资源描述：
《this report also appears as human-computer interaction》由会员上传分享，免费在线阅读，更多相关内容在教育资源-天天文库。
1、TheRADARTestMethodology:EvaluatingaMulti-TaskMachineLearningSystemwithHumansintheLoopAaronSteinfeld,RachaelBennett,KyleCunningham,MattLahut,Pablo-AlejandroQuinones,DjangoWexler,DanSiewiorek,1233PaulCohen,JulieFitzgerald,OtharHansson,JordanHayes,45MikePool,andMarkDrummondOctober2006CMU-CS-06-125
2、SchoolofComputerScienceCarnegieMellonUniversityPittsburgh,PA15213ThisreportalsoappearsasHuman-ComputerInteractionInstituteTechnicalReportCMU-HCII-06-102AbstractTheRADARprojectinvolvesacollectionofmachinelearningresearchthruststhatareintegratedintoacognitivepersonalassistant.Progressisexaminedwi
3、thatestdevelopedtomeasuretheimpactoflearningwhenusedbyahumanuser.Threeconditions(conventionaltools,Radarwithoutlearning,andRadarwithlearning)areevaluatedinalarge-scale,between-subjectsstudy.ThispaperdescribestheRADARTestwithafocusontestdesign,testharnessdevelopment,experimentexecution,andanalys
4、is.Resultsforthe1.1versionofRadarillustratethemeasurementanddiagnosticcapabilityofthetest.Generallessonsonsucheffortsarealsodiscussed.14UniversityofSouthernCaliforniaIET,Inc(Formerlyat)25JSFConsultingSRIInternational3Thinkbank,IncThismaterialisbaseduponworksupportedbytheDefenseAdvancedResearchP
5、rojectsAgency(DARPA)underContractNo.NBCHD030010.Anyopinions,findingsandconclusionsorrecommendationsexpressedinthismaterialarethoseoftheauthorsanddonotnecessarilyreflecttheviewsoftheDARPAortheDepartmentofInterior-NationalBusinessCenter(DOI-NBC).Keywords:machinelearning,human-computerinteraction,
6、artificialintelligence,multi-agentsystems,evaluation,humansubjectexperimentsTheRADARTestMethodology:EvaluatingaMulti-TaskMachineLearningSystemwithHumansintheLoop1Overview1TheRADAR(ReflectiveAgentswithDistributedAdaptiveReasoning)projectwithintheDARPAPAL(PersonalizedAssistantthatLearns)programis
7、centeredonresearchanddevelopmenttowardsapersonalcognitiveassistant.Theunderlyingscientificadvanceswithintheprojectarepredominantlywithintherealmofmachinelearning(ML).TheseMLapproachesarevariedandtheresultingtechnologiesarediverse.
