求解器研究中的实验设计

最近收集了部分文献中实验设计如下：

	1. 文献：Community Structure in Industrial SAT Instances A. 段落：第7节第4段后半段。讲解了一个实验，很有意思。 笔记：（1）关于SAT结论对应于可能有多个不同的赋值序列。特别是：Notice that this experiment is equivalent to removing all activitycounters used by the heuristic after timep t, and this may dramatically worsen the perfor-mance of the solver when the formula is satisfiable. （2）结论为UNSAT的样例，中途保留学习子句集而去掉过程累计的信息后重新求解，不影响整体求解时间。这说明UNSAT对过程量的依赖性很低，学习子句集的重要性比较突出。   B. 段落：第7节第4段后半段。讲解了第二个实验，也很有意思,并成为算法3，作为该文章的核心技术。 （1）Notice that the  algorithm imposes a very strong condition, which is solving all subformulas between two connected communities and keepingalllearned clauses foundin this process.  Notice that a solver not aware of thecommunity structure may remove them, unless, as we do, these clauses are added in apreprocessing step as original clauses, so the solver is forced to keep them （2）This could be further refined. 在子句中加入社区的数量作为决定子句删除的试探法，而不是在预处理中使用它，我们推测改进甚至会比我们在这个实验中观察到的更大。 （3）此外，这一预处理步骤可以在搜索过程中以内部处理方法的方式进行启发式应用.
	2. 实验设计：已经求出结果的情况下，保留最后单元子句、决策序列、充当reason子句的所有学习子句。交给不同的求解器评价其某项策略的优劣。
	3. 关于求解backbone的算法实验 （1）基于一次一测的思想： 文献：On computing backbones of propositional theories. J. Marques-Silva, M. Janota, I. Lynce. On computing backbones of propositional theories{C}. In Proc of the 19th ECAI, 2010, 15--20   （2）借鉴了启发式思想，设计了评分机制和过滤策略，并对测试集进行了调整。基于这一思想，提出了启发式主干算法，提高了主干求解的速度。 文献：Backbone solving algorithm based on heuristic thinking. Shaohua Guo, Junru He, Xueting Song, Weiting Liu: Backbone solving algorithm based on heuristic thinking. ICMLSC 2019: 44-48   (3)  In this paper, we propose a new backbone algorithm which makes use of a "SAT with preferences" solver, i.e., a SAT solver which is guaranteed to output a most preferred satisfying assignment w.r.t. a given preference over literals of the SAT instance at hand. 文献：A preference-based approach to backbone computation with application to argumentation.
	4. 回归分析 regression analysis  Newsham et al. [84] performed a regression analysis to give evidence of the correlation between SAT structure and the running time of the CDCL SAT solver MiniSAT [35] over all instances in the SAT 2013 competition [1].  特别是，研究了变量、子句、社区、模块化值的数量以及这些因素之间的比率。实验表明，社区的数量、模块化以及包括这两种度量的任何组合确实可以预测求解器的性能。 [1]2002. SAT Competitions. http://www.satcompetition.org. (2002) [35]Eén Niklas and Sörensson Niklas. 2003. An extensible SAT-solver. In SAT (Lecture Notes in Computer Science), Giunchiglia Enrico and Tacchella Armando (Eds.), Vol. 2919. Springer, 502–518 [84]Newsham Zack, Ganesh Vijay, Fischmeister Sebastian, Audemard Gilles, and Simon Laurent. 2014. Impact of community structure on SAT solver performance. In Theory and Applications of Satisfiability Testing–SAT 2014, Sinz Carsten and Egly Uwe (Eds.). Springer International Publishing, Cham, 252–268.
	各种特征和解决时间之间的Spearman相关性【99】结果表明，社区测量似乎与许多子类别密切相关。
	5.  Understand Restart of SAT Solver Using Search Similarity Index Yoichiro Iida, Tomohiro Sonobe, Mary Inaba: Understand Restart of SAT Solver Using Search Similarity Index (Student Abstract). AAAI 2023: 16236-16237   Abstract SAT solvers are widely used to solve many industrial problems because of their high performance, which is achieved by various heuristic methods. Understanding why these methods are effective is essential to improving them. One approach to this is analyzing them using qualitative measurements. In our previous study, we proposed search similarity index (SSI), a metric to quantify the similarity between searches. SSI significantly improved the performance of the parallel SAT solver. Here, we apply SSI to analyze the effect of restart, a key SAT solver technique. Experiments using SSI reveal the correlation between the difficulty of instances and the search change effect by restart, and the reason behind the effectiveness of the state-of-the-art restart method is also explained.