surrog rank expens similar queri mixtur model learn low ddimension role high ddimension data datapath system data dcentric analyt process engin larg data warehous evalu probabilist threshold queri mcdb mcdb dr risk analysi databas enterpris assess manag risk aris uncertainti data uncertainti typic model probabl distribut uncertain data valu mean complex predict stochast model probabl distribut data valu lead probabl distribut databas queri result risk assess amount explor upper lower tail queri result distribut paper extend mont carlo databas system effici obtain set sampl tail queri result distribut adapt recent gibb clone idea simul literatur databas set lrt framework fast spatial anomali detect lrt framework fast spatial anomali detect turbo dcharg estim converg dbo dbo databas system util random algorithm give statist meaning estim final answer multi tabl disk base queri start finish queri execut dbo time til util ttu time dbo give estim overli larg case databas tabl join queri case join queri includ select predic tabl data skew paper describ turbo dbo prototyp databas system answer multi tabl join queri scalabl fashion dbo turbo dbo lower ttu dbo kei innov turbo dbo make algorithm rememb partial match tupl random fashion tupl satisfi boolean predic queri possibl grown tupl contribut final queri result time sampl dbase estim subset dbase queri guess extrem valu data set bayesian method applic robust stratifi sampl plan low select queri dbo databas system demonstr prototyp dbo databas system dbo design facilit scalabl analyt process larg data archiv dbo analyt process perform competit databas system unlik exist research industri system dbo maintain statist meaning guess final answer queri start finish queri process guess accur second minut answer queri hour result signific save user comput time user abort queri happi guess accuraci bayesian mixtur model linear regress mix proport dbo databas system demonstr prototyp dbo databas system dbo design facilit scalabl analyt process larg data archiv dbo analyt process perform competit databas system unlik exist research industri system dbo maintain statist meaning guess final answer queri start finish queri process guess accur second minut answer queri hour result signific save user comput time user abort queri happi guess accuraci mcdb mont carlo approach manag uncertain data confid bound sampl dbase group estim scalabl approxim queri process dbo engin materi sampl view databas approxim problem creat sampl view databas tabl sampl view index materi view permit effici sampl arbitrari rang queri view sampl view applic requir random sampl databas approxim queri process onlin aggreg data mine random algorithm exampl core technic contribut file organ call ac tree suitabl organ index sampl view import aspect ac tree support onlin random sampl view time set record return ac tree constitut statist random sampl databas record satisfi relat select predic view paper present experiment result demonstr util ac tree learn correl mixtur dof dsubset model refer dbase index metric space costli distanc measur maintain larg random sampl geometr file scalabl approxim queri process dbo engin bayesian method guess extrem valu data set larg number data manag problem obtain sampl data set sampl guess largest smallest entir data set min max onlin aggreg top queri process outlier detect distanc join applic paper detail statist rigor bayesian approach attack problem importantli demonstr util approach show appli specif problem aris context data manag random algorithm data reconcili wide area aggreg queri process aspect data integr problem consid literatur match schema data sourc decid record refer entiti effici perform requir entiti resolut batch fashion larg effici deploi exist method realist distribut enterpris integr environ straightforward exist method requir data ship coordin clean unaccept develop set random algorithm effici applic exist entiti resolut method answer aggreg queri data distribut multipl site method effici gener aggreg queri result account duplic inconsist valu scatter feder system onlin random shuffl larg databas tabl condit anomali detect abstract anomali detect method identifi unusu interest pattern data recent propos condit anomali detect framework extend anomali detect problem identifi anomal pattern subset attribut data anomali depend condit remain attribut work present paper focus instanc base method detect condit anomali method reli distanc metric identifi exampl dataset critic detect anomali investig metric metric learn method optim perform instanc base anomali detect method show benefit instanc base method real world detect problem detect unusu admiss decis patient commun acquir pneumonia detect unusu order hpf test confirm heparin induc thrombocytopenia life threaten condit caus heparin therapi partit exponenti file databas storag manag materi sampl view databas approxim problem creat sampl view databas tabl sampl view index materi view permit effici sampl arbitrari rang queri view sampl view applic requir random sampl databas approxim queri process onlin aggreg data mine random algorithm exampl core technic contribut file organ call ac tree suitabl organ index sampl view import aspect ac tree support onlin random sampl view time set record return ac tree constitut statist random sampl databas record satisfi relat select predic view paper present experiment result demonstr util ac tree sampl dbase estim olap queri import sampl approxim queri process databas environ differ tradit applic sampl databas feasibl collect accur summari statist data addit sampl paper describ set sampl base estim approxim queri process make simpl summari statist greatli increas accuraci sampl base estim estim give tight probabilist guarante estim accuraci suitabl low high dimension data work categor numer attribut inform estim easili gather singl pass make suitabl stream environ closest dpoint dof dapproach join move object histori applic produc larg amount data describ path move object question interact object long record histori paper problem comput join massiv move object histori join studi closest point ofapproach join ask massiv move object histori object approach distanc carefulli obviou strategi comput answer join propos adapt join algorithm natur alter comput join respons characterist underli data outlier detect sampl accuraci guarante effect approach detect anomal point data set distanc base outlier detect paper describ simpl sampl algorithm effici detect distancebas outlier domain distanc comput expens unlik exist algorithm sampl algorithm requir fix number distanc comput return good result accuraci guarante computation expens aspect estim accuraci result sort distanc comput sampl algorithm enabl interact speed perform expens distanc comput paper algorithm test domain requir expens distanc function ten addit real data set experiment studi demonstr effici