Detection and Classification of Interacting Persons

Size: px
Start display at page:

Download "Detection and Classification of Interacting Persons"

Transcription

1 Deecon and Classfcaon of Ineracng Persons Sco Blunsden (1), Rober Fsher (2) (1) European Commsson Jon Research Cenre, Ialy (2) Unversy of Ednburgh, Scoland ABSTRACT Ths chaper presens a way o classfy neracons beween people. Examples of he neracons we nvesgae are; people meeng one anoher, walkng ogeher and fghng. A new feaure se s proposed along wh a correspondng classfcaon mehod. Resuls are presened whch show he new mehod performng sgnfcanly beer han he prevous sae of he ar mehod as proposed by [Olver e al., 2000]. INTRODUCTION Ths chaper presens an nvesgaon no classfcaon of mulple person neracons. There has been much prevous work upon denfyng wha acvy ndvdual people are engaged n. [Davs and Bobck, 2001] used a momen based represenaon based on exraced slhouees and [Efros e al., 2003] modeled human acvy by generang opcal flow descrpons of a person s acon. Descrpons were generaed by frs hand-rackng an ndvdual, re-scalng o a sandard sze and hen akng he opcal flow of a persons acons over several frames. A daabase of hese descrpons was creaed and mached o novel suaons. Ths mehod was exended by [Roberson, 2006] who also ncluded locaon nformaon o help gve conexual nformaon o a scene. Locaon nformaon s of asssance when ryng o deermne f someone s loerng or merely wang a a road crossng. Followng on from flow based feaures [Dollar e al., 2005] exraced spaal-emporal feaures o denfy sequences of acons. Rbero and Sanos-Vcor [Rbero and Sanos-Vcor, 2005] ook a dfferen approach o classfy an ndvdual s acons n ha hey used mulple feaures calculaed from rackng (such as speed, egenvecors of flow) and seleced hose feaures whch bes classfed he persons acons usng a classfcaon ree wh each branch usng a mos 3 feaures o classfy he example. The classfcaon of neracng ndvduals was suded by [Olver e al., 2000] who used rackng o exrac he speed, algnmen and dervave of he dsance beween wo ndvduals. Ths nformaon was hen used o classfy sequences usng a coupled hdden Markov model (CHMM). [Lu and Chua, 2006] expanded he wo person classfcaon o hree person sequences usng a hdden Markov model (HMM) wh an explc role arbue. Informaon derved from rackng was used o provde feaures such as he relave angle beween wo persons o classfy complee sequences. Xang and Gong [Gong and Xang, 2003] agan used a CHMM o model

2 neracons beween vehcles on an arcraf runway. These feaures are calculaed by deecng sgnfcanly changed pxels over several frames. The correc model for represenng he sequence s deermned by he connecons beween he separae models. Goodness of f s calculaed by he Bayesan nformaon creron. Usng hs mehod a model represenng he sequences acons s deermned. Mul-person neracons whn a rgd formaon was also he goal of [Khan and Shah, 2005] who used a geomerc model o deec rgd formaons beween people, such an example would be a marchng band. [Inlle and Bobck, 2001] used a pre-defned Bayesan nework o descrbe planned moons durng Amercan fooball games. Ohers such as [Perse e al., 2007] also use a pre-specfed emplae o evaluae he curren acon beng performed by many ndvduals. Prespecfed emplaes have been used by [Van Vu e al., 2003, Hongeng and Nevaa, 2001] whn he conex of survellance applcaons. SPECIFICALLY WHAT ARE WE TRYING TO DO? Gven an npu vdeo sequence he goal s o auomacally deermne f any neracons beween wo people are akng place. If any are akng place hen we wan o denfy he class of he neracon. Here we lm ourselves o pre-defned classes whch have been prevously labeled. To make he suaon more realsc here s also a no neracon class. We seek o gve each frame a label from a predefned se. For example a label may be ha person 1 and person 2 are walkng ogeher n frame 56. The ably o auomacally classfy such neracons would be useful n cases whch are ypcal of many survellance suaons. Such an ably o auomacally recognze neracons would also be useful n vdeo summarzaon where could be possble o focus only on specfc neracons. FEATURES AND VARIABLES Vdeo daa s rch n nformaon wh he resoluon of modern survellance cameras capable of delverng megapxel resoluon a a susaned frame rae of greaer han 10fps. Such daa s overwhelmng and mosly unnecessary for classfcaon of neracons. As a frs sep, rackng of he ndvduals s employed. There s a rch body of work upon rackng of people and obecs whn he leraure. See he revew by Ylmaz e al. for a good survey of curren rackng echnology [Ylmaz, A e al. 2006]. For all expermens performed n hs chaper he boundng box mehod of denfyng a person was used. The posonal nformaon was calculaed based upon he cenre of hs box. Ths process s llusraed n Fgure 1. Such rackng nformaon s ypcal of he oupu of many rackng procedures and wll be assumed ha such a racker s avalable hroughou all expermens carred ou n hs chaper.

3 Fgure 1 - Boundng box rackng. Colored lnes show he prevous poson of he cenre of he racked obec. Here hree ypes of feaures are used as npu o a classfer. We make use of movemen, algnmen and dsance based feaures. More deals are gven n he followng secons. Movemen Based feaures Movemen plays an mporan role n recognzng neracons. The speed of an ndvdual s calculaed as shown n equaon (1.1). The double vercal bar (.. ) represens a vecor L2 norm as gven by x = sqr( x1+ x x n ) where x n refers o he nh componen of vecor x. s w = p p (1.1) w Here p refers o he poson of he racked obec a me for obec. Whn hs work only he wo dmensonal ( p, = x y ) case s consdered due o rackng nformaon only beng avalable n wo dmensons. The emporal offse w s nroduced due o he hgh raes whch ypfy many modern vdeo cameras. Wh frame raes of around 25 frames per second hs can mean ha dfferences beween he curren and las frame (w=1) are very small and may be domnaed by nose. The absolue dfference n speed ( ε ) beween wo racks s also calculaed ( s s ). The vorcy ( v ) s measured as a devaon from a lne. Ths lne s calculaed by fng a lne o a w se of prevous posons of he raecory P = p,.., p. The wndow sze w s he same as ha used n equaon (1.1). A each pon he orhogonal dsance o he lne s found. The oal dsance of all pons are hen summed and normalzed by wndow lengh and so gve a measure of he vorcy.

4 Algnmen Based feaures The algnmen of wo racks can gve valuable nformaon as o how hey are neracng. The degree of algnmen s common o [Ggerenzer e al., 1999; Olver e al., 2000; and Lu and Chua, 2006] who all make use of such nformaon when classfyng raecory nformaon. To calculae he do produc he headng (h) of he obec s aken as n equaon (1.2) and he do produc (1.3) s calculaed from he drecons of racks and. hˆ p p w = w p p a ˆ ˆ = (1.2) h h (1.3) In addon o he algnmen beween wo people he poenal nersecon ( ) of wo γ Traecores s also calculaed. Such feaures are suggesed n [Ggerenzer e al., 1999] and [Lu and Chua, 2006]. We frs es for an nersecon of he headngs. Ths s acheved as shown n Algorhm 1. d = p p c f c δ l = h h c elsef c f / /dsance beween arges and a me / /cross produc of he headngs of and a me 0 / /non zero cross produc d = h = p + δ h / /hey nesec a hs pon = 0 / /parallel bu may be n oppose drecon d > 0 and h = h and h h > 0 = h + h r δ l d = h r c = p + δ h / /hey nesec a hs pon else hey do no mee end Algorhm 1- Algorhm o deermne he meeng of wo raecores

5 Dsance Based feaures Dsance s a good measure for many ypes of neracon, for example meeng s no possble whou beng n close physcal proxmy. Frs he Eucldean dsance s aken and s used as gven n equaon (1.4) below. d = p p (1.4) The dervave of he dsance was also calculaed. Ths s he dfference n dsance a conguous me seps. I s calculaed as shown n equaon (1.5) below. dˆ d w (1.5) = ( w) An nsananeous measure such as he dsance and he dervave of he dsance can boh be prone o shor erm rackng errors. In an effor o remove hs effec a wndow sze conanng w pons was averaged (as n P n he prevous secon). The dsance was calculaed for every pon (as n equaon (1.4)) n hs wndow. Fnal feaure vecor The fnal feaure vecor for each par of people s gven n equaon (1.6) below: ˆ ˆ ˆ r = s s, s s, ε, a, d, d, v v, γ (1.6) The vecor beween persons and a me s made up of he speed of each person ( s s ) and he change n speed ˆ, ˆ. The algnmen, dsance and change n dsance a a parcular s s pon n me s gven by and a, d ˆ. The vorcy a a parcular pon n me for each d person s gven by whls he possbly of an nersecon beween wo raecores s gven v v by. Ths gves a fnal feaure vecor conanng 11 elemens. A furher processng sep of γ, normalzng he ranng daa o have zero mean and un varance was also aken.

6 Observaon Wndow Sze Throughou hese expermens we nvesgaed he role of varyng he number of vdeo frames used before makng a decson as o wha s happenng whn he frame. Fgure 2 (below) shows how hs s acheved. Throughou hs work we used nformaon from before and afer he curren frame n order o classfy. Typcally he frames relae o a small quany of me (1-2 seconds) and help wh he lag problem when your decson s based by he large amoun of prevous nformaon used n he classfcaon. The wndow sze varaon hroughou hs work s equvalen o a few seconds delay. Ths was no foreseen as a problem f such an approach was aken n a real survellance applcaon. The fac ha here would be a lag n classfcaon f makng use of only prevous nformaon seems an approprae rade-off for an ncrease n accuracy. Fgure 2 - The frame o classfy () uses nformaon from w frames around he curren frame n order o classfy he frame. CLASSIFICATION Ths secon nroduces he classfers whch are used hroughou subsequen expermens. We make use of a smple lnear dscrmnan classfer (LDA) whch s non-probablsc and provdes a baselne for performance. Ths s nroduced n he nex secon. We hen brefly nroduce he hdden Markov model (HMM) whch s wdely used hroughou he leraure. We hen nroduce a newer model, he condonal random feld (CRF). Fnally he prevous bes mehod as suggesed by [Olver, 2000] s brefly revewed.

7 Lnear Dscrmnan Classfer Lnear dscrmnan analyss (LDA) seeks o maxmze he obecve funcon gven n equaon (1.7) below. Here SW s he whn class scaer marx wh S B beng he beween class scaer marx. T B J ( w) = wsw T ws w (1.7) The obecve funcon (equaon (1.7)) s ofen referred o as he sgnal o nose rao. Wha we are ryng o acheve s a proecon (w) whch maxmzes he dsance of he class means relave o he (sum of) varances of a parcular class. To generae a soluon s noed ha equaon (1.7) has a propery whereby s nvaran wh respec o scalng of he w vecors (eg T w αwwhere α s some arbrary scalng). Therefore w can be chosen such ha wsw w = 1. I s also common (and ndeed he case here) o perform a whenng sep (zero mean and un varance) on he daa pror o npu no hs mehod. Ths maxmzaon can be urned no a regular egenvalue problem. Proecons are found for each class (e one class vs all ohers) and he mean and varance (C) of he class proecon are found. In order o classfy a novel pon he new pon s proeced wh S. The class label s deermned by akng he smalles Mahalonobs dsance beween he calculaed class model s mean and varance and he new es pon. Hdden Markov Model Hdden Markov models (HMM s) have been nroduced by (among ohers) Rabner [Rabner,1990]. The model s parameerzed by he pror dsrbuon wh each elemen π represenng π = p( x= ) across all hdden saes [1,.., N]. The saonary sae ranson marx A s used o represen he probably of a ranson from one sae () o anoher () hrough me. An enry whn he sochasc marx A s referenced by a = p( x= x 1 = ). Whn hs work we are concerned wh connuous real valued npus ( r ) whch can be accommodaed whn he model by usng a Gaussan mxure model o represen he npu dsrbuon p( r x = ). W 1 W M b ( r) = c N( r, μ, C )(1.8), m, m, m m= 1 Here he observed daa R s he vecor beng modeled, c, m mh mxure n sae. N s a Gaussan dsrbuon wh mean vecor for he mh mxure componen n sae. The mxure coeffcen s he mxure coeffcen for he c μ, m and covarance mus sum o 1. C, m

8 The hdden Markov model s parameers can hus be represened as λ = ( Π, A, Θ) where Θ represen he parameers of he mxure model. Condonal Random Feld In hs secon he workngs of a condonal random feld (CRF) are explaned and hen he specfc formulaon as appled n hs chaper s gven. The srucure of he CRF we use s shown n Fgure 3. The CRF can be confgured o resemble HMM lke models. However hey can be more expressve n ha arbrary dependences on he observaons are allowed. Usng he feaure funcons of he CRF allows any par of he npu sequence o be used a any me durng he nference sage. I s also possble ha dfferen sae s (classes) can have dfferng feaure funcons (hough we do no make use of hs here). The feaure funcons descrbe he srucure of he model. The CRF s also a dscrmnave model where as he HMM s a generave one. A poenal advanage of he dscrmnave CRF over generave models s ha hey have been shown o be more robus o volaons of ndependence assumpons [Laffery e al., 2001]. Fgure 3- CRF model. The observaons (r) are shown for each mesep. The class label x s also shown. The dscree emporal sae a a parcular mesep s gven by x whch akes a value from he se of all possble class labels x X = {1,2,.., C}. Here C s he maxmum number of class labels whls represens he me wh T beng he maxmum lengh of he sequence. Observaons a me are denoed as r wh he on observaons gven as R = ( r1,.., r ). Lkewse he on sae s gven by X = ( x1,.., x). For noaonal compacness we shall refer o X as X and R as R n accordance wh oher auhor s [Smnchsescu e al.,2005, and Wallach, 2004]. The dsrbuon over on labels X gven observaons R and parameers Θ are gven by: 1 c pθ( X R) = φθ( Xc, Rc) Z ( R) θ c C( XR, ) (1.9)

9 Whn hs equaon φ s a real valued poenal funcon of he clque c and ( R ) s he c θ observaon dependen normalzaon (somemes referred o as a paron funcon). C(X,R) s he se of maxmal clques. Here a frs order lnear chan s used (as show n Fgure 3). The clques are pars f neghborng saes ( x, x + 1), whls he connecvy among observaons s resrced o ha shown n he graph (Fgure 3- CRF model. The observaons (r) are shown for each mesep. The class label x s also shown.). A CRF model wh T meseps, as used here, can be re-wren n erms of exponenal feaure funcons F θ compued n erms of weghed sums over he feaures of he clques. Ths exponenal feaure formulaon s gven below n equaon (1.10) Z θ T 1 pθ( X R) = exp( Fθ( x, x 1, R)) Z ( R) θ = 1 (1.10) The condonal log lkelhood of he CRF s gven below. Assumng ha he ranng daa s X d d, R he parameers of he model are obaned by opmzaon of he fully labeled { } 1,.., followng funcon: d= D D D T L = log p ( X R ) = ( F ( x, x, R ) log Z ( R )) (1.11) d d d d d d θ θ θ 1 θ d= 1 d= 1 = In order o make parameer opmzaon more sable he problem ofen makes use of a regularzed erm ( R θ ). The problem hen becomes one of opmzng he penalzed lkelhood 2 ( Lθ + Rθ ). The regularzng erm used here was chosen o be R θ = θ. Once raned novel npu s gven o a CRF model and a probably dsrbuon s gven hroughou all meseps for all classes. In hs case we choose he hghes probably as beng he classfcaon label of he new example. A Gaussan pror over he npu daa was used hroughou all expermens. Olver s Coupled Hdden Markov Model Here we brefly cover Olver s mehod ([Olver e al., 2000]) of classfyng neracng ndvduals. Ths work s revewed as provdes a sae of he ar mehod o compare our resuls wh. Olver used coupled hdden Markov models o model fve dfferen neracve behavors. Olver s work s used for comparson wh he work presened here. Olver e al use wo feaure vecors (one for each chan). These feaure vecors are made up of he velocy of he person, he change n dsance beween he wo people and he algnmen beween he wo people. Ths gves wo feaure vecors, one for each chan. For each class he parameers of a wo chan coupled hdden Markov model are raned. When classfyng he model whch produces he hghes lkelhood for a es sequence s aken as he class label.

10 RESULTS Ths secon presens he resuls obaned by usng he mehods descrbed n he precedng chapers. Resuls usng a condonal random feld (CRF), hdden Markov model (HMM) and s coupled varaon (CHMM) are presened. Resuls usng a lnear dscrmnan model (LDA) are also presened and used as a baselne non-probablsc mehod o whch resuls are compared. We presen he resul of classfcaon over many ranng and esng subses o gve an ndcaon of he sandard devaon and he expeced performance when usng a mehod. Resuls are presened over many dfferen wndow szes. The graphs show he averaged performance of he classfer over 50 runs. The sandard devaon s gven by he shaded regons. Expermenal seup The CAVIAR daase has been prevously used n [Dee and Hogg, 2004a, Wu and Nevaa, 2007] however here s no a unversally agreed ranng and esng se. Therefore was deemed ha n order o characerse an algorhm s rue performance upon a daase should be esed wh dfferen subses of he enre daa. Ths wll gve an ndcaon of he expeced performance of he algorhm raher han fndng a parcularly good (n erms of classfcaon accuracy) subse. We are neresed n comparng he four mehods as descrbed n he prevous secon. Furhermore he role of me s nvesgaed. We seek o nvesgae wha s he opmal lengh of me a sequence should be wached before a decson s made. Resuls comparng each mehod and he effecs of me are gven n he followng secons. Throughou he ranng procedure he esng se was kep separae from he ranng daa and was unseen by he learnng algorhm unl esng me. Therefore only he ranng se was used when deermnng he parameers of he learnng model. Tranng and esng ses were spl 50/50 on a per class bass. Paronng was done per-class raher han over he whole daase due o he uneven dsrbuon of classes. Such a sep means ha n he ranng sage he learnng algorhm wll have examples of every ype of class. We would no expec a correc classfcaon on unseen classes and so hs measure can sop msleadng resuls. In order o show he average performance hs procedure was repeaed over 50 dfferen parons of he ranng and es daa. The daase conans examples of complee sequences, for example a sequence conssng of wo people walkng ogeher may be hundreds of frames long. Our goal s o classfy each frame correcly. If we were o ake hs sequence and spl up as ranng and esng frames hen he classfcaon ask would be much smpler as ranng and esng pons would be smply a maer of nerpolaon beween hghly smlar pons. I s for hs reason ha when decdng on he ranng and esng daa we paron based on he complee sequences. Ths means ha an enre walkng ogeher sequence wll be assgned o he ranng se whls anoher complee walkng ogeher sequence wll be assgned o he esng se. Ths should avod he pfall of havng ranng and esng daa whch s essenally he same. Ths s especally rue as he daa has a very srong emporal couplng.

11 Wha we are amng o do s ry o gve a class label o wo neracng ndvduals. Ths s shown n Fgure 4-. The wo people are approachng one anoher. Beween he frs and second frame he dsance beween hem decreases. The classfer would assgn one label for hs neracon (coverng boh people) o ndcae approachng. Classfcaon Resuls Fgure 4-Two people approachng one anoher. CAVIAR Daase The CAVIAR daase [EC Funded CAVIAR.. (2004)] conans 11,415 frames n whch some have labeled examples of neracons. Whn hs se here are 5 dsnc classes whch we seek o denfy and classfy. The 5 classes conss of examples of people: walkng ogeher (2,543), approachng (942), gnorng (4,916), meeng one anoher (1,801), splng up (879) and fghng (334). The numbers n brackes ndcaes how many frames conan hs behavor. Overall resuls for he daase are shown n Fgure 5. These resuls are he furher broken down and shown for each class n Fgure 6, Fgure 7 and Fgure 8.

12 Fgure 5- Overall resuls on he CAVIAR daase for each mehod. Lnes show averaged resuls (over 50 runs) whls he shaded regons show one sandard devaon. I s vsble ha for small wndow szes he CRF mehod offers he bes performance (excep perhaps when classfyng he approachng behavor). However n hs daase he HMM model gves he bes possble average performance. Boh he HMM and he CRF usng he new proposed feaures show superor performance o Olver s mehod. A parcular problem for all mehods s n he classfcaon of fghng behavor. A small sample sze and he shor mescale where any fghng acually occurs conrbue owards hs. We see ha for some cases he wndow s oo small (such as walkng ogeher and gnore) or oo large (such as gnore and approach). A per class me wndow or an enhanced feaure se would help hs problem

13 Fgure 6 Resuls on he approachng and fghng class, for he CAVIAR daa. Fgure 7 Resuls on he gnore and mee class, for he CAVIAR daa.

14 Fgure 8 Resuls on he spl and walk ogeher class, for he CAVIAR daa One of he oher feaures of hs daase s ha for approachng, splng and fghng here were perhaps no enough examples o ge a buld a suffcenly generalsable model. A smple answer would be o say ha more daa s requred. However n many real survellance applcaons such an approach s no possble so showng how a mehod performs usng only lmed daa s sll of value. BEHAVE Daase The BEHAVE daase [Blunsden e al. 2007] conans 134,457 frames whch have labeled examples of neracons. Whn hs se here are 5 dsnc classes whch we seek o denfy and classfy. The 8 classes conss of examples of people: In a group (91,421), Approachng (8,991), walkng ogeher (14,261), splng (11,046), gnorng (1,557), fghng (4,772), runnng (1,870) and chasng (539) one anoher. The numbers n brackes ndcaes how many frames conan hs behavor.

15 Fgure 9- Overall performance on he BEHAVE daase for each mehod. Lnes show averaged resuls (over 50 runs) whls he shaded regons show one sandard devaon. The overall averaged classfcaon s shown n Fgure 9. The CRF clearly performs beer han all oher mehods on hs daase for all wndow szes. There s a slgh ncrease n performance when he wndow sze s ncreased when usng a CRF. However he effec s more dramac for boh he CHMM, HMM and he LDA mehod. All hree of hese mehods (HMM, CHMM, LDA) ncrease n performance as he wndow sze s ncreased. Sgnfcan performance ncreases are observed beween wndow szes of 1 and 20. Per class resuls (fgures Fgure 10, Fgure 11 and Fgure 12) dsplay a smlar sory where ncreasng he wndow sze has lle effec upon CRF classfcaon. When classfyng splng, approachng (Fgure Fgure 10) and fghng (Fgure 11) ncreasng he wndow sze mproves he performance of he HMM, CHMM and he LDA classfers. The HMM classfer gves he bes performance of all mehods when classfyng fghng (Fgure 6). Increasng wndow sze also gves a smlar ncrease n performance for boh he n group classfcaon and he walkng ogeher class (Fgure 12). However when classfyng people n a group he LDA mehod decreases n performance.

16 Fgure 10-Resuls on he spl and approach classes for he BEHAVE daase. Fgure 11 - Resuls on he fgh and gnore classes for he BEHAVE daase.

17 Fgure 12- Resuls on he walk ogeher and n group classes for he BEHAVE daase. CONCLUSION Over all daases he CRF classfer performs well (80-95%) when usng lmed nformaon. The prevous bes mehod suggesed by Olver [Olver, 2000] s mproved upon usng he CRF classfer n conuncon wh he new proposed feaure se. The new proposed feaure se also ouperforms Olver s mehod when usng a HMM model for a grea many cases. The CRF classfer dsplays an ably o beer classfy daa n he shor erm compared o he HMM. In conras he HMM model mproves more rapdly when he observaon wndow sze s ncreased suggesng s beer a smoohng he sgnal over longer sequences. The forward algorhm used o deermne he lkelhood seems o smooh he sgnal much beer han he CRF. The case of he CHMM he more gradual mprovemen could be arbuable o he larger number of parameers whch requres more daa n order o represen he daa adequaely. A suggeson for fuure work would be herefore o mprove he long erm emporal model of he CRF ha we are usng. I should be noed ha hese commens abou he CRF model apply for he sngle chan srucure whch s used here. There are many confguraons whch can be used whn he CRF framework. A hgher order CRF may produce a beer emporal model and so we would expec o see larger mprovemens when he observaon wndow sze s ncreased. However a CRF model wll always be dscrmnave compared o he HMM and CHMM s generave ably. The ably o generae samples may be mporan n ceran cases (such as esmang a model s complexy [Gong and Xang, 2003]) however hs ably s no requred for he classfcaon asks as presened here. Throughou all expermens on all of he daases s vsble ha here seems o be an opmal wndow sze for classfcaon of a parcular class. For some acves such as fghng n he CAVIAR daase (Fgure 6) he wndow sze s que shor (due o he speed of a fgh) where as

18 for oher classes such as he gnore behavor (Fgure 7) a longer wndow sze mproves performance. FUTURE WORK The sgnfcance of he role of me whn hs work has been demonsraed. Fuure work should seek o explo hs n a prncpled way. I would also be fruful o nvesgae n a prncpled way auomac feaure selecon. I would be envsaged ha such a procedure could ncorporae los of feaures and chose hem auomacally raher han, as s he case here, where s only possble o use a lmed number of feaures. By havng a comparave sudy of he classfers on he daase s also possble o know how much you should rely on hem, f for nsance you were usng some classfcaon crera n feaure selecon or o esablsh an opmal me wndow. REFERENCES S. Blunsden, E. Andrade, A. Laghaee, and R. Fsher. Behave neracons es case scenaros, epsrc proec gr/s98146, On Lne, Sepember URL: hp://groups.nf.ed.ac.uk/vson/behavedata/interactions/ndex. EC Funded CAVIAR proec/ist found a url: hp://homepages.nf.ed.ac.uk/rbf/cavar/, J. W. Davs and A. F. Bobck. The represenaon and recognon of acon usng emporal emplaes. In IEEE Transacons on Paern Analyss and Machne Inellgence, volume 23, pages IEEE Compuer Socey, H. M. Dee and D. C. Hogg. Deecng nexplcable behavour. In Brsh Machne Vson Conference, pages , Sepember P. Dollar, V. Rabaud, G. Corell, and S. Belonge. Behavor recognon va sparse spaoemporal feaures. In PETS, pages 65 72, Chna, A. Efros, A. Berg, G. Mor and J. Malk. Recognsng acon a a dsance. In In 9 h Inernaonal Conference on Compuer Vson, volume 2, pages , G. Ggerenzer, P. M. Todd, and ABC Research Group. Smple Heurscs Tha Make Us Smar. Evoluon and Cognon Seres. Oxford Unversy Press, S. Gong and T. Xang. Recognon of group acves usng a dynamc probablsc nework. In IEEE Inernaonal Conference on Compuer Vson, pages , Ocober 2003b. S. S. Inlle and A. F. Bobck. Recognzng planned, mulperson acon. CVIU, 81(3): , March X. Lu and C. S. Chua. Mul-agen acvy recognon usng observaon decomposed hdden markov models. Image and Vson Compung, pages , 2006.

19 N. M. Olver, B. Rosaro, and A. P. Penland. A Bayesan compuer vson sysem for modelng human neracons. IEEE Transacons on Paern Analyss and Machne Inellgence, 22(8): , M. Perse, M. Krsan, J. Pers, and S. Kovacc. A emplae-based mul-player acon recognon of he baskeball game. In J. Pers and D. R. Magee, edors, ECCV Workshop on Compuer Vson Based Analyss n Spor Envronmens, pages 71 82, Graz, Ausra, May L. R. Rabner. A uoral on hdden markov models and seleced applcaons n speech recognon. Readngs n speech recognon, pages , P. Rbero and J. Sanos-Vcor. Human acves recognon from vdeo: modelng, feaure selecon and classfcaon archecure. In Workshop on Human Acvy Recognon and Modellng (HAREM n conuncon wh BMVC 2005), pages 61 70, Oxford, Sepember N. M. Roberson. Auomac Causal Reasonng for Vdeo Survellance. PhD hess, Hearford College, Unversy of Oxford, June C. Smnchsescu, A. Kanaua, Z. L and D. Meaxas. Condonal models for conexual human moon recognon. In Inernaonal Conference on Compuer Vson, T. Van Vu, F. Bremond, and M. Thonna. Auomac vdeo nerpreaon: A recognon algorhm for emporal scenaros based on precompled scenaro models. In ICVS, H. M. Wallach. Condonal random felds: An nroducon. Techncal repor, Unversy of Pennsylvana, CIS Techncal Repor MS-CIS Bo Wu and Ram Nevaa. Deecon and rackng of mulple, parally occluded humans by bayesan combnaon of edgele based par deecors. Inernaonal Journal of Compuer Vson, 75(2): , Ylmaz, A., Javed, O., and Shah, M. (2006). Obec rackng: A survey. ACM Compu. Surv. 38, 4 (Dec. 2006)

Bayes rule for a classification problem INF Discriminant functions for the normal density. Euclidean distance. Mahalanobis distance

Bayes rule for a classification problem INF Discriminant functions for the normal density. Euclidean distance. Mahalanobis distance INF 43 3.. Repeon Anne Solberg (anne@f.uo.no Bayes rule for a classfcaon problem Suppose we have J, =,...J classes. s he class label for a pxel, and x s he observed feaure vecor. We can use Bayes rule

More information

CS434a/541a: Pattern Recognition Prof. Olga Veksler. Lecture 4

CS434a/541a: Pattern Recognition Prof. Olga Veksler. Lecture 4 CS434a/54a: Paern Recognon Prof. Olga Veksler Lecure 4 Oulne Normal Random Varable Properes Dscrmnan funcons Why Normal Random Varables? Analycally racable Works well when observaon comes form a corruped

More information

Robust and Accurate Cancer Classification with Gene Expression Profiling

Robust and Accurate Cancer Classification with Gene Expression Profiling Robus and Accurae Cancer Classfcaon wh Gene Expresson Proflng (Compuaonal ysems Bology, 2005) Auhor: Hafeng L, Keshu Zhang, ao Jang Oulne Background LDA (lnear dscrmnan analyss) and small sample sze problem

More information

Variants of Pegasos. December 11, 2009

Variants of Pegasos. December 11, 2009 Inroducon Varans of Pegasos SooWoong Ryu bshboy@sanford.edu December, 009 Youngsoo Cho yc344@sanford.edu Developng a new SVM algorhm s ongong research opc. Among many exng SVM algorhms, we wll focus on

More information

Clustering (Bishop ch 9)

Clustering (Bishop ch 9) Cluserng (Bshop ch 9) Reference: Daa Mnng by Margare Dunham (a slde source) 1 Cluserng Cluserng s unsupervsed learnng, here are no class labels Wan o fnd groups of smlar nsances Ofen use a dsance measure

More information

Outline. Probabilistic Model Learning. Probabilistic Model Learning. Probabilistic Model for Time-series Data: Hidden Markov Model

Outline. Probabilistic Model Learning. Probabilistic Model Learning. Probabilistic Model for Time-series Data: Hidden Markov Model Probablsc Model for Tme-seres Daa: Hdden Markov Model Hrosh Mamsuka Bonformacs Cener Kyoo Unversy Oulne Three Problems for probablsc models n machne learnng. Compung lkelhood 2. Learnng 3. Parsng (predcon

More information

( ) () we define the interaction representation by the unitary transformation () = ()

( ) () we define the interaction representation by the unitary transformation () = () Hgher Order Perurbaon Theory Mchael Fowler 3/7/6 The neracon Represenaon Recall ha n he frs par of hs course sequence, we dscussed he chrödnger and Hesenberg represenaons of quanum mechancs here n he chrödnger

More information

V.Abramov - FURTHER ANALYSIS OF CONFIDENCE INTERVALS FOR LARGE CLIENT/SERVER COMPUTER NETWORKS

V.Abramov - FURTHER ANALYSIS OF CONFIDENCE INTERVALS FOR LARGE CLIENT/SERVER COMPUTER NETWORKS R&RATA # Vol.) 8, March FURTHER AALYSIS OF COFIDECE ITERVALS FOR LARGE CLIET/SERVER COMPUTER ETWORKS Vyacheslav Abramov School of Mahemacal Scences, Monash Unversy, Buldng 8, Level 4, Clayon Campus, Wellngon

More information

Solution in semi infinite diffusion couples (error function analysis)

Solution in semi infinite diffusion couples (error function analysis) Soluon n sem nfne dffuson couples (error funcon analyss) Le us consder now he sem nfne dffuson couple of wo blocks wh concenraon of and I means ha, n a A- bnary sysem, s bondng beween wo blocks made of

More information

CHAPTER 10: LINEAR DISCRIMINATION

CHAPTER 10: LINEAR DISCRIMINATION CHAPER : LINEAR DISCRIMINAION Dscrmnan-based Classfcaon 3 In classfcaon h K classes (C,C,, C k ) We defned dscrmnan funcon g j (), j=,,,k hen gven an es eample, e chose (predced) s class label as C f g

More information

( ) [ ] MAP Decision Rule

( ) [ ] MAP Decision Rule Announcemens Bayes Decson Theory wh Normal Dsrbuons HW0 due oday HW o be assgned soon Proec descrpon posed Bomercs CSE 90 Lecure 4 CSE90, Sprng 04 CSE90, Sprng 04 Key Probables 4 ω class label X feaure

More information

Robustness Experiments with Two Variance Components

Robustness Experiments with Two Variance Components Naonal Insue of Sandards and Technology (NIST) Informaon Technology Laboraory (ITL) Sascal Engneerng Dvson (SED) Robusness Expermens wh Two Varance Componens by Ana Ivelsse Avlés avles@ns.gov Conference

More information

This document is downloaded from DR-NTU, Nanyang Technological University Library, Singapore.

This document is downloaded from DR-NTU, Nanyang Technological University Library, Singapore. Ths documen s downloaded from DR-NTU, Nanyang Technologcal Unversy Lbrary, Sngapore. Tle A smplfed verb machng algorhm for word paron n vsual speech processng( Acceped verson ) Auhor(s) Foo, Say We; Yong,

More information

An introduction to Support Vector Machine

An introduction to Support Vector Machine An nroducon o Suppor Vecor Machne 報告者 : 黃立德 References: Smon Haykn, "Neural Neworks: a comprehensve foundaon, second edon, 999, Chaper 2,6 Nello Chrsann, John Shawe-Tayer, An Inroducon o Suppor Vecor Machnes,

More information

Machine Learning 2nd Edition

Machine Learning 2nd Edition INTRODUCTION TO Lecure Sldes for Machne Learnng nd Edon ETHEM ALPAYDIN, modfed by Leonardo Bobadlla and some pars from hp://www.cs.au.ac.l/~aparzn/machnelearnng/ The MIT Press, 00 alpaydn@boun.edu.r hp://www.cmpe.boun.edu.r/~ehem/mle

More information

Introduction ( Week 1-2) Course introduction A brief introduction to molecular biology A brief introduction to sequence comparison Part I: Algorithms

Introduction ( Week 1-2) Course introduction A brief introduction to molecular biology A brief introduction to sequence comparison Part I: Algorithms Course organzaon Inroducon Wee -2) Course nroducon A bref nroducon o molecular bology A bref nroducon o sequence comparson Par I: Algorhms for Sequence Analyss Wee 3-8) Chaper -3, Models and heores» Probably

More information

Advanced Machine Learning & Perception

Advanced Machine Learning & Perception Advanced Machne Learnng & Percepon Insrucor: Tony Jebara SVM Feaure & Kernel Selecon SVM Eensons Feaure Selecon (Flerng and Wrappng) SVM Feaure Selecon SVM Kernel Selecon SVM Eensons Classfcaon Feaure/Kernel

More information

Introduction to Boosting

Introduction to Boosting Inroducon o Boosng Cynha Rudn PACM, Prnceon Unversy Advsors Ingrd Daubeches and Rober Schapre Say you have a daabase of news arcles, +, +, -, -, +, +, -, -, +, +, -, -, +, +, -, + where arcles are labeled

More information

Linear Response Theory: The connection between QFT and experiments

Linear Response Theory: The connection between QFT and experiments Phys540.nb 39 3 Lnear Response Theory: The connecon beween QFT and expermens 3.1. Basc conceps and deas Q: ow do we measure he conducvy of a meal? A: we frs nroduce a weak elecrc feld E, and hen measure

More information

Lecture 11 SVM cont

Lecture 11 SVM cont Lecure SVM con. 0 008 Wha we have done so far We have esalshed ha we wan o fnd a lnear decson oundary whose margn s he larges We know how o measure he margn of a lnear decson oundary Tha s: he mnmum geomerc

More information

Dynamic Team Decision Theory. EECS 558 Project Shrutivandana Sharma and David Shuman December 10, 2005

Dynamic Team Decision Theory. EECS 558 Project Shrutivandana Sharma and David Shuman December 10, 2005 Dynamc Team Decson Theory EECS 558 Proec Shruvandana Sharma and Davd Shuman December 0, 005 Oulne Inroducon o Team Decson Theory Decomposon of he Dynamc Team Decson Problem Equvalence of Sac and Dynamc

More information

Detection of Waving Hands from Images Using Time Series of Intensity Values

Detection of Waving Hands from Images Using Time Series of Intensity Values Deecon of Wavng Hands from Images Usng Tme eres of Inensy Values Koa IRIE, Kazunor UMEDA Chuo Unversy, Tokyo, Japan re@sensor.mech.chuo-u.ac.jp, umeda@mech.chuo-u.ac.jp Absrac Ths paper proposes a mehod

More information

Econ107 Applied Econometrics Topic 5: Specification: Choosing Independent Variables (Studenmund, Chapter 6)

Econ107 Applied Econometrics Topic 5: Specification: Choosing Independent Variables (Studenmund, Chapter 6) Econ7 Appled Economercs Topc 5: Specfcaon: Choosng Independen Varables (Sudenmund, Chaper 6 Specfcaon errors ha we wll deal wh: wrong ndependen varable; wrong funconal form. Ths lecure deals wh wrong ndependen

More information

FTCS Solution to the Heat Equation

FTCS Solution to the Heat Equation FTCS Soluon o he Hea Equaon ME 448/548 Noes Gerald Reckenwald Porland Sae Unversy Deparmen of Mechancal Engneerng gerry@pdxedu ME 448/548: FTCS Soluon o he Hea Equaon Overvew Use he forward fne d erence

More information

Math 128b Project. Jude Yuen

Math 128b Project. Jude Yuen Mah 8b Proec Jude Yuen . Inroducon Le { Z } be a sequence of observed ndependen vecor varables. If he elemens of Z have a on normal dsrbuon hen { Z } has a mean vecor Z and a varancecovarance marx z. Geomercally

More information

John Geweke a and Gianni Amisano b a Departments of Economics and Statistics, University of Iowa, USA b European Central Bank, Frankfurt, Germany

John Geweke a and Gianni Amisano b a Departments of Economics and Statistics, University of Iowa, USA b European Central Bank, Frankfurt, Germany Herarchcal Markov Normal Mxure models wh Applcaons o Fnancal Asse Reurns Appendx: Proofs of Theorems and Condonal Poseror Dsrbuons John Geweke a and Gann Amsano b a Deparmens of Economcs and Sascs, Unversy

More information

TSS = SST + SSE An orthogonal partition of the total SS

TSS = SST + SSE An orthogonal partition of the total SS ANOVA: Topc 4. Orhogonal conrass [ST&D p. 183] H 0 : µ 1 = µ =... = µ H 1 : The mean of a leas one reamen group s dfferen To es hs hypohess, a basc ANOVA allocaes he varaon among reamen means (SST) equally

More information

Video-Based Face Recognition Using Adaptive Hidden Markov Models

Video-Based Face Recognition Using Adaptive Hidden Markov Models Vdeo-Based Face Recognon Usng Adapve Hdden Markov Models Xaomng Lu and suhan Chen Elecrcal and Compuer Engneerng, Carnege Mellon Unversy, Psburgh, PA, 523, U.S.A. xaomng@andrew.cmu.edu suhan@cmu.edu Absrac

More information

Notes on the stability of dynamic systems and the use of Eigen Values.

Notes on the stability of dynamic systems and the use of Eigen Values. Noes on he sabl of dnamc ssems and he use of Egen Values. Source: Macro II course noes, Dr. Davd Bessler s Tme Seres course noes, zarads (999) Ineremporal Macroeconomcs chaper 4 & Techncal ppend, and Hamlon

More information

[ ] 2. [ ]3 + (Δx i + Δx i 1 ) / 2. Δx i-1 Δx i Δx i+1. TPG4160 Reservoir Simulation 2018 Lecture note 3. page 1 of 5

[ ] 2. [ ]3 + (Δx i + Δx i 1 ) / 2. Δx i-1 Δx i Δx i+1. TPG4160 Reservoir Simulation 2018 Lecture note 3. page 1 of 5 TPG460 Reservor Smulaon 08 page of 5 DISCRETIZATIO OF THE FOW EQUATIOS As we already have seen, fne dfference appromaons of he paral dervaves appearng n he flow equaons may be obaned from Taylor seres

More information

Computing Relevance, Similarity: The Vector Space Model

Computing Relevance, Similarity: The Vector Space Model Compung Relevance, Smlary: The Vecor Space Model Based on Larson and Hears s sldes a UC-Bereley hp://.sms.bereley.edu/courses/s0/f00/ aabase Managemen Sysems, R. Ramarshnan ocumen Vecors v ocumens are

More information

Reactive Methods to Solve the Berth AllocationProblem with Stochastic Arrival and Handling Times

Reactive Methods to Solve the Berth AllocationProblem with Stochastic Arrival and Handling Times Reacve Mehods o Solve he Berh AllocaonProblem wh Sochasc Arrval and Handlng Tmes Nsh Umang* Mchel Berlare* * TRANSP-OR, Ecole Polyechnque Fédérale de Lausanne Frs Workshop on Large Scale Opmzaon November

More information

CHAPTER 5: MULTIVARIATE METHODS

CHAPTER 5: MULTIVARIATE METHODS CHAPER 5: MULIVARIAE MEHODS Mulvarae Daa 3 Mulple measuremens (sensors) npus/feaures/arbues: -varae N nsances/observaons/eamples Each row s an eample Each column represens a feaure X a b correspons o he

More information

Learning Objectives. Self Organization Map. Hamming Distance(1/5) Introduction. Hamming Distance(3/5) Hamming Distance(2/5) 15/04/2015

Learning Objectives. Self Organization Map. Hamming Distance(1/5) Introduction. Hamming Distance(3/5) Hamming Distance(2/5) 15/04/2015 /4/ Learnng Objecves Self Organzaon Map Learnng whou Exaples. Inroducon. MAXNET 3. Cluserng 4. Feaure Map. Self-organzng Feaure Map 6. Concluson 38 Inroducon. Learnng whou exaples. Daa are npu o he syse

More information

Lecture 6: Learning for Control (Generalised Linear Regression)

Lecture 6: Learning for Control (Generalised Linear Regression) Lecure 6: Learnng for Conrol (Generalsed Lnear Regresson) Conens: Lnear Mehods for Regresson Leas Squares, Gauss Markov heorem Recursve Leas Squares Lecure 6: RLSC - Prof. Sehu Vjayakumar Lnear Regresson

More information

ECE 366 Honors Section Fall 2009 Project Description

ECE 366 Honors Section Fall 2009 Project Description ECE 366 Honors Secon Fall 2009 Projec Descrpon Inroducon: Muscal genres are caegorcal labels creaed by humans o characerze dfferen ypes of musc. A muscal genre s characerzed by he common characerscs shared

More information

. The geometric multiplicity is dim[ker( λi. number of linearly independent eigenvectors associated with this eigenvalue.

. The geometric multiplicity is dim[ker( λi. number of linearly independent eigenvectors associated with this eigenvalue. Lnear Algebra Lecure # Noes We connue wh he dscusson of egenvalues, egenvecors, and dagonalzably of marces We wan o know, n parcular wha condons wll assure ha a marx can be dagonalzed and wha he obsrucons

More information

Hidden Markov Models

Hidden Markov Models 11-755 Machne Learnng for Sgnal Processng Hdden Markov Models Class 15. 12 Oc 2010 1 Admnsrva HW2 due Tuesday Is everyone on he projecs page? Where are your projec proposals? 2 Recap: Wha s an HMM Probablsc

More information

Ordinary Differential Equations in Neuroscience with Matlab examples. Aim 1- Gain understanding of how to set up and solve ODE s

Ordinary Differential Equations in Neuroscience with Matlab examples. Aim 1- Gain understanding of how to set up and solve ODE s Ordnary Dfferenal Equaons n Neuroscence wh Malab eamples. Am - Gan undersandng of how o se up and solve ODE s Am Undersand how o se up an solve a smple eample of he Hebb rule n D Our goal a end of class

More information

Lecture VI Regression

Lecture VI Regression Lecure VI Regresson (Lnear Mehods for Regresson) Conens: Lnear Mehods for Regresson Leas Squares, Gauss Markov heorem Recursve Leas Squares Lecure VI: MLSC - Dr. Sehu Vjayakumar Lnear Regresson Model M

More information

THE PREDICTION OF COMPETITIVE ENVIRONMENT IN BUSINESS

THE PREDICTION OF COMPETITIVE ENVIRONMENT IN BUSINESS THE PREICTION OF COMPETITIVE ENVIRONMENT IN BUSINESS INTROUCTION The wo dmensonal paral dfferenal equaons of second order can be used for he smulaon of compeve envronmen n busness The arcle presens he

More information

J i-1 i. J i i+1. Numerical integration of the diffusion equation (I) Finite difference method. Spatial Discretization. Internal nodes.

J i-1 i. J i i+1. Numerical integration of the diffusion equation (I) Finite difference method. Spatial Discretization. Internal nodes. umercal negraon of he dffuson equaon (I) Fne dfference mehod. Spaal screaon. Inernal nodes. R L V For hermal conducon le s dscree he spaal doman no small fne spans, =,,: Balance of parcles for an nernal

More information

Machine Learning Linear Regression

Machine Learning Linear Regression Machne Learnng Lnear Regresson Lesson 3 Lnear Regresson Bascs of Regresson Leas Squares esmaon Polynomal Regresson Bass funcons Regresson model Regularzed Regresson Sascal Regresson Mamum Lkelhood (ML)

More information

In the complete model, these slopes are ANALYSIS OF VARIANCE FOR THE COMPLETE TWO-WAY MODEL. (! i+1 -! i ) + [(!") i+1,q - [(!

In the complete model, these slopes are ANALYSIS OF VARIANCE FOR THE COMPLETE TWO-WAY MODEL. (! i+1 -! i ) + [(!) i+1,q - [(! ANALYSIS OF VARIANCE FOR THE COMPLETE TWO-WAY MODEL The frs hng o es n wo-way ANOVA: Is here neracon? "No neracon" means: The man effecs model would f. Ths n urn means: In he neracon plo (wh A on he horzonal

More information

2.1 Constitutive Theory

2.1 Constitutive Theory Secon.. Consuve Theory.. Consuve Equaons Governng Equaons The equaons governng he behavour of maerals are (n he spaal form) dρ v & ρ + ρdv v = + ρ = Conservaon of Mass (..a) d x σ j dv dvσ + b = ρ v& +

More information

Single and Multiple Object Tracking Using a Multi-Feature Joint Sparse Representation

Single and Multiple Object Tracking Using a Multi-Feature Joint Sparse Representation Sngle and Mulple Objec Trackng Usng a Mul-Feaure Jon Sparse Represenaon Wemng Hu, We L, and Xaoqn Zhang (Naonal Laboraory of Paern Recognon, Insue of Auomaon, Chnese Academy of Scences, Bejng 100190) {wmhu,

More information

Hidden Markov Models Following a lecture by Andrew W. Moore Carnegie Mellon University

Hidden Markov Models Following a lecture by Andrew W. Moore Carnegie Mellon University Hdden Markov Models Followng a lecure by Andrew W. Moore Carnege Mellon Unversy www.cs.cmu.edu/~awm/uorals A Markov Sysem Has N saes, called s, s 2.. s N s 2 There are dscree meseps, 0,, s s 3 N 3 0 Hdden

More information

. The geometric multiplicity is dim[ker( λi. A )], i.e. the number of linearly independent eigenvectors associated with this eigenvalue.

. The geometric multiplicity is dim[ker( λi. A )], i.e. the number of linearly independent eigenvectors associated with this eigenvalue. Mah E-b Lecure #0 Noes We connue wh he dscusson of egenvalues, egenvecors, and dagonalzably of marces We wan o know, n parcular wha condons wll assure ha a marx can be dagonalzed and wha he obsrucons are

More information

DEEP UNFOLDING FOR MULTICHANNEL SOURCE SEPARATION SUPPLEMENTARY MATERIAL

DEEP UNFOLDING FOR MULTICHANNEL SOURCE SEPARATION SUPPLEMENTARY MATERIAL DEEP UNFOLDING FOR MULTICHANNEL SOURCE SEPARATION SUPPLEMENTARY MATERIAL Sco Wsdom, John Hershey 2, Jonahan Le Roux 2, and Shnj Waanabe 2 Deparmen o Elecrcal Engneerng, Unversy o Washngon, Seale, WA, USA

More information

An Effective TCM-KNN Scheme for High-Speed Network Anomaly Detection

An Effective TCM-KNN Scheme for High-Speed Network Anomaly Detection Vol. 24, November,, 200 An Effecve TCM-KNN Scheme for Hgh-Speed Nework Anomaly eecon Yang L Chnese Academy of Scences, Bejng Chna, 00080 lyang@sofware.c.ac.cn Absrac. Nework anomaly deecon has been a ho

More information

Sampling Procedure of the Sum of two Binary Markov Process Realizations

Sampling Procedure of the Sum of two Binary Markov Process Realizations Samplng Procedure of he Sum of wo Bnary Markov Process Realzaons YURY GORITSKIY Dep. of Mahemacal Modelng of Moscow Power Insue (Techncal Unversy), Moscow, RUSSIA, E-mal: gorsky@yandex.ru VLADIMIR KAZAKOV

More information

January Examinations 2012

January Examinations 2012 Page of 5 EC79 January Examnaons No. of Pages: 5 No. of Quesons: 8 Subjec ECONOMICS (POSTGRADUATE) Tle of Paper EC79 QUANTITATIVE METHODS FOR BUSINESS AND FINANCE Tme Allowed Two Hours ( hours) Insrucons

More information

Chapter Lagrangian Interpolation

Chapter Lagrangian Interpolation Chaper 5.4 agrangan Inerpolaon Afer readng hs chaper you should be able o:. dere agrangan mehod of nerpolaon. sole problems usng agrangan mehod of nerpolaon and. use agrangan nerpolans o fnd deraes and

More information

Discrete Markov Process. Introduction. Example: Balls and Urns. Stochastic Automaton. INTRODUCTION TO Machine Learning 3rd Edition

Discrete Markov Process. Introduction. Example: Balls and Urns. Stochastic Automaton. INTRODUCTION TO Machine Learning 3rd Edition EHEM ALPAYDI he MI Press, 04 Lecure Sldes for IRODUCIO O Machne Learnng 3rd Edon alpaydn@boun.edu.r hp://www.cmpe.boun.edu.r/~ehem/ml3e Sldes from exboo resource page. Slghly eded and wh addonal examples

More information

Lecture Slides for INTRODUCTION TO. Machine Learning. ETHEM ALPAYDIN The MIT Press,

Lecture Slides for INTRODUCTION TO. Machine Learning. ETHEM ALPAYDIN The MIT Press, Lecure Sldes for INTRDUCTIN T Machne Learnng ETHEM ALAYDIN The MIT ress, 2004 alpaydn@boun.edu.r hp://www.cmpe.boun.edu.r/~ehem/2ml CHATER 3: Hdden Marov Models Inroducon Modelng dependences n npu; no

More information

2. SPATIALLY LAGGED DEPENDENT VARIABLES

2. SPATIALLY LAGGED DEPENDENT VARIABLES 2. SPATIALLY LAGGED DEPENDENT VARIABLES In hs chaper, we descrbe a sascal model ha ncorporaes spaal dependence explcly by addng a spaally lagged dependen varable y on he rgh-hand sde of he regresson equaon.

More information

Relative controllability of nonlinear systems with delays in control

Relative controllability of nonlinear systems with delays in control Relave conrollably o nonlnear sysems wh delays n conrol Jerzy Klamka Insue o Conrol Engneerng, Slesan Techncal Unversy, 44- Glwce, Poland. phone/ax : 48 32 37227, {jklamka}@a.polsl.glwce.pl Keywor: Conrollably.

More information

GENERATING CERTAIN QUINTIC IRREDUCIBLE POLYNOMIALS OVER FINITE FIELDS. Youngwoo Ahn and Kitae Kim

GENERATING CERTAIN QUINTIC IRREDUCIBLE POLYNOMIALS OVER FINITE FIELDS. Youngwoo Ahn and Kitae Kim Korean J. Mah. 19 (2011), No. 3, pp. 263 272 GENERATING CERTAIN QUINTIC IRREDUCIBLE POLYNOMIALS OVER FINITE FIELDS Youngwoo Ahn and Kae Km Absrac. In he paper [1], an explc correspondence beween ceran

More information

Chapter 6: AC Circuits

Chapter 6: AC Circuits Chaper 6: AC Crcus Chaper 6: Oulne Phasors and he AC Seady Sae AC Crcus A sable, lnear crcu operang n he seady sae wh snusodal excaon (.e., snusodal seady sae. Complee response forced response naural response.

More information

e-journal Reliability: Theory& Applications No 2 (Vol.2) Vyacheslav Abramov

e-journal Reliability: Theory& Applications No 2 (Vol.2) Vyacheslav Abramov June 7 e-ournal Relably: Theory& Applcaons No (Vol. CONFIDENCE INTERVALS ASSOCIATED WITH PERFORMANCE ANALYSIS OF SYMMETRIC LARGE CLOSED CLIENT/SERVER COMPUTER NETWORKS Absrac Vyacheslav Abramov School

More information

Using Fuzzy Pattern Recognition to Detect Unknown Malicious Executables Code

Using Fuzzy Pattern Recognition to Detect Unknown Malicious Executables Code Usng Fuzzy Paern Recognon o Deec Unknown Malcous Execuables Code Boyun Zhang,, Janpng Yn, and Jngbo Hao School of Compuer Scence, Naonal Unversy of Defense Technology, Changsha 40073, Chna hnxzby@yahoo.com.cn

More information

A Bayesian algorithm for tracking multiple moving objects in outdoor surveillance video

A Bayesian algorithm for tracking multiple moving objects in outdoor surveillance video A Bayesan algorhm for racng mulple movng obecs n oudoor survellance vdeo Manunah Narayana Unversy of Kansas Lawrence, Kansas manu@u.edu Absrac Relable racng of mulple movng obecs n vdes an neresng challenge,

More information

CS 536: Machine Learning. Nonparametric Density Estimation Unsupervised Learning - Clustering

CS 536: Machine Learning. Nonparametric Density Estimation Unsupervised Learning - Clustering CS 536: Machne Learnng Nonparamerc Densy Esmaon Unsupervsed Learnng - Cluserng Fall 2005 Ahmed Elgammal Dep of Compuer Scence Rugers Unversy CS 536 Densy Esmaon - Cluserng - 1 Oulnes Densy esmaon Nonparamerc

More information

5th International Conference on Advanced Design and Manufacturing Engineering (ICADME 2015)

5th International Conference on Advanced Design and Manufacturing Engineering (ICADME 2015) 5h Inernaonal onference on Advanced Desgn and Manufacurng Engneerng (IADME 5 The Falure Rae Expermenal Sudy of Specal N Machne Tool hunshan He, a, *, La Pan,b and Bng Hu 3,c,,3 ollege of Mechancal and

More information

THEORETICAL AUTOCORRELATIONS. ) if often denoted by γ. Note that

THEORETICAL AUTOCORRELATIONS. ) if often denoted by γ. Note that THEORETICAL AUTOCORRELATIONS Cov( y, y ) E( y E( y))( y E( y)) ρ = = Var( y) E( y E( y)) =,, L ρ = and Cov( y, y ) s ofen denoed by whle Var( y ) f ofen denoed by γ. Noe ha γ = γ and ρ = ρ and because

More information

Polymerization Technology Laboratory Course

Polymerization Technology Laboratory Course Prakkum Polymer Scence/Polymersaonsechnk Versuch Resdence Tme Dsrbuon Polymerzaon Technology Laboraory Course Resdence Tme Dsrbuon of Chemcal Reacors If molecules or elemens of a flud are akng dfferen

More information

ELASTIC MODULUS ESTIMATION OF CHOPPED CARBON FIBER TAPE REINFORCED THERMOPLASTICS USING THE MONTE CARLO SIMULATION

ELASTIC MODULUS ESTIMATION OF CHOPPED CARBON FIBER TAPE REINFORCED THERMOPLASTICS USING THE MONTE CARLO SIMULATION THE 19 TH INTERNATIONAL ONFERENE ON OMPOSITE MATERIALS ELASTI MODULUS ESTIMATION OF HOPPED ARBON FIBER TAPE REINFORED THERMOPLASTIS USING THE MONTE ARLO SIMULATION Y. Sao 1*, J. Takahash 1, T. Masuo 1,

More information

On One Analytic Method of. Constructing Program Controls

On One Analytic Method of. Constructing Program Controls Appled Mahemacal Scences, Vol. 9, 05, no. 8, 409-407 HIKARI Ld, www.m-hkar.com hp://dx.do.org/0.988/ams.05.54349 On One Analyc Mehod of Consrucng Program Conrols A. N. Kvko, S. V. Chsyakov and Yu. E. Balyna

More information

Anomaly Detection. Lecture Notes for Chapter 9. Introduction to Data Mining, 2 nd Edition by Tan, Steinbach, Karpatne, Kumar

Anomaly Detection. Lecture Notes for Chapter 9. Introduction to Data Mining, 2 nd Edition by Tan, Steinbach, Karpatne, Kumar Anomaly eecon Lecure Noes for Chaper 9 Inroducon o aa Mnng, 2 nd Edon by Tan, Senbach, Karpane, Kumar 2/14/18 Inroducon o aa Mnng, 2nd Edon 1 Anomaly/Ouler eecon Wha are anomales/oulers? The se of daa

More information

New M-Estimator Objective Function. in Simultaneous Equations Model. (A Comparative Study)

New M-Estimator Objective Function. in Simultaneous Equations Model. (A Comparative Study) Inernaonal Mahemacal Forum, Vol. 8, 3, no., 7 - HIKARI Ld, www.m-hkar.com hp://dx.do.org/.988/mf.3.3488 New M-Esmaor Objecve Funcon n Smulaneous Equaons Model (A Comparave Sudy) Ahmed H. Youssef Professor

More information

WiH Wei He

WiH Wei He Sysem Idenfcaon of onlnear Sae-Space Space Baery odels WH We He wehe@calce.umd.edu Advsor: Dr. Chaochao Chen Deparmen of echancal Engneerng Unversy of aryland, College Par 1 Unversy of aryland Bacground

More information

A Novel Object Detection Method Using Gaussian Mixture Codebook Model of RGB-D Information

A Novel Object Detection Method Using Gaussian Mixture Codebook Model of RGB-D Information A Novel Objec Deecon Mehod Usng Gaussan Mxure Codebook Model of RGB-D Informaon Lujang LIU 1, Gaopeng ZHAO *,1, Yumng BO 1 1 School of Auomaon, Nanjng Unversy of Scence and Technology, Nanjng, Jangsu 10094,

More information

Comb Filters. Comb Filters

Comb Filters. Comb Filters The smple flers dscussed so far are characered eher by a sngle passband and/or a sngle sopband There are applcaons where flers wh mulple passbands and sopbands are requred Thecomb fler s an example of

More information

Let s treat the problem of the response of a system to an applied external force. Again,

Let s treat the problem of the response of a system to an applied external force. Again, Page 33 QUANTUM LNEAR RESPONSE FUNCTON Le s rea he problem of he response of a sysem o an appled exernal force. Agan, H() H f () A H + V () Exernal agen acng on nernal varable Hamlonan for equlbrum sysem

More information

CH.3. COMPATIBILITY EQUATIONS. Continuum Mechanics Course (MMC) - ETSECCPB - UPC

CH.3. COMPATIBILITY EQUATIONS. Continuum Mechanics Course (MMC) - ETSECCPB - UPC CH.3. COMPATIBILITY EQUATIONS Connuum Mechancs Course (MMC) - ETSECCPB - UPC Overvew Compably Condons Compably Equaons of a Poenal Vecor Feld Compably Condons for Infnesmal Srans Inegraon of he Infnesmal

More information

Algorithm Research on Moving Object Detection of Surveillance Video Sequence *

Algorithm Research on Moving Object Detection of Surveillance Video Sequence * Opcs and Phooncs Journal 03 3 308-3 do:0.436/opj.03.3b07 Publshed Onlne June 03 (hp://www.scrp.org/journal/opj) Algorhm Research on Movng Objec Deecon of Survellance Vdeo Sequence * Kuhe Yang Zhmng Ca

More information

Chapters 2 Kinematics. Position, Distance, Displacement

Chapters 2 Kinematics. Position, Distance, Displacement Chapers Knemacs Poson, Dsance, Dsplacemen Mechancs: Knemacs and Dynamcs. Knemacs deals wh moon, bu s no concerned wh he cause o moon. Dynamcs deals wh he relaonshp beween orce and moon. The word dsplacemen

More information

Genetic Algorithm in Parameter Estimation of Nonlinear Dynamic Systems

Genetic Algorithm in Parameter Estimation of Nonlinear Dynamic Systems Genec Algorhm n Parameer Esmaon of Nonlnear Dynamc Sysems E. Paeraks manos@egnaa.ee.auh.gr V. Perds perds@vergna.eng.auh.gr Ah. ehagas kehagas@egnaa.ee.auh.gr hp://skron.conrol.ee.auh.gr/kehagas/ndex.hm

More information

Mechanics Physics 151

Mechanics Physics 151 Mechancs Physcs 5 Lecure 9 Hamlonan Equaons of Moon (Chaper 8) Wha We Dd Las Tme Consruced Hamlonan formalsm H ( q, p, ) = q p L( q, q, ) H p = q H q = p H = L Equvalen o Lagrangan formalsm Smpler, bu

More information

CS286.2 Lecture 14: Quantum de Finetti Theorems II

CS286.2 Lecture 14: Quantum de Finetti Theorems II CS286.2 Lecure 14: Quanum de Fne Theorems II Scrbe: Mara Okounkova 1 Saemen of he heorem Recall he las saemen of he quanum de Fne heorem from he prevous lecure. Theorem 1 Quanum de Fne). Le ρ Dens C 2

More information

Mechanics Physics 151

Mechanics Physics 151 Mechancs Physcs 5 Lecure 9 Hamlonan Equaons of Moon (Chaper 8) Wha We Dd Las Tme Consruced Hamlonan formalsm Hqp (,,) = qp Lqq (,,) H p = q H q = p H L = Equvalen o Lagrangan formalsm Smpler, bu wce as

More information

Analyzing Human Movements from Silhouettes using Manifold Learning

Analyzing Human Movements from Silhouettes using Manifold Learning Analyzng Human Movemens from Slhouees usng Manfold Learnng Lang Wang and Davd Suer ARC Cenre for Percepve and Inellgen Machnes n Complex Envronmens Monash Unversy, Clayon, VIC, 38, Ausrala {lang.wang,

More information

3. OVERVIEW OF NUMERICAL METHODS

3. OVERVIEW OF NUMERICAL METHODS 3 OVERVIEW OF NUMERICAL METHODS 3 Inroducory remarks Ths chaper summarzes hose numercal echnques whose knowledge s ndspensable for he undersandng of he dfferen dscree elemen mehods: he Newon-Raphson-mehod,

More information

doi: info:doi/ /

doi: info:doi/ / do: nfo:do/0.063/.322393 nernaonal Conference on Power Conrol and Opmzaon, Bal, ndonesa, -3, June 2009 A COLOR FEATURES-BASED METHOD FOR OBJECT TRACKNG EMPLOYNG A PARTCLE FLTER ALGORTHM Bud Sugand, Hyoungseop

More information

II. Light is a Ray (Geometrical Optics)

II. Light is a Ray (Geometrical Optics) II Lgh s a Ray (Geomercal Opcs) IIB Reflecon and Refracon Hero s Prncple of Leas Dsance Law of Reflecon Hero of Aleandra, who lved n he 2 nd cenury BC, posulaed he followng prncple: Prncple of Leas Dsance:

More information

Part II CONTINUOUS TIME STOCHASTIC PROCESSES

Part II CONTINUOUS TIME STOCHASTIC PROCESSES Par II CONTINUOUS TIME STOCHASTIC PROCESSES 4 Chaper 4 For an advanced analyss of he properes of he Wener process, see: Revus D and Yor M: Connuous marngales and Brownan Moon Karazas I and Shreve S E:

More information

( t) Outline of program: BGC1: Survival and event history analysis Oslo, March-May Recapitulation. The additive regression model

( t) Outline of program: BGC1: Survival and event history analysis Oslo, March-May Recapitulation. The additive regression model BGC1: Survval and even hsory analyss Oslo, March-May 212 Monday May 7h and Tuesday May 8h The addve regresson model Ørnulf Borgan Deparmen of Mahemacs Unversy of Oslo Oulne of program: Recapulaon Counng

More information

Graduate Macroeconomics 2 Problem set 5. - Solutions

Graduate Macroeconomics 2 Problem set 5. - Solutions Graduae Macroeconomcs 2 Problem se. - Soluons Queson 1 To answer hs queson we need he frms frs order condons and he equaon ha deermnes he number of frms n equlbrum. The frms frs order condons are: F K

More information

Density Matrix Description of NMR BCMB/CHEM 8190

Density Matrix Description of NMR BCMB/CHEM 8190 Densy Marx Descrpon of NMR BCMBCHEM 89 Operaors n Marx Noaon Alernae approach o second order specra: ask abou x magnezaon nsead of energes and ranson probables. If we say wh one bass se, properes vary

More information

Building Temporal Models for Gesture Recognition

Building Temporal Models for Gesture Recognition Buldng Temporal Models for Gesure Recognon Rchard Bowden, Mansoor Sarhad Vson and VR Group Dep of Sysems Engneerng, Brunel Unversy Uxbrdge, Mddlesex, UB8 3PH, UK rchard.bowden@brunel.ac.uk Absrac Ths work

More information

/99 $10.00 (c) 1999 IEEE

/99 $10.00 (c) 1999 IEEE Recognzng Hand Gesure Usng Moon Trajecores Mng-Hsuan Yang and Narendra Ahuja Deparmen of Compuer Scence and Beckman Insue Unversy of Illnos a Urbana-Champagn, Urbana, IL 611 fmhyang,ahujag@vson.a.uuc.edu

More information

Volatility Interpolation

Volatility Interpolation Volaly Inerpolaon Prelmnary Verson March 00 Jesper Andreasen and Bran Huge Danse Mares, Copenhagen wan.daddy@danseban.com brno@danseban.com Elecronc copy avalable a: hp://ssrn.com/absrac=69497 Inro Local

More information

Fall 2010 Graduate Course on Dynamic Learning

Fall 2010 Graduate Course on Dynamic Learning Fall 200 Graduae Course on Dynamc Learnng Chaper 4: Parcle Flers Sepember 27, 200 Byoung-Tak Zhang School of Compuer Scence and Engneerng & Cognve Scence and Bran Scence Programs Seoul aonal Unversy hp://b.snu.ac.kr/~bzhang/

More information

EEL 6266 Power System Operation and Control. Chapter 5 Unit Commitment

EEL 6266 Power System Operation and Control. Chapter 5 Unit Commitment EEL 6266 Power Sysem Operaon and Conrol Chaper 5 Un Commmen Dynamc programmng chef advanage over enumeraon schemes s he reducon n he dmensonaly of he problem n a src prory order scheme, here are only N

More information

Cubic Bezier Homotopy Function for Solving Exponential Equations

Cubic Bezier Homotopy Function for Solving Exponential Equations Penerb Journal of Advanced Research n Compung and Applcaons ISSN (onlne: 46-97 Vol. 4, No.. Pages -8, 6 omoopy Funcon for Solvng Eponenal Equaons S. S. Raml *,,. Mohamad Nor,a, N. S. Saharzan,b and M.

More information

Chapter 4. Neural Networks Based on Competition

Chapter 4. Neural Networks Based on Competition Chaper 4. Neural Neworks Based on Compeon Compeon s mporan for NN Compeon beween neurons has been observed n bologcal nerve sysems Compeon s mporan n solvng many problems To classfy an npu paern _1 no

More information

Lecture 2 L n i e n a e r a M od o e d l e s

Lecture 2 L n i e n a e r a M od o e d l e s Lecure Lnear Models Las lecure You have learned abou ha s machne learnng Supervsed learnng Unsupervsed learnng Renforcemen learnng You have seen an eample learnng problem and he general process ha one

More information

Kernel-Based Bayesian Filtering for Object Tracking

Kernel-Based Bayesian Filtering for Object Tracking Kernel-Based Bayesan Flerng for Objec Trackng Bohyung Han Yng Zhu Dorn Comancu Larry Davs Dep. of Compuer Scence Real-Tme Vson and Modelng Inegraed Daa and Sysems Unversy of Maryland Semens Corporae Research

More information

Time-interval analysis of β decay. V. Horvat and J. C. Hardy

Time-interval analysis of β decay. V. Horvat and J. C. Hardy Tme-nerval analyss of β decay V. Horva and J. C. Hardy Work on he even analyss of β decay [1] connued and resuled n he developmen of a novel mehod of bea-decay me-nerval analyss ha produces hghly accurae

More information

Fault Diagnosis in Industrial Processes Using Principal Component Analysis and Hidden Markov Model

Fault Diagnosis in Industrial Processes Using Principal Component Analysis and Hidden Markov Model Faul Dagnoss n Indusral Processes Usng Prncpal Componen Analyss and Hdden Markov Model Shaoyuan Zhou, Janmng Zhang, and Shuqng Wang Absrac An approach combnng hdden Markov model (HMM) wh prncpal componen

More information