A Proactive Fault-detection Mechanism in Large-scale Cluster Systems

Size: px
Start display at page:

Download "A Proactive Fault-detection Mechanism in Large-scale Cluster Systems"

Transcription

1 A Poactve Fault-detecto Mechasm age-scale Cluste Systems Wu pg Meg Da Gao We ad Zha Jafeg Isttute of Computg Techology Chese Academy of Sceces Bejg Cha {wlp md gw Gaduate School of the Chese Academy of Sceces Bejg Cha Abstact To mpove the whole depedablty of lage-scale cluste systems a ole fault detecto mechasm s poposed ths pape. Ths mechasm ca detect the fault tme befoe ode fals ad eables the poactve fault maagemet. The poposed mechasm s summazed as follows: Fst the dyamc chaactestcs of cluste system ug omal actvty ae bult usg Tme Sees Aalyss methods. Secod the fault detecto pocess s mplemeted by compag the cuet ug state of cluste system wth omal ug model. The fault alam decso s made mmedately whe the cuet ug state devates the omal ug model. The expemet esults show that ths mechasm ca detect the fault cluste system good tme.. Itoducto The depedablty ad avalablty ae the majo challeges fo the lage-scale cluste systems. The eseaches elated to cluste system depedablty ae dvded to two categoes appoxmately. Oe s Hgh Avalablty HA techque based o edudat compoets ad ams at povdg cotuous sevce. Whe oe of the odes fals the fal-ove pocedues wll beg mmedately ad the applcatos ug o faled ode ae mgated to aothe edudat ode[]. The othe s based o the pedcto mechasm. The faults ae detected tme ad the coectly poactve mateace pogam s mplemeted befoe ode fals. Ths method avods the costs of uplaed dowtme of odes. Ths method s amed as Poactve Fault Maagemet seveal elated eseaches[]. The key questo fo Ths wok s suppoted by the Natoal 863 Hgh-Tech Pogam of Cha No. 4AA66 ad the 5th key poject of Cha No. 4BA8B9-. poactve fault maagemet s how to detect the fault system quckly that's effcet fault detecto mechasm. I fact the fault detecto mechasm has bee appled to some msso-ctcal ad safety-ctcal felds such as the acaft fault detecto[4] ad uclea powe plat fault detecto[5]. Now two elated woks compute systems ae dscussed. Oe s detecto of softwae agg pheomea ad the othe s SMART techology fo had dsk falue pedcto. Softwae agg[6] s oe kd of softwae fault whch s a pheomeo that the state of the softwae system degades o cashes wth tme. The measuemet-based ejuveato appoach deals wth detecto of the exstece of softwae agg ad pedctg agg-elated falues. I[7] the opeatg system esouce usage ad system actvty data ae collected at egula tevals fom etwoked UNIX wokstatos. A statstcal ted detecto techque s appled to the collected data to detect the exstece of agg ad the estmated tme to exhausto s calculated usg a o-paametc slope estmato techque. I[8] the dyamc chaactestc of web seve s bult usg tme-sees ARMA model ad the ARMA model s used to detect agg ad estmate esouce exhausto tme. To mpove the elablty of had dsk the SMART[9] Self-Motog Aalyss ad Repotg Techology falue pedcto system s cuetly mplemeted dsk dves. The SMART ca detect the fault had dves befoe dsk fals. The eseaches elated to SMART focus o the fault detecto ad falue pedcto techques. Fo example [9][] the geeal famewok s to detect aomales o vaatos fom omal behavo usg a ak-sum ull hypothess test; [] the fault detecto pocess s mplemeted usg dffeet statstcal tests based o ave Bayesa classfes. Both softwae agg detecto ad SMART methods focus o pat of system o pat of fault models. I ths pape we gve a poactve fault detecto mechasm fo lage-scale cluste systems fom system vew. System motog s a ecessay compoet fo lage-scale cluste systems. Fo example the system motog BlueGee/ s accomplshed though a combato of I/O ode ad sevce ode fuctoalty[]; fo /6/$. 6 IEEE

2 Dawg4A[3] the system motog fucto s accomplshed by the detecto ad data bullet sevce[4] the cluste opeatg system Phoex[5]. The pefomace paametes elated to system actvty whch s collected by system moto at egula tevals ae vaous such as the usage of CPU ad Memoy o Swap the voltage of odes the otate speed of fa ad the tempeatue of odes. Whe faults occu cluste system the actual obsevato of those paametes by system moto wll devate fom the omal ug model. So ole aalyss the state fomato deved fom cluste motog system ca atta the symptom of fault. Cuetly the aalyss of state fomato s smple theshold detecto method: f oe of the pefomace paametes volates the acceptable theshold the fault maagemet system wll cosde that the fault occued cluste system ad sed out a alam. Fo example a alam wll be set to admstato whe the load of CPU exceeds 9%. The met of theshold detecto s smple to mplemet ad appopate whe the sgals ae statc o slowly vayg. I cluste system the pefomace paametes ae vayg wth tme so theshold detecto method s appopate. Ths pape poposed a poactve ole fault detecto mechasm based o the actvty data collected by cluste system moto. The actvty data descbes the system ug state. Fst the omal ug models ae bult usg tme sees aalyss o collected data whe cluste us omally. Secod the dffeece betwee a paamete s pedcted value by omal ug model ad ts dectly obseved value s temed a esdual sees. All paametes esdual sees should be whte ose sees whe the cluste system s behavg omally. So the esdual sees gves the symptom of faults. I the ed the kelhood Rato Test I the emade RT fo shot o the esdual sees s used to detect faults. The pape s ogazed as follows. Secto gves the fault detecto pocedue ovevew. The ma algothms cludg modelg the cluste omal actvtes ad the RT method ae gve secto 3. I secto 4 we peset the expemet esults ad secto 5 cocludes wth futue wok.. Fault detecto pocedue The fault detecto pocedue s llustated fgue. Supposed that oe applcato exclusve use of the odes that t us o whe oe job maked AppM s ug o oe ode maked N wthout faults maked <N AppM> the pefomace paamete S s sampled at egula tevals ad the tme sees SN t s fomed. The omal ug model of <N AppM S> s bult usg the tme sees model of SN t. The pocedue of buldg the tme sees model s dscussed secto 3. detal. Fgue. The ole fault detecto pocedue Node N Nomal ug models Cluste system moto f Actvty data Resdual sees The fault detecto fucto f Result of f? f Fault alam The tme sees model ca pedct the futue value of paamete S. Afte buldg the omal ug models of evey <N AppM S> we popose a ole fault detecto method based o Slde Wdow. Defe the wdow W<o p > t-l+ <o p > t l fo evey <N AppM S> whee the tple <o p > t-l+ deotes the actual obseved value o the pedcted value p by the omal ug model ad the esdual o-p at t-l+ tme ad <o p > t deotes the actual obseved value o the pedcted value p by the omal ug model ad the esdual o-p at t tme; l s the legth of W. Now the fault detecto pocedue s descbed as follows: Fst step: omal ug model selecto. Select the omal ug model maked M based o tple <N AppM S> whee N s ode ame AppM s job ame ad S s oe of the pefomace paametes. Secod step: wdow W moves fowad oe pace. At t tme the wdow s W<o p > t-l+ <o p > t l. At t+ t tme the wdow moves fowad oe pace ad s W<o p > t+ t-l+ <o p > t+ t l whee p t+ t s the pedcted value by model M o t+ t s the obseved value collected fom system moto the esdual at t+ t tme s t+ t o t+ t p t+ t. Thd step: fault detecto. Rl t s oe l-legth esdual sees W<o p > t+ t-l+ <o p > t+ t l. Usg Rl t as the put of fault detecto fucto f ad calculate the esult of f. If f oe fault alam decso s made. Othewse cotue the fault detecto pocess ad etu to the secod step. The defto of fault detecto fucto s gve 3.3.