effect sampl algorithm comparison state theart algorithm reliabl accuraci guarante refer dbase index sequenc databas problem similar search larg sequenc databas edit distanc similar measur limit main memori goal develop refer base index reduc number costli edit distanc comput order answer queri idea refer base index select small set refer sequenc serv surrog sequenc databas strategi select refer strategi assign refer databas sequenc experiment result show select assign method outperform competit method method prune time sequenc omni method time sequenc frequenc vector method scale nice databas long sequenc sort dmerg dshrink join common oper analyt queri process applic aggreg function result relat join describ algorithm call sort merg shrink sm join comput answer queri larg disk base input tabl kei innov sm join input data cluster statist random fashion disk time join onlin statist estim eventu answer queri probabilist confid bound user monitor progress join execut stop join satisfi estim accuraci run algorithm complet total time requir longer common join algorithm contrast onlin join algorithm offer statist guarante offer guarante long input data fit main memori disk dbase join probabilist guarante common oper analyt queri process applic aggreg function result relat join describ algorithm comput answer queri larg disk base input tabl kei innov algorithm time onlin statist estim eventu answer queri probabilist confid bound user monitor progress join execut stop join satisfi estim accuraci run algorithm complet total time requir longer common join algorithm contrast onlin join algorithm offer statist guarante offer guarante long input data fit core memori relat confid bound easi bootstrap statist estim approxim queri process increasingli preval applic databas system approxim sort guarante estim accuraci confid bound analyt deriv probabilist guarante databas queri sampl data daunt task suitabl faint heart expertis typic databas system end user paper consid problem incorpor databas system power plug method comput confid bound answer relat databas queri sampl incomplet data statist tool call bootstrap simpl databas programm rudimentari mathemat background gener appli statist infer problem power eas bootstrap argu algorithm present support bootstrap incorpor databas system intend support analyt process onlin estim subset dbase sql queri largest databas todai larg answer queri minut hour dai address problem make approxim algorithm previou work onlin aggreg consid give onlin estim everincreas accuraci aggreg function relat join select queri exist work applic onlin estim subset base sql queri queri correl subqueri link outer queri exist exist orin claus queri asexcept intersect subset base queri paper develop algorithm onlin estim queri difficult problem provid probabilist accuraci guarante time queri execut find interest correl databas hard onlin mainten larg random sampl random sampl fundament data manag tool current research involv sampl consid problem sampl comput implicit assumpt sampl small data structur easili maintain data encount simpl statist argument demonstr larg sampl gigabyt terabyt size provid high accuraci exist work tackl problem maintain larg disk base sampl data manag perspect techniqu exist maintain larg sampl onlin manner stream data paper present onlin algorithm maintain disk sampl gigabyt terabyt size algorithm design stream data environ larg sampl maintain onlin singl pass data set algorithm meet strict requir sampl true statist random sampl replac data process algorithm suitabl bias unequ probabl sampl plai hide dand dseek correl robust estim sampl approxim pre daggreg major data reduct techniqu approxim queri process wavelet histogram kernel applic categor data disconnect research area realiti databas data recent research focus approxim queri process order numer attribut arguabl major databas attribut categor countri state job titl color sex depart paper consid problem approxim aggreg function categor data mix categor numer data propos method base random sampl call approxim pre aggreg apa biggest drawback sampl aggreg function estim sensit sampl attribut skew apa techniqu overcom sensit increas accuraci apa compar plain vanilla sampl dramat sum avg queri rel error random sampl greater sampl apa stratifi sampl techniqu error greater apa bridg gap respons time energi deffici broadcast schedul design tree work motiv observ hard disk seek capabl byte storag declin exponenti decad popular access structur base basic idea propos year ago access structur typic seek intens paper propos paradigm data access organ call linear file aim drastic reduc disk seek queri updat evalu organ data set ongo disk sort linear file updat process quickli structur pack data carefulli organ disk reduc seek disk page linear file gener templat data organ paper benchmark sm sm specif instanti dimension spatial data contact author paper id introduct tree close relat structur foundat physi lossi reduct high dimension data data reduct techniqu problem approxim queri answer focu applic accur answer select queri requir data high dimension hundr thousand dimens carefulli examin assumpt underli exist reduct techniqu ensur speed accuraci show method assum statist characterist high dimension dataset gener possess present data reduct method suffer limit call rs kernel demonstr effect method answer difficult highli select queri high dimension data real dataset contact author introduct statist model larg databas effect tool effici accur approxim queri answer ubiqu da maintain larg spatial databas sm effici data alloc multipl channel broadcast server maintain larg spatial databas sm comput complex high ddimension correl search approxim queri answer high ddimension data cube data mine work success compress model larg data set valu problem high dimens suffici address work develop data reduct method aim speed subsequ data analysi effici construct high dimension joint probabl distribut distribut summar data identifi subspac unexpect find region fewer data point expect propos data reduct method defin inher comput complex high dimension search mind lend effici approxim present result illustr effect method find accur approxim answer high dimension data cube queri problem statement huge high dimension numer categor dataset provid difficult set problem knowledg di index support high volum data warehous insert desir support fast ad hoc queri process larg data warehous motiv recent introduct index structur notabl except lsm tree step merg method attent develop index scheme fast insert addit larg warehous number million dai indic requir disk seek signific fraction seek insert accept paper offer altern tree call tree index huge warehous frequent insert tree index structur support point rang queri singl attribut retriev perform compar tree process insert tree exhibit speedup time batch insert tree comput program modular dcut method 