discrimin laplacian embed multi dlabel classif inconsist class balanc dnearest neighbor exploit user interest collabor filter interest expans person rank imag categor direct graph multi dlabel featur transform imag classif multi dlabel linear discrimin analysi improv minmax cut graph cluster nonneg relax direct graph learn high dorder dlinkag analysi collabor filter weight nonneg matrix factor incorpor user item graph bridg domain word opinion analysi matrix tri dfactor close form solut similar algorithm featur subset dneg matrix factor applic document understand convex semi dnonneg matrix factor present variat theme nonneg matrix factor nmf factor form focu algorithm restrict nonneg entri data matrix mix sign extend applic rang nmf method algorithm basi vector constrain convex combin data point kernel extens nmf provid algorithm comput factor provid support theoret analysi analyz relationship algorithm cluster algorithm implic spars solut final present experiment result explor properti method binari matrix factor analyz gene express data convex semi dnonneg matrix factor present variat theme nonneg matrix factor nmf factor form focu algorithm restrict nonneg entri data matrix mix sign extend applic rang nmf method algorithm basi vector constrain convex combin data point kernel extens nmf provid algorithm comput factor provid support theoret analysi analyz relationship algorithm cluster algorithm implic spars solut final present experiment result explor properti method proceed acm sigkdd workshop data mine matric tensor symmetr dimension linear discrimin analysi dlda dsubspac cluster abstract wide mean cluster deal ball shape spheric gaussian cluster paper extend mean cluster accommod extend cluster subspac lineshap cluster plane shape cluster ball shape cluster algorithm retain mean cluster flavor easi implement fast converg model select procedur incorpor determin cluster shape result algorithm recogn wide rang subspac cluster studi literatur global ball shape cluster live dimens carri extens experi synthet real world dataset result demonstr effect algorithm dneg laplacian embed label propag dpartit graph consensu group stabl featur select stabil import address issu featur select high dimension small sampl data paper show stabil featur select strong depend sampl size propos framework stabl featur select identifi consensu featur group subsampl train sampl perform featur select treat consensu featur group singl entiti experi synthet real world data set show algorithm develop framework effect allevi problem small sampl size lead stabl featur select result compar gener perform state art featur select algorithm synthet data set algorithm sourc code integr kl dmean laplacian cluster cluster approach combin attribut data pairwis relat knowledg transform cross ddomain sentiment classif dimens reduct techniqu cluster abstract paper compar perform dimens reduct techniqu svd nmf sdd qualit comparison evalu collect bar compar qualiti method base visual impact compar dimens reduct techniqu svd ho svd tensor bar tensor decomposit solut uniqu global converg hosvd parafac algorithm robust tensor factor norm kdd workshop report dmmt data mine matric tensor robust tensor factor norm tensor reduct error analysi applic video compress classif tensor base dimension reduct recent extens studi comput vision applic knowledg exist rigor error analysi method provid error analysi method provid error bound result similar eckart young theorem plai critic role develop applic singular decomposit svd perform guarante error bound subspac size determin requir video imag reconstruct error video surveil retriev medic imag analysi comput vision applic requir reduct spatio tempor space data index dimens motiv tensor reduct standard method high order svd hosvd compress data index dimens perform classif pattern recognit task provid algorithm error bound analysi factor spatio tempor data dimension experi video sequenc demonstr approach outperform previou dimension deduct method spatiotempor data simultan tensor subspac select cluster equival high order svd dmean cluster stabl featur select dens featur group featur select algorithm propos past focus improv classif accuraci work point import stabl featur select knowledg discoveri high dimension data identifi instabl featur select algorithm select minimum subset redund featur small sampl size propos gener framework stabl featur select emphas good gener stabil featur select result framework identifi dens featur group base kernel densiti estim treat featur dens group coher entiti featur select effici algorithm drag dens relev attribut group selector develop framework introduc gener measur assess stabil featur select algorithm empir studi base microarrai data verifi dens featur group remain stabl random sampl hold drag algorithm effect identifi set featur group exhibit high classif accuraci stabil weight consensu cluster consensu cluster emerg import extens classic cluster problem propos weight consensu cluster input cluster weight weight determin final consensu cluster qualiti solut cluster separ compar standard consensu cluster theoret show reformul wellknown regular lasso problem equival weight optim weight consensu cluster approach spars solut resolv difficult situat input cluster diverg significantli show weight consensu cluster