A Method to Avoid Duplicative Flipping in Local Search for SAT

Duong, Thach-Thao ORCID: https://orcid.org/0000-0003-2294-3619 and Pham, Duc Nghia and Sattar, Abdul (2012) A Method to Avoid Duplicative Flipping in Local Search for SAT. In: 25th Australasian Joint Conference on Artificial Intelligence (AI 2012), 4 Dec - 7 Dec 2012, Sydney, Australia.


Stochastic perturbation on variable flipping is the key idea of local search for SAT. Observing that variables are flipped several times in an attempt to escape from a local minimum, this paper presents a duplication learning mechanism in stagnation stages to minimise duplicative variable flipping. The heuristic incorporates the learned knowledge into a variable weighting scheme to effectively prevent the search from selecting duplicative variables. Additionally, probability-based and time window smoothing techniques are adopted to eliminate the effects of redundant information. The integration of the heuristic and gNovelty +  was compared with the original solvers and other state-of-the-art local search solvers. The experimental results showed that the new solver outperformed other solvers on the full set of SAT 2011 competition instances and three sets of real-world verification problems.

Statistics for USQ ePrint 46979
Statistics for this ePrint Item
Item Type: Conference or Workshop Item (Commonwealth Reporting Category E) (Paper)
Refereed: Yes
Item Status: Live Archive
Faculty/School / Institute/Centre: No Faculty
Faculty/School / Institute/Centre: No Faculty
Date Deposited: 30 Mar 2022 05:22
Last Modified: 30 Mar 2022 05:22
Uncontrolled Keywords: Learning mechanism; Local minimums; Local search; Redundant informations; Smoothing techniques; Stochastic perturbations; Time windows; Variable weighting; Verification problems
Fields of Research (2020): 46 INFORMATION AND COMPUTING SCIENCES > 4602 Artificial intelligence > 460210 Satisfiability and optimisation
Identification Number or DOI: https://doi.org/10.1007/978-3-642-35101-3_19
URI: http://eprints.usq.edu.au/id/eprint/46979

Actions (login required)

View Item Archive Repository Staff Only