Westudytheruntimeprofilesofcompletebacktrackstylesearchmethodsappliedtohardschedulingproblems.Suchsearchmethodsoftenexhibita largevariabilityinperformanceduetothenon-standardnature oftheirunderlyingcostdistributions.Thedistributionsgenerallyexhibitverylongtailsor"heavytails" andarebestcharacterizedbya generalclassofdistributionsthathavenomoments(i.e.,aninfinitemean, variance,etc.).Weshowhowonecanexploitthespecialnatureofsuchdistributionstosignificantlyimproveupondeterministic completesearch procedures.
Carla P. Gomes, Bart Selman, Ken McAloon, Carol Tr