resolv redund problem input cluster correl highli detail algorithm experi carri demonstr effect weight consensu cluster knowledg transform word space document space multi ddocument summar sentenc dlevel semant analysi symmetr matrix factor multi document summar aim creat compress summari retain main characterist origin set document approach statist machin learn techniqu extract sentenc document paper propos multi document summar framework base sentenc level semant analysi symmetr neg matrix factor calcul sentenc sentenc similar semant analysi construct similar matrix symmetr matrix factor shown equival normal spectral cluster group sentenc cluster final inform sentenc select group form summari experiment result duc duc data set demonstr improv propos framework implement exist summar system studi factor benefit high perform conduct posterior probabilist cluster nmf equival dneg matrix factor probabilist latent semant index effici parallel fo commun atmospher model cam effici parallel fo commun atmospher model cam kdd workshop report dmmt data mine matric tensor kdd workshop report dmmt data mine matric tensor dstage gene select algorithm combin relieff mrmr abstract gene express data larg number gene small number sampl featur select gene express data aim find set gene discrimin biolog sampl type paper present stage select algorithm combin relieff mrmr stage relieff appli find candid gene set stage mrmr method appli directli explicitli reduc redund select compact effect gene subset candid set perform comprehens experi compar mrmr relieff select algorithm relieff mrmr featur select method classifi svm naiv bay dataset experiment result show mrmr relieff gene select algorithm effect index term gene select algorithm relieff mrmr mrmr relieff binari matrix factor applic interest problem nonneg matrix factor nmf factor matrix specif class binari matrix paper extend standard nmf binari matrix factor bmf short binari matrix factor binari matric conserv import integ properti object matrix satisfi wh algorithm studi compar method reli fundament bounded properti nmf propos prove properti natur normal scheme elimin bia factor matric experi synthet real world dataset conduct show compet effect bmf solv consensu semi dsupervis cluster problem nonneg matrix factor adapt dimens reduct discrimin analysi dmean cluster combin linear discrimin analysi lda mean cluster coher framework adapt select discrimin subspac mean cluster gener class label lda subspac select cluster process integr subspac select process data simultan cluster featur subspac select show rich structur gener lda km framework examin variant relationship earlier approach extens experiment result real world dataset show effect approach find hotspot document collect learn framework green function kernel regular applic recommend system nonneg matrix factor probabilist latent semant index equival chi dsquar statist biclust protein complex interact bicliqu find algorithm biclust protein complex interact bicliqu find algorithm relationship nonneg matrix factor method cluster nonneg matrix factor nmf shown recent cluster extens nmf propos paper present overview theoret analyz relationship addit clarifi previous unaddress issu nmf normal cluster posterior probabilti nmf algoritm converg rate experi conduct empir evalu compar factor method dpca rotat invari dnorm princip compon analysi robust subspac factor supernova recognit support vector machin introduc applic support vector machin svm problem identifi potenti supernova photometr geometr featur comput astronom imageri challeng supervis learn applic signific noisi corrupt imageri result high level featur uncertainti featur heavi tail peak distribut extrem imbalanc overlap posit neg data set reach high posit classif rate find potenti supernova reduc burdensom workload manual examin fals posit high accuraci achiev sign preserv shift log transform appli featur peak heavi tail distribut imbalanc data problem handl oversampl posit exampl select sampl misclassifi neg exampl iter train multipl svm improv supernova recognit unseen test data present crossvalid result demonstr impact largescal supernova survei svm decis rank order potenti supernova night orthogon nonneg matrix dfactor cluster nmf plsi equival hybrid algorithm psol posit sampl learn algorithm find dcode rna gene dynam cluster format level set method transit closur metric inequ weight graph detect protein interact modul cliqu transit closur metric inequ weight multi dlevel approach scop fold recognit classif protein base structur plai import role deduct discoveri protein function low number solv protein structur unknown relationship structur sequenc requir altern method represent classif effect larg number potenti fold problem classif strategi increas likelihood classifi reach local optima distinguish structur categori present hierarch strategi structur classif partit protein base scop class attempt assign protein fold dataset deriv popul scop fold sequenc base descriptor properti input featur test number classif method includ na ve bay boost strategi achiev averag fold recognit significantli higher previous report literatur indic effect multi level approach nonneg lagrangian relax dmean spectral cluster probabilist approach optim spectral cluster spectral cluster enjoi success data cluster