SAT-based big-step local search

Morad Muslimany, Michael Codish

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

This paper introduces a hybrid search method for optimization problems which combines techniques from Local Search methods and from SAT-based methods. At each iteration, the method performs a 'big-step' move on a subset of variables of the current solution. This step is achieved by encoding the big-step itself as an optimization problem and solving it using a SAT (MaxSAT) solver such that the solution of the big-step results in a higher-quality solution to the entire problem. Experimentation illustrates a clear benefit of the approach over both methods: Local Search methods and SAT-based methods.

Original languageEnglish
Title of host publicationProceedings - 2018 20th International Symposium on Symbolic and Numeric Algorithms for Scientific Computing, SYNASC 2018
PublisherInstitute of Electrical and Electronics Engineers
Pages109-116
Number of pages8
ISBN (Electronic)9781728106250
DOIs
StatePublished - 1 Sep 2018
Event20th International Symposium on Symbolic and Numeric Algorithms for Scientific Computing, SYNASC 2018 - Timisoara, Romania
Duration: 20 Sep 201823 Sep 2018

Publication series

NameProceedings - 2018 20th International Symposium on Symbolic and Numeric Algorithms for Scientific Computing, SYNASC 2018

Conference

Conference20th International Symposium on Symbolic and Numeric Algorithms for Scientific Computing, SYNASC 2018
Country/TerritoryRomania
CityTimisoara
Period20/09/1823/09/18

Keywords

  • Constraints
  • Examination timetabling
  • Local search
  • Maxsat
  • Optimization
  • Sat
  • Sat solver
  • Scheduling
  • Simulated annealing

ASJC Scopus subject areas

  • Computational Theory and Mathematics
  • Software
  • Computational Mathematics
  • Modeling and Simulation
  • Numerical Analysis

Fingerprint

Dive into the research topics of 'SAT-based big-step local search'. Together they form a unique fingerprint.

Cite this