3 Fouth step: f AppM s ove ad aothe ew job s load etu to the fst step. 3. The ma algothms Fom above aalyss the key questo s buldg the omal ug models of evey job o evey ode ad usg the esdual sees betwee obseved value ad pedcted value of pefomace paametes as the symptom of fault detecto. Based o the tme sees aalyss techque the esdual sees should be a whte ose sees ad the dstbuto of the esdual sees follows the omal dstbuto N. If fault occus cluste system the esdual sees wll devate fom the whte ose sees ad the dstbuto of the esdual sees does t follow the omal dstbuto N. The RT method s used to costuct the fault detecto fucto f. 3. Modelg the omal actvtes Whe job AppM s ug o ode N wthout faults oe of the pefomace paametes maked S such as usage of CPU ad Memoy etwok badwdth etc s sampled at egula tevals by the cluste motog system. The sample cycle s t ad oe tme sees maked SN t s fomed whee N s the legth of sees. The dyamc model s bult usg tme sees aalyss o SN t ad ths model descbes the omal ug model of <N AppM S>. The followg gves the detaled pocedue of buldg the omal ug model of <N AppM S> usg tme sees aalyss o SN t. Fst step: Data Pepocess. Fo some upedctable easos SN t s ot a statoay tme sees. To mpove the accuacy of fault detecto two data pepocess pocedues ae ecessay fo SN t. Oe s movg aveage smoothg method ad the othe s computg the zeo aveage of SN t. Movg aveage smoothg: the -pot movg aveage of SN t s S N-+ t ad s + s. j j Zeo aveage: suppose µ s the aveage of tme sees S N-+ t that s µ s N +. The zeo aveage of S N-+ t s ZN-+ t whee evey sample s z s µ. Secod step: buldg the ARIMA model. ZN-+ t s t a statoay tme sees geeally. Thee s ted t. Hece we wll fst emove the ted by seveal dffeecg pocedues. Defe the oe pace dffeecg opeatos s z t z t -z t- -Bz t whee B s the backshft opeato so the d-pace dffeecg s d z t -B d z t. The d-pace dffeecg of ZN-+ t s XN-+-d t whee the sample of ew sees s d j x d d t z t C z t j t > d j d Fo each d3 calculate the Auto Coelato Fucto ACF ad Patal Auto Coelato Fucto PACF of XN-+-d t espectvely tll the XN-+-d t ca be egaded as a statoay tme sees Whe the ACF of the tme sees s quckly deceasg ths s a dcato that the tme sees s statoay. Buldg the ARMAp q model fo XN-+-d t: p q x t j j x t a t j a t ϕ whee p s autoegessve ode of the model q s the movg aveage ode of the model a t s a whte ose sees wth mea zeo ad vaace ad s ad s ae costats. If q the model educes to ARp. Ad f p the model p educes to MAq. Suppose ϕ B ϕ B ad q B B. So the above model ca be descbed as: Bx t Ba t The key step s detemg the ode of the model that s the value of p q. The ACF ad PACF povde mpotat fomato of p q. Table gves the behavo of the ACF ad PACF fo ARp MAq ad ARMAp q models. Table. ACF ad PACF fo AR MA ARMA ARp MAq ARMApq ACF Tals off Cuts off afte lag q Tals off PACF Cuts off afte lag p Tals off Tals off Fo ARp ad MAq the value of p o q ca be detemed by the cuts off of PACF o ACF. Fo ARMAp q two key pots to deteme p q ae: The ACF of ARMAp q begs decay at lag q. The PACF of ARMAp q decays statg at lag p. Afte deteme the ode of ARMAp q the values of s ad s ae estmated by east Squaes method. Fom ad the ARIMA model of ZN-+ t s B d z t Ba t t N-+ 3 Thd step: the omal ug model. Fom the aalyss of fst ad secod steps the omal ug model of <N AppM S> s descbed as follows: ϕ B d s t µ B at t N-+ 4 whee s t s the sample of pefomace paamete S at tme t. Fo d s d s t µ t fomula 4 ca be descbed as: ϕ B d s t B at t N-+ 5

4 The model 5 gves the omal actvtes fo <N AppM S>. If AppM s ug o ode N wthout faults the pefomace paamete S should follow model 5 ad the esdual sees {a t } betwee actual obseved sees ad pedcted sees should be whte omal ose. Othewse S wll devate model 5 ad the esdual sees {a t } wll ot be whte omal ose aymoe. So the fault detecto questo cluste system s tasfomed as testg whethe the esdual sees s whte omal ose o ot. We use RT method to vefy whethe the esdual sees s whte omal ose o ot. 3. kelhood Rato Test Fo a -legth esdual sees R t the ull hypothess ad alteatve hypothess ae defed as: H : R t s whte omal ose wth mea zeo cluste system s ug omally; H : The mea of R t s ot zeo thee ae faults cluste system. The level of sgfcace P{eject H H } s the pobablty of type I eo the pobablty of ejectg ull hypothess that s tue ad s the false-alam ato. Assume oe tme sees whose samples come fom omal dstbuto N whee s mea ad s vaace paamete space s { ; < < < < }. So the above hypothess test ca be descbed as follows: H : R t~ N ad > H : R t~ N ad > The subset { ; < < } of s the paamete space fo H. It s a composte vesus composte test ad we use RT method to do ths. The lkelhood fuctos of R t o ad ae: exp. ; π ω Ω exp. ; π x et ωˆ max ad Ωˆ max that s exp ˆ e π π ω Ω exp ˆ π e π whee. So the lkelhood ato s ˆ ˆ + Ω ω ad the test wll eject H f s small. The ctcal ego of fo H s. I aothe wods we wll eject H ad accept that thee ae faults cluste system whe. The value of s calculated as: + that s c 6 The left of fomula 6 s a adom vaable maked t ad t s a t-dstbuto wth - degees of feedom. Fo the gve ad the sgfcace level the value of c ca be foud the t-table based o P[ t c; H ]. So c +. Fo the esdual sees R t we eject H f. Othewse we accept H. 3.3 The fault detecto fucto Fo the esdual sees Rl t wdow W<o p > t+ t-l+ <o p > t+ t l we use RT method to test whethe Rl t s whte omal ose wth mea zeo o ot. Fo Rl t wdow W<o p > t+ t-l+ <o p > t+ t l defe the fucto > + f f t m t whee ad ae defed secto 3.. To get d of the acto of ose we toduce the pesstece checkg paamete w to the Fault Detecto Fucto FDF. The FDF s defed as s w s m s f +. If f oe fault alam decso s made. It s obvously that oe fault alam s tggeed ff the umbe of cotuous m s w. 4. Expemet

5 I ths secto we wll pove the utlty of ou poactve fault detecto mechasm by expemetal method. I the expemets the fault jecto s used to smulate the fault evomet. 4. Buldg the omal ug models The expemetal setup ths pape cossts of fou odes ode ode ode3 ode4 whose cofguatos ae dual P3 GHz CPU G Memoy 8GB had dsk ad ug ux.4.-8smp keel. Fst thee MPI jobs App App App3 ae ug o these odes wthout faults tu ad the ug tme s secods espectvely. At the same tme the cluste motog system collects the pefomace paametes of evey ode at egula tevals ad the sample cycle s 5 secods. Based o the coss coelato aalyss we select 9 pmay pefomace paametes fom about 8 pefomace paametes of ux opeatg system ad these 9 pefomace paametes ae show table. We buld the omal ug models of evey oe of 9 pefomace paametes ad use these models as the bass of fault detecto. Fo each <N AppM S> whee N {ode ode ode3 ode4} ad AppM {App App App3} ad S s oe of pefomace paametes table we buld the omal ug model usg the collected data. The follow gves a example: whe App s ug o ode wthout faults the omal ug model of actvepg s bult usg the collected data. The pocess of buldg the omal ug model of <ode App actvepg> s gve Fgue : a s the ogal sample data the umbe s 48; b s the data afte 3-pot movg aveage ad zeo aveage the aveage µ the fot 3 sample data ae goed ad the umbe of emade data s 45; c s the ACF k 5 fo b data the ACF of b s ot quckly deceasg so b s ot oe statoay tme sees; d s the data afte oe pace dffeecg of b; e ad f ae the ACF k ad PACF k of d espectvely accodg to table the d ca be modeled Table. Pefomace paametes Num Name Descpto Fle_sz Numbe of used fle hadles. Iode_sz Numbe of used ode hadles. 3 Supe_sz Numbe of supe block hadles allocated by the keel. 4 Dquot_sz Numbe of allocated dsk quota etes. 5 Rtsg_sz Numbe of queued RT sgals. 6 actvepg Numbe of actve ecetly touched pages memoy 7 Txpck_ps Total umbe of packets tasmtted pe secod 8 kbmemused Amout of used memoy klobytes. Ths does ot take to accout memoy used by the keel tself. 9 kbswpused Amout of used swap space klobytes. Cswch_ps Total umbe of cotext swtches pe secod. Ruq_sz Ru queue legth umbe of pocesses watg fo u tme davg_oe System load aveage fo the last mute 3 Pecetage_use Pecetage of CPU utlzato that occued whle excutg at the use level. 4 Pecetage_system Pecetage of CPU utlzato that occued whle excutg at the system level. 5 Fmpg_ps Numbe of memoy pages feed by the system pe secod. A egatve value epesets a umbe of pages allocated by the system a page has a sze of 4 kb o 8 kb accodg to the mache achtectue. 6 Iteupt_ps Total umbe of teupts eceved pe secod. 7 Tps Total umbe of tasfes pe secod that wee ssued to the physcal dsk. A tasfe s a I/O equest to the physcal dsk. Multple logcal equests ca be combed to a sgle I/O equest to the dsk. 8 Txe_ps Total umbe of eos that happeed pe secod whle tasmttg packets. 9 Txdop_ps Numbe of tasmtted packets dopped pe secod because of a lack of space lux buffes.