semisupervis learn spectral cluster algorithm handl multi class cluster problem directli addit strategi need extend spectral cluster algorithm multi class cluster problem spectral cluster algorithm emploi hard cluster membership trap local optimum paper present spectral cluster algorithm name soft cut improv normal cut algorithm introduc soft membership effici comput bound optim algorithm experi varieti dataset shown promis perform propos cluster algorithm dynam cluster format level set method cluster aggreg inequ multi dlevel hierarch cluster identif function modul protein complex hypercliqu pattern discoveri ddimension singular decomposit map imag equival nonneg matrix factor spectral cluster abstract current nonneg matrix factor nmf dealswith gt type provid systemat analysisand extens nmf symmetr hht weight hsht show hht equival kernel mean cluster ing laplacian base spectral cluster gt equival simultan cluster ofrow column bipartit graph algorithm comput symmetr nmf compar map sequenc dbase structur dbase protein domain evolv featur select describ evolutionari approach activ vision system dynam featur select summar recent work evolut simul activ retina complex shape discrimin describ detail experi extend approach terrain mobil robot equip mobil camera show evolv robot capabl select simpl visual featur activ maintain retin posit larg simplifi recognit task order gener effici navig trajectori extrem simpl neural control system analysi evolv solut robot develop simpl featur select base mutual inform criteria max ddepend abstract featur select import problem pattern classif system studi select good featur maxim statist depend criterion base mutual inform difficulti directli implement maxim depend condit deriv equival form call minim redund maxim relev criterion mrmr order increment featur select present stage featur select algorithm combin mrmr sophist featur selector wrapper select compact set superior featur low cost perform extens experiment comparison algorithm method classifi naiv bay support vector machin linear discrimin analysi data set handwritten digit arrhythmia nci cancer cell line lymphoma tissu result confirm mrmr lead promis improv featur select classif accuraci index term featur select mutual inform minim redund maxim relev maxim depend classif cpl extens coupl multicompon model mph distribut memori comput architectur term norm distribut effect latent semant index probabilist model latent semant index dimens reduct method latent semant index lsi appli semant space built text collect improv inform retriev inform filter word sens disambigu dual probabl model base similar concept introduc provid deeper understand lsi semant associ quantit character statist signific likelihood semant dimens redund noisi inform separ contribut statist signific neg lsi optim solut model peak likelihood curv exist intrins semant dimens import lsi dimens zipf distribut indic lsi dimens repres latent concept document frequenc word follow zipf distribut number distinct word log normal distribut experi standard document collect confirm illustr analysi posit sampl learn psol predict rna gene coli contract graph represent analysi rna secondari structur dmean cluster princip compon analysi princip compon analysi pca wide statist techniqu unsupervis dimens reduct mean cluster commonli data cluster unsupervis learn task prove princip compon continu solut discret cluster membership indic mean cluster equival show subspac span cluster centroid spectral expans data covari matrix truncat term result unsupervis dimens reduct close relat unsupervis learn dimens reduct result insight observ effect pca base data reduct convent nois reduct explan map data point higher dimension space kernel show solut kernel mean kernel pca learn result suggest effect techniqu mean cluster dna gene express internet newsgroup analyz illustr result experi newli deriv lower bound mean object optim valu linear cluster assign spectral order spectral cluster eigenvector laplacian similar matrix conveni appli cluster problem appli multi cluster spectral cluster recurs appli embed spectral space method cluster point propos studi cluster assign method method transform problem find vallei peak quantiti call cluster cross measur symmetr cluster overlap cut point linear order data point method determin cluster shot recurs split current cluster smaller show linear order base distanc sensit object continu solut eigenvector laplacian show close relationship cluster order method reli connect matrix construct truncat spectral expans similar matrix reveal cluster structur method appli newsgroup illustr introduc concept experi show outperform recurs cluster standard mean cluster integr program compon execut distribut memori architectur mph cluster structur dmean cluster princip compon analysi dnearest dneighbor consist data cluster incorpor local inform global optim princip compon analysi effect dmean cluster abstract wide adopt mean cluster algorithm sum squar error object function detail analysi show close relationship mean cluster princip compon analysi pca extens util unsupervis dimens reduct prove continu solut discret mean cluster membership indic data project princip direct princip eigenvector covari matrix lower bound mean object function deriv relat directli eigenvalu covari matrix experi