Reactive anddynamiclocalsearchformax-clique:Engineeringeffectivebuilding blocks

This paperpresentstheresultsofanongoinginvestigationabouthowdifferentalgorithmicbuilding blocks contributetosolvingthemaximumcliqueproblem.Weconsidergreedyconstructions,plateau searches, andmorecomplexschemesbasedondynamicpenaltiesand/orprohibitions,inparticularthe recently proposedtechniqueofdynamiclocalsearchandthepreviouslyproposedreactivelocalsearch (RLS). WedesignavariationoftheoriginalRLSalgorithmwheretheroleoflong-termmemory(LTM) is increased(RLS-LTM).Inaddition,weconsiderindetailtheeffectofthelow-levelimplementation choices ontheCPUtimeperiteration.Wepresentexperimentalresultsonrandomlygeneratedgraphs with differentstatisticalproperties,showingthecrucialeffectsoftheimplementation,therobustnessof different techniques,andtheirempiricalscalability.