6 as ARMA64 ad B.49B B B B B B -6 ; B+.7579B B B B -4 ; g s the esdual sees whose mea s zeo ad vaace s 58; The QQ plot h shows that the esduals g s whte omal ose. I the ed the omal ug model of <ode App actvepg> ca be bult as B actvepg t Ba t : whee actvepg t s the sample of actvepg at tme t. The omal ug models of each <N AppM S> ae bult by the same meas. These models ae used fault detecto ad the esults of fault detecto ae show secto 4.. Fgue. Buldg the omal ug model of <ode App actvepg>.3 x 5 The Ogal sample data 5 The Sample data pocessed actvepg actvepg a The ogal sample data b The data afte pepocess Sample Autocoelato Fucto ACF 8 The Sample data oe degee dffeecg 6 actvepg.8 4 Sample Autocoelato ags b The ACF of b d Data afte -pace dffeecg of b Sample Autocoelato Fucto ACF Sample Patal Autocoelato Fucto PACF.8 Sample Autocoelato.5 Sample Patal Autocoelatos ags e The ACF k of d ags f The PACF k of d

7 The Resdual sees betwee Obseved data ad Pedcted value 8 Stadad Nomal Q-Q Plot of Sample Data 8 esduals Quatles of Resdual sees lags g The esduals Stadad Nomal Dstbuto h The QQ plot fo g 4. Results of fault detecto The Resdual sees betwee obseved value ad Pedcted value 8 Resduals 6 Oce the omal ug models ae bult the fault detecto pocess ca be mplemeted by compag the cuet ug state wth the omal ug models. Dug the expemet we wll ecod the tme whe the FDF s ad the tme whe the fault s jected to system. The dffeece betwee these two tmes s the fault detecto latecy. The latecy epesets the sestvty of fault detecto mechasm fo faults. The fault jecto techque s used to smulate the eal-wold faults cluste system. Fst u App ad ject memoy faults o ode. The jecto tme s the 3-th lag 5 secods afte App begs. Fgue 3 gves the fault detecto esults based o the bthe esdual sees Fault Alam Idex Fault Detecto Results.6 Fgue 3. Fault detecto esults by omal ug model of <ode App kbmemused> The kbmeused obseved data ad pedcted value x 6.3 Obseved value Pedcted value c The fault detecto fucto a kbmemused sees omal ug model of <ode App kbmemused>: a the blue s the actual obseved value ad gee s the pedcted value by the omal ug model; b s the esdual sees betwee obseved value ad pedcted value; c s the esults of fault detecto fucto ad the pesstece

8 checkg paamete w5 the sze of slde wdow. Fom c the fault detecto fucto s fom 43-th lag o. So the fault detecto latecy s 3 lags 565 secods. Secod u App ad ject I/O faults o ode. The faults ae jected whe App begs. Fgue 4 gves the fault detecto esults based o the omal ug model of <ode App Iteupt_ps>: a the blue s the actual obseved value ad gee s the pedcted value by the omal ug model; b s the esdual sees betwee obseved value ad pedcted value; c s the esults of fault detecto fucto ad the pesstece checkg paamete w5 the sze of slde wdow. Fom c the fault detecto fucto s fom 7-th lag o ad the fault detecto latecy s 7 lags 86 secods. The esults of fst ad secod expemets show the valdty of ou fault detecto mechasm. But the fault detecto latecy s hgh about mutes that's to say afte the fault exst about te mutes the Fgue 4. Fault detecto esults by omal ug model of <ode App Iteupt_ps> 5 The Iteupt-ps obseved data ad pedcted value Fault Detecto Results Fault Alam Idex c The fault detecto fucto fault alam just bee sed out. The ma cause fo the hgh latecy s the sze of slde wdow ad the ext expemet esults gve the elato betwee latecy ad the sze of slde wdow. I the last expemet u App3 ad ject I/O faults o ode3. The faults ae jected whe App3 begs. Nomal ug model of <ode3 App3 Cswch_ps> s B s t a t whee B.5439B - Fgue 5. The esdual sees by omal ug model of <ode3 App3 Cswch_ps> 5 The Cswch-ps Obseved data ad Pedcted value Obseved value Pedcted value Obseved value Pedcted value a Iteupt_ps sees Sample lags a Cswch_ps sees The Resdual sees betwee obseved value ad Pedcted value The Resdual sees betwee obseved value ad Pedcted Value 8 8 Resduals b The esdual sees b The esdual sees

9 ad s used to fault detecto. The fault detecto esults fo dffeet sze of wdows 46 ad w5 show the mpact of o fault detecto latecy. I fgue 5 a shows the obseved value blue ad the pedcted value gee; b s the esdual sees. Table 3 gves the fault detecto esults fo dffeet sze of wdows 46 ad w5. The esults show that the fault detecto latecy deceases whle the sze of wdow ceases. The ma easo s that the RT method s moe sestve wth a lage sample space. Table 3. The fault detecto latecy fo dffeet sze of wdows The sze of wdow Fault detecto latecy lags lags 5 secods Cocluso I lage-scale cluste system the fault detecto befoe ode falue s mpotat. Ths pape poposes a poactve ole fault detecto method usg the fomato collected fom cluste motog system. Fst buldg the cluste omal ug models of pefomace paametes usg the actvty data whe cluste us wthout faults; Secod the dffeece betwee actual obseved value ad pedcted value by omal ug model s esdual ad use the esdual tme sees as the symptom of fault; Whe the cluste system s behavg omally all the esdual sees ae whte ose sees; I the ed usg the RT method to test whethe the esdual sees ae whte ose sees ad the fault detecto fucto f s costucted based o the esults of RT ad the pesstece checkg paamete w. If f oe fault alam decso s made. The expemet esults show that the method ths pape ca detect the fault system good tme ad povdes the ablty to poactve fault maagemet. But the method ths pape s just a fault detecto ad ot fault dagoss. If abomal behavo s detected the ext step s locatg the accuate cause of fault. I the futue wok we wll pay moe atteto o two questos: Extact the moe sestve symptom of fault cluste system. I ths pape oly usg the esdual sees of pefomace paametes as the symptom of faults. I the futue we wll select the moe sestve symptom fo faults ad educe the fault detecto latecy educe the computato complexty at same tme. Ole fault dagoss method. Fdg the oot cause of fault ad gve the coectly healg advce to system admstato. 6. Refeeces []. Gao We. The desg ad aalyss method of hgh avalablty the seve cosoldato system: [Ph.D. dssetato]. Bejg: Isttute of Computg Techology Chese Academy of Sceces.. []. Vttoo Castell Rchad E. Hape Phlp Hedelbege Steve W. Hute Ksho S. Tved Kalyaaama Vadyaatha Wllam P. Zegget: Poactve maagemet of softwae agg. IBM Joual of Reseach ad Developmet 45: 3-33 [3]. R. Sahoo A. Ole I. Rsh M. Gupta J. Moea S. Ma R. Vlalta A. Svasubamaam. Ctcal Evet Pedcto fo Poactve Maagemet age-scale Compute Clustes. I Poceedgs of the ACM SIGKDD Iteatoal Cofeece o Kowledge Dscovey ad Data Mg pages August 3. [4]. D. Dasgupta K. KshaKuma D. Wog M. Bey: Negatve Selecto Algothm fo Acaft Fault Detecto. 3d Iteatoal Cofeece o Atfcal Immue Systems Cataa Scly.Italy Septembe [5]. J. Wesley Hes D. W. Mlle ad B. K. Hajek. Megg Pocess Models Wth Neual Netwoks fo Nuclea Powe Plat Fault Detecto ad Isolato The poceedgs of the 9th Powe Plat Dyamcs Cotol ad Testg Symposum Koxvlle TN 995. [6]. Y. Huag C. Ktala N. Koletts N.D. Fulto. Softwae ejuveato: aalyss module ad applcatos.