internet newsgroup bound optim valu pca effect solut mean cluster princip compon analysi princip compon analysi pca multivari statist wide adopt effect unsupervis dimens reduct method extend direct main justif dimens reduct pca singular decomposit svd low rank approxim origin data norm due eckart young theorem essenti nois reduct perspect inadequ explain effect pca paper provid perspect pca base close relationship mean cluster algorithm show princip compon relax cluster membership indic background pca origin data point dimension space contain data matrix xn gener data center origin defin center data matrix yn yi xi xi covar matrix xi xi minimum redund featur select microarrai gene express data motiv select small subset thousand gene microarrai data import accur classif phenotyp wide method typic rank gene differenti express phenotyp pick top rank gene observ featur set obtain redund studi method minim result propos minimum redund maximum relev mrmr featur select framework gene select mrmr provid balanc coverag space captur broader characterist phenotyp lead significantli improv class predict extens experi gene express data set nci structur search stabil enhanc bayesian network data cluster princip compon hopfield daggreg network unsupervis featur select dwai order gene express analysi cluster merg split hierarch cluster algorithm hierarch cluster construct hierarchi cluster repeatedli merg smaller cluster larger split larger cluster smaller crucial step select cluster split merg provid comprehens analysi select method propos method perform extens cluster experi test select method find averag similar method divis cluster minmax linkag agglom cluster cluster balanc kei factor achiev good perform introduc concept object function satur cluster target distanc ectiv assess qualiti cluster adapt dimens reduct cluster high dimension data high dimension data cluster standard algorithm em mean trap local minimum initi method propos tackl problem limit success paper propos approach resolv problem repeat dimens reduct mean em perform low dimens cluster membership util bridg reduc dimension subspac origin space provid flexibl eas implement cluster analysi perform highli overlap gaussian dna gene express profil internet newsgroup demonstr ectiv propos algorithm unsupervis learn daggreg scale princip compon space analysi gene express profil class discoveri leaf order abstract approach class discoveri leaf order problem spectral graph partit methodolog class discoveri cluster present rain max cut hierarch cluster method show produc subtyp close human expert label lymphoma dataset class optim leaf order displai gene express data present sequenti dere method comput tz time preserv cluster structur show statist method statist test princip compon analysi gene express analysi mpi openmp paradigm cluster smp architectur vacanc track algorithm multi ddimension arrai transposit investig remap multi dimension arrai cluster smp architectur openmp mpi hybrid paradigm tradit method arrai transpos auxiliari arrai size copi back stage recent develop place method vacanc track cycl web document cluster hyperlink structur exponenti growth inform world wide web great demand develop effici effect method organ retriev inform document cluster plai import role inform retriev taxonomi manag world wide web remain interest challeng problem field web comput paper document cluster method explor textual inform hyperlink structur citat relat appli normal cut cluster method develop comput vision task hyperdocu cluster explor theoret connect normal cut method mean method experi normal cut method context cluster queri result set web search engin pagerank hit unifi framework link analysi popular webpag rank algorithm hit pagerank hit emphas mutual reinforc author hub webpag pagerank emphas hyperlink weight normal web surf base random walk model systemat gener combin concept unifi framework rank framework larg algorithm space evalu mpi openmp paradigm multi ddimension data remap web document cluster hyperlink structur exponenti growth inform world wide web great demand develop effici effect method organ retriev inform document cluster plai import role inform retriev taxonomi manag world wide web remain interest challeng problem field web comput paper document cluster method explor textual inform hyperlink structur citat relat appli normal cut cluster method develop comput vision task hyperdocu cluster explor theoret connect normal cut method mean method experi normal cut method context cluster queri result set web search engin bipartit graph partit data cluster data type aris data mine applic model bipartit graph exampl includ term document text corpu custom purchas item market basket analysi review movi movi recommend system paper propos data cluster method base partit underli bipartit graph partit construct minim normal sum edg weight unmatch pair vertic bipartit graph show approxim solut minim problem obtain comput partial singular decomposit svd edg weight matrix bipartit graph point connect cluster algorithm correspond analysi multivari analysi briefli discuss issu assign data object multipl cluster experiment result appli cluster algorithm problem document cluster illustr effect effici min dmax cut algorithm graph partit data cluster import applic graph partit data cluster graph model