Saving constraint checks in maintaining coarse-grained generalized arc consistency

被引:7
|
作者
Li, Hongbo [1 ]
Li, Ruizhi [1 ]
Yin, Minghao [1 ]
机构
[1] Northeast Normal Univ, Sch Comp Sci & Informat Technol, Changchun 130117, Jilin, Peoples R China
来源
NEURAL COMPUTING & APPLICATIONS | 2019年 / 31卷 / Suppl 1期
关键词
Constraint satisfaction; Local consistency; Backtracking;
D O I
10.1007/s00521-017-3015-7
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Constraint check plays a central role in establishing generalized arc consistency which is widely used to solve constraint satisfaction problems. In this paper, we propose a new generalized arc consistency algorithm, called GTR, which ensures that the tuples that have been checked to be allowed by a constraint will never be checked again. For each constraint, GTR maintains a dynamic list of the tuples that were checked to be allowed by this constraint and check their validities to identify some values with supports. It is equipped with a mechanism avoiding redundant validity checks. The basic GAC3 algorithm is employed to find a support for the rest values and to add new tuples to the dynamic list. The experiments show that maintaining GTR during search saves a number of constraint checks. It also brings some improvements over cpu time while solving some CSPs with tight constraints.
引用
收藏
页码:499 / 508
页数:10
相关论文
共 50 条