10 Poc. 5th Iteatoal Symposum o Fault-Toleat Computg Jue 995 pp [7]. S. Gag A. va Moosel K. Vadyaatha ad K. Tved A Methodology fo Detecto ad Estmato of Softwae Agg Poceedgs of the 9th Iteatoal Symposum o Softwae Relablty Egeeg Padebo Gemay Novembe 998 pp [8]. e Kalyaaama Vadyaatha Ksho S. Tved: A Appoach fo Estmato of Softwae Agg a Web Seve. Poc. It'l Symp. o Empcal Softwae Eg. pp.9 IEEE CS Pess. [9]. Hughes G.F. Muay J.F. Keutz-Delgado K. ad Elka C. Impoved Dsk Dve Falue Wags. IEEE Tasactos o Relablty Septembe. [].J. F. Muay G F Hughes Had Dve Falue Pedcto Usg No-paametc Statstcal Methods coautho Iteatoal Cofeece o Atfcal Neual Netwoks Istabul Jue [].Geg Hamely ad Chales Elka. Bayesa appoaches to falue pedcto fo dsk dves. I Poceedgs of the eghteeth teatoal cofeece o mache leag pages -9. Moga Kaufma Sa Facsco CA. [].J. Moea "System Maagemet The BlueGee/ Supecompute" Poc. It'l. Paallel ad Dstbuted Pocessg Symposum. os Alamtos CA IEEE Compute Socety. 3 p. 8. Apl 3. [3].Top 5 Supecopute Stes. System Ifo: Dawg 4A Opteo. GHz Myet [4].Che Y Meg Da Zha Ja-Feg Ze Ng Desg ad mplemet of fedeato based tadg sevce Compute Egeeg ad Applcatos 4. [5].Meg Da Zha Jafeg Wag e Tu Bbo Zhag Zhhog. Fully tegated cluste opeatg system: Phoex. Joual of Compute Reseach ad Developmet 4.6.

Registration of Multiple Laser Scans Based on 3D Contour Features

Registration of Multiple Laser Scans Based on 3D Contour Features Regstato of Multple Lase Scas Based o 3D Cotou Featues st Shaoxg HU, d Hogb ZHA, 3 d Awu ZHANG st School of Mechacal Egeeg & Automato, Beg Uvesty of Aeoautcs ad Astoautcs, Beg 83, d Natoal Laboatoy o Mache

More information

Fuzzy Probability Approximation Space and Its Information Measures

Fuzzy Probability Approximation Space and Its Information Measures Fuzzy Pobablty Appomato Space ad Its Ifomato Measues Qghua Hu, Dae Yu Hab Isttute of Techology, Cha Abstact ough set theoy has attacted much atteto modelg wth mpecse ad complete fomato A geealzed appomato

More information

Pairwise comparisons in the analysis of carcinogenicity data *

Pairwise comparisons in the analysis of carcinogenicity data * Vol4, No1, 91-918 (1) http://xoog/1436/health141139 ealth Pawse compasos the aalyss of cacogecty ata Mohamma A Rahma 1#, Ram C Twa 1 Dvso of Bometcs-6, Offce of Bostatstcs, Cete fo Dug Evaluato a Reseach,

More information

SOLVING METHOD FOR FUZZY MULTIPLE OBJECTIVE INTEGER OPTIMIZATION

SOLVING METHOD FOR FUZZY MULTIPLE OBJECTIVE INTEGER OPTIMIZATION SOLVING METHOD FOR FUZZY MULTIPLE OBJECTIVE INTEGER OPTIMIZATION BOGDANA POP Taslvaa Uvesty of Basov Romaa Abstact Statg fom the dea of Wag ad Lao (00 fo solvg fuzzy o-lea tege ogammg oblem ad tag to accout

More information

Computer Technology MSIS 22:198:605 Homework 1

Computer Technology MSIS 22:198:605 Homework 1 Compute Techology MSIS 22:198:605 Homewok 1 Istucto: Faid Alizadeh Due Date: Moday Septembe 30, 2002 by midight Submissio: by e-mail See below fo detailed istuctios) last updated o Septembe 27, 2002 Rules:

More information

Mixture Model of the Exponential, Gamma and Weibull Distributions to Analyse Heterogeneous Survival Data

Mixture Model of the Exponential, Gamma and Weibull Distributions to Analyse Heterogeneous Survival Data Joual of Scetfc Reseach & Repots 5: -9, 05; Atcle o.jsrr.05.080 ISSN: 0-07 SCIENCEDOMAIN teatoal www.scecedoma.og Mxtue Model of the Expoetal, Gamma ad Webull Dstbutos to Aalyse Heteogeeous Suvval Data

More information

Risk Evaluation in Auto Spare Parts Transport Based on the

Risk Evaluation in Auto Spare Parts Transport Based on the MAEC Web of Cofeeces 100, 05060 Rsk Evaluato Auto Spae Pats aspot Based o the AHP Method Rog Zeg 1 ad Chag Xu 2 1 Wuha Huaxa Uvesty of echology, Wuha, Cha 2 Wuha Uvesty of echology, Wuha, Cha Abstact By

More information

The Search for Coalition Formation in Costly Environments 1

The Search for Coalition Formation in Costly Environments 1 The Seach fo Coalto Fomato Costly Evomets 1 Davd Sae 1 ad Sat Kaus 1,2 1 Depatmet of Compute Scece, Ba-Ila Uvesty, Ramat-Ga, 52900 Isael {saed, sat} @ macs.bu.ac.l 2 Isttute fo Advaced Compute Studes Uvesty

More information

RECOGNITION OF COMMON BUILDINGS IN CARTOGRAPHIC FILES

RECOGNITION OF COMMON BUILDINGS IN CARTOGRAPHIC FILES RECOGITIO OF COMMO BUIIGS I CRTOGRPHIC FIES Ha-We Hsao, Kam W. Wog epatmet of Cvl Egeeg Uvesty of Illos at Ubaa-Champag 5. Mathews ve. Ubaa, Illos 68, US Emal: h-hsao@studets.uuc.edu, -wog@staff.uuc.edu

More information

Adaptive Naïve Bayesian Anti-Spam Engine

Adaptive Naïve Bayesian Anti-Spam Engine Wold Academy of Scece, Egeeg ad Techology 7 2005 Adaptve Naïve Bayesa At-Spam Ege Wojcech P. Gajewsk Abstact The poblem of spam has bee seously toublg the Iteet commuty dug the last few yeas ad cuetly

More information

A Comparison of the Optimal Classification Rule and Maximum Likelihood Rule for Binary Variables

A Comparison of the Optimal Classification Rule and Maximum Likelihood Rule for Binary Variables Joual of Mathematcs Reseach; Vol. 6 No. 4; 04 ISSN 96-9795 E-ISSN 96-9809 Publshed by Caada Cete of Scece ad Educato A Comaso of the Otmal Classfcato Rule ad Mamum Lelhood Rule fo ay Vaables I. Egbo S.

More information

arxiv: v1 [math.co] 21 Jan 2016

arxiv: v1 [math.co] 21 Jan 2016 PROOF OF BIJECTION FOR COMBINATORIAL NUMBER SYSTEM axv:60.05794v [math.co] Jan 06 ABU BAKAR SIDDIQUE, SAADIA FARID, AND MUHAMMAD TAHIR Abstact. Combnatoal numbe system epesents a non-negatve natual numbes

More information

Visually Built Task Models for Robot Teams in Unstructured Environments Abstract 1. Introduction

Visually Built Task Models for Robot Teams in Unstructured Environments Abstract 1. Introduction Vsuall Bult as Models fo Robot eams Ustuctued Evomets Vve A. Suja ad Steve Dubows (vasuja dubows@mt.edu} Depatmet of Mechacal Egeeg Massachusetts Isttute of echolog Cambdge, MA 039 Abstact I feld evomets

More information

A Multi-Categorization Method of Text Documents using Fuzzy Correlation Analysis

A Multi-Categorization Method of Text Documents using Fuzzy Correlation Analysis Poceedg of the 0th WE Iteatoal Cofeece o PPLIED MTHEMTIC, Dalla, Texa, U, Novembe -3, 006 9 Mult-Categozato Method of Text Documet ug Fuzzy Coelato aly NNCY P. LIN, HO-EN CHUEH Deatmet of Comute cece ad

More information

Bidding Strategies for Generation Companies in a Day-ahead Market using Fuzzy Adaptive Particle Swarm Optimization

Bidding Strategies for Generation Companies in a Day-ahead Market using Fuzzy Adaptive Particle Swarm Optimization J. Vaya Kuma, D. M. Vod Kuma, K. Eduodalu Bddg Stateges fo Geeato Compaes a Day-ahead Maet usg Fuzzy Adaptve Patcle Swam Optmzato J. VIJAYA KUMAR *, D. M. VINOD KUMAR ad K EDUKONDALU Depatmet of Electcal

More information

Optically adjustable display color gamut in time-sequential displays using LED/Laser light sources

Optically adjustable display color gamut in time-sequential displays using LED/Laser light sources Optcall adjustale dspla colo amut tme-sequetal dsplas us LE/Lase lht souces splas vol. 7 006 Moo-Cheol Km School of Electcal Eee ad Compute Scece Kupook Natoal Uv. Astact evelopmet of vaous wde colo amut

More information

CS 2710 Foundations of AI Lecture 22. Machine learning. Machine Learning

CS 2710 Foundations of AI Lecture 22. Machine learning. Machine Learning CS 7 Foudatos of AI Lecture Mache learg Mlos Hauskrecht mlos@cs.ptt.edu 539 Seott Square Mache Learg The feld of mache learg studes the desg of computer programs (agets) capable of learg from past eperece

More information

A COMPARISON OF MULTIVARIATE DISCRIMINATION OF BINARY DATA

A COMPARISON OF MULTIVARIATE DISCRIMINATION OF BINARY DATA Iteatoal Joual of Mathematcs ad Statstcs Studes Vol.No.4.40-6 Setembe 04 Publshed by Euoea Cete fo eseach Tag ad Develomet UK www.eaouals.og A COMPAISON OF MULTIVAIATE DISCIMINATION OF INAY DATA. I. Egbo;.