pairwis similar data object form weight graph adjac matrix inform cluster propos algorithm graph partit object function min max cluster principl relax version optim min max cut object function lead fiedler vector spectral graph partit theoret analys min max cut lead balanc partit lower bond deriv min max cut algorithm test newsgroup dataset found outperform current popular partit cluster method linkagebas nement algorithm improv qualiti cluster substanti demonstr linear search order base linkag di erenti base fiedler vector provid ectiv partit method automat topic identif webpag cluster group webpag distinct topic organ larg amount retriev inform web paper report base similar metric incorpor textual inform hyperlink structur citat relat unsupervis cluster method automat effect identifi relev topic shown experi retriev set webpag cluster method state art spectral graph partit method base normal cut criterion develop imag segment spectral method separ disconnect ddisconnect web graph compon separ connect compon graph disconnect graph compon breadth search bf depth search df graph algorithm propos algebra method separ disconnect disconnect compon method base spectral graph partit kei observ disconnect compon show properli sort step function curv lowest eigenvector laplacian matrix graph perturb analysi framework systemat analyz graph structur rst disconnect subgraph case ect ad edg spars connect di erent subgraph perturb result deriv provid insight spectral method relat cluster object function exampl illustr concept result method compar standard graph algorithm method jej jv jlog jv complex easier implement readili eigensolv method easili identifi articul point bridg disconnect graph segment real web graph queri amazon found disconnect disconnect compon form cluster clear topic spectral relax dmean cluster popular mean cluster partit data set minim sum squar cost function coordin descend method find local minima paper show minim reformul trace maxim problem gram matrix data vector show relax version trace maxim problem possess global optim solut obtain comput partial eigendecomposit gram matrix cluster assign data vector found comput pivot qr decomposit eigenvector matrix product deriv lower bound minimum sum squar cost function ghost cell expans method reduc commun solv pde problem solv partial differenti equat barotrop equat ocean model distribut memori comput finit differ method commonli processor subdomain boundari updat time step boundari updat process involv messag small size larg commun overhead propos approach expand ghost cell layer updat boundari frequent reduc total messag volum group small messag bigger techniqu elimin diagon commun method speedup commun substanti upto explain method implement detail provid systemat time result perform analysi crai ibm sp multi dclass protein fold recognit support vector machin neural network motiv protein fold recognit import approach structur discoveri reli sequenc similar studi approach multi class classic method examin issu import practic recognit system result current discrimin method protein fold predict againstoth method fals posit problem investig method uniqu method improv predict accuraci dataset scop fold support vector machin neural network learn method base classier svm converg fast lead high accuraci score multipl paramet dataset combin major vote reduc nois increas recognit accuraci examin issu involv larg number class includ depend predict accuraci number fold number repres fold recognit system achiev fold predict accuraci protein test dataset protein sequenc ident protein train contact chqding lbl gov ildubchak lbl gov supplementari inform protein paramet dataset paper onlin http www nersc gov cding protein keyword protein fold recognit protein structur multi class classic support vection machin neural network correspond address bipartit graph partit data cluster data type aris data mine applic model bipartit graph exampl includ term document text corpu custom purchas item market basket analysi review movi movi recommend system paper propos data cluster method base partit underli bipartit graph partit construct minim normal sum edg weight unmatch pair vertic bipartit graph show approxim solut minim problem obtain comput partial singular decomposit svd edg weight matrix bipartit graph point connect cluster algorithm correspond analysi multivari analysi briefli discuss issu assign data object multipl cluster experiment result appli cluster algorithm problem document cluster illustr effect effici high perform comput larg scale simul subsurfac multiphas fluid heat flow tough wide reservoir simul solv subsurfac flow relat problem nuclear wast geolog isol environment remedi soil groundwat contamin geotherm reservoir engin solv set coupl mass energi balanc equat finit volum method contribut present parallel version tough parallel implement partit unstructur comput domain time step set coupl linear equat solv newton iter newton step jacobian matrix calcul ill condit symmetr linear system solv precondit iter solver commun requir converg test data exchang partit border parallel perform result crai present real applic problem aris yucca mountain nuclear wast site studi execut time reduc second processor second processor problem involv equat larger problem equat time decreas second processor