作者: Marc Schoenauer , Cyril Furtlehner
DOI:
关键词:
摘要: An original approach to multi-objective optimization is introduced, using a message-passing algorithm sample the Pareto set, i.e. set of Pareto-non-dominated solutions. Several heuristics are proposed and tested on simple bi-objective 3-SAT problem. The first one based straightforward deformation Survey-Propagation (SP) equation locally encode trade-off. A heuristic then tested, which combines an elimination procedure clauses with usual decimation variables used in SP algorithm, able different regions Pareto-front. We study more details compliance these deformed equations basic Belief-Propagation (BP) properties. This leads explicit Markov Random Field (MRF) valid warning configuration, for BP equations. observation generalized context. Numerical experiments artificial problems up 100000 presented discussed.