integr data mine busi process svm classifi estim group probabl learn problem recent gain attent machin learn commun learn classifi group probabl learn task li task supervis unsupervis learn sens set observ label group observ frequenc distribut label learn problem import practic applic privaci preserv data mine paper present approach learn classifi group probabl base support vector regress idea invert classifi calibr process detail analysi show approach outperform exist approach privaci dpreserv data dmine paper address issu privaci preserv data mine specif scenario parti own confidenti databas run data mine algorithm union databas reveal unnecessari inform work motiv protect privileg inform enabl research purpos build system advanc clinico dgenom trial cancer abstract analysi clinico genom data pose complex comput problem project acgt grid base softwar system support clinician bio statistician daili work develop start detail user requir analysi continu integr usabl analysi develop process project strive develop architectur substanti improv clinico genom trial conduct todai paper result initi requir analysi approach address requir present discuss import metadata tailor system user subgroup discoveri numer domain subgroup discoveri numer domain gridr dbase tool scientif data analysi grid environ journal homepag www elsevi locat fgc tight optimist estim fast subgroup discoveri abstract subgroup discoveri task find subgroup popul exhibit distribut unusu high gener due monoton evalu function standard prune techniqu subgroup discoveri requir optimist estim techniqu optimist estim prune consid extrem simpl case binari target attribut attempt made move suboptim heurist optimist estim paper show optimist estim prune develop sound highli effect prune approach subgroup discoveri base precis definit optim show previou estim tight special case present tight optimist estim popular binari multi class qualiti function present famili increasingli effici approxim optim function show empir experi newli propos optimist estim lead speed order magnitud compar previou approach semant grid servic architectur support effici knowledg discoveri multilevel clinic genom dataset paper present architectur consider advanc clinico genom trial cancer acgt project aim deliv european biomed grid support effici knowledg discoveri context post genom clinic trial cancer main research challeng acgt requir develop infrastructur produc deploi knowledg basic element advanc applic constitut biomed knowledg grid approach offer semant model servic data sourc support high level servic dynam servic discoveri composit present ontolog metadata basic element grid intellig servic develop current achievmnet project domain discuss support parallel code clinic trial grid dbase approach paper describ extens acgt gridr environ parallel loop script view distribut execut comput grid acgt gridr servic extend compon set preprocessor direct organ distribut calcul parallel direct special comment user potenti acceler lengthi calcul preexist code gridr servic extens develop compon acgt platform aim facilit data mine clinic trial involv larg dataset acgt gridr script execut framework specif develop workflow environ briefli outlin present articl benchmark domain specif processor case studi evalu smart card processor design procur fraud discoveri similar measur learn abstract paper describ approach detect hint procur fraud develop context european union project fraud prevent procur fraud special kind fraud occur employe cheat employ execut trigger bogu payment approach present base idea learn similar measur compar employe payrol stand data record creditor record order detect creditor suspici similar employe end combin simpl similar measur address similar spacial similar weight scheme weight similar function learn user input pair payrol creditor data record similar kei word similar measur learn fraud detect health care facilit clinico dgenom knowledg discoveri automat select kdd process analysi clinico genom data pose complex problem machin learn high volum data gener easili select suitabl kdd process problem hand increasingli hard experienc research main idea paper facilit process select repres data set graph base ontolog describ data set attribut appli graph mine method perform similar search measur effect comparison data set graph induc ontolog propos effect propos approach evalu dataset result show ontolog base characterist lead improv character data set knowledg discoveri scientif workflow clinico dgenom robust probabilist calibr probabilist calibr task produc reliabl estim condit class probabl class observ output numer classifi recent compar studi reveal isoton regress platt calibr ectiv probabilist calibr techniqu wide rang classifi grid dbase knowledg discoveri clinico dgenom data knowledg discoveri clinico genom data task requir integr highli heterogen kind data requir interest di erent user group technolog grid comput promis ectiv tool combin requir singl architectur paper describ scenario futur research direct relat grid base knowledg discoveri clinico genom data introduc approach recent launch acgt project endeavor consid context biomed informat research aim realiz integr grid enabl biomed infrastructur present integr clinico genom knowledg discoveri icgkd scenario process realiz base multi strategi data mine approach seamlessli integr distinct data mine compon cluster associ rule mine featur select preliminari experiment result indic ration reliabl approach simpl method estim condit probabl svm support vector machin svm popular learn algorithm larg high dimension classif problem learn local model contribut deal activ learn learner power select train sampl propos queri algorithm local learn model class learner consid context activ learn queri algorithm base idea select queri borderlin actual classif draw geometr properti local model typic induc voronoi tessel input space voronoi vertic tessel offer prospect queri point perform queri algorithm test spiral problem promis result keyword activ learn local model queri base learn vector quantiz introduct supervis learn interest train student set input output pair gener unknown target function student concept drift import learn task data collect extend period time underli distribut chang typic inform ltere adapt classi cation document respect user interest interest user document content chang time ltere system adapt concept multistrategi approach classif phase busi cycl classif busi cycl perform macro econom equat linear discrimin analysi hard import problem classifi econom phase govern busi decis reli assess current busi cycl paper investig economist support combin machin learn techniqu successfulli appli induct logic program ilp applic ilp requir pre process order establish time interv end top induct decis tree rule set learn experi analys respect correl order find concept drift shift support vector machin relat databas todai data busi applic store relat databas support vector machin relat databas todai data busi applic store relat databas increment learn support vector machin support vector machin svm popular tool learn larg amount high dimension data prefer learn increment previou svm result comput svm costli term time memori consumpt svm onlin learn set paper approach increment learn support vector machin present improv exist approach empir evid prove approach effect deal target concept result increment learn set som acceler system high perform sofm hardwar dsystem applic selforgan featur map sofm high perform hardwar system order effici regular modular structur sofm hardwar realiz obviou base idea massiv parallel system chip design manufactur test author paper high perform system latest nbisom chip present nbisom integr circuit process element arrai due scalabl chip vme bu board built ic control vme bu sofm hardwar realiz fpga system run sofm applic element parallel mode map weight vector weight bit accuraci maximum perform board system gcp recal gcup learn high perform sofm hardwar dsystem applic selforgan featur map sofm high perform hardwar system order effici regular modular structur sofm hardwar realiz obviou base idea massiv parallel system chip design manufactur test author paper high perform system latest nbisom chip present nbisom integr circuit process element arrai due scalabl chip vme bu board built ic control vme bu sofm hardwar realiz fpga system run sofm applic element parallel mode map weight vector weight bit accuraci maximum perform board system gcp recal gcup learn hardwar design dorgan featur map binari input vector number applic organ featur map requir power hardwar algorithm sofm multipl larg chip area fast implement hardwar paper restict class organ featur map investig hardwar aspect fundament idea restict chip area processor element map smaller element chip work parallel binari input vector manhatten distanc special treatment adapt factor effici implement hardwar design algorithm present vhdl simul show perform mcp million connect recal phase mcup million connect updat learn phase map standard cell layout processor element full custom design import part 