second processor accur arithmet improv numer reproduc stabil parallel applic accur arithmet improv numer reproduc stabil parallel applic parallel algorithm fft real number similar dbase probabl model latent semant index dual probabl model construct latent semant index lsi cosin similar measur document document similar matrix term term similar matrix natur aris maximum likelihood estim model paramet optim solut latent semant vector lsi dimension reduct justifi statist signific latent semant vector measur likelihood model lead statist criterion optim semant dimens answer critic open question lsi practic import model establish statist framework lsi ambigu relat statist model lsi clarifi high perform fortran practic scientif algorithm dto ddate evalu suit hpf code exampl practic scientif algorithm examin detail idea simpl trivial exampl fairli understand issu relat data distribut parallel construct program style static dynam alloc code exampl includ stencil solut pde bodi problem lu factor vector matrix librari routin arrai redistribut perform hpf code compar hand written fortran code messag pass librari hpf compil improv significantli hpf code perform fortran mpi code exampl investig import peculiar hpf code exist keyword hpf distribut memori architectur parallel program languag mpi stencil comput bodi problem lu factor arrai redistribut fft introduct high atmosper data assimil distribut dmemori parallel supercomput evalu hpf practic scientif algorithm hpf code exampl practic scientif algorithm examin detail idea simpl trivial exampl fairli understand issu relat data distribut parallel construct program style static dynam alloc code exampl includ stencil solut pde bodi problem lu factor vector matrix librari routin arrai redistribut perform hpf code close hand written messag pass mpi code lu factor vector matrix routin arrai redistribut regular data commun pattern stencil calcul bodi problem hpf code perform consider effici time slower scale hpf code good mpi code hpf code perform highli inconsist minor chang code result factor chang perform peculia element dbase concurr partition unstructur finit element mesh abstract concurr partition partit unstructur finit element mesh distribut memori architectur develop partition element base partit strategi main advantag convent node base partit strategi modular program approach develop parallel applic partition partit element centroid recurs inerti bisect algorithm element node migrat partit centroid data request commun templat unpredict incom messag scalabl implement contrast scalabl implement straightforward parallel sequenti partition algorithm adopt partition scale logarithm confirm actual time measur intel delta processor scale size problem gener purpos spars matrix parallel solver packag introduct common problem scientif engin comput constrcut solut spars linear system aris solv partial differenti equat base unstructur grid exampl occur analysi finit element finit differ finit volum method scale complex calcul grow massiv parallel comput increasingli wide calcul exist parallel spars linear system solver classifi distinct class refer iner conjug gradient type solver kei parallel part matrix vector product parallel choleski factor solver column row base matrix distribut solver restrict posit definit matric larger classs symmetr complex matrix system fe gener purpos spars matrix parallel solver packag introduct common problem scientif engin comput constrcut solut spars linear system aris solv partial differenti equat base unstructur grid exampl occur analysi finit element finit differ finit volum method scale complex calcul grow massiv parallel comput increasingli wide calcul exist parallel spars linear system solver classifi distinct class refer iner conjug gradient type solver kei parallel part matrix vector product parallel choleski factor solver column row base matrix distribut solver restrict posit definit matric larger classs symmetr complex matrix system fe slice scalabl partition finit element mesh parallel partition partit unstructur finit element mesh distribut memori architectur develop element base partition handl mixtur element type algorithm adopt partition scalabl includ commun templat unpredict incom messag shown actual time measur introduct partit finit element mesh processor parallel supercomput set stage finit element analysi problem domain partit achiev load balanc preserv proper data local reduc commun solut problem handl larg mesh avoid excess io partition partit mesh concurr parallel comput contrast exist partition partit mesh sequenti comput download partit mesh distribut memori comput mesh decomposit element base parti portabl fft packag distribut dmemori parallel architectur parallel algorithm fft implement seri local fft combin data transpos vendor suppli fulli optim sequenti fft fft carri place inplac data transpos processor introduct multi dimension fft frequent engin scientif calcul imag process parallel implement fft gener follow approach binari exchang approach data exchang place pair processor processor number differ bit transpos approach multi dimension fft fft carri success local sequenti fft data transpos occur inter processor commun place data transpos advantag approach vendor suppli fft fulli optim furth mont carlo simul quantum system massiv parallel comput 