confid dbase featur acquisit minim train test cost progress refin support vector machin model learn user prefer set dmean space radiat sensit evalu spacecraft increasingli emploi onboard data analysi inform data collect priorit decis spacecraft oper high radiat environ reliabl dataintens comput paper present studi radiat sensit mean cluster kei find mean data structur differ sensit determin sole amount memori expos special radiat protect need dataset depend radiat threshold enabl faster smaller cheaper onboard memori subsampl improv radiat toler slightli kdtree reduc toler conclus tailor mean futur high radiat environ aaai workshop report multipl dinstanc regress structur data dboard analysi uncalibr data spacecraft mar report sigart doctor consortium constrain cluster benefici research illustr constraint improv result varieti cluster algorithm larg variat improv fix number constraint data set present attempt provid insight phenomenon character constraint set properti inconsist incoher show measur strongli anti correl cluster algorithm perform comput prior cluster measur aid decid constraint practic activ learn irrelev exampl abstract activ learn algorithm attempt acceler learn process request label inform item real world problem exist unlabel item irrelev user classif goal queri point slow learn provid inform problem interest observ irrelev item present activ learn perform wors random select requir time queri achiev level accuraci propos approach relev bia activ learner combin default select heurist output simultan train relev classifi favor item inform relev experi real world problem benchmark dataset relev bia approach significantli improv learn rate activ learn approach learn user prefer set object work prefer learn focus pairwis prefer rank individu item paper present method learn prefer set item learn method take input collect posit exampl set identifi user desir kernel densiti estim estim function individu item desir set divers estim averag set divers observ collect learn problem introduc evalu methodolog evalu learn method data collect synthet block world data real world music data collect gather cost share open issu constrain cluster cluster import tool data mine identifi major pattern trend supervis label data measur constraint dset util partit cluster algorithm abstract cluster constraint activ area machin learn data mine research previou empir work convincingli shown ad constraint cluster improv perform varieti algorithm experi result averag randomli chosen constraint set set label mask interest properti individu set demonstr constraint set vari significantli constrain cluster constraint set decreas algorithm perform creat quantit measur inform coher identifi constraint set show measur explain differ perform constrain cluster algorithm sigart faaai doctor consortium dd dpref languag express prefer set applic domain repres reason user prefer set object present represent languag dd pref divers depth prefer desir divers depth set object object repres vector featur valu strong divers prefer featur user set includ object valu evenli dispers rang valu featur strong depth prefer featur user interest specif target valu rang divers depth complementari necessarili opposit defin object function maxim identifi subset object satisfi statement prefer dd pref exhaust search space subset intract larg problem space present effici greedi algorithm gener prefer object subset demonstr express power dd pref perform greedi algorithm encod appli qualit prefer multipl task block world data set final provid experiment result collect mar rover imag demonstr successfulli captur individu prefer user retriev high qualiti imag subset mine gp trace map refin constrain dmean cluster background knowledg cluster tradition view unsupervis method data analysi case inform problem domain addit data instanc paper demonstr popular mean cluster algorithm pro tabli modi ed make inform experi arti cial constraint data set observ improv cluster accuraci appli method real world problem automat detect road lane gp data observ dramat increas perform cluster instanc dlevel constraint cluster algorithm conduct search space organ data set paper propos type instanc level cluster constraint link link constraint show incorpor cluster algorithm aid search data set test result incorpor surprisingli constraint increas cluster accuraci decreas runtim investig rel effect type constraint find type contribut accuraci improv depend behavior cluster algorithm constraint gener cluster supervis learn data assign cluster algorithm increasingli import handl analyz data consider work devis effect increasingli specif cluster algorithm contrast develop gener framework accommod divers cluster algorithm systemat framework view cluster gener process iter optim includ modul supervis learn instanc assign framework suggest cluster method paper investig experiment efficaci algorithm test hypothes relat unsupervis techniqu supervis method embed cluster instanc dlevel constraint cluster algorithm conduct search space organ data set paper propos type instanc level cluster constraint link link constraint show incorpor cluster algorithm aid search data set test result incorpor surprisingli constraint increas cluster accuraci decreas runtim investig rel effect type constraint find type contribut accuraci improv depend behavior cluster algorithm constraint refin induct bia unsupervis learn constraint refin induct bia unsupervis learn constraint cluster instanc dlevel constraint cluster algorithm conduct search space organ data set paper propos type instanc level cluster constraint link link constraint show incorpor cluster algorithm aid search data set test result incorpor surprisingli constraint increas cluster accuraci decreas runtim investig rel effect type constraint find type contribut accuraci improv depend behavior cluster algorithm constraint alpha seed support vector machin kei practic obstacl appli support vector machin larg scale data mine task svm gener scale quadrat wors number exampl support vector complex compound specif svm train leav cross valid loocv determin optim svm kernel paramet wrapper base featur select paper explor techniqu reduc amort cost svm train seed success svm train result previou similar train 