More information

Segmentation of Casting Defects in X-Ray Images Based on Fractal Dimension

Segmentation of Casting Defects in X-Ray Images Based on Fractal Dimension 17th Wold Confeence on Nondestuctive Testing, 25-28 Oct 2008, Shanghai, China Segmentation of Casting Defects in X-Ray Images Based on Factal Dimension Jue WANG 1, Xiaoqin HOU 2, Yufang CAI 3 ICT Reseach

More information

Chap 5: Product-Form Queuing Networks (QN)

Chap 5: Product-Form Queuing Networks (QN) CPU disk Chap 5: Poduct-Fom Queuig Netwoks (QN) Etities: ) sevice cetes with diffeet sevice disciplies ) customes (obs) sigle class multiple classes (each/w a diffeet wokload) ) liks coectig sevice cetes

More information

A Neural Network Model for Storing and Retrieving 2D Images of Rotated 3D Object Using Principal Components

A Neural Network Model for Storing and Retrieving 2D Images of Rotated 3D Object Using Principal Components A Neual Netwok Model fo Stong and Reteving 2D Images of Rotated 3D Object Using Pncipal Components Tsukasa AMANO, Shuichi KUROGI, Ayako EGUCHI, Takeshi NISHIDA, Yasuhio FUCHIKAWA Depatment of Contol Engineeng,

More information

ADDING REALISM TO SOURCE CHARACTERIZATION USING A GENETIC ALGORITHM

ADDING REALISM TO SOURCE CHARACTERIZATION USING A GENETIC ALGORITHM ADDING REALISM TO SOURCE CHARACTERIZATION USING A GENETIC ALGORITHM Luna M. Rodiguez*, Sue Ellen Haupt, and Geoge S. Young Depatment of Meteoology and Applied Reseach Laboatoy The Pennsylvania State Univesity,

More information

Research of Education Evaluation Information Mining Technology Based on Gray Clustering Analysis and Fuzzy Evaluation Method

Research of Education Evaluation Information Mining Technology Based on Gray Clustering Analysis and Fuzzy Evaluation Method Compute a Ifomato Scece Reseach of Eucato Evaluato Ifomato Mg Techology Base o Gay Clusteg Aalyss a Fuzzy Evaluato Metho Yag Lu College of Compute a Automatzato, Taj Polytechc Uvesty Taj, 30060, Cha E-mal:

More information

Controlled Information Maximization for SOM Knowledge Induced Learning

Controlled Information Maximization for SOM Knowledge Induced Learning 3 Int'l Conf. Atificial Intelligence ICAI'5 Contolled Infomation Maximization fo SOM Knowledge Induced Leaning Ryotao Kamimua IT Education Cente and Gaduate School of Science and Technology, Tokai Univeisity

More information

Adaptive Clustering Algorithm for Mining Subspace Clusters in High-Dimensional Data Stream *

Adaptive Clustering Algorithm for Mining Subspace Clusters in High-Dimensional Data Stream * ISSN 673-948 CODEN JKYTA8 E-mal: fcst@vp.63.com Joural of Froters of Computer Scece ad Techology http://www.ceaj.org 673-948/200/04(09)-0859-06 Tel: +86-0-566056 DOI: 0.3778/j.ss.673-948.200.09.009 *,2,

More information

Positioning of a robot based on binocular vision for hand / foot fusion Long Han

Positioning of a robot based on binocular vision for hand / foot fusion Long Han 2nd Intenational Confeence on Advances in Mechanical Engineeing and Industial Infomatics (AMEII 26) Positioning of a obot based on binocula vision fo hand / foot fusion Long Han Compute Science and Technology,

More information

A SAS Macro for Finding Optimal k-means Clustering in One Dimension with Size Constraints

A SAS Macro for Finding Optimal k-means Clustering in One Dimension with Size Constraints Pape SD-02 A SAS Maco fo Fndng Optmal k-means Clusteng n One Dmenson wth Sze Constants Fengao Hu, Geoga Regents Unvesty; Robet E. Johnson, Vandeblt Unvesty ABSTRACT Wang and Song (2011) poposed a k-means

More information

Journal of World s Electrical Engineering and Technology J. World. Elect. Eng. Tech. 1(1): 12-16, 2012

Journal of World s Electrical Engineering and Technology J. World. Elect. Eng. Tech. 1(1): 12-16, 2012 2011, Scienceline Publication www.science-line.com Jounal of Wold s Electical Engineeing and Technology J. Wold. Elect. Eng. Tech. 1(1): 12-16, 2012 JWEET An Efficient Algoithm fo Lip Segmentation in Colo

More information

Any modern computer system will incorporate (at least) two levels of storage:

Any modern computer system will incorporate (at least) two levels of storage: 1 Any moden compute system will incopoate (at least) two levels of stoage: pimay stoage: andom access memoy (RAM) typical capacity 32MB to 1GB cost pe MB $3. typical access time 5ns to 6ns bust tansfe

More information

Slotted Random Access Protocol with Dynamic Transmission Probability Control in CDMA System

Slotted Random Access Protocol with Dynamic Transmission Probability Control in CDMA System Slotted Random Access Potocol with Dynamic Tansmission Pobability Contol in CDMA System Intaek Lim 1 1 Depatment of Embedded Softwae, Busan Univesity of Foeign Studies, itlim@bufs.ac.k Abstact In packet

More information

Performance Evaluation of Neighborhood Signature Techniques for Peer-to-Peer Search

Performance Evaluation of Neighborhood Signature Techniques for Peer-to-Peer Search Pefomance Evaluaton of Neghbohood gnatue Technques fo Pee-to-Pee each e L Wang-Chen Lee * Anand vasubamanam Depatment of Compute cence and Engneeng Pennsylvana tate Unvesty Unvesty Pak, PA 16802, UA E-al:

More information

ANALYTIC PERFORMANCE MODELS FOR SINGLE CLASS AND MULTIPLE CLASS MULTITHREADED SOFTWARE SERVERS

ANALYTIC PERFORMANCE MODELS FOR SINGLE CLASS AND MULTIPLE CLASS MULTITHREADED SOFTWARE SERVERS ANALYTIC PERFORMANCE MODELS FOR SINGLE CLASS AND MULTIPLE CLASS MULTITHREADED SOFTWARE SERVERS Daniel A Menascé Mohamed N Bennani Dept of Compute Science Oacle, Inc Geoge Mason Univesity 1211 SW Fifth

More information

SYSTEM LEVEL REUSE METRICS FOR OBJECT ORIENTED SOFTWARE : AN ALTERNATIVE APPROACH

SYSTEM LEVEL REUSE METRICS FOR OBJECT ORIENTED SOFTWARE : AN ALTERNATIVE APPROACH I J C A 7(), 202 pp. 49-53 SYSTEM LEVEL REUSE METRICS FOR OBJECT ORIENTED SOFTWARE : AN ALTERNATIVE APPROACH Sushil Goel and 2 Rajesh Vema Associate Pofesso, Depatment of Compute Science, Dyal Singh College,

More information

Frequency Domain Approach for Face Recognition Using Optical Vanderlugt Filters

Frequency Domain Approach for Face Recognition Using Optical Vanderlugt Filters Optics and Photonics Jounal, 016, 6, 94-100 Published Online August 016 in SciRes. http://www.scip.og/jounal/opj http://dx.doi.og/10.436/opj.016.68b016 Fequency Domain Appoach fo Face Recognition Using

More information

Communication vs Distributed Computation: an alternative trade-off curve

Communication vs Distributed Computation: an alternative trade-off curve Communication vs Distibuted Computation: an altenative tade-off cuve Yahya H. Ezzeldin, Mohammed amoose, Chistina Fagouli Univesity of Califonia, Los Angeles, CA 90095, USA, Email: {yahya.ezzeldin, mkamoose,

