NeuroBack与NeuroCore方法学习-未完继续

涉及两类文献

Daniel Selsam, Nikolaj S. Bjørner: NeuroCore: Guiding High-Performance SAT Solvers with Unsat-Core Predictions. CoRR abs/1903.04671 (2019)     @article{DBLP:journals/corr/abs-1903-04671, author = {Daniel Selsam and Nikolaj S. Bj{\o}rner}, title = {NeuroCore: Guiding High-Performance {SAT} Solvers with Unsat-Core Predictions}, journal = {CoRR}, volume = {abs/1903.04671}, year = {2019}, url = {http://arxiv.org/abs/1903.04671}, eprinttype = {arXiv}, eprint = {1903.04671}, timestamp = {Thu, 14 Apr 2022 20:26:11 +0200}, biburl = {https://dblp.org/rec/journals/corr/abs-1903-04671.bib}, bibsource = {dblp computer science bibliography, https://dblp.org} }

NEUROBACK: IMPROVING CDCL SAT SOLVING USING GRAPH NEURAL NETWORKS @inproceedings{DBLP:conf/iclr/WangHTKMM24, author = {Wenxi Wang and Yang Hu and Mohit Tiwari and Sarfraz Khurshid and Kenneth L. McMillan and Risto Miikkulainen}, title = {NeuroBack: Improving {CDCL} {SAT} Solving using Graph Neural Networks}, booktitle = {The Twelfth International Conference on Learning Representations, {ICLR} 2024, Vienna, Austria, May 7-11, 2024}, publisher = {OpenReview.net}, year = {2024}, url = {https://openreview.net/forum?id=samyfu6G93}, timestamp = {Wed, 07 Aug 2024 17:11:53 +0200}, biburl = {https://dblp.org/rec/conf/iclr/WangHTKMM24.bib}, bibsource = {dblp computer science bibliography, https://dblp.org} }

 

 

	NeuroCore
	NeuroCore Selsam & Bjørner (2019), the most closely related approach to this paper, aims to make the solving more effective especially for large-scale problems as in SAT competitions. It enhances the branching heuristic for CDCL using supervised learning to map unsat problems to unsat core variables (i.e., the variables involved in the unsat core). NeuroCore Selsam和Bjørner（2019年）的方法与本文最为相关，旨在提高解决效率，特别是对于SAT竞赛中大规模问题。它通过监督学习增强CDCL的分支启发式算法，将未满足问题映射到未满足核心变量（即参与未满足核心的变量）。
	Based on the dynamically learned clauses during the solving process, NeuroCore performs frequent online model inferences to tune the predictions. 基于在解决过程中动态学习的条款，NeuroCore执行频繁的在线模型推断以调整预测。
	NeuroBack
	However, this online inference is computationally demanding. NeuroBack is distinct from NeuroCore in two main aspects. One, while NeuroCore is designed to refine the branching heuristic in CDCL SAT solvers, NeuroBack is invented to enhance their phase selection heuristics. Two, while NeuroCore extracts dynamic unsat core information from unsat formulas through online model inferences, NeuroBack captures static backbone information from sat formulas using offline model inference. Details of NeuroBack are introduced in the following section. 然而，这种在线推理在计算上是要求很高的。NeuroBack与NeuroCore在两个主要方面有所不同。首先，虽然NeuroCore旨在改进CDCL SAT求解器中的分支启发式算法，但NeuroBack是为了增强它们的相位选择启发式算法而发明的。其次，虽然NeuroCore通过在线模型推理从不满足的公式中提取动态的不满足核心信息，但NeuroBack使用离线模型推理从满足的公式中捕获静态的主干信息。NeuroBack的细节将在下一节中介绍。