资源描述:
《Knowledge Representation for Stochastic Decision Processes》由会员上传分享,免费在线阅读,更多相关内容在学术论文-天天文库。
1、KnowledgeRepresentationforStochasticDecisionProcessesCraigBoutilierUept.ofConiputerScience,UniversityofBritishColumbia,Vancouver,BCV6T124,CAKADAcebly@cs.ubc.cuAbstract.Reasoningaboutstochasticdynamicalsystemsandplanningunderuncertaintyhascometoplayafundamentalrolei
2、nA1researchandapplications.Therepresentationofsuchsystems,inparticular,ofactionswithstochasticeffects,hasaccordinglybeengivenincreasingat-tentioninrecentyears.Inthisarticle,wesurveyanumberoftechniquesforrepresentingstochasticprocessesandactionswithstochasticeffects
3、usingdyriilrriicBayesiannetworksandinfluencediagrams,andbrieflydescribehowt,hesesupporteffectiveinferencefortaskssuchasmoni-toring,forecasting,explanationanddecisionmaking.Wealsocomparethesetechniquestoseveralact,ionrepresentationsadoptedintheclassicalreasoningabou
4、tactionandplanningcommunities,describinghowtra-dit,ionalproblemssuchastheframeandramificationproblemsaredealtwithiristochasticsettings,andhowthesesolutionscomparetorecentapproachestothisproblemintheclassical(deterministic)literature.Wearguethatwhilestochasticdynami
5、csintroducecertaincomplicationswhenitcomestosuchissues,forthemostpart,intuitionsunderlyingclassicalmodelscanbeextendedtothestochasticsetting.1IntroductionWithinartificialintelligence,increasingattcntionhasbeenpaidt.otheproblemsofthemonitoring,forecastingandcontrolo
6、fcomplexstochasticprocesses.WhileclassicalplanninghashistoricallybeentheprimefocusofthoseinA1interestedincontrollingdynamicalsystems,researchershavecometorea.lizethatmany(ormost)realisticproblemscannotbeadequatelymodeledusingtheassumptionsofclassicalplanning.Specif
7、ically,oneisgenerallyforcedtoconsideractionswithnondet,erministicorstochasticeffects,processesinwhichexogenouseventsoccur,incompleteoruncertainknowledgeofthesystemstate,impreciseobservationsofthesystemstate,problemswithill-definedgoalsormultiple,possiblyconflict-in
8、gobjectives,andon-going(possiblynonterrninating)processeswithindefinitehorizon.St,ochasticanddecisiontheoret,icplanning[17,19,51attemptstoincorpo