Better Decision Heuristics in CDCL through Local Search and Target Phases
|
Our first contribution is to maximize in a local search fashion
the assignment trail in CDCL, by sticking to and extending promising assignments
via a technique called target phases.
我们的第一个贡献是通过一种称为目标阶段的技术来坚持和扩展有希望的分配，以局部搜索的方式最大化CDCL中的分配轨迹。
Second, we relax the CDCL framework by again extending promising branches to complete assignments while ignoring conflflicts.
These assignments are then used as starting point of local search which tries to fifind improved assignments with fewer unsatisfified clauses.
其次，我们通过再次扩展有希望的分支来完成任务，同时忽略冲突，从而放松CDCL框架。然后，这些赋值被用作局部搜索的起点，试图找到不满意子句较少的改进赋值。 Third, these improved assignments are imported back to the CDCL loop where they are used to determine the value assigned
to decision variables.
第三，这些改进的赋值被导入到CDCL循环中，用于确定分配给决策变量的值。
Finally, the conflflict frequency of variables in local search can be exploited during variable selection in branching heuristics of CDCL.
最后，在分支启发式算法的变量选择中，可以利用变量在局部搜索中的冲突频率。
|
|
|
The satisfifiability problem (SAT) asks to determine whether a given propositional formula
is satisfifiable or not. Propositional formulas are usually represented in conjunctive normal
form (CNF). A growing number of problem domains are successfully tackled by SAT solvers,
including electronic design automation (EDA) (Silva & Sakallah, 2000), particularly hard
ware verifification (Prasad, Biere, & Gupta, 2005) and model checking (Vizel, Weissenbacher,
& Malik, 2015; Biere & Kr¨oning, 2018), mathematical theorem proving (Heule, Kullmann,
& Marek, 2016), AI planning (Kautz & Selman, 1992), and spectrum allocation (Newman,
Fr´echette, & Leyton-Brown, 2018), among others.
Additionally, SAT solvers are also often used as a core component of more complex tools such as solvers for satisfifiability modulo therory (SMT) (Barrett, Sebastiani, Seshia, & Tinelli, 2021), which form a crucial component of state-of-the-art program analysis and software verifification.
此外，SAT求解器也经常被用作更复杂工具的核心组件，如满足度模理论(SMT)求解器(Barrett, Sebastiani, Seshia， & Tinelli, 2021)，这构成了最先进的程序分析和软件验证的关键组件。 |
|
|
Many techniques, including clause learning (Fang & Ruml, 2004) and unit propaga
tion (Hirsch & Kojevnikov, 2005), have been tried to improve local search algorithms but
they are still not competitive. Recent studies show that given a promising initial solution
for local search helps to improve the performance on some benchmarks (Zhang, Sun, Zhu,
Li, Cai, Xiong, & Zhang, 2020; Cai, Luo, Zhang, & Zhang, 2021). In this paper we go one
step further and the two components exchange information.
许多技术，包括子句学习(Fang & Ruml, 2004)和单位传播(Hirsch & Kojevnikov, 2005)，已经尝试改进局部搜索算法，但它们仍然没有竞争力。最近的研究表明，为局部搜索提供一个有希望的初始解决方案有助于提高在某些基准上的性能(Zhang, Sun, Zhu，李、蔡、熊、张，2020;蔡，罗，张，&张，2021)。在本文中，我们更进一步，两个组件交换信息。 |
|
|
There have been several attempts to combine both approaches. However, in previous hybrid solvers, both solvers, the CDCL and the local search solver, are opaque to each other, at most exchange some partial information in one direction and therefore usually see each other as a black box. 人们曾多次尝试将这两种方法结合起来。然而，在以前的混合解算器中，CDCL和局部搜索解算器彼此是不透明的，最多只能在一个方向上交换部分信息，因此通常将对方视为一个黑盒子。
These early hybrid solvers invoke the respective solver according to
difffferent situations (Mazure, Sais, & Gr´egoire, 1998; Habet, Li, Devendeville, & Vasquez,
2002; Letombe & Marques-Silva, 2008; Balint, Henn, & Gableske, 2009; Audemard, Lagniez,
Mazure, & Sais, 2010) as discussed in Section 8 on related work.
这些早期的混合求解器会根据不同的情况调用相应的求解器.在第8节的相关工作中讨论。 |
|