More information

Robust Proper Clustering Structure Fuzzy Modeling for Function Approximation

Robust Proper Clustering Structure Fuzzy Modeling for Function Approximation Robust Pope lusteng Stuctue Fuzzy Modelng fo Functon Appoxmaton hh-hng Hsao Depatment of Electcal Engneeng Kao Yuan Unvesty Kaohsung ounty,tawan RO hsao@ooneentustedutw Shun-Feng Su Depatment of Electcal

More information

A Novel Estimation-Based Backoff Algorithm in the IEEE Based Wireless Network

A Novel Estimation-Based Backoff Algorithm in the IEEE Based Wireless Network A ovel Estimatio-Based Algoithm i the EEE 80. Based Wieless etwok Seok-Wo Kag, Jae-Ryog Cha ad Jae-Hyu Kim School of Electical Egieeig, Ajou Uivesity Sa 5 Wocheo-og, Yougtog-Gu, Suwo 443-749, Koea Abstact

More information

4.2. Co-terminal and Related Angles. Investigate

4.2. Co-terminal and Related Angles. Investigate .2 Co-teminal and Related Angles Tigonometic atios can be used to model quantities such as

More information

COMSC 2613 Summer 2000

COMSC 2613 Summer 2000 Programmg II Fal Exam COMSC 63 Summer Istructos: Name:. Prt your ame the space provded Studet Id:. Prt your studet detfer the space Secto: provded. Date: 3. Prt the secto umber of the secto whch you are

More information

Differentiated Service of Streaming Media Playback Technology

Differentiated Service of Streaming Media Playback Technology Iteratoal Coferece o Advaced Iformato ad Commucato Techology for Educato (ICAICTE 2013) Dfferetated Servce of Streamg Meda Playback Techology CHENG Z-ao 1 MENG Bo 1 WANG Da-hua 1 ZHAO Yue 1 1 Iformatzato

More information

A Two-stage and Parameter-free Binarization Method for Degraded Document Images

A Two-stage and Parameter-free Binarization Method for Degraded Document Images A Two-stage and Paamete-fee Binaization Method fo Degaded Document Images Yung-Hsiang Chiu 1, Kuo-Liang Chung 1, Yong-Huai Huang 2, Wei-Ning Yang 3, Chi-Huang Liao 4 1 Depatment of Compute Science and

More information

Detection and Recognition of Alert Traffic Signs

Detection and Recognition of Alert Traffic Signs Detection and Recognition of Alet Taffic Signs Chia-Hsiung Chen, Macus Chen, and Tianshi Gao 1 Stanfod Univesity Stanfod, CA 9305 {echchen, macuscc, tianshig}@stanfod.edu Abstact Taffic signs povide dives

More information

Local vs. Global Illumination & Radiosity

Local vs. Global Illumination & Radiosity Local vs. Global Illumato & Radosty Last Tme? Ray Castg & Ray-Object Itesecto Recusve Ray Tacg Dstbuted Ray Tacg A ealy applcato of adatve heat tasfe stables. Local Illumato BRDF Ideal Dffuse Reflectace

More information

Optimal Allocation of Complex Equipment System Maintainability

Optimal Allocation of Complex Equipment System Maintainability Optmal Allocato of Complex Equpmet System ataablty X Re Graduate School, Natoal Defese Uversty, Bejg, 100091, Cha edcal Protecto Laboratory, Naval edcal Research Isttute, Shagha, 200433, Cha Emal:rexs841013@163.com

More information

Keith Dalbey, PhD. Sandia National Labs, Dept 1441 Optimization & Uncertainty Quantification

Keith Dalbey, PhD. Sandia National Labs, Dept 1441 Optimization & Uncertainty Quantification SAND 0-50 C Effective & Efficient Handling of Ill - Conditioned Coelation atices in Kiging & adient Enhanced Kiging Emulatos hough Pivoted Cholesky Factoization Keith Dalbey, PhD Sandia National Labs,

More information

LP: example of formulations

LP: example of formulations LP: eample of formulatos Three classcal decso problems OR: Trasportato problem Product-m problem Producto plag problem Operatos Research Massmo Paolucc DIBRIS Uversty of Geova Trasportato problem The decso

More information

For all questions, answer choice E) NOTA" means none of the above answers is correct. A) 50,500 B) 500,000 C) 500,500 D) 1,001,000 E) NOTA

For all questions, answer choice E) NOTA means none of the above answers is correct. A) 50,500 B) 500,000 C) 500,500 D) 1,001,000 E) NOTA For all questos, aswer choce " meas oe of the above aswers s correct.. What s the sum of the frst 000 postve tegers? A) 50,500 B) 500,000 C) 500,500 D),00,000. What s the sum of the tegers betwee 00 ad

More information

Fitting. We ve learned how to detect edges, corners, blobs. Now what? We would like to form a. compact representation of

Fitting. We ve learned how to detect edges, corners, blobs. Now what? We would like to form a. compact representation of Fttg Fttg We ve leared how to detect edges, corers, blobs. Now what? We would lke to form a hgher-level, h l more compact represetato of the features the mage b groupg multple features accordg to a smple

More information

2. PROPELLER GEOMETRY

2. PROPELLER GEOMETRY a) Fames of Refeence 2. PROPELLER GEOMETRY 10 th Intenational Towing Tank Committee (ITTC) initiated the pepaation of a dictionay and nomenclatue of ship hydodynamic tems and this wok was completed in

More information

1-D matrix method. U 4 transmitted. incident U 2. reflected U 1 U 5 U 3 L 2 L 3 L 4. EE 439 matrix method 1

1-D matrix method. U 4 transmitted. incident U 2. reflected U 1 U 5 U 3 L 2 L 3 L 4. EE 439 matrix method 1 -D matrx method We ca expad the smple plae-wave scatterg for -D examples that we ve see to a more versatle matrx approach that ca be used to hadle may terestg -D problems. The basc dea s that we ca break

More information

A Bayesian Approach toward Active Learning for Collaborative Filtering

A Bayesian Approach toward Active Learning for Collaborative Filtering A Bayesan Appoach towad Actve Leanng fo Collaboatve Flteng Rong Jn Depatment of Compute Scence and Engneeng Mchgan State Unvesty ong@cse.cmu.edu Abstact Collaboatve flteng s a useful technque fo eplotng

More information

Bilateral Filter Based Selective Unsharp Masking Using Intensity and/or Saturation Components

Bilateral Filter Based Selective Unsharp Masking Using Intensity and/or Saturation Components Intenational Jounal of Electonics and Electical Engeeg ol. 1, No. 1, Mach 013 Bilateal Filte Based Selective Unshap Maskg Usg Intensity and/o Satuation Components Noiaki Suetake, Yohei Kuamoto, and Eii

More information

Recursion. Announcements. Follow-up from yesterday. Recursion Overview. A Recursive Program. The Factorial Function (n!)

Recursion. Announcements. Follow-up from yesterday. Recursion Overview. A Recursive Program. The Factorial Function (n!) Aoucemets Recusio Lectue CS Summe 7 Fo exta Java help Lots of cosultig/office-hous available Ca set up idividual meetigs with TAs via email Check soo that you ae i CMS Academic Itegity Note Assigmet #

More information

Machine Learning. CS 2750 Machine Learning. Administration. Lecture 1. Milos Hauskrecht 5329 Sennott Square, x4-8845

Machine Learning. CS 2750 Machine Learning. Administration. Lecture 1. Milos Hauskrecht 5329 Sennott Square, x4-8845 CS 75 Mache Learg Lecture Mache Learg Mlos Hauskrecht mlos@cs.ptt.edu 539 Seott Square, 5 people.cs.ptt.edu/~mlos/courses/cs75/ Admstrato Istructor: Prof. Mlos Hauskrecht mlos@cs.ptt.edu 539 Seott Square,

More information

An Object Based Auto Annotation Image Retrieval System

An Object Based Auto Annotation Image Retrieval System Poceedngs of the 5th WSEAS Intenatonal Confeence on Telecommuncatons and Infomatcs, Istanbul, Tukey, May 27-29, 2006 (pp509-54) An Obect Based Auto Annotaton Image Reteval System Pe-Cheng Cheng, Been-Chan

More information

Machine Learning: Algorithms and Applications

Machine Learning: Algorithms and Applications /03/ Mache Learg: Algorthms ad Applcatos Florao Z Free Uversty of Boze-Bolzao Faculty of Computer Scece Academc Year 0-0 Lecture 3: th March 0 Naïve Bayes classfer ( Problem defto A trag set X, where each

More information

MapReduce Optimizations and Algorithms 2015 Professor Sasu Tarkoma

MapReduce Optimizations and Algorithms 2015 Professor Sasu Tarkoma apreduce Optimizations and Algoithms 2015 Pofesso Sasu Takoma www.cs.helsinki.fi Optimizations Reduce tasks cannot stat befoe the whole map phase is complete Thus single slow machine can slow down the

