Meta-Learning for Recommending Metaheuristics for the MaxSAT Problem

Enrico S. Miranda, Fabio Fabris, Chrystian G. M. Nascimento, Alex A. Freitas, Alexandre C. M. Oliveira
2018 2018 7th Brazilian Conference on Intelligent Systems (BRACIS)  
Solving even moderately-sized Maximum Satisfiability (Max-SAT) problems exactly can be unfeasible due to their NP-Hardness. This leads to the use of metaheuristics that find a solution without exactness guarantees but run in a reasonable time. Yet, choosing the best metaheuristic to solve a MaxSAT problem is hard, justifying the use of meta-learning algorithms for metaheuristic recommendation. These meta-learning algorithms use past experience to choose the best metaheuristic to solve an unseen
more » ... problem. As far as we know, this is the first time a meta-learning approach is proposed to select metaheuristics for solving a MaxSAT problem. Our approach includes the creation of new meta-features derived from graph representations of MaxSAT problems and an interpretation of part of a meta-model. Our approach successfully selected the best metaheuristic to solve the problems 87% of the time, the new meta-features have shown to be as good as the state-of-theart meta-features, and the meta-model interpretation found interesting problem-specific knowledge.
doi:10.1109/bracis.2018.00037 dblp:conf/bracis/MirandaFNFO18 fatcat:nxh36vg6ojbqhazbu3ftib7oui