digit wall power deffici solut locat dbase data share sensor dbase abnorm human dactiv detect avail afford sensor sensor network sensor base human activ recognit attract attent artifici intellig ubiquit comput paper present phase approach detect abnorm activ base wireless sensor attach human bodi detect abnorm activ import task secur monitor healthcar applic sensor network tradit approach problem suffer high fals posit rate collect sensor data bias normal data abnorm event rare lack train data tradit data mine method appli solv problem approach emploi class support vector machin svm train commonli normal activ filter activ high probabl normal deriv abnorm activ model gener normal model kernel nonlinear regress knlr reduc fals posit rate unsupervis manner show approach good tradeoff abnorm detect rate fals alarm rate abnorm activ model automat deriv explicitli label abnorm train data scarc demonstr onlin dlocal indoor wireless network dimens reduct adapt local dynam wifi environ multi dview learn accur locat user wireless environ import task pervas comput ai applic activ recognit wifi environ mobil devic local signal receiv transmitt access point ap local approach build map signal space physic locat space offlin phase receiv signal strength rss map estim locat onlin phase map outdat signal strength valu chang time due environment dynam infeas expens repeat data calibr reconstruct rss map case import adapt model learnt time period time period recalibr paper present locat estim approach base manifold regular machin learn techniqu build map function data describ lemancor system adapt map function signal space physic locat space time period base manifold regular show lemancor effect transfer knowledg time period requir calibr effort illustr leman cor effect real wifi environ dlocal label unlabel data graph laplacian domain dconstrain semi dsupervis mine track model sensor network manifold regular approach calibr reduct sensor dnetwork base track abil accur detect locat mobil node sensor network import artifici intellig ai task rang robot context awar comput previou approach locat estim problem assum avail calibr data obtain data requir great effort paper present manifold regular approach leman calibr effort reduct track mobil node wireless sensor network comput subspac map function signal space physic space small amount label data larg amount unlabel data map function onlin determin locat mobil node sensor network base signal receiv crossbow mica setup network usb camera arrai obtain ground truth experiment result show achiev higher accuraci calibr effort compar previou system cost dsensit ddata preprocess mine custom relationship manag databas manifold regular approach calibr reduct sensor dnetwork base track abil accur detect locat mobil node sensor network import artifici intellig ai task rang robot context awar comput previou approach locat estim problem assum avail calibr data obtain data requir great effort paper present manifold regular approach leman calibr effort reduct track mobil node wireless sensor network comput subspac map function signal space physic space small amount label data larg amount unlabel data map function onlin determin locat mobil node sensor network base signal receiv crossbow mica setup network usb camera arrai obtain ground truth experiment result show achiev higher accuraci calibr effort compar previou system queri enrich web dqueri classif web search queri typic short ambigu classifi queri target categori difficult import problem paper present techniqu call queri enrich take short queri map intermedi object base collect intermedi object queri map target categori build map function ensembl search engin produc enrich queri techniqu appli acm knowledg discoveri data mine competit acm kddcup won championship evalu metric precis measur combin precis recal creativ judg organ total team worldwid paper show difficulti abund ambigu queri lack train data queri enrich techniqu solv problem satisfactorili phase classif framework present detail descript algorithm experiment evalu result precis multidimension vector regress accur low dcost locat estim pervas comput abstract paper present algorithm multidimension vector regress data highli uncertain nonlinear appli problem indoor locat estim wireless local area network wlan aim obtain accur map signal space physic space requir human calibr effort locat estim problem tradition tackl probabilist model train manual label data expens obtain contrast algorithm adopt kernel canon correl analysi kcca build nonlinear map signal vector space physic locat space transform data space canon featur pairwis similar sampl space maxim correl kernel gaussian kernel adapt noisi characterist signal strength mat rn kernel sens physic locat real data collect wireless lan environ achiev accur locat estim pervas comput requir smaller set label train data previou method index term locat depend sensit correl regress analysi pervas comput compet driven case dbase mine accur low dcost locat estim kernel ust win solut queri classif kddcup 