More information

A comparison of variance estimators with known and unknown population means

A comparison of variance estimators with known and unknown population means Ng ad Hoocks Studies by Udegaduate Reseaches at Guelph Vol., No., Wite 008, 4-48 A compaiso of vaiace estimatos with kow ad ukow populatio meas Jeife Ng ad Julie Hoocks This mauscipt was pepaed while Jeife

More information

ITEM ToolKit Technical Support Notes

ITEM ToolKit Technical Support Notes ITEM ToolKt Notes Fault Tree Mathematcs Revew, Ic. 2875 Mchelle Drve Sute 300 Irve, CA 92606 Phoe: +1.240.297.4442 Fax: +1.240.297.4429 http://www.itemsoft.com Page 1 of 15 6/1/2016 Copyrght, Ic., All

More information

Coordinates Transformation by Zoning Method for Parameters Computing between WGS84 and North Sahara

Coordinates Transformation by Zoning Method for Parameters Computing between WGS84 and North Sahara Coodates Tasfomato b og Method fo Paametes Computg betwee WGS84 ad Noth Sahaa Nouedde KHELOUFI, Algea Kewods: Catogaph GPS, Lad eadjustmet, postog SUMMAR The global tasfoms model (busa wolf, molodeskbadekas)

More information

Optical Flow for Large Motion Using Gradient Technique

Optical Flow for Large Motion Using Gradient Technique SERBIAN JOURNAL OF ELECTRICAL ENGINEERING Vol. 3, No. 1, June 2006, 103-113 Optical Flow fo Lage Motion Using Gadient Technique Md. Moshaof Hossain Sake 1, Kamal Bechkoum 2, K.K. Islam 1 Abstact: In this

More information

Hierarchically Clustered P2P Streaming System

Hierarchically Clustered P2P Streaming System Hieachically Clusteed P2P Steaming System Chao Liang, Yang Guo, and Yong Liu Polytechnic Univesity Thomson Lab Booklyn, NY 11201 Pinceton, NJ 08540 Abstact Pee-to-pee video steaming has been gaining populaity.

More information

Efficient End-to-End Communication Services for Mixed Criticality Avionics Systems

Efficient End-to-End Communication Services for Mixed Criticality Avionics Systems 214 IEEE 22nd Intenatonal Symposum of Qualty of Sevce (IWQoS) Effcent End-to-End Communcaton Sevces fo Mxed Ctcalty Avoncs Systems Yng Fang Yu Hua Xue Lu McGll Unvesty Huazhong Unv. of Sc. and Tech. McGll

More information

Image Enhancement in the Spatial Domain. Spatial Domain

Image Enhancement in the Spatial Domain. Spatial Domain 8-- Spatial Domain Image Enhancement in the Spatial Domain What is spatial domain The space whee all pixels fom an image In spatial domain we can epesent an image by f( whee x and y ae coodinates along

More information

Weighting Cache Replace Algorithm for Storage System

Weighting Cache Replace Algorithm for Storage System Weghtg Cache Replace Algorthm for Storage System Yhu Luo 2 Chagsheg Xe 2 Chegfeg Zhag 2 School of mathematcs ad Computer Scece, Hube Uversty, Wuha 430062, P.R. Cha 2 Natoal Storage System Laboratory, School

More information

Query Language #1/3: Relational Algebra Pure, Procedural, and Set-oriented

Query Language #1/3: Relational Algebra Pure, Procedural, and Set-oriented Quey Language #1/3: Relational Algeba Pue, Pocedual, and Set-oiented To expess a quey, we use a set of opeations. Each opeation takes one o moe elations as input paamete (set-oiented). Since each opeation

More information

Effective Missing Data Prediction for Collaborative Filtering

Effective Missing Data Prediction for Collaborative Filtering Effective Missing Data Pediction fo Collaboative Filteing Hao Ma, Iwin King and Michael R. Lyu Dept. of Compute Science and Engineeing The Chinese Univesity of Hong Kong Shatin, N.T., Hong Kong { hma,

More information

Color Correction Using 3D Multiview Geometry

Color Correction Using 3D Multiview Geometry Colo Coection Using 3D Multiview Geomety Dong-Won Shin and Yo-Sung Ho Gwangju Institute of Science and Technology (GIST) 13 Cheomdan-gwagio, Buk-ku, Gwangju 500-71, Republic of Koea ABSTRACT Recently,

More information

TDT-2004: ADAPTIVE TOPIC TRACKING AT MARYLAND

TDT-2004: ADAPTIVE TOPIC TRACKING AT MARYLAND TDT-2004: ADAPTIVE TOPIC TRACKING AT MARYLAND Tamer Elsayed, Douglas W. Oard, Davd Doerma Isttute for Advaced r Studes Uversty of Marylad, College Park, MD 20742 Cotact author: telsayed@cs.umd.edu Gary

More information

Software reliability is defined as the probability of failure

Software reliability is defined as the probability of failure Evolutoary Regresso Predcto for Software Cumulatve Falure Modelg: a comparatve study M. Beaddy, M. Wakrm & S. Aljahdal 2 : Dept. of Math. & Ifo. Equpe MMS, Ib Zohr Uversty Morocco. beaddym@yahoo.fr 2:

More information

A Mathematical Implementation of a Global Human Walking Model with Real-Time Kinematic Personification by Boulic, Thalmann and Thalmann.

A Mathematical Implementation of a Global Human Walking Model with Real-Time Kinematic Personification by Boulic, Thalmann and Thalmann. A Mathematical Implementation of a Global Human Walking Model with Real-Time Kinematic Pesonification by Boulic, Thalmann and Thalmann. Mashall Badley National Cente fo Physical Acoustics Univesity of

More information

SVM Classification Method Based Marginal Points of Representative Sample Sets

SVM Classification Method Based Marginal Points of Representative Sample Sets P P College P P College P Iteratoal Joural of Iformato Techology Vol. No. 9 005 SVM Classfcato Method Based Margal Pots of Represetatve Sample Sets Wecag ZhaoP P, Guagrog JP P, Ru NaP P, ad Che FegP of

More information

Coded Distributed Computing

Coded Distributed Computing Coded Distibuted Computing Salman Avestimeh USC joint wok with Songze Li (USC), Qian Yu (USC), and Mohammad Maddah-Ali (Bell-Labs) Asiloma Confeence Nov. 2016 Infastuctues fo (Big) Data Analytics How to

More information

Bezier curves. 1. Defining a Bezier curve. A closed Bezier curve can simply be generated by closing its characteristic polygon

Bezier curves. 1. Defining a Bezier curve. A closed Bezier curve can simply be generated by closing its characteristic polygon Curve represetato Copyrght@, YZU Optmal Desg Laboratory. All rghts reserved. Last updated: Yeh-Lag Hsu (--). Note: Ths s the course materal for ME55 Geometrc modelg ad computer graphcs, Yua Ze Uversty.

More information

Spiral Recognition Methodology and Its Application for Recognition of Chinese Bank Checks

Spiral Recognition Methodology and Its Application for Recognition of Chinese Bank Checks Spial Recognition Methodology and Its Application fo Recognition of Chinese Bank Checks Hanshen Tang 1, Emmanuel Augustin 2, Ching Y. Suen 1, Olivie Baet 2, Mohamed Cheiet 3 1 Cente fo Patten Recognition

More information

Guaranteeing Quality of Service to Peering Traffic

Guaranteeing Quality of Service to Peering Traffic Guaanteeng Qualty of Sevce to Peeng Taffc Ru Zhang-Shen Depatment of Electcal Engneeng Pnceton Unvesty Emal: z@pnceton.edu ck McKeown Compute Systems Laboatoy Stanfod Unvesty Emal: nckm@stanfod.edu Abstact

More information

Fault Tolerant Routing For Wireless Sensor Grid Networks

Fault Tolerant Routing For Wireless Sensor Grid Networks Fault oleant Routng Fo Weless Senso Gd etwoks Xn-ng Huang, Jng Deng 2, Jng a and Zeyu Wu Depatment of Electcal Engneeng 2 Depatment of Compute Scence Unvesty of ew Oleans, ew Oleans, A 7048, USA Abstact

More information

Point-Biserial Correlation Analysis of Fuzzy Attributes

Point-Biserial Correlation Analysis of Fuzzy Attributes Appl Math Inf Sci 6 No S pp 439S-444S (0 Applied Mathematics & Infomation Sciences An Intenational Jounal @ 0 NSP Natual Sciences Publishing o Point-iseial oelation Analysis of Fuzzy Attibutes Hao-En hueh

More information

COEN-4730 Computer Architecture Lecture 2 Review of Instruction Sets and Pipelines

COEN-4730 Computer Architecture Lecture 2 Review of Instruction Sets and Pipelines 1 COEN-4730 Compute Achitectue Lectue 2 Review of nstuction Sets and Pipelines Cistinel Ababei Dept. of Electical and Compute Engineeing Maquette Univesity Cedits: Slides adapted fom pesentations of Sudeep

More information

And Ph.D. Candidate of Computer Science, University of Putra Malaysia 2 Faculty of Computer Science and Information Technology,

And Ph.D. Candidate of Computer Science, University of Putra Malaysia 2 Faculty of Computer Science and Information Technology, (IJCSIS) Intenational Jounal of Compute Science and Infomation Secuity, Efficient Candidacy Reduction Fo Fequent Patten Mining M.H Nadimi-Shahaki 1, Nowati Mustapha 2, Md Nasi B Sulaiman 2, Ali B Mamat

More information

IP Network Design by Modified Branch Exchange Method

IP Network Design by Modified Branch Exchange Method Received: June 7, 207 98 IP Netwok Design by Modified Banch Method Kaiat Jaoenat Natchamol Sichumoenattana 2* Faculty of Engineeing at Kamphaeng Saen, Kasetsat Univesity, Thailand 2 Faculty of Management

More information

Administrivia. CMSC 411 Computer Systems Architecture Lecture 5. Data Hazard Even with Forwarding Figure A.9, Page A-20

Administrivia. CMSC 411 Computer Systems Architecture Lecture 5. Data Hazard Even with Forwarding Figure A.9, Page A-20 Administivia CMSC 411 Compute Systems Achitectue Lectue 5 Basic Pipelining (cont.) Alan Sussman als@cs.umd.edu as@csu dedu Homewok poblems fo Unit 1 due today Homewok poblems fo Unit 3 posted soon CMSC

More information

COSC 6385 Computer Architecture. - Pipelining

COSC 6385 Computer Architecture. - Pipelining COSC 6385 Compute Achitectue - Pipelining Sping 2012 Some of the slides ae based on a lectue by David Culle, Pipelining Pipelining is an implementation technique wheeby multiple instuctions ae ovelapped

More information

Topic -3 Image Enhancement

Topic -3 Image Enhancement Topic -3 Image Enhancement (Pat 1) DIP: Details Digital Image Pocessing Digital Image Chaacteistics Spatial Spectal Gay-level Histogam DFT DCT Pe-Pocessing Enhancement Restoation Point Pocessing Masking

More information

An Improved Resource Reservation Protocol

An Improved Resource Reservation Protocol Jounal of Compute Science 3 (8: 658-665, 2007 SSN 549-3636 2007 Science Publications An mpoved Resouce Resevation Potocol Desie Oulai, Steven Chambeland and Samuel Piee Depatment of Compute Engineeing

More information

THE THETA BLOCKCHAIN

THE THETA BLOCKCHAIN THE THETA BLOCKCHAIN Theta is a decentalized video steaming netwok, poweed by a new blockchain and token. By Theta Labs, Inc. Last Updated: Nov 21, 2017 esion 1.0 1 OUTLINE Motivation Reputation Dependent

More information

Reliable Surface Extraction from Point-Clouds using Scanner-Dependent Parameters

Reliable Surface Extraction from Point-Clouds using Scanner-Dependent Parameters 1 Relable Surface Extracto from Pot-Clouds usg Scaer-Depedet Parameters Hrosh Masuda 1, Ichro Taaka 2, ad Masakazu Eomoto 3 1 The Uversty of Tokyo, masuda@sys.t.u-tokyo.ac.jp 2 Tokyo Dek Uversty, taaka@cck.deda.ac.jp

More information

Point Estimation-III: General Methods for Obtaining Estimators

Point Estimation-III: General Methods for Obtaining Estimators Pot Estmato-III: Geeral Methods for Obtag Estmators RECAP 0.-0.6 Data: Radom Sample from a Populato of terest o Real valued measuremets: o Assumpto (Hopefully Reasoable) o Model: Specfed Probablty Dstrbuto

More information

A Recommender System for Online Personalization in the WUM Applications

A Recommender System for Online Personalization in the WUM Applications A Recommende System fo Online Pesonalization in the WUM Applications Mehdad Jalali 1, Nowati Mustapha 2, Ali Mamat 2, Md. Nasi B Sulaiman 2 Abstact foeseeing of use futue movements and intentions based

More information

A New and Efficient 2D Collision Detection Method Based on Contact Theory Xiaolong CHENG, Jun XIAO a, Ying WANG, Qinghai MIAO, Jian XUE

A New and Efficient 2D Collision Detection Method Based on Contact Theory Xiaolong CHENG, Jun XIAO a, Ying WANG, Qinghai MIAO, Jian XUE 5th Intenational Confeence on Advanced Mateials and Compute Science (ICAMCS 2016) A New and Efficient 2D Collision Detection Method Based on Contact Theoy Xiaolong CHENG, Jun XIAO a, Ying WANG, Qinghai

More information

Adjusting the Contact Surface of Forming Tools in Order to Compensate for Elastic Deformations during the Process

Adjusting the Contact Surface of Forming Tools in Order to Compensate for Elastic Deformations during the Process Adjustng the Contact Suface of Fomng Tools n Ode to Compensate fo Elastc Defomatons dung the Pocess Knut Goßmann, Hajo Weme, Andè Hadtmann, Las Pente, Sebastan Kechenbaue Insttute fo Machne Tools and Contol

More information

Assessment of Track Sequence Optimization based on Recorded Field Operations

Assessment of Track Sequence Optimization based on Recorded Field Operations Assessment of Tack Sequence Optimization based on Recoded Field Opeations Matin A. F. Jensen 1,2,*, Claus G. Søensen 1, Dionysis Bochtis 1 1 Aahus Univesity, Faculty of Science and Technology, Depatment

More information

Chapter 3 Descriptive Statistics Numerical Summaries

Chapter 3 Descriptive Statistics Numerical Summaries Secto 3.1 Chapter 3 Descrptve Statstcs umercal Summares Measures of Cetral Tedecy 1. Mea (Also called the Arthmetc Mea) The mea of a data set s the sum of the observatos dvded by the umber of observatos.

More information

Area and Power Efficient Modulo 2^n+1 Multiplier

Area and Power Efficient Modulo 2^n+1 Multiplier Iteratoal Joural of Moder Egeerg Research (IJMER) www.jmer.com Vol.3, Issue.3, May-Jue. 013 pp-137-1376 ISSN: 49-6645 Area ad Power Effcet Modulo ^+1 Multpler K. Ptambar Patra, 1 Saket Shrvastava, Sehlata

More information

All lengths in meters. E = = 7800 kg/m 3

All lengths in meters. E = = 7800 kg/m 3 Poblem desciption In this poblem, we apply the component mode synthesis (CMS) technique to a simple beam model. 2 0.02 0.02 All lengths in metes. E = 2.07 10 11 N/m 2 = 7800 kg/m 3 The beam is a fee-fee

More information

ANALYSIS OF VARIANCE WITH PARETO DATA

ANALYSIS OF VARIANCE WITH PARETO DATA Proceedgs of the th Aual Coferece of Asa Pacfc Decso Sceces Isttute Hog Kog, Jue -8, 006, pp. 599-609. ANALYSIS OF VARIANCE WITH PARETO DATA Lakhaa Watthaacheewakul Departmet of Mathematcs ad Statstcs,

More information

A Novel Automatic White Balance Method For Digital Still Cameras

A Novel Automatic White Balance Method For Digital Still Cameras A Novel Automatic White Balance Method Fo Digital Still Cameas Ching-Chih Weng 1, Home Chen 1,2, and Chiou-Shann Fuh 3 Depatment of Electical Engineeing, 2 3 Gaduate Institute of Communication Engineeing

More information

CS 2461: Computer Architecture 1 Program performance and High Performance Processors

CS 2461: Computer Architecture 1 Program performance and High Performance Processors Couse Objectives: Whee ae we. CS 2461: Pogam pefomance and High Pefomance Pocessos Instucto: Pof. Bhagi Naahai Bits&bytes: Logic devices HW building blocks Pocesso: ISA, datapath Using building blocks

More information

Development and Analysis of a Real-Time Human Motion Tracking System

Development and Analysis of a Real-Time Human Motion Tracking System Development and Analysis of a Real-Time Human Motion Tacking System Jason P. Luck 1,2 Chistian Debunne 1 William Hoff 1 Qiang He 1 Daniel E. Small 2 1 Coloado School of Mines 2 Sandia National Labs Engineeing

More information

Topological Characteristic of Wireless Network

Topological Characteristic of Wireless Network Topological Chaacteistic of Wieless Netwok Its Application to Node Placement Algoithm Husnu Sane Naman 1 Outline Backgound Motivation Papes and Contibutions Fist Pape Second Pape Thid Pape Futue Woks Refeences

More information