ART2BasedClassification ofSparse High Dimensional Parameter SetsForA Simulation Parameter Selection Assistant
G.A. Klotz · 2005
Thispaper presentsthedesign andcreation of a Simulation Parameter Selection Assistant (SPSA)thathelps modeling researchers choose meaningful values fortheir com- plexsimulations, andencouragescollaboration between teams searching through highdimensional parameter spaces.Proposed simulation parameters arecompared topastrunsusing Adaptive Resonance Theory tomeasure sinmilarity withthegoals ofpre- venting repetitive exploitations ofparameters andofencouraging theexploration ofnew regions oftheparameterspace.The Assistant was designed tobeused aspartofa highperformance animaldisease spread simulator butisgeneral andmodular enoughtobeeasily adapted toothersimulation andsearch domains. I. INTRODUCTION Complex simulations suchaspredicting thespread ofdis- ease can takedaystocomplete, even on thelatest super- computers. Tomakethebest useofavailable resourcesand tosavetime, itisimportant fortheresearchers running the simulations toexplore theparameter spaceoftheir problem efficiently. Thisisaccomplished byhelping researchers choose parameters fortheir simulation thathavenotalready been modeled bythemselves or others (except inspecial cases, suchas validation). Byrecognizing similarities between the proposed simulation parameters andparametersfrompast simulation runs,theresearcher isshownthelikely results of their simulation andwhether or notthey arefocusing on an unexplored partoftheparameter set. Ourresearch focusison thedomainofcomplex, high dimensional parameter spacesencountered inlarge-scale, high performance simulations such asthose foreconomic modeling, weather simulation anddisease spread modeling. We have looked attheenvironment ofa highperformance (super- computer/cluster) animal disease spread simulator tomotivate ourresearch directions. Characteristics oftheanimal disease spread simulator that are relevant tothisdiscussion are the following: * a verylarge setofparameters necessarytodescribe the model/simulation (ontheorder ofhundreds tothousands) * a dynamic setofparameters(since thesimulation has todealwitha variety ofanimal production types, e.g. DeborahA.Stacey