL CONT
  • [3] THE RANDOM FIELD ISING-MODEL
    BELANGER, DP
    YOUNG, AP
    [J]. JOURNAL OF MAGNETISM AND MAGNETIC MATERIALS, 1991, 100 (1-3) : 272 - 291
  • [4] Survey propagation:: An algorithm for satisfiability
    Braunstein, A
    Mézard, M
    Zecchina, R
    [J]. RANDOM STRUCTURES & ALGORITHMS, 2005, 27 (02) : 201 - 226
  • [5] Encoding for the Blackwell channel with reinforced Belief Propagation
    Braunstein, Alfredo
    Kayhan, Farbod
    Montorsi, Guido
    Zecchina, Riccardo
    [J]. 2007 IEEE INTERNATIONAL SYMPOSIUM ON INFORMATION THEORY PROCEEDINGS, VOLS 1-7, 2007, : 1891 - +
  • [6] COUNTING INDEPENDENT SETS USING THE BETHE APPROXIMATION
    Chandrasekaran, Venkat
    Chertkov, Misha
    Gamarnik, David
    Shah, Devavrat
    Shin, Jinwoo
    [J]. SIAM JOURNAL ON DISCRETE MATHEMATICS, 2011, 25 (02) : 1012 - 1034
  • [7] Bethe States of Random Factor Graphs
    Coja-Oghlan, Amin
    Perkins, Will
    [J]. COMMUNICATIONS IN MATHEMATICAL PHYSICS, 2019, 366 (01) : 173 - 201
  • [8] THE COMPUTATIONAL-COMPLEXITY OF PROBABILISTIC INFERENCE USING BAYESIAN BELIEF NETWORKS
    COOPER, GF
    [J]. ARTIFICIAL INTELLIGENCE, 1990, 42 (2-3) : 393 - 405
  • [9] ON CONTINUITY OF MINIMUM SET OF A CONTINUOUS FUNCTION
    DANTZIG, GB
    FOLKMAN, J
    SHAPIRO, N
    [J]. JOURNAL OF MATHEMATICAL ANALYSIS AND APPLICATIONS, 1967, 17 (03) : 519 - &
  • [10] ISING MODELS ON LOCALLY TREE-LIKE GRAPHS
    Dembo, Amir
    Montanari, Andrea
    [J]. ANNALS OF APPLIED PROBABILITY, 2010, 20 (02) : 565 - 592