Quadratization and Roof Duality of Markov Logic Networks

被引:1
作者
de Nijs, Roderick [1 ]
Landsiedel, Christian [2 ]
Wollher, Dirk [2 ]
Buss, Martin [2 ]
机构
[1] Inst Comp Sci, Albrechtstr 28, D-49076 Osnabruck, Germany
[2] Univ Munich, Lehrstuhl Steuerungs & Regelungstech, Theresienster 90, D-80333 Munich, Germany
基金
欧洲研究理事会;
关键词
GRAPH CUTS; INFERENCE; ALGORITHM; BINARY; MINIMIZATION;
D O I
10.1613/jair.5023
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
This article discusses the quadratization of Markov Logic Networks, which enables efficient approximate MAP computation by means of maximum flows. The procedure relies on a pseudo-Boolean representation of the model, and allows handling models of any order. The employed pseudo-Boolean representation can be used to identify problems that are guaranteed to be solvable in low polynomial-time. Results on common benchmark problems show that the proposed approach finds optimal assignments for most variables in excellent computational time and approximate solutions that match the quality of ILP-based solvers.
引用
收藏
页码:685 / 714
页数:30
相关论文
共 41 条
[1]  
[Anonymous], 2008, P 23 AAAI C ARTIFICI
[2]  
[Anonymous], 2003, P 18 INT JOINT C ART
[3]  
[Anonymous], 15 GI S DAT SYST BUS
[4]  
Apsel U., 2012, UAI 12 P 28 C UNCERT, P74
[5]   MAXIMIZING A SUPERMODULAR PSEUDOBOOLEAN FUNCTION - A POLYNOMIAL ALGORITHM FOR SUPERMODULAR CUBIC FUNCTIONS [J].
BILLIONNET, A ;
MINOUX, M .
DISCRETE APPLIED MATHEMATICS, 1985, 12 (01) :1-11
[6]   Pseudo-Boolean optimization [J].
Boros, E ;
Hammer, PL .
DISCRETE APPLIED MATHEMATICS, 2002, 123 (1-3) :155-225
[7]  
Boros E., 2006, PREPROCESSING UNCONS
[8]   RECOGNITION PROBLEMS FOR SPECIAL CLASSES OF POLYNOMIALS IN 0-1 VARIABLES [J].
CRAMA, Y .
MATHEMATICAL PROGRAMMING, 1989, 44 (02) :139-155
[9]   A differential approach to inference in Bayesian networks [J].
Darwiche, A .
JOURNAL OF THE ACM, 2003, 50 (03) :280-305
[10]  
Domingos P., 2012, P C ART INT AAAI