资源描述：
《A Preference Optimization Based基于偏好的优化》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、APreferenceOptimizationBasedUnifyingFrameworkforSupervisedLearningProblemsFabioAiolliandAlessandroSperdutiAbstractSupervisedlearningischaracterizedbyabroadspectrumoflearningproblems,ofteninvolvingstructuredtypesofprediction,includingclassification,ranking-basedpredictions(la
2、belandinstanceranking),and(ordinal)regressioninitsvariousforms.Allthesedifferentlearningproblemsaretypicallyaddressedbyspecificalgorithmicsolutions.Inthischapter,weproposeageneralpreferencelearningmodel(GPLM),whichgivesaneasywaytotranslateanysupervisedlearningproblemandtheas
3、sociatedcostfunctionsintosetsofpreferencestolearnfrom.Alargemarginprincipledapproachtosolvethisproblemisalsoproposed.Examplesofhowtheproposedframeworkhasbeeneffectivelyusedbyustoaddressnon-standardreal-worldapplicationsarereportedshowingtheflexibilityandeffectivenessoftheapp
4、roach.1IntroductionSupervisedlearningisprobablythemostcommonlyusedlearningparadigmandalargespectrumoflearningalgorithmshavebeendevisedfordifferentlearningtasksinthelastdecades.Theneedforsuchalargespectrumoflearningalgorithmsis,inpart,duetothemanyreal-worldlearningproblems,t
5、hatarecharacterizedbyhet-erogeneoustasksandproblem-specificlearningalgorithmsfortheirsolution.Theseincludeclassificationandregressionproblems(includingmultilabelandmulticlassclassification,andmultivariateregression),aswellasranking-based(eitherlabelorinstanceranking)andordinal
6、regressionproblems.Typically,theapproachfollowedtodealwithanonstandardproblemistomapitintoaseriesofsimpler,well-knownproblemsandthentocombinetheresultingpredictions.Often,however,thistypeF.Aiolli(B)andA.SperdutiDepartmentofPureandAppliedMathematics-Padova-Italy,ViaTrieste63
7、,35131Padova,Italye-mail:aiolli@math.unipd.it,sperduti@math.unipd.itJ.FürnkranzandE.Hüllermeier(eds.),PreferenceLearning,19DOI10.1007/978-3-642-14125-6_2,cSpringer-VerlagBerlinHeidelberg201020F.AiolliandA.Sperdutiofmethodologylacksaprincipledtheorysupportingitand/orrequire
8、stoomuchcomputationalresourcestobepracticalforreal-worldapplications.Inthischapter
