DIFFUSION ESTIMATION OF MIXTURE MODELS WITH LOCAL AND GLOBAL PARAMETERS

Size: px
Start display at page:

Download "DIFFUSION ESTIMATION OF MIXTURE MODELS WITH LOCAL AND GLOBAL PARAMETERS"

Transcription

1 DIFFUSION ESTIMATION OF MIXTURE MODES WITH OCA AND GOBA PARAMETERS Kaml Dedecus and Vladmíra Sečkárová Insue of Informaon Theory and Auomaon Czech Academy of Scences Pod Vodárenskou věží 1143/ Prague 8 Czech Republc ABSTRACT The sae-of-ar mehods for dsrbued esmaon of mxures assume he exsence of a common mxure model. In many praccal suaons hs assumpon may be oo resrcve as a subse of parameers may be purely local e.g. f he numbers of observable componens dffer across he nework. To reflec hs ssue we propose a new onlne Bayesan mehod for smulaneous esmaon of local parameers and dffuson esmaon of global parameers. The algorhm consss of wo seps. Frs he nodes perform local esmaon from own observaons by means of facorzed pror/poseror dsrbuons. Second a dffuson opmzaon sep s used o merge he nodes global parameers esmaes. A smulaon example demonsraes mproved performance n esmaon of boh parameers ses. Index Terms Dffuson esmaon dsrbued esmaon exponenal famly mxure models message passng. 1. INTRODUCTION Dsrbued esmaon of parameers of sochasc models has araced a sgnfcan aenon n he las wo decades parcularly due o a rapd developmen of cheap ad-hoc wreless sensor neworks conssng of nodes endowed wh sensng daa processng and communcaon capables. Ther applcaons range from localsaon arge rackng nruson deecon dconary learnng ec. [1]. Accordng o he communcaon sraeges among sources several groups of mehods can be recognzed. Frs he ncremenal algorhms passng and processng nformaon n a Hamlonan cyclc pah. These algorhms are prone o falures as each node and lnk are sngle-pons of falure and recovery s an NP-hard problem [2]. Ths ssue can be solved by he dffuson and consensus sraeges. The former perform only a lmed amoun of communcaon among neghborng nodes ypcally represened by an exchange of observaons durng he adapaon sep and an exchange of esmaes durng he combnaon sep [1 3]. The sandard consensus sraeges nvolved nermedae seps o reach consensus bu he recen runnng consensus algorhms removes hs overhead e.g. [4]. In hs paper we specfcally focus on sequenal esmaon of mxure models. Under roughly common dsrbuon of observaons y across he nework here exs several mehods for collaborave esmaon of mxures. One branch of soluons s based on expecaon-maxmzaon EM algorhm. For nsance Gu proposed a verson of a decenralzed EM algorhm wh a consensus sep evaluang global nework sascs beween he sandard local E K. Dedecus s suppored by he Czech Scence foundaon gran No P. V. Sečkárová s suppored by he Czech Scence Foundaon gran No S. and M seps [5]. A smlar approach called dsrbued expecaonmaxmzaon was proposed by Safarnejadan Menha and Karrar [6] where beween he E and M seps a dsrbued averagng approach s used o dffuse local suffcen sascs o neghborng nodes and esmae global suffcen sascs n each node. In he M-sep each node updaes parameers of he normal mxure model usng he esmaed global suffcen sascs. A fully dffuson-orened EM algorhm was proposed by Perera Pages- Zamora and opez-valcarce n [7]. Anoher branch of soluons semmng from he Bayesan heory s mosly based on varaonal nference. The dsrbued varaonal Bayesan algorhm DVBA of Safarnejadan Menhaj and Karrar allows dsrbued esmaon of normal mxure models [8]. However DVBA s an ncremenal algorhm. In order o remove he robusness ssue of hs opology he auhors laer proposed a peer-o-peer DVBA algorhm where he nodes communcae wh her neghbors [9]. Recenly Dedecus Rechl and Djurć proposed a dsrbued quas-bayesan algorhm wh pon-esmaed componen ndcaor applcable o a wde class of mxure models wh componen dsrbuons from he exponenal famly and suable for onlne esmaon [10]. Wh he excepon of componen weghs n [5] he lsed algorhms rely on he assumpon of a leas roughly