Maximum Likelihood Directed Enumeration Method in Piecewise-Regular Object Recognition

Size: px
Start display at page:

Download "Maximum Likelihood Directed Enumeration Method in Piecewise-Regular Object Recognition"

Transcription

1 Maxmum Lkelhood Dected Enumeaton Method n Pecewse-Regula Obect Recognton Andey Savchenko Abstact We exploe the poblems of classfcaton of composte obect (mages, speech sgnals wth low numbe of models pe class. We study the queston of mpovng ecognton pefomance fo medum-szed database (thousands of classes. The key ssue of fast appoxmate neaest-neghbo methods wdely appled n ths task s the heustc natue. It s possble to stongly pove the effcency by usng the theoy of algothms only fo smple smlaty measues and atfcally geneated tasks. On the contay, n ths pape we popose an altenatve, statstcally optmal geedy algothm. At each step of ths algothm ont densty (lkelhood of dstances to pevously checked models s estmated fo each class. The next model to check s selected fom the class wth the maxmal lkelhood. The latte s estmated based on the asymptotc popetes of the Kullback-Leble nfomaton dscmnaton and mathematcal model of pecewse-egula obect wth dstbuton of each egula segment of exponental type. Expemental esults n face ecognton fo FERET dataset pove that the poposed method s much moe effectve than not only bute foce and the baselne (dected enumeaton method but also appoxmate neaest neghbo methods fom FLANN and NonMetcSpaceLb lbaes (andomzed kd-tee, composte ndex, pem-sot.. Intoducton Conventonal machne leanng technques (suppot vecto machnes, multlayeed feed-fowad neual netwoks, deep neual netwoks, etc [] eque lage epesentatve tanng sample to estmate the class bode. These methods ae known to be chaactezed wth low accuacy f only few models ae avalable fo each class []. Ths ssue s qute acute n, e.g., face ecognton task n whch t s sometmes dffcult to gathe vaous photos of the nteestng peson [, 3]. The poblem of nsuffcent accuacy becomes moe complcated f the numbe of classes s lage (hundeds o even thousands of classes. As a esult, thee s pactcally no altenatve to the neaest neghbo (NN methods n ths task []. Howeve, f the complex obects should be ecognzed n eal-tme (e.g., vdeo-based face ecognton [3] and only standad hadwae s avalable, the pefomance of bute-foce mplementaton of the NN seach s not enough. It seems that conventonal fast appoxmate NN methods fo mage ecognton, e.g. tangle tee [4], composte kd-tee [5], andomzed kd-tee [6], Best- Bn Fst [7], etc. can be appled. Unfotunately, t s known that these technques show good pefomance only f the fst NN s qute dffeent fom othe models [7]. Such estcton has much n common wth many eal-wold applcatons, fo nstance, faces have smla shape and common featues. The othe lmtaton s the applcaton wth smlaty measues whch satsfy metc popetes (sometmes, tangle nequalty and, usually, symmety [4, 7, 8]. Moeove, these methods ae usually developed to appoxmately match vey-lage numbe ( of mage

2 descptos of extacted keyponts [9]. Hence, the pefomance s compaable wth bute-foce method fo medum-szed vocabulaes (thousands of classes. To decease the ecognton speed fo such tanng sets, odeng pemutatons (pemsot method has ecently been poposed [0]. Anothe nteestng appoach, namely, the dected enumeaton method (DEM outpefoms the known appoxmate NN methods n face ecognton []. Fnal ssue s the heustc natue of most popula appoxmate NN methods. It s pactcally mpossble to pove that patcula algothm s optmal (n some sense and nothng can be done to mpove t. In ths pape we popose an altenatve soluton on the bass of the statstcal appoach - whle lookng fo the NN fo patcula quey obect, conventonal pobablty of belongng of pevously checked models to each class s estmated. The next model fom the database s selected fom the class wth maxmal pobablty. Thus, ou task s to estmate ths pobablty and to clafy the mentoned geedy-seach algothm. The est of the pape s oganzed as follows. In Secton we exploe the task of ecognton of pecewse-egula obects and pesent the statstcal paametc cteon based on the Kullback-Leble mnmum dscmnaton pncple []. In Secton 3 we befly evew the baselne method (DEM, emnd the asymptotc popetes of the Kullback-Leble dscmnaton and popose the novel Maxmum-Lkelhood DEM (ML-DEM. In Secton 4 we demonstate expemental esults of compason of ou method wth seveal appoxmate NN algothms n face ecognton wth FERET dataset. Fnally, concludng comments ae gven n Secton 5.. Statstcal ecognton of pecewse-egula obect In the classfcaton task t s equed to assgn the quey obect (facal photo, speech sgnal, mage of natual scenes, text to one of R> classes. Most pat of contempoay eseach assumes that each class s specfed by the gven database { }, {,..., R} of R cases (models. Let the quey obect be epesented as a sequence of K egula (homogeneous pats [3] extacted by any segmentaton pocedue: { ( k=, K} =. Evey k-th segment ( { ( =, n( } a sequence of (pmtve featue vectos ( { x,..., ( } = x s put n coespondence wth x = ; ( x ; p wth fxed dmenson p=const, whee n( s the numbe of featues n the k-th segment. Smlaly, evey -th model s epesented as a sequence { ( k=, K } K segments and the k-th segment s defned as the numbe of featues n the k-th segment of the -th model. To apply statstcal appoach, let's assume that:. Vectos x (, x ( k ae andom. = of ( = x ( =, n ( of featue vectos x (. Hee n ( s. Segments (, k=, K and (, k =, K ae goups - andom samples of..d. featue vectos x ( and x (, espectvely.

3 3. Featue vectos of patcula segment of one class ae dentcally dstbuted. As the pocedue of automatc segmentaton s naccuate, evey segment ( should be compaed wth a set ( N of numbes of closed to k segments of the -th model. Ths neghbohood s detemned fo a specfc task ndvdually. If t s assumed that segmentaton pocedue s always coect, we may put N ( { k}, =, K = K. K K Thee ae two possble appoaches to estmate unknown class denstes, namely, paametc and nonpaametc []. Let's dscove paametc appoach n detal. It s assumed that dstbutons of vectos x ( and x ( ae of multvaate exponental type f θ ; n [] geneated by the fxed (fo all classes functon f 0( wth p-dmensonal paamete vecto θ: ( exp( τ( θ θˆ( f ( / M( fθ; n = 0 τ ( whee ˆ( θ s an estmaton of paamete θ usng avalable data (andom sample of sze n, ( τ( θ θˆ( f ( M( τ = exp 0 d ( and τ (θ s a nomalzng functon (p-dmensonal paamete vecto defned by the followng equaton f the paamete estmaton ˆ( θ s unbased (see [] fo detals d θˆ( fθ ( d ln ( τ = θ ; n M (3 dτ Each -th class of each k-th segment s detemned by paamete vecto θ (. Ths assumpton about exponental famly f θ ˆ( ( ; n( n whch paamete θ (k s estmated by usng the obseved (gven sample (, coves wde ange of known dstbutons (polynomal, nomal, etc. []. Hence, the ecognton task s educed to a poblem of statstcal testng of R smple hypothess about paamete vecto θ (. In ths pape we focus on the case of full po uncetanty and assume that the po pobabltes of each class ae equal. In such case, Bayesan appoach wll be equvalent to the maxmum lkelhood cteon. Fo ou task, evey segment s ecognzed wth the followng ule max k N max,..., f ( ( θˆ ( k ; n( k { R}. (4 It can be shown that eq. (4 s equvalent to the Kullback-Leble mnmum nfomaton dscmnaton pncple [] whee and (, ρ = K = mn nk k = k N (, ρ mn, (5 =, R I ˆ (*: f ; ( θˆ ( ( k ; k n( (6 3

4 Iˆ *: fˆ θ ( ( k ; = f ( θˆ ( ( ; n( ; ( = n( f θˆ ln f θˆ ( ( ; ( ( k ( n( d ( ; n( (7 s the Kullback-Leble dvegence between segments ( and ( k ; and n= K n(. k= Thus, cteon (5-(7 s an obvous mplementaton of Bayesan appoach to composte obect ecognton f the pobablstc mathematcal model of pecewse-egula obect [3] s used. 3. Maxmum-lkelhood dected enumeaton method Let's use an appoach known fom atfcal ntellgence to ceate an appoxmate NN algothm fo measue of smlaty (6, (7. Namely, we pmaly focus on geedy algothms: on each step t exploes the model whch s the NN of the quey obect wth the hghest pobablty. Such choose of the geedy class of algothms s explaned not only by ts smplcty, but by the fact that pactcally all known appoxmate NN methods ae geedy n some sense. 3.. Baselne: dected enumeaton method As a baselne method we use the DEM [] whch was based on the metc popetes of the Kullback-Leble dvegence and egads the models' smlaty ρ, = ρ(, as an aveage nfomaton fom an obsevaton to dstnct class fom an altenatve class. Hence, at the pelmnaly step of the DEM, the model dstance matx Ρ = [ ρ, ] s calculated as t s done n the AESA (Appoxmatng and Elmnatng Seach Algothm method [4]. Ths tme-consumng pocedue should be epeated only once fo a patcula task and tanng set. Ognal DEM used the followng heustc: f thee exsts a model * fo whch ρ *, < ρ0 <<, then fo an abtay -th model the followng condton holds (, ρ,. << (5 can be smplfed ρ * wth hgh pobablty. Hence, the ctea ρ, * < ρ = const 0. (8 Eq. (8 defnes the temnaton condton of the appoxmate NN method. If false-accept ate (FAR s fxed then 0 β = const, ρ s evaluated as a β -quantle of the dstances between mages fom dstnct classes { ρ =, R, =, R, } a matte of fact, the optmzaton task (5 s eplaced to an exhaustve seach whch temnates f condton (8 holds fo the cuently checked model. Accodng to the DEM [], at fst, the model {,..., R}, s andomly chosen so that {,..., R} ρ, ρ,,, (9,. As 4

5 and the dstance ρ, s calculated. If the dstance s lowe than a theshold ρ 0 (8, the seach s temnated. Othewse, t s put nto the poty queue of models soted by the dstance to. Next, the hghest poty tem s pulled fom the queue and the set of models whee ρ( = ρ (, the set (M s detemned fom ( M M k, (0 ρ( ρ(, ρ s the devaton of ρ, elatve to the dstance between and. Fo all models fom (M the dstance to the quey obect s calculated and the condton (9 s vefed. Afte that, evey pevously unchecked model fom ths set s put nto the poty queue. The method s temnated f fo one model obect condton (9 holds o afte checkng fo E max = const models. As we stated eale, ths method s heustc as most popula appoxmate NN algothms. Howeve, the pobablty that the model s the NN of can be dectly calculated fo the Kullback-Leble dscmnaton by usng ts asymptotc popetes. Let's descbe them befly. 3.. Asymptotc popetes θ ˆ ν ν, It s known [] that f the segment ( has dstbuton of exponental type wth paamete ( (, {,..., R} then the -tmes Kullback-Leble dvegence (6 I *: fˆ ; ( ; k θ k n( ˆ s asymptotcally dstbuted as a noncental χ wth p degees of feedom and noncentalty paamete I *: fˆ ; ( k θ k ; n( ν ˆ. By assumng the ndependence of all K segments (, we can conclude that f the quey obect coesponds to class ν, then the dstance ( nk ρ, ν s asymptotcally dstbuted as a has asymptotc non-cental χ dstbuton wth χ wth K p degees of feedom. Smlaly, nk ρ(,, ν K p degees of feedom and noncentalty paamete K p s hgh, then, by usng the cental lmt theoem we obtan the nomal dstbuton nk ρν,. If of the dstance (, ρ. p 8nK ρν, + K p N ρ ν, + ; n nk. ( 5

6 3.3. Poposed method Based on the asymptotc dstbuton ( we eplace the step (0 of the ognal DEM to the pocedue of choosng the maxmum lkelhood model. Let's assume that the models dstances,..., l have been checked befoe the l-th step,.e. the ρ,,..., ρ, have been calculated. By assumng the equal po pobablty of each class and l ndependence of the models fom dffeent classes, let's choose the next most pobable model lkelhood method []: l l+ = ν l wth the maxmum l+ ag max f ρ (, W ν, ( {,..., R} {,..., } = whee f (, s the condtonal densty (lkelhood of the dstance (, ρ W ν ρ f the hypothess W ν s tue (the class label of the quey obect s ν. To estmate ths lkelhood, asymptotc dstbuton ( s used. Hence, the lkelhood n ( can be wtten n the followng fom ( ρ(, W ( nk ( ρ(, ρ K p nk ν, f = = ν exp π (8nK ρ ν, + K p 8nK ρν, + K p (3. ( nk = + nk ( ρ(, ρ K p ν, exp ln8nk ρν, K p exp π 8nK ρν, + K p By dvdng ( by a constant ( nk / π l, takng a natual logathm, dvdng by nk / and addng l, expesson ( can be fnally tansfomed to whee l ag mn ϕ µ. (4 µ {,..., R} {,..., } = + = l l p ρ(, ρ µ, = n p ϕ µ + ln4ρµ, +. (5 p 4nK n 4ρµ, + n As the aveage segment's sze s usually much hghe the numbe of paametes smplfed ϕ µ ( ( ρ(, ρ n>> N, then the functon n (5 can be µ, (6 4ρµ, Ths equaton s n good ageement wth the heustc fom the ognal DEM [] - the close ae the dstances ρ (, and ρ µ, and the hghe s the dstance between models µ and, the lowe s ϕ. µ 6

7 Next, the temnaton condton (8 s checked fo the model. If the dstance ρ, l+ s lowe than a l + theshold ρ 0, then the seach pocedue s stopped on the L checks = l+ step. Othewse the model set of pevously checked models and the pocedue (4, (6 s epeated. Let's etun to the ntalzaton of ou method. We would lke to choose the fst model s put nto the l+ to obtan the decson (8 n a shotest (n tems of numbe of calculatons L checks way. Let's maxmze an aveage pobablty to obtan the decson on the second step To estmate the condtonal pobablty Pϕ ( (: Pϕ ν = ( µ {,..., Rρ ρ } Fnally, based on ( one can wte R = ag max P ϕ ν µ W { } = { },..., R R mn ϕ µ ν. (7 µ ν,... R ν mn ϕ Wν n (7 we use agan the asymptotc dstbuton {,... R} ( ρ(, µ ρ, µ ρν, µ Wν ( ρ(, ρ ( ρ(, ρ R µ ν, µ µ mn ϕ µ Wν = P { },... R = ρν, µ ρ, µ { Rρ < } P P, µ ν, µ,...,, µ ρν, µ, µ Wν ( ρ(, µ ρ, µ ρν, µ Wν R { } nk P ϕ = +Φ ν µ mn ϕ µ Wν ρ, µ ρν, µ, (0,... R = whee Φ ( s the cumulatve densty functon of the nomal dstbuton. As a esult, the fst model to check obtaned fom the followng expesson { } R R = ag max +Φ µ,..., R ν= = nk (9 s ρ, µ ρν, µ. ( Thus, the poposed ML-DEM (9, (4, (6, ( s an optmal (maxmal lkelhood geedy algothm fo an appoxmate NN seach wth temnaton condton (8 fo the Kullback-Leble dscmnaton (6, (7. As a matte of fact, ths method can be appled wth an abtay complex smlaty measue. The next secton povdes expemental evdence to suppot ths clam. 4. Expemental study In ths secton we pesent expemental study of poposed ML-DEM n the poblem of face ecognton. 4.. Face ecognton It s equed to assgn a quey mage to one of R classes (pesons specfed by efeence mages { }, {,..., R}. 7

8 We assume that the obect of nteest (face s pelmnaly detected by an abtay algothm (e.g., Vola-Jones method [4]. To mplement hee statstcal appoach descbed n Secton, the mage s dvded nto a egula gd of S S blocks, S ows and S columns (n ou notaton, S S = K = K =... = KR. Next, the hstogam H ( s, s = [ h ( s, s,..., hn ( s, s] of appopate smple featues s sepaately evaluated fo each block ( s, s of the efeence mage. Hee N s the numbe of bns n the hstogam, s { S }, s { S },...,,...,. The most popula mage pont's smple featue s the gadent oentaton (pobably, weghted wth the gadent magntude,.e., H ( s, s s the hstogam of oented gadents (HOG [5]. In ths pape we assume, that each hstogam H ( s, s nomalzed, so that t may be teated as a pobablty dstbuton [5]. The unted vecto [ H (,,..., H (, S, H (,,..., H ( S, S] s amounted the desed descpto of the whole efeence mage. The same pocedue s epeated to evaluate the hstogams H( s, s = [ h( s, s,..., hn( s, s] coespondng to the quey mage. The neghbohood ( ( N of block (, s s contans the cells (, s s fo whch s s <, s s <, whee = const s chosen based on the concete task (usually =0 o = [6]. In such case, the dstance n the neaest neghbo ule (5, (6 wll be calculated wth the mutual algnment of the hstogams n the - neghbohood as follows ρ S S (, = mn ρ ( H H( s s, H ( s + s + whee SS s = s =,,, s, ( ρ ( H H s,, +, + s H s s s an appopate dstance between HOGs. If the Kullback-Leble dscmnaton (7 s appled, the dstance between HOGs can be wtten n the followng fom N hk; ρ KLH, H = hk; ln. (3 = hk; Hee we mssed ndces ( s, s fo smplcty and use the convoluton of the HOGs wth any kenel K N N hk; = Kh, hk; = Kh (4 = = to pevent dvson by zeo n (3 f the hstogam value fo seveal bns s equal to zeo. In the expement we use the conventonal Gaussan Pazen wndow [7]. In ths study we also exploe the homogenety-testng pobablstc neual netwok (HT-PNN showed good accuacy wth HOGs n face ecognton and known to be equvalent to the statstcal appoach f the patten ecognton poblem s efeed as a task of testng fo homogenety of segments [8]: HT PNN ( H, H = N hk; h K; h ln + h ln = h + + K; hk; hk; hk; ρ. (5 8

9 4.. Expemental esults In ths expement FERET dataset was used ( R=43 fontal mages of 994 pesons populate the database (.e. a tanng set, othe 88 fontal photos of the same pesons fomed a test set. The faces wee detected wth the OpenCV lbay. The medan flte wth wndow sze (3x3 was appled to emove nose n detected faces. The faces wee dvded nto 00 fagments ( S = S = 0. The numbe of bns n the HOG N=8. To obtan theshold ρ 0, the FAR s fxed to be β =%. These paametes povde the best accuacy n ou expements. The eo ate obtaned by coss-valdaton wth the NN ule and smlaty measue ( wth Eucldean and the PNNH ( dstances s shown n Table n the fomat aveage eo ate ± ts standad devaton. Table. Eo ate (n % of the NN method ( =0 = Kullback-Leble (3 8.9±.3 7.0±.3 HT-PNN (5 7.8±. 6.6±.3 Fom ths table one could notce that, fst, algnment of HOGs ( wth = mpoves the ecognton accuacy. And, second, we expementally suppots the fact [8] that the eo ate fo the Kullback-Leble dstance (3 exceeds the eo fo the HT-PNN (5. In the next expement we compae the pefomance of the poposed ML-DEM wth an ognal DEM [], bute foce and seveal appoxmate NN methods fom FLANN [5] and NonMetcSpaceLb [9] lbaes showed the best speed, namely. Randomzed kd-tee fom FLANN wth 4 tees [6]. Composte ndex fom FLANN whch combnes the andomzed kd-tees (wth 4 tees and the heachcal k-means tee [5]. 3. Odeng pemutatons (pem-sot fom NonMetcSpaceLb whch s known to decease the ecognton speed fo medum-szed databases (thousands of models [0]. We evaluate the eo ate (n % and the aveage tme (n ms to ecognze one test mage wth a moden laptop (4 coe 7, 6 Gb RAM and Vsual C++ 03 Expess comple (x64 envonment and optmzaton by speed. We exploe an obvous way to mpove pefomance by usng paallel computng [6]. Namely, the whole tanng set was dvded nto T=const non-ovelapped pats and each pat s pocessed n ts own task. All tasks wok n paallel and temnate ght afte any task fnds the soluton. Each task s mplemented as a sepaate thead by usng the Wndows TheadPool API. We analyze both conventonal nonpaallel case (T= and the paallel one (T=8. Afte seveal expements the best (n tems of ecognton speed value of paamete M of ognal DEM (0 was chosen M=64 fo nonpaallel case and M=6 fo paallel one. Paamete E max was chosen to acheve the ecognton 9

10 accuacy whch s not 0.5% less than the accuacy of bute foce (Table. If such accuacy could not be acheved, E max was set to be equal to the count of models assgned to each task. The aveage ecognton tme pe one test mage (n ms fo the Kullback-Leble dscmnaton (3 fo =0 and = s shown n Fg. and Fg., espectvely. 4 Aveage ecognton tme, ms Numbe of paallel theads T Bute foce Randomzed KD tee Composte Pem-sot DEM ML-DEM Fgue : Aveage ecognton tme, Kullback-Leble dscmnaton, =0 Aveage ecognton tme, ms Numbe of paallel theads T Bute foce Randomzed KD tee Composte Pem-sot DEM ML-DEM Fgue : Aveage ecognton tme, Kullback-Leble dscmnaton, = Hee one can notce that modfcatons of kd-tee fom FLANN (andomzed and composte ndces do not show supeo pefomance even ove bute foce as the numbe of models n the database s not vey hgh. Howeve, as t was expected, pem-sot method s chaactezed wth -3.5 tmes lowe ecognton speed n compason wth an exhaustve seach. Moeove, pem-sot seems to be bette than the ognal DEM fo nonpaallel case (T=, though the DEM's paallel mplementaton s a bt bette. The most mpotant concluson hee s that the poposed ML-DEM shows the hghest speed n all expements. To clafy the dffeence n pefomance of the ognal DEM and the poposed ML-DEM, we show the dependence of the eo ate and the numbe of checked models L checks / R 00% on the maxmum numbe of models to be checked E max n Fg. 3 and Fg. 4, espectvely. We descbe hee the case = fo whch the DEM s the best among all othe methods. Fg. 3 demonstates that the speed of convegence to an optmal soluton fo the ML-DEM s much hghe than the same 0

11 ndcato of the DEM. Even when Emax = 0. R we can get an appopate soluton. Eo ate, % DEM ML-DEM 0, 0,3 0,5 0,7 0,9 Emax/R Fgue 3: Dependence of eo ate on E max, Kullback-Leble dscmnaton, = Count of model checks pe database sze, % , 0,3 0,5 0,7 0,9 DEM ML-DEM Emax/R Fgue 4: Count of models checks pe database sze L checks / R 00%, Kullback-Leble dscmnaton, = Fg. 4 poves that the poposed ML-DEM s an optmal geedy algothm n tems of the numbe of calculated dstances L checks. Howeve, addtonal computatons of the ML-DEM (4, (6 whch nclude the calculatons fo evey non pevously checked model, ae qute complex. Hence, the dffeence n pefomance wth the DEM and othe appoxmate NN methods s hgh only fo vey complex smlaty measues (e.g., fo the case of HOG's algnment, =. The aveage ecognton tme fo the HT-PNN (5 fo =0 and = s shown n Fg. 5 and Fg. 6, espectvely. 0 Aveage ecognton tme, ms Numbe of paallel theads T Bute foce Randomzed KD tee Composte Pem-sot DEM ML-DEM Fgue 5: Aveage ecognton tme, HT-PNN, =0

12 Aveage ecognton tme, ms Numbe of paallel theads T Bute foce Randomzed KD tee Composte Pem-sot DEM ML-DEM Fgue 6: Aveage ecognton tme, HT-PNN, = The esults of ths expement ae vey smla to the Kullback-Leble esults (Fg., though the eo ate hee s 0.5- % lowe (see also Table. Howeve, the ognal DEM s hee a bt faste than the pem-sot fo conventonal dstance ( =0, Fg. 5 but s not so effectve fo algnment ( =, Fg. 6. FLANN's kd-tees ae 0-5% faste than the bute foce. And agan, the poposed ML-DEM s the best choce hee especally fo most complex case (T=8, = fo whch only 6 ms (n aveage s necessay to ecognze a quey face wth 93% accuacy. 5. Concluson and futue wok We have shown that usng the asymptotc popetes ( of the Kullback-Leble dscmnaton fo ecognton pecewse-egula obects (5-(7 n the DEM [] gves vey good esults n face ecognton wth medum-szed database, educng the ecognton speed by moe than tmes n compason wth bute foce and by.-.5 tmes n compason wth othe appoxmate NN methods fom FLANN and NonMetcSpaceLb lbaes. We studed the nfluence of vaous dstance paametes (dstance type, neghbohood sze and the maxmal numbe E max of dstances to calculate. In contast to the most popula fast algothms, ou method s not heustc (except the temnaton condton (8. Moeove, t does not buld data stuctue based on an algothmc popetes of appled smlaty measue (e.g., tangle nequalty of Mnkowsk metc n the AESA [4], Begman ball fo Begman dvegences [8]. The poposed ML-DEM s an optmal (maxmum lkelhood geedy method n tems of the numbe of dstance calculatons (see Fg. 3 fo NN ule (5 wth the sum of Kullback-Leble dscmnatons (5, (6. Moeove, as we showed n the last pat of ou expemental study, the ML-DEM can be successfully appled (Fg. 5, 6 wth othe dstances, e.g., not popula but vey accuate HT- PNN (5 [8]. The man decton fo futhe eseach n the ML-DEM can be elated to mpovng the pefomance of each step (4, (6 by ts smplfcaton o the usage of deas of pvot-based appoxmate NN methods [0]. As a matte of fact, t s the man obstacle to use ou method wth vaous smlaty measues. We ae also wokng on exploaton the nfluence of the popula dstances (Eucldean, ch-squaed, Jensen-Shannon, etc. on the pefomance of ou method. Refeences [] S. Theodods, and K. Koutoumbas (eds.. Patten ecognton. Boston: Academc Pess, p.

13 []. Tan, S. Chen, Z. H. Zhou, and F. Zhang. Face ecognton fom a sngle mage pe peson: a suvey. Patten Recognton, 39(9:75 745, 006 [3] C. Shan. Face ecognton and eteval n vdeo // Vdeo Seach and Mnng, Studes n Computatonal Intellgence, 87:35-60, 00 [4] E. Vdal. An algothm fo fndng neaest neghbous n (appoxmately constant aveage tme. Patten ecognton Lettes, 4(3:45 57, 986. [5] M. Mua, and D. G. Lowe. Fast appoxmate neaest neghbos wth automatc algothm confguaton. 4th Intenatonal Confeence on Compute Vson Theoy & Applcatons (VISAPP, : , 009. [6] C. Slpa-Anan, and R. Hatley. Optmsed KD-tees fo fast mage descpto matchng. CVPR, Anchoage, Alaska, USA, pages -8, 008 [7] J. Bes, and D. G. Lowe. Shape ndexng usng appoxmate neaest neghbou seach n hgh dmensonal spaces. CVPR, San Juan, Pueto Rco, pages , 997. [8] L. Cayton. Effcent Begman Range Seach. Advances n Neual Infomaton Pocessng Systems, Y. Bengo, D. Schuumans, J.D. Laffety, C.K.I. Wllams, and A. Culotta (Eds., :43-5, 009 [9] D. Lowe. Dstnctve mage featues fom scale nvaant keyponts. Intenatonal Jounal of Compute Vson, 60(:9 0, 004 [0] E. C. Gonzalez, K. Fgueoa, and G. Navao. Effectve Poxmty Reteval by Odeng Pemutatons. IEEE Tansactons on Patten Analyss and Machne Intellgence, 30(9: , 008. [] A. V. Savchenko, Dected enumeaton method n mage ecognton. Patten Recognton, 45(8:95 96, 0. [] S. Kullback. Infomaton Theoy and Statstcs. Mneola, N.Y: Dove Publcatons, 997. [3] P. Pandon, and M.Vettel. Appoxmaton and compesson of pecewse smooth functons. Phlosophcal Tansactons of the Royal Socety, 357(760:573-59, 999. [4] P. Vola, and M. Jones. Rapd obect detecton usng a boosted cascade of smple featues. CVPR, Kaua, HI, USA, pages 5 58, 00. [5] N. Dalal, and B. Tggs. Hstogams of oented gadents fo human detecton. CVPR, San Dego, CA, USA, pages , 005. [6] A. V. Savchenko. Real-Tme Image Recognton wth the Paallel Dected Enumeaton Method. The 9th ICVS (Intenatonal Confeence on Vson Systems, M. Chen, B. Lebe, and B. Neumann (Eds., LNCS, 7963:3-3, 03. [7] D. F. Specht. Pobablstc neual netwoks. Neual netwoks, 3(:09 8, 990. [8] A. V. Savchenko. Pobablstc neual netwok wth homogenety testng n ecognton of dscete pattens set. Neual Netwoks, 46:7 4, 03. [9] L. Boytsov, and N. Blegsakhan. Engneeng Effcent and Effectve Non-Metc Space Lbay. The 6th SISAP (Smlaty Seach and Applcatons, N. Bsaboa, O. Pedea, P. Zezula (Eds., LNCS 899: 80-93, 03 [0] B. Bustos, E. Ch avez, and G. Navao. Pvot selecton technques fo poxmty seachng n metc spaces. Patten Recognton Lettes, 4-4: , 003 3

Multistage Median Ranked Set Sampling for Estimating the Population Median

Multistage Median Ranked Set Sampling for Estimating the Population Median Jounal of Mathematcs and Statstcs 3 (: 58-64 007 ISSN 549-3644 007 Scence Publcatons Multstage Medan Ranked Set Samplng fo Estmatng the Populaton Medan Abdul Azz Jeman Ame Al-Oma and Kamaulzaman Ibahm

More information

Learning the structure of Bayesian belief networks

Learning the structure of Bayesian belief networks Lectue 17 Leanng the stuctue of Bayesan belef netwoks Mlos Hauskecht mlos@cs.ptt.edu 5329 Sennott Squae Leanng of BBN Leanng. Leanng of paametes of condtonal pobabltes Leanng of the netwok stuctue Vaables:

More information

Exact Simplification of Support Vector Solutions

Exact Simplification of Support Vector Solutions Jounal of Machne Leanng Reseach 2 (200) 293-297 Submtted 3/0; Publshed 2/0 Exact Smplfcaton of Suppot Vecto Solutons Tom Downs TD@ITEE.UQ.EDU.AU School of Infomaton Technology and Electcal Engneeng Unvesty

More information

Machine Learning. Spectral Clustering. Lecture 23, April 14, Reading: Eric Xing 1

Machine Learning. Spectral Clustering. Lecture 23, April 14, Reading: Eric Xing 1 Machne Leanng -7/5 7/5-78, 78, Spng 8 Spectal Clusteng Ec Xng Lectue 3, pl 4, 8 Readng: Ec Xng Data Clusteng wo dffeent ctea Compactness, e.g., k-means, mxtue models Connectvty, e.g., spectal clusteng

More information

Dirichlet Mixture Priors: Inference and Adjustment

Dirichlet Mixture Priors: Inference and Adjustment Dchlet Mxtue Pos: Infeence and Adustment Xugang Ye (Wokng wth Stephen Altschul and Y Kuo Yu) Natonal Cante fo Botechnology Infomaton Motvaton Real-wold obects Independent obsevatons Categocal data () (2)

More information

3. A Review of Some Existing AW (BT, CT) Algorithms

3. A Review of Some Existing AW (BT, CT) Algorithms 3. A Revew of Some Exstng AW (BT, CT) Algothms In ths secton, some typcal ant-wndp algothms wll be descbed. As the soltons fo bmpless and condtoned tansfe ae smla to those fo ant-wndp, the pesented algothms

More information

8 Baire Category Theorem and Uniform Boundedness

8 Baire Category Theorem and Uniform Boundedness 8 Bae Categoy Theoem and Unfom Boundedness Pncple 8.1 Bae s Categoy Theoem Valdty of many esults n analyss depends on the completeness popety. Ths popety addesses the nadequacy of the system of atonal

More information

Correspondence Analysis & Related Methods

Correspondence Analysis & Related Methods Coespondence Analyss & Related Methods Ineta contbutons n weghted PCA PCA s a method of data vsualzaton whch epesents the tue postons of ponts n a map whch comes closest to all the ponts, closest n sense

More information

P 365. r r r )...(1 365

P 365. r r r )...(1 365 SCIENCE WORLD JOURNAL VOL (NO4) 008 www.scecncewoldounal.og ISSN 597-64 SHORT COMMUNICATION ANALYSING THE APPROXIMATION MODEL TO BIRTHDAY PROBLEM *CHOJI, D.N. & DEME, A.C. Depatment of Mathematcs Unvesty

More information

Set of square-integrable function 2 L : function space F

Set of square-integrable function 2 L : function space F Set of squae-ntegable functon L : functon space F Motvaton: In ou pevous dscussons we have seen that fo fee patcles wave equatons (Helmholt o Schödnge) can be expessed n tems of egenvalue equatons. H E,

More information

Khintchine-Type Inequalities and Their Applications in Optimization

Khintchine-Type Inequalities and Their Applications in Optimization Khntchne-Type Inequaltes and The Applcatons n Optmzaton Anthony Man-Cho So Depatment of Systems Engneeng & Engneeng Management The Chnese Unvesty of Hong Kong ISDS-Kolloquum Unvestaet Wen 29 June 2009

More information

Distinct 8-QAM+ Perfect Arrays Fanxin Zeng 1, a, Zhenyu Zhang 2,1, b, Linjie Qian 1, c

Distinct 8-QAM+ Perfect Arrays Fanxin Zeng 1, a, Zhenyu Zhang 2,1, b, Linjie Qian 1, c nd Intenatonal Confeence on Electcal Compute Engneeng and Electoncs (ICECEE 15) Dstnct 8-QAM+ Pefect Aays Fanxn Zeng 1 a Zhenyu Zhang 1 b Lnje Qan 1 c 1 Chongqng Key Laboatoy of Emegency Communcaton Chongqng

More information

The Greatest Deviation Correlation Coefficient and its Geometrical Interpretation

The Greatest Deviation Correlation Coefficient and its Geometrical Interpretation By Rudy A. Gdeon The Unvesty of Montana The Geatest Devaton Coelaton Coeffcent and ts Geometcal Intepetaton The Geatest Devaton Coelaton Coeffcent (GDCC) was ntoduced by Gdeon and Hollste (987). The GDCC

More information

Thermodynamics of solids 4. Statistical thermodynamics and the 3 rd law. Kwangheon Park Kyung Hee University Department of Nuclear Engineering

Thermodynamics of solids 4. Statistical thermodynamics and the 3 rd law. Kwangheon Park Kyung Hee University Department of Nuclear Engineering Themodynamcs of solds 4. Statstcal themodynamcs and the 3 d law Kwangheon Pak Kyung Hee Unvesty Depatment of Nuclea Engneeng 4.1. Intoducton to statstcal themodynamcs Classcal themodynamcs Statstcal themodynamcs

More information

A Brief Guide to Recognizing and Coping With Failures of the Classical Regression Assumptions

A Brief Guide to Recognizing and Coping With Failures of the Classical Regression Assumptions A Bef Gude to Recognzng and Copng Wth Falues of the Classcal Regesson Assumptons Model: Y 1 k X 1 X fxed n epeated samples IID 0, I. Specfcaton Poblems A. Unnecessay explanatoy vaables 1. OLS s no longe

More information

CS649 Sensor Networks IP Track Lecture 3: Target/Source Localization in Sensor Networks

CS649 Sensor Networks IP Track Lecture 3: Target/Source Localization in Sensor Networks C649 enso etwoks IP Tack Lectue 3: Taget/ouce Localaton n enso etwoks I-Jeng Wang http://hng.cs.jhu.edu/wsn06/ png 006 C 649 Taget/ouce Localaton n Weless enso etwoks Basc Poblem tatement: Collaboatve

More information

Physics 2A Chapter 11 - Universal Gravitation Fall 2017

Physics 2A Chapter 11 - Universal Gravitation Fall 2017 Physcs A Chapte - Unvesal Gavtaton Fall 07 hese notes ae ve pages. A quck summay: he text boxes n the notes contan the esults that wll compse the toolbox o Chapte. hee ae thee sectons: the law o gavtaton,

More information

N = N t ; t 0. N is the number of claims paid by the

N = N t ; t 0. N is the number of claims paid by the Iulan MICEA, Ph Mhaela COVIG, Ph Canddate epatment of Mathematcs The Buchaest Academy of Economc Studes an CECHIN-CISTA Uncedt Tac Bank, Lugoj SOME APPOXIMATIONS USE IN THE ISK POCESS OF INSUANCE COMPANY

More information

PHYS 705: Classical Mechanics. Derivation of Lagrange Equations from D Alembert s Principle

PHYS 705: Classical Mechanics. Derivation of Lagrange Equations from D Alembert s Principle 1 PHYS 705: Classcal Mechancs Devaton of Lagange Equatons fom D Alembet s Pncple 2 D Alembet s Pncple Followng a smla agument fo the vtual dsplacement to be consstent wth constants,.e, (no vtual wok fo

More information

Tian Zheng Department of Statistics Columbia University

Tian Zheng Department of Statistics Columbia University Haplotype Tansmsson Assocaton (HTA) An "Impotance" Measue fo Selectng Genetc Makes Tan Zheng Depatment of Statstcs Columba Unvesty Ths s a jont wok wth Pofesso Shaw-Hwa Lo n the Depatment of Statstcs at

More information

Rigid Bodies: Equivalent Systems of Forces

Rigid Bodies: Equivalent Systems of Forces Engneeng Statcs, ENGR 2301 Chapte 3 Rgd Bodes: Equvalent Sstems of oces Intoducton Teatment of a bod as a sngle patcle s not alwas possble. In geneal, the se of the bod and the specfc ponts of applcaton

More information

Experimental study on parameter choices in norm-r support vector regression machines with noisy input

Experimental study on parameter choices in norm-r support vector regression machines with noisy input Soft Comput 006) 0: 9 3 DOI 0.007/s00500-005-0474-z ORIGINAL PAPER S. Wang J. Zhu F. L. Chung Hu Dewen Expemental study on paamete choces n nom- suppot vecto egesson machnes wth nosy nput Publshed onlne:

More information

A. Thicknesses and Densities

A. Thicknesses and Densities 10 Lab0 The Eath s Shells A. Thcknesses and Denstes Any theoy of the nteo of the Eath must be consstent wth the fact that ts aggegate densty s 5.5 g/cm (ecall we calculated ths densty last tme). In othe

More information

Robust Feature Induction for Support Vector Machines

Robust Feature Induction for Support Vector Machines Robust Featue Inducton fo Suppot Vecto Machnes Rong Jn Depatment of Compute Scence and Engneeng, Mchgan State Unvesty, East Lansng, MI4884 ROGJI@CSE.MSU.EDU Huan Lu Depatment of Compute Scence and Engneeng,

More information

Engineering Mechanics. Force resultants, Torques, Scalar Products, Equivalent Force systems

Engineering Mechanics. Force resultants, Torques, Scalar Products, Equivalent Force systems Engneeng echancs oce esultants, Toques, Scala oducts, Equvalent oce sstems Tata cgaw-hll Companes, 008 Resultant of Two oces foce: acton of one bod on anothe; chaacteed b ts pont of applcaton, magntude,

More information

VParC: A Compression Scheme for Numeric Data in Column-Oriented Databases

VParC: A Compression Scheme for Numeric Data in Column-Oriented Databases The Intenatonal Aab Jounal of Infomaton Technology VPaC: A Compesson Scheme fo Numec Data n Column-Oented Databases Ke Yan, Hong Zhu, and Kevn Lü School of Compute Scence and Technology, Huazhong Unvesty

More information

LASER ABLATION ICP-MS: DATA REDUCTION

LASER ABLATION ICP-MS: DATA REDUCTION Lee, C-T A Lase Ablaton Data educton 2006 LASE ABLATON CP-MS: DATA EDUCTON Cn-Ty A. Lee 24 Septembe 2006 Analyss and calculaton of concentatons Lase ablaton analyses ae done n tme-esolved mode. A ~30 s

More information

Event Shape Update. T. Doyle S. Hanlon I. Skillicorn. A. Everett A. Savin. Event Shapes, A. Everett, U. Wisconsin ZEUS Meeting, October 15,

Event Shape Update. T. Doyle S. Hanlon I. Skillicorn. A. Everett A. Savin. Event Shapes, A. Everett, U. Wisconsin ZEUS Meeting, October 15, Event Shape Update A. Eveett A. Savn T. Doyle S. Hanlon I. Skllcon Event Shapes, A. Eveett, U. Wsconsn ZEUS Meetng, Octobe 15, 2003-1 Outlne Pogess of Event Shapes n DIS Smla to publshed pape: Powe Coecton

More information

Constraint Score: A New Filter Method for Feature Selection with Pairwise Constraints

Constraint Score: A New Filter Method for Feature Selection with Pairwise Constraints onstant Scoe: A New Flte ethod fo Featue Selecton wth Pawse onstants Daoqang Zhang, Songcan hen and Zh-Hua Zhou Depatment of ompute Scence and Engneeng Nanjng Unvesty of Aeonautcs and Astonautcs, Nanjng

More information

Chapter Fifiteen. Surfaces Revisited

Chapter Fifiteen. Surfaces Revisited Chapte Ffteen ufaces Revsted 15.1 Vecto Descpton of ufaces We look now at the vey specal case of functons : D R 3, whee D R s a nce subset of the plane. We suppose s a nce functon. As the pont ( s, t)

More information

APPLICATIONS OF SEMIGENERALIZED -CLOSED SETS

APPLICATIONS OF SEMIGENERALIZED -CLOSED SETS Intenatonal Jounal of Mathematcal Engneeng Scence ISSN : 22776982 Volume Issue 4 (Apl 202) http://www.mes.com/ https://stes.google.com/ste/mesounal/ APPLICATIONS OF SEMIGENERALIZED CLOSED SETS G.SHANMUGAM,

More information

Re-Ranking Retrieval Model Based on Two-Level Similarity Relation Matrices

Re-Ranking Retrieval Model Based on Two-Level Similarity Relation Matrices Intenatonal Jounal of Softwae Engneeng and Its Applcatons, pp. 349-360 http://dx.do.og/10.1457/sea.015.9.1.31 Re-Rankng Reteval Model Based on Two-Level Smlaty Relaton Matces Hee-Ju Eun Depatment of Compute

More information

Vibration Input Identification using Dynamic Strain Measurement

Vibration Input Identification using Dynamic Strain Measurement Vbaton Input Identfcaton usng Dynamc Stan Measuement Takum ITOFUJI 1 ;TakuyaYOSHIMURA ; 1, Tokyo Metopoltan Unvesty, Japan ABSTRACT Tansfe Path Analyss (TPA) has been conducted n ode to mpove the nose

More information

Approximate Abundance Histograms and Their Use for Genome Size Estimation

Approximate Abundance Histograms and Their Use for Genome Size Estimation J. Hlaváčová (Ed.): ITAT 2017 Poceedngs, pp. 27 34 CEUR Wokshop Poceedngs Vol. 1885, ISSN 1613-0073, c 2017 M. Lpovský, T. Vnař, B. Bejová Appoxmate Abundance Hstogams and The Use fo Genome Sze Estmaton

More information

UNIT10 PLANE OF REGRESSION

UNIT10 PLANE OF REGRESSION UIT0 PLAE OF REGRESSIO Plane of Regesson Stuctue 0. Intoducton Ojectves 0. Yule s otaton 0. Plane of Regesson fo thee Vaales 0.4 Popetes of Resduals 0.5 Vaance of the Resduals 0.6 Summay 0.7 Solutons /

More information

An Approach to Inverse Fuzzy Arithmetic

An Approach to Inverse Fuzzy Arithmetic An Appoach to Invese Fuzzy Athmetc Mchael Hanss Insttute A of Mechancs, Unvesty of Stuttgat Stuttgat, Gemany mhanss@mechaun-stuttgatde Abstact A novel appoach of nvese fuzzy athmetc s ntoduced to successfully

More information

Composite Hypotheses testing

Composite Hypotheses testing Composte ypotheses testng In many hypothess testng problems there are many possble dstrbutons that can occur under each of the hypotheses. The output of the source s a set of parameters (ponts n a parameter

More information

Integral Vector Operations and Related Theorems Applications in Mechanics and E&M

Integral Vector Operations and Related Theorems Applications in Mechanics and E&M Dola Bagayoko (0) Integal Vecto Opeatons and elated Theoems Applcatons n Mechancs and E&M Ι Basc Defnton Please efe to you calculus evewed below. Ι, ΙΙ, andιιι notes and textbooks fo detals on the concepts

More information

Physics 11b Lecture #2. Electric Field Electric Flux Gauss s Law

Physics 11b Lecture #2. Electric Field Electric Flux Gauss s Law Physcs 11b Lectue # Electc Feld Electc Flux Gauss s Law What We Dd Last Tme Electc chage = How object esponds to electc foce Comes n postve and negatve flavos Conseved Electc foce Coulomb s Law F Same

More information

Using DP for hierarchical discretization of continuous attributes. Amit Goyal (31 st March 2008)

Using DP for hierarchical discretization of continuous attributes. Amit Goyal (31 st March 2008) Usng DP fo heachcal dscetzaton of contnos attbtes Amt Goyal 31 st Mach 2008 Refeence Chng-Cheng Shen and Yen-Lang Chen. A dynamc-pogammng algothm fo heachcal dscetzaton of contnos attbtes. In Eopean Jonal

More information

24-2: Electric Potential Energy. 24-1: What is physics

24-2: Electric Potential Energy. 24-1: What is physics D. Iyad SAADEDDIN Chapte 4: Electc Potental Electc potental Enegy and Electc potental Calculatng the E-potental fom E-feld fo dffeent chage dstbutons Calculatng the E-feld fom E-potental Potental of a

More information

If there are k binding constraints at x then re-label these constraints so that they are the first k constraints.

If there are k binding constraints at x then re-label these constraints so that they are the first k constraints. Mathematcal Foundatons -1- Constaned Optmzaton Constaned Optmzaton Ma{ f ( ) X} whee X {, h ( ), 1,, m} Necessay condtons fo to be a soluton to ths mamzaton poblem Mathematcally, f ag Ma{ f ( ) X}, then

More information

4 SingularValue Decomposition (SVD)

4 SingularValue Decomposition (SVD) /6/00 Z:\ jeh\self\boo Kannan\Jan-5-00\4 SVD 4 SngulaValue Decomposton (SVD) Chapte 4 Pat SVD he sngula value decomposton of a matx s the factozaton of nto the poduct of thee matces = UDV whee the columns

More information

COMPLEMENTARY ENERGY METHOD FOR CURVED COMPOSITE BEAMS

COMPLEMENTARY ENERGY METHOD FOR CURVED COMPOSITE BEAMS ultscence - XXX. mcocd Intenatonal ultdscplnay Scentfc Confeence Unvesty of skolc Hungay - pl 06 ISBN 978-963-358-3- COPLEENTRY ENERGY ETHOD FOR CURVED COPOSITE BES Ákos József Lengyel István Ecsed ssstant

More information

The M 2 -tree: Processing Complex Multi-Feature Queries with Just One Index

The M 2 -tree: Processing Complex Multi-Feature Queries with Just One Index The M -tee: Pocessng Complex Mult-Featue Quees wth Just ne Index Paolo Cacca, Maco Patella DEIS - CSITE-CNR, Unvesty of Bologna, Italy fpcacca,mpatellag@des.unbo.t Abstact Motvated by the needs fo effcent

More information

Detection and Estimation Theory

Detection and Estimation Theory ESE 54 Detecton and Etmaton Theoy Joeph A. O Sullvan Samuel C. Sach Pofeo Electonc Sytem and Sgnal Reeach Laboatoy Electcal and Sytem Engneeng Wahngton Unvety 411 Jolley Hall 314-935-4173 (Lnda anwe) jao@wutl.edu

More information

Bayesian Assessment of Availabilities and Unavailabilities of Multistate Monotone Systems

Bayesian Assessment of Availabilities and Unavailabilities of Multistate Monotone Systems Dept. of Math. Unvesty of Oslo Statstcal Reseach Repot No 3 ISSN 0806 3842 June 2010 Bayesan Assessment of Avalabltes and Unavalabltes of Multstate Monotone Systems Bent Natvg Jøund Gåsemy Tond Retan June

More information

Energy in Closed Systems

Energy in Closed Systems Enegy n Closed Systems Anamta Palt palt.anamta@gmal.com Abstact The wtng ndcates a beakdown of the classcal laws. We consde consevaton of enegy wth a many body system n elaton to the nvese squae law and

More information

A NOTE ON ELASTICITY ESTIMATION OF CENSORED DEMAND

A NOTE ON ELASTICITY ESTIMATION OF CENSORED DEMAND Octobe 003 B 003-09 A NOT ON ASTICITY STIATION OF CNSOD DAND Dansheng Dong an Hay. Kase Conell nvesty Depatment of Apple conomcs an anagement College of Agcultue an fe Scences Conell nvesty Ithaca New

More information

4 Recursive Linear Predictor

4 Recursive Linear Predictor 4 Recusve Lnea Pedcto The man objectve of ths chapte s to desgn a lnea pedcto wthout havng a po knowledge about the coelaton popetes of the nput sgnal. In the conventonal lnea pedcto the known coelaton

More information

State Estimation. Ali Abur Northeastern University, USA. Nov. 01, 2017 Fall 2017 CURENT Course Lecture Notes

State Estimation. Ali Abur Northeastern University, USA. Nov. 01, 2017 Fall 2017 CURENT Course Lecture Notes State Estmaton Al Abu Notheasten Unvesty, USA Nov. 0, 07 Fall 07 CURENT Couse Lectue Notes Opeatng States of a Powe System Al Abu NORMAL STATE SECURE o INSECURE RESTORATIVE STATE EMERGENCY STATE PARTIAL

More information

Optimal System for Warm Standby Components in the Presence of Standby Switching Failures, Two Types of Failures and General Repair Time

Optimal System for Warm Standby Components in the Presence of Standby Switching Failures, Two Types of Failures and General Repair Time Intenatonal Jounal of ompute Applcatons (5 ) Volume 44 No, Apl Optmal System fo Wam Standby omponents n the esence of Standby Swtchng Falues, Two Types of Falues and Geneal Repa Tme Mohamed Salah EL-Shebeny

More information

ON THE FRESNEL SINE INTEGRAL AND THE CONVOLUTION

ON THE FRESNEL SINE INTEGRAL AND THE CONVOLUTION IJMMS 3:37, 37 333 PII. S16117131151 http://jmms.hndaw.com Hndaw Publshng Cop. ON THE FRESNEL SINE INTEGRAL AND THE CONVOLUTION ADEM KILIÇMAN Receved 19 Novembe and n evsed fom 7 Mach 3 The Fesnel sne

More information

Some Approximate Analytical Steady-State Solutions for Cylindrical Fin

Some Approximate Analytical Steady-State Solutions for Cylindrical Fin Some Appoxmate Analytcal Steady-State Solutons fo Cylndcal Fn ANITA BRUVERE ANDRIS BUIIS Insttute of Mathematcs and Compute Scence Unvesty of Latva Rana ulv 9 Rga LV459 LATVIA Astact: - In ths pape we

More information

Generating Functions, Weighted and Non-Weighted Sums for Powers of Second-Order Recurrence Sequences

Generating Functions, Weighted and Non-Weighted Sums for Powers of Second-Order Recurrence Sequences Geneatng Functons, Weghted and Non-Weghted Sums fo Powes of Second-Ode Recuence Sequences Pantelmon Stăncă Aubun Unvesty Montgomey, Depatment of Mathematcs Montgomey, AL 3614-403, USA e-mal: stanca@studel.aum.edu

More information

Chapter 23: Electric Potential

Chapter 23: Electric Potential Chapte 23: Electc Potental Electc Potental Enegy It tuns out (won t show ths) that the tostatc foce, qq 1 2 F ˆ = k, s consevatve. 2 Recall, fo any consevatve foce, t s always possble to wte the wok done

More information

Optimization Methods: Linear Programming- Revised Simplex Method. Module 3 Lecture Notes 5. Revised Simplex Method, Duality and Sensitivity analysis

Optimization Methods: Linear Programming- Revised Simplex Method. Module 3 Lecture Notes 5. Revised Simplex Method, Duality and Sensitivity analysis Optmzaton Meods: Lnea Pogammng- Revsed Smple Meod Module Lectue Notes Revsed Smple Meod, Dualty and Senstvty analyss Intoducton In e pevous class, e smple meod was dscussed whee e smple tableau at each

More information

an application to HRQoL

an application to HRQoL AlmaMate Studoum Unvesty of Bologna A flexle IRT Model fo health questonnae: an applcaton to HRQoL Seena Boccol Gula Cavn Depatment of Statstcal Scence, Unvesty of Bologna 9 th Intenatonal Confeence on

More information

Machine Learning 4771

Machine Learning 4771 Machne Leanng 4771 Instucto: Tony Jebaa Topc 6 Revew: Suppot Vecto Machnes Pmal & Dual Soluton Non-sepaable SVMs Kenels SVM Demo Revew: SVM Suppot vecto machnes ae (n the smplest case) lnea classfes that

More information

Amplifier Constant Gain and Noise

Amplifier Constant Gain and Noise Amplfe Constant Gan and ose by Manfed Thumm and Wene Wesbeck Foschungszentum Kalsuhe n de Helmholtz - Gemenschaft Unvestät Kalsuhe (TH) Reseach Unvesty founded 85 Ccles of Constant Gan (I) If s taken to

More information

SOME NEW SELF-DUAL [96, 48, 16] CODES WITH AN AUTOMORPHISM OF ORDER 15. KEYWORDS: automorphisms, construction, self-dual codes

SOME NEW SELF-DUAL [96, 48, 16] CODES WITH AN AUTOMORPHISM OF ORDER 15. KEYWORDS: automorphisms, construction, self-dual codes Факултет по математика и информатика, том ХVІ С, 014 SOME NEW SELF-DUAL [96, 48, 16] CODES WITH AN AUTOMORPHISM OF ORDER 15 NIKOLAY I. YANKOV ABSTRACT: A new method fo constuctng bnay self-dual codes wth

More information

Efficiency of the principal component Liu-type estimator in logistic

Efficiency of the principal component Liu-type estimator in logistic Effcency of the pncpal component Lu-type estmato n logstc egesson model Jbo Wu and Yasn Asa 2 School of Mathematcs and Fnance, Chongqng Unvesty of Ats and Scences, Chongqng, Chna 2 Depatment of Mathematcs-Compute

More information

On Maneuvering Target Tracking with Online Observed Colored Glint Noise Parameter Estimation

On Maneuvering Target Tracking with Online Observed Colored Glint Noise Parameter Estimation Wold Academy of Scence, Engneeng and Technology 6 7 On Maneuveng Taget Tacng wth Onlne Obseved Coloed Glnt Nose Paamete Estmaton M. A. Masnad-Sha, and S. A. Banan Abstact In ths pape a compehensve algothm

More information

PARAMETER ESTIMATION FOR TWO WEIBULL POPULATIONS UNDER JOINT TYPE II CENSORED SCHEME

PARAMETER ESTIMATION FOR TWO WEIBULL POPULATIONS UNDER JOINT TYPE II CENSORED SCHEME Sept 04 Vol 5 No 04 Intenatonal Jounal of Engneeng Appled Scences 0-04 EAAS & ARF All ghts eseed wwweaas-ounalog ISSN305-869 PARAMETER ESTIMATION FOR TWO WEIBULL POPULATIONS UNDER JOINT TYPE II CENSORED

More information

Physics Exam 3

Physics Exam 3 Physcs 114 1 Exam 3 The numbe of ponts fo each secton s noted n backets, []. Choose a total of 35 ponts that wll be gaded that s you may dop (not answe) a total of 5 ponts. Clealy mak on the cove of you

More information

THE REGRESSION MODEL OF TRANSMISSION LINE ICING BASED ON NEURAL NETWORKS

THE REGRESSION MODEL OF TRANSMISSION LINE ICING BASED ON NEURAL NETWORKS The 4th Intenatonal Wokshop on Atmosphec Icng of Stuctues, Chongqng, Chna, May 8 - May 3, 20 THE REGRESSION MODEL OF TRANSMISSION LINE ICING BASED ON NEURAL NETWORKS Sun Muxa, Da Dong*, Hao Yanpeng, Huang

More information

Scalars and Vectors Scalar

Scalars and Vectors Scalar Scalas and ectos Scala A phscal quantt that s completel chaacteed b a eal numbe (o b ts numecal value) s called a scala. In othe wods a scala possesses onl a magntude. Mass denst volume tempeatue tme eneg

More information

Monte Carlo comparison of back-propagation, conjugate-gradient, and finite-difference training algorithms for multilayer perceptrons

Monte Carlo comparison of back-propagation, conjugate-gradient, and finite-difference training algorithms for multilayer perceptrons Rocheste Insttute of Technology RIT Schola Woks Theses Thess/Dssetaton Collectons 20 Monte Calo compason of back-popagaton, conugate-gadent, and fnte-dffeence tanng algothms fo multlaye peceptons Stephen

More information

19 The Born-Oppenheimer Approximation

19 The Born-Oppenheimer Approximation 9 The Bon-Oppenheme Appoxmaton The full nonelatvstc Hamltonan fo a molecule s gven by (n a.u.) Ĥ = A M A A A, Z A + A + >j j (883) Lets ewte the Hamltonan to emphasze the goal as Ĥ = + A A A, >j j M A

More information

Closed-loop adaptive optics using a CMOS image quality metric sensor

Closed-loop adaptive optics using a CMOS image quality metric sensor Closed-loop adaptve optcs usng a CMOS mage qualty metc senso Chueh Tng, Mchael Gles, Adtya Rayankula, and Pual Futh Klpsch School of Electcal and Compute Engneeng ew Mexco State Unvesty Las Cuces, ew Mexco

More information

SURVEY OF APPROXIMATION ALGORITHMS FOR SET COVER PROBLEM. Himanshu Shekhar Dutta. Thesis Prepared for the Degree of MASTER OF SCIENCE

SURVEY OF APPROXIMATION ALGORITHMS FOR SET COVER PROBLEM. Himanshu Shekhar Dutta. Thesis Prepared for the Degree of MASTER OF SCIENCE SURVEY OF APPROXIMATION ALGORITHMS FOR SET COVER PROBLEM Hmanshu Shekha Dutta Thess Pepaed fo the Degee of MASTER OF SCIENCE UNIVERSITY OF NORTH TEXAS Decembe 2009 APPROVED: Fahad Shahokh, Mao Pofesso

More information

A STUDY OF SOME METHODS FOR FINDING SMALL ZEROS OF POLYNOMIAL CONGRUENCES APPLIED TO RSA

A STUDY OF SOME METHODS FOR FINDING SMALL ZEROS OF POLYNOMIAL CONGRUENCES APPLIED TO RSA Jounal of Mathematcal Scences: Advances and Applcatons Volume 38, 06, Pages -48 Avalable at http://scentfcadvances.co.n DOI: http://dx.do.og/0.864/jmsaa_700630 A STUDY OF SOME METHODS FOR FINDING SMALL

More information

V. Principles of Irreversible Thermodynamics. s = S - S 0 (7.3) s = = - g i, k. "Flux": = da i. "Force": = -Â g a ik k = X i. Â J i X i (7.

V. Principles of Irreversible Thermodynamics. s = S - S 0 (7.3) s = = - g i, k. Flux: = da i. Force: = -Â g a ik k = X i. Â J i X i (7. Themodynamcs and Knetcs of Solds 71 V. Pncples of Ievesble Themodynamcs 5. Onsage s Teatment s = S - S 0 = s( a 1, a 2,...) a n = A g - A n (7.6) Equlbum themodynamcs detemnes the paametes of an equlbum

More information

OPTIMISED PERMUTATION FILTER

OPTIMISED PERMUTATION FILTER 50 Acta Electotechnca et Infomatca o, Vol, 200 OPTIMISED PERMUTATIO FILTER * Rastslav LUKÁČ, ** Ján LIZÁK * Depatment of Electoncs and Multmeda Communcatons, Techncal Unvesty of Košce, Pak Komenského 3,

More information

Test 1 phy What mass of a material with density ρ is required to make a hollow spherical shell having inner radius r i and outer radius r o?

Test 1 phy What mass of a material with density ρ is required to make a hollow spherical shell having inner radius r i and outer radius r o? Test 1 phy 0 1. a) What s the pupose of measuement? b) Wte all fou condtons, whch must be satsfed by a scala poduct. (Use dffeent symbols to dstngush opeatons on ectos fom opeatons on numbes.) c) What

More information

Order Reduction of Continuous LTI Systems using Harmony Search Optimization with Retention of Dominant Poles

Order Reduction of Continuous LTI Systems using Harmony Search Optimization with Retention of Dominant Poles Ode Reducton of Contnuous LTI Systems usng Hamony Seach Optmzaton wth Retenton of Domnant Poles Ode Reducton of Contnuous LTI Systems usng Hamony Seach Optmzaton wth Retenton of Domnant Poles a Akhlesh

More information

CSJM University Class: B.Sc.-II Sub:Physics Paper-II Title: Electromagnetics Unit-1: Electrostatics Lecture: 1 to 4

CSJM University Class: B.Sc.-II Sub:Physics Paper-II Title: Electromagnetics Unit-1: Electrostatics Lecture: 1 to 4 CSJM Unvesty Class: B.Sc.-II Sub:Physcs Pape-II Ttle: Electomagnetcs Unt-: Electostatcs Lectue: to 4 Electostatcs: It deals the study of behavo of statc o statonay Chages. Electc Chage: It s popety by

More information

Part V: Velocity and Acceleration Analysis of Mechanisms

Part V: Velocity and Acceleration Analysis of Mechanisms Pat V: Velocty an Acceleaton Analyss of Mechansms Ths secton wll evew the most common an cuently pactce methos fo completng the knematcs analyss of mechansms; escbng moton though velocty an acceleaton.

More information

A NOVEL DWELLING TIME DESIGN METHOD FOR LOW PROBABILITY OF INTERCEPT IN A COMPLEX RADAR NETWORK

A NOVEL DWELLING TIME DESIGN METHOD FOR LOW PROBABILITY OF INTERCEPT IN A COMPLEX RADAR NETWORK Z. Zhang et al., Int. J. of Desgn & Natue and Ecodynamcs. Vol. 0, No. 4 (205) 30 39 A NOVEL DWELLING TIME DESIGN METHOD FOR LOW PROBABILITY OF INTERCEPT IN A COMPLEX RADAR NETWORK Z. ZHANG,2,3, J. ZHU

More information

VQ widely used in coding speech, image, and video

VQ widely used in coding speech, image, and video at Scalar quantzers are specal cases of vector quantzers (VQ): they are constraned to look at one sample at a tme (memoryless) VQ does not have such constrant better RD perfomance expected Source codng

More information

Stellar Astrophysics. dt dr. GM r. The current model for treating convection in stellar interiors is called mixing length theory:

Stellar Astrophysics. dt dr. GM r. The current model for treating convection in stellar interiors is called mixing length theory: Stella Astophyscs Ovevew of last lectue: We connected the mean molecula weght to the mass factons X, Y and Z: 1 1 1 = X + Y + μ 1 4 n 1 (1 + 1) = X μ 1 1 A n Z (1 + ) + Y + 4 1+ z A Z We ntoduced the pessue

More information

VISUALIZATION OF THE ABSTRACT THEORIES IN DSP COURSE BASED ON CDIO CONCEPT

VISUALIZATION OF THE ABSTRACT THEORIES IN DSP COURSE BASED ON CDIO CONCEPT VISUALIZATION OF THE ABSTRACT THEORIES IN DSP COURSE BASED ON CDIO CONCEPT Wang L-uan, L Jan, Zhen Xao-qong Chengdu Unvesty of Infomaton Technology ABSTRACT The pape analyzes the chaactestcs of many fomulas

More information

Pattern Analyses (EOF Analysis) Introduction Definition of EOFs Estimation of EOFs Inference Rotated EOFs

Pattern Analyses (EOF Analysis) Introduction Definition of EOFs Estimation of EOFs Inference Rotated EOFs Patten Analyses (EOF Analyss) Intoducton Defnton of EOFs Estmaton of EOFs Infeence Rotated EOFs . Patten Analyses Intoducton: What s t about? Patten analyses ae technques used to dentfy pattens of the

More information

A Novel Approach to Expression Recognition from Non-frontal Face Images

A Novel Approach to Expression Recognition from Non-frontal Face Images A Novel Appoach to Expesson Recognton fom Non-fontal Face Images Wenmng Zheng 1,2, HaoTang 1, Zhouchen Ln 3, Thomas S. Huang 1 1 Beckman Insttute, Unvesty of Illnos at Ubana-Champagn, USA 2 Reseach Cente

More information

PHY126 Summer Session I, 2008

PHY126 Summer Session I, 2008 PHY6 Summe Sesson I, 8 Most of nfomaton s avalable at: http://nngoup.phscs.sunsb.edu/~chak/phy6-8 ncludng the sllabus and lectue sldes. Read sllabus and watch fo mpotant announcements. Homewok assgnment

More information

2E Pattern Recognition Solutions to Introduction to Pattern Recognition, Chapter 2: Bayesian pattern classification

2E Pattern Recognition Solutions to Introduction to Pattern Recognition, Chapter 2: Bayesian pattern classification E395 - Pattern Recognton Solutons to Introducton to Pattern Recognton, Chapter : Bayesan pattern classfcaton Preface Ths document s a soluton manual for selected exercses from Introducton to Pattern Recognton

More information

On the Distribution of the Product and Ratio of Independent Central and Doubly Non-central Generalized Gamma Ratio random variables

On the Distribution of the Product and Ratio of Independent Central and Doubly Non-central Generalized Gamma Ratio random variables On the Dstbuton of the Poduct Rato of Independent Cental Doubly Non-cental Genealzed Gamma Rato om vaables Calos A. Coelho João T. Mexa Abstact Usng a decomposton of the chaactestc functon of the logathm

More information

Announcements. Stereo (Part 3) Summary of Stereo Constraints. Features on same epipolar line. Stereo matching. Truco Fig. 7.5

Announcements. Stereo (Part 3) Summary of Stereo Constraints. Features on same epipolar line. Stereo matching. Truco Fig. 7.5 Announcements Steeo (Pat ) Homewok s due Nov, :59 PM Readng: Chapte 7: Steeopss CSE 5A Lectue 0 Featues on same eppola lne Summay of Steeo Constants CONSRAIN BRIEF DESCRIPION -D Eppola Seach Abtay mages

More information

Lecture Notes on Linear Regression

Lecture Notes on Linear Regression Lecture Notes on Lnear Regresson Feng L fl@sdueducn Shandong Unversty, Chna Lnear Regresson Problem In regresson problem, we am at predct a contnuous target value gven an nput feature vector We assume

More information

q-bernstein polynomials and Bézier curves

q-bernstein polynomials and Bézier curves Jounal of Computatonal and Appled Mathematcs 151 (2003) 1-12 q-bensten polynomals and Béze cuves Hall Ouç a, and Geoge M. Phllps b a Depatment of Mathematcs, Dokuz Eylül Unvesty Fen Edebyat Fakültes, Tınaztepe

More information

A Study about One-Dimensional Steady State. Heat Transfer in Cylindrical and. Spherical Coordinates

A Study about One-Dimensional Steady State. Heat Transfer in Cylindrical and. Spherical Coordinates Appled Mathematcal Scences, Vol. 7, 03, no. 5, 67-633 HIKARI Ltd, www.m-hka.com http://dx.do.og/0.988/ams.03.38448 A Study about One-Dmensonal Steady State Heat ansfe n ylndcal and Sphecal oodnates Lesson

More information

Accurate Evaluation Schemes for Triangular Domain Integrals

Accurate Evaluation Schemes for Triangular Domain Integrals IOSR Jounal of Mechancal and Cvl Engneeng (IOSRJMCE) ISSN : 78-68 Volume, Issue 6 (Sep-Oct 0), PP 38-5 Accuate Evaluaton Schemes fo Tangula Doman Integals Fazana Hussan, M. S. Kam, (Depatment of Mathematc

More information

A Tutorial on Low Density Parity-Check Codes

A Tutorial on Low Density Parity-Check Codes A Tutoal on Low Densty Paty-Check Codes Tuan Ta The Unvesty of Texas at Austn Abstact Low densty paty-check codes ae one of the hottest topcs n codng theoy nowadays. Equpped wth vey fast encodng and decodng

More information

For now, let us focus on a specific model of neurons. These are simplified from reality but can achieve remarkable results.

For now, let us focus on a specific model of neurons. These are simplified from reality but can achieve remarkable results. Neural Networks : Dervaton compled by Alvn Wan from Professor Jtendra Malk s lecture Ths type of computaton s called deep learnng and s the most popular method for many problems, such as computer vson

More information

Goodness-of-fit for composite hypotheses.

Goodness-of-fit for composite hypotheses. Section 11 Goodness-of-fit fo composite hypotheses. Example. Let us conside a Matlab example. Let us geneate 50 obsevations fom N(1, 2): X=nomnd(1,2,50,1); Then, unning a chi-squaed goodness-of-fit test

More information

Contact, information, consultations

Contact, information, consultations ontact, nfomaton, consultatons hemsty A Bldg; oom 07 phone: 058-347-769 cellula: 664 66 97 E-mal: wojtek_c@pg.gda.pl Offce hous: Fday, 9-0 a.m. A quote of the week (o camel of the week): hee s no expedence

More information

Applied Statistical Mechanics Lecture Note - 13 Molecular Dynamics Simulation

Applied Statistical Mechanics Lecture Note - 13 Molecular Dynamics Simulation Appled Statstcal Mechancs Lectue Note - 3 Molecula Dynamcs Smulaton 고려대학교화공생명공학과강정원 Contents I. Basc Molecula Dynamcs Smulaton Method II. Popetes Calculatons n MD III. MD n Othe Ensembles I. Basc MD Smulaton

More information

Optimization Algorithms for System Integration

Optimization Algorithms for System Integration Optmzaton Algothms fo System Integaton Costas Papadmtou 1, a and Evaggelos totsos 1,b 1 Unvesty of hessaly, Depatment of Mechancal and Industal Engneeng, Volos 38334, Geece a costasp@uth.g, b entotso@uth.g

More information

33. 12, or its reciprocal. or its negative.

33. 12, or its reciprocal. or its negative. Page 6 The Point is Measuement In spite of most of what has been said up to this point, we did not undetake this poject with the intent of building bette themometes. The point is to measue the peson. Because

More information

Information Retrieval

Information Retrieval Clusteng Technques fo Infomaton Reteval Beln Chen Depatment t of Compute Scence & Infomaton Engneeng Natonal Tawan Nomal Unvesty Refeences:. Chstophe D. Mannng, Pabhaa Raghavan and Hnch Schütze, Intoducton

More information