common model parameers. However he parameers may dffer across he nework e.g. f he nodes observe dfferen arges maneuverng n a formaon. Then he collaboraon may be based on an assumpon of correlaon or smlary of nferred parameers [11]. These mulask dffuson algorhms are mosly MS-based e.g. [11 12] wh a few modfcaons reflecng for nsance sparsy [13]. Some more general recen algorhms allow unsupervsed deermnaon of neghbors belongng o he same cluser [ ]. These algorhms allow sac or sequenal collaborave esmaon of local and global parameers and parameers shared by clusers of nodes. We would lke o conrbue o hs opc from he probablscally conssen and versale Bayesan vewpon allowng absrac formulaon of nference asks. In parcular we propose a new mehod for collaborave esmaon of mxure models wh local and global parameer ses. The confguraon of parameers s very general he mxures may dffer n he number of componens her weghs and he componens may have dfferen parameers oo. The mehod s generc and applcable o a wde class of mxure models. Under ceran condons smplfes o an analycally racable mean-feld varaonal mehod. The resulng algorhm consss of wo phases. Frs he local esmaon where he nodes perform nference of boh local and global parameers ses locally from own observaons. Second he dffuson opmzaon sep where he nodes exchange he dsrbuons of global parameers wh her neghbors whn 1 hop dsance. If he mxure componens are exponenal 362

2 famly dsrbuons he whole algorhm can be reduced o varaonal message passng e.g. [17] exended by message-passng dffuson opmzaon. 2. PROBEM STATEMENT e us consder a nework a dreced or undreced graph of nodes I = {1... I} conneced by a se of edges deermnng he graph opology. Each node I acqures observaons y where = s a dscree me ndex. These observaons can be modelled by mxure models wh a common global subse of parameers. Each node also communcaes wh s adjacen neghbors whn 1 hop dsance formng s neghborhood I. We emphasze ha I oo. The adoped communcaon sraegy s dffuson [1 3] however compared o he ordnary dffuson algorhms he adapaon sep for an observaons exchange among nodes s no applcable here for her poenally dfferen dsrbuons. The combnaon sep mergng he parameers esmaes s preserved n a specal form. The ncomng observaons y of node follow a mxure dsrbuon.e. a convex combnaon of K probably dsrbuons called componens. Ther probably denses are denoed p k y θ k where θ k are componen parameers and k = 1... K are componen ndces. Denong by Θ all unknown parameers of he consdered mxure we have K p y Θ = φ k p k y θ k 1 k=1 where φ = [φ 1... φ K] s a vecor of componen weghs ha s posve real numbers akng values n he K 1-dmensonal probablsc smplex. Noe ha Θ may be any combnaon of φ θ or K. As a mnor smplfcaon we assume he local numbers of componens K o be known a pror. We face he problem of sequenal collaborave esmaon of nodes mxures parameers wh he assumpon ha here exss a global subse of parameers Θ G ha s common o all nework nodes and a local subse ha s s local complemen o Θ Θ G = I Θ and Θ = Θ \ Θ G. The goal s o explo he exsng nework n order o collaboravely arrve a beer esmaes of Θ G under modes communcaon requremens. In addon we conjecure ha an mprovemen n he esmaon of Θ G wll nduce an mproved esmaon of Θ. 3. SEQUENTIA DIFFUSION ESTIMATION The ordnary sequenal Bayesan nference of unknown parameers Θ reles on a jon pror dsrbuon π Θ y 0: 1 = π Θ Θ G y0: 1 quanfyng he curren sascal knowledge abou Θ and Θ G up o me 1 whch s based on he prevous observaons y 1... y 1 and he pseudo-observaons y 0 deermnng any nal knowledge e.g. from hsorcal daa or an exper s opnon. The ncorporaon of new observaon y s performed by vrue of he Bayes heorem π Θ y 0: = π Θ y 0: 1p y Θ πθ y 0: 1p y Θ dθ. 2 Unforunaely hs rgorous Bayesan approach o nference of Θ s mpraccal no only compuaonally bu also from aspec of he dsrbued esmaon. In order o combne nformaon from several sources nodes he dsrbuon of Θ G mus have he same form across he nework condonally ndependen of any local parameers Θ. Tha means π Θ y 0: = π Θ Θ G y0: ρ Θ y0: ρ Θ G y0:. 3 In oher words he rue dsrbuon of Θ and Θ G a each node I mus be replaced by wo lower-dmensonal dsrbuons ρ Θ and ρ Θ G he laer wh he same funconal form for all and whose produc s as close o he orgnal dsrbuon as possble n he Kullback-ebler sense. Somemes hs facorzaon s naural e.g. n mxure esmaon where he componen weghs may be modelled as ndependen on componen parameers. More on possble facorzaons can be found n [18]. The second sep s he dffuson opmzaon of he dsrbuon of Θ G. From node I vewpon s possble o vew he neghbors dsrbuons ρ jθ G as hypoheses abou Θ G. The goal s o opmally merge hese hypoheses no a sngle dsrbuon ρ Θ G ha s as close o ndvdual hypoheses as possble. Conssenly wh he Bayesan heory boh seps wll explo he Kullback-ebler dvergence as he measure of proxmy of probably dsrbuons. For wo probably denses fx and gx he laer absoluely connuous wh respec o he former he Kullback- ebler dvergence s defned as [ D [fx gx] = E fx log fx ]. 4 gx Ths dvergence s a premerc: s nonnegave and equal o zero f f = g almos everywhere bu s neher symmerc nor does sasfy he rangle nequaly. For he sake of smplcy we adop he assumpon ha he nodes know whch parameers are local and whch are global. Ths suaon occurs e.g. f dfferen ypes of sensors are used for akng observaons. The proposed algorhms s as follows: Algorhm formulaon We am o desgn a dsrbued algorhm for onlne mxures esmaon performng he followng wo seps: 1. ocal esmaon: A each node I locally esmae parameers Θ and Θ G n a Kullback-ebler-opmal facorzed form D [ρ Θ ρ Θ G π Θ Θ G ] mn. 2. Dffuson opmzaon: A each node I fnd he Kullback-ebler-opmally merged dsrbuon ρθ G usng he neghbors dsrbuon ρ jθ G as hypoheses abou Θ G ρ Θ G = arg mn I 1 D ρ [ ρ Θ G ρj Θ G ]

3 3.1. ocal esmaon ocal esmaon assers approxmaon of he nracable rue densy π Θ Θ G by ndvdual lower-dmensonal denses ρ Θ and ρ Θ G as n 5. Tha s he goal s o perform mnmzaon of he Kullback-ebler dvergence D [ρ Θ Θ G π Θ Θ G ] y 0: [ ρ Θ Θ G ] = E ρ Θ ΘG log π Θ Θ G y0: [ ] = E ρ Θ log ρ Θ + E ρ Θ G [log ρ Θ G ] ] E ρ Θ E ρ Θ G [log π Θ Θ G. 6 I s sraghforward o see ha he mnmzaon of 6 wh respec o Θ and Θ G leads o mnmzaons of expeced Kullback-ebler dvergences wh lower-dmensonal denses. Because he mnmum s reached under equal argumens of he Kullback-ebler dvergence he soluon s gven by he sysem ρ Θ G ]} = c exp {E ρθ [log π Θ Θ G { ]} ρ Θ = c G exp E ρθ G [log π Θ Θ G where c and c G are normalzaon consans. The presened facorzaon s closely relaed o he mean-feld varaonal Bayesan nference [19 20]. Indeed s possble o furher facorze he parcular denses ρ Θ and/or ρ Θ G and proceed wh he ordnary varaonal Bayes mehod sequenally seekng a conssen soluon by revsng he lower-dmensonal denses n a crcular way unl a convergence creron s me. The mean-feld varaonal algorhms are guaraneed o converge [21]. In he sequenal esmaon framework usually a few eraons are performed a each me sep usng prevous observaons sored n memory. The poseror dsrbuons from he prevous me sep hen serve as he pror dsrbuons. More elaboraed mehods for onlne varaonal nference suable for large daa can be found n [22 23]. In he nex secon he descrbed local esmaon va facorzed denses wll be exended by dffuson opmzaon of he poseror dsrbuon of Θ G. Then wll be shown ha f he dsrbuons of he mxure componens belong o he exponenal famly he proposed mehod s analycally racable as a message passng algorhm Dffuson opmzaon Each node I has access o denses ρ jθ G of s neghbors j I represenng hypoheses abou he rue parameer Θ G. Insead of workng wh he whole sysem of ndvdual denses we wan o replace hem by a sngle densy ρ Θ G. In parcular we seek an elemen from he se of all admssble denses mnmzng he average dvergence o all ndvdual denses ρ Θ G = arg mn I [ 1 D ρ ρ Θ G ρ jθ G ] = arg mn D ρ ρ Θ G = [ ρ jθ G ] 1/ I [ρ j Θ G ] 1/ I. 7 The resulng Kullback-ebler-opmal dsrbuon ρ Θ G s hus a geomerc average of neghbors dsrbuons. In he Bayesan dffuson framework hs mergng concdes wh he combne sep preferably performed on neghbors poseror dsrbuons [25]. However n our case he local esmaon mehod s erave and 7 may be used beween any subsequen eraons o speed up convergence. Ths promsng opc s posponed o furher research. Generally he varaonal esmaon of he poseror dsrbuon s naccurae n he early sage of he onlne learnng and gradually becomes accurae as learnng proceeds [24]. Therefore s reasonable o sar wh a hgher number of local eraons o speed up hs convergence and o decrease laer down o one eraon beween wo dffuson seps. The proposed collaboraon by fuson of poseror dsrbuons can be seen as a weghed Bayesan learnng ha conrbues o he nference process hs wll become more apparen n he ongong secon. 4. COMPONENTS FROM THE EXPONENTIA FAMIY If he mxure componens belong o he exponenal famly of dsrbuons he local esmaon can ake he form of a message passng algorhm. In general any random varable y has an exponenal famly dsrbuon wh a parameer ϑ f s probably densy funcon can be wren n he form py ϑ = exp [η T yy Aη ky ] 8 where η ηϑ s a naural parameer T yy s a suffcen sasc Aη s a log-paron normalzng funcon and ky s a funcon of y. The parcular form s no unque. The Bayesan esmaon of ϑ s analycally racable f he pror dsrbuon of ϑ s conjugae.e. parameerzed by hyperparameers ξ 1 of he same sze as T yy and real scalar ν 1 and wh a densy of he form π ϑ ϑ ξ 1 ν 1 = exp [η ξ 1 ν 1Aη] exp [hξ 1 ν 1] 9 where Aη s he same funcon as n he exponenal famly dsrbuon and hξ 1 ν 1 s a known funcon. The poseror dsrbuon of ϑ s hen gven by updaed hyperparameers ξ = ξ 1 + T yy and ν = ν In he local esmaon sep he componen denses p k y θ k from Equaon 1 and he convenen pror dsrbuons for θ k are rewren o compable forms accordng o 8 and 9. Smlarly he componen ndcaors for y deermnng whch componen generaed he acual y are modelled by mulnomal dsrbuons wh he probables.e. componen weghs φ k provded by he conjugae Drchle dsrbuon. The local esmaon algorhm hen eraes by passng messages expecaons from he pror dsrbuons o he componens and mulnomal ndcaor models whch n urn reurn messages conanng he suffcen sascs updang he pror hyperparameers n a sense 10. Ths s known as he varaonal message passng [17]. From 9 and he dffuson opmzaon sep 7 follows ha mergng of poseror dsrbuons akes he form ξ = 1 ξ j and ν = 1 ν j. I I 364

4 Remnd ha can be performed beween any wo subsequen eraons of he message passng algorhm e.g. o speed up convergence or afer he local esmaon sep o save communcaon resources. Also remark s prncpal smlary wh he Bayesan updae under conjugacy 10 hs s where he cornersone of hs mergng les. 5. SIMUATION EXAMPE Ths example demonsraes he sequenal dffuson esmaon of a normal mxure model by a nework conssng of 16 nodes. Is opology s depced n Fgure 1. The mxure has he form K y Θ Θ G φ k N µ k Σ k k=1 where µ k and Σ k are he mean vecors and covarance marces. The frs componen observed by all nework nodes has parameers [ ] [ ] ε 0.9 µ 1 = Σ 5 1 = ε where ε ε Exp0.5 are..d. random samples from he exponenal dsrbuon. The second componen s observed by nodes { } only. Ther parameers are [ [ ] ε 0.8 µ 2 = Σ 0] 2 = ε where ε ε Exp0.5 are..d. random samples from he exponenal dsrbuon. The componen weghs n nodes { } are φ = [ ]. The parameers ses are Θ G = {µ 1} and Θ = {µ 2 Σ 1 Σ 2 π }. Noe ha here s a poenal for cooperaon n esmaon of µ 2 and φ. The esmaon sars from he nal normal N nverse- Wshar W and Drchle Dr pror dsrbuons µ 1 N [ 7 7] 15 I [2 2] µ 2 N [7 7] 15 I [2 2] Σ 1 Σ 2 W I [2 2] π Dr[1 1]. The sngle-componen nodes provde her nformaon o all neghbors bu ncorporae ρ Θ G only from oher sngle-componen nodes observaons were generaed for each node he esmaon sars when he frs 100 observaons are receved. A each me 5 eraons of he local esmaon sep are performed hen he dffuson opmzaon follows. As a performance measure we employ MSE averaged over he nework denoed by AMSE. The evoluon of AMSEs under cooperaon and no-cooperaon are depced n Fg. 2 for µ 1 and µ 2 Fg. 3 for Σ 1 and Σ 2 and Fg. 4 for π. We conclude ha he proposed mehod leads o a sgnfcan mprovemen n esmaon of µ 1. Moreover smulaneously helps he esmaon of oher parameers. Snce he memory lengh for y would be lmed n pracce we also performed a smulaon wh a queue-ype memory for 100 observaons. The resuls were very smlar o hose presened here he mehod performed very slghly worse n erms of AMSE. Fnally we remark ha [5] provdes an algorhm ha allows dsrbued mxure esmaon wh nhomogeneous componen weghs across he nework bu assumes dencal componens Fg. 1. Topology of he dffuson nework Coop. No coop. µ 1 µ Fg. 2. Evoluon of logarhm of AMSE of µ 1 and µ 2 under cooperaon Coop. and no cooperaon No coop Coop. No coop. 0.5 Fg. 3. Evoluon of logarhm of AMSE of Σ 1 and Σ 2 under cooperaon Coop. and no cooperaon No coop Σ Σ2 Coop. No coop. Fg. 4. Evoluon of logarhm of AMSE of π under cooperaon Coop. and no cooperaon No coop.. 365

5 6. REFERENCES [1] A. H. Sayed Adapve neworks Proceedngs of he IEEE vol. 102 no. 4 pp Apr [2] C. H. Papadmrou Compuaonal Complexy Addson Wesley [3] A. H Sayed Adapaon learnng and opmzaon over neworks Foundaons and Trends n Machne earnng vol. 7 no. 4-5 pp [4] P. Braca S. Marano and V. Maa Runnng consensus n wreless sensor neworks n 11h Inernaonal Conference on Informaon Fuson June 2008 pp [5] D. Gu Dsrbued EM Algorhm for Gaussan Mxures n Sensor Neworks IEEE Transacons on Neural Neworks vol. 19 no. 7 pp July [6] B. Safarnejadan M. B. Menhaj and M. Karrar Dsrbued unsupervsed Gaussan mxure learnng for densy esmaon n sensor neworks IEEE Transacons on Insrumenaon and Measuremen vol. 59 no. 9 pp Sep [7] S. S. Perera A. Pages-Zamora and R. opez-valcarce A dffuson-based dsrbued EM algorhm for densy esmaon n wreless sensor neworks n 2013 IEEE Inernaonal Conference on Acouscs Speech and Sgnal Processng. May 2013 pp [8] B. Safarnejadan M.B. Menhaj and M. Karrar Dsrbued varaonal Bayesan algorhms for Gaussan mxures n sensor neworks Sgnal Processng vol. 90 no. 4 pp Apr [9] B. Safarnejadan and M. B. Menhaj Dsrbued densy esmaon n sensor neworks based on varaonal approxmaons Inernaonal Journal of Sysems Scence Apr [10] K. Dedecus J. Rechl and P.M. Djurć Sequenal esmaon of mxures n dffuson neworks IEEE Sgnal Processng eers vol. 22 no. 2 pp [11] J. Chen C. Rchard and A.H. Sayed Dffuson MS over mulask neworks IEEE Transacons on Sgnal Processng vol. 63 no. 11 pp [12] J. Plaa-Chaves N. Bogdanovć and K. Berberds Dsrbued dffuson-based MS for node-specfc adapve parameer esmaon IEEE Transacons on Sgnal Processng vol. 63 no. 13 pp July [13] R. Nassf C. Rchard A. Ferrar and A.H. Sayed. Mulask dffuson MS wh sparsy-based regularzaon. Proc. IEEE Inernaonal Conference on Acouscs Speech and Sgnal Processng ICASSP [14] J. Chen C. Rchard A. O. Hero and A. H. Sayed Dffuson MS for mulask problems wh overlappng hypohess subspaces n 2014 IEEE Inernaonal Workshop on Machne earnng for Sgnal Processng MSP Sep pp. 1 6 IEEE. [15] X. Zhao and A. H. Sayed Dsrbued cluserng and learnng over neworks IEEE Transacons on Sgnal Processng vol. 63 no. 13 pp July [16] S. Khawam A.M. Zoubr and A.H. Sayed. Decenralzed cluserng over adapve neworks. Proc European Sgnal Processng Conference [17] J. Wnn and C.M. Bshop Varaonal message passng Journal of Machne earnng Research vol. 6 pp [18] C.M. Bshop Paern Recognon and Machne earnng Sprnger [19] M. I. Jordan Z. Ghahraman T. S. Jaakkola and. K. Saul An nroducon o varaonal mehods for graphcal models Machne earnng vol. 37 no. 2 pp [20] T. S. Jaakkola Tuoral on varaonal approxmaon mehods n Advanced Mean Feld Mehods: Theory and Pracce 2000 pp [21] S. Boyd and. Vandenberghe Convex Opmzaon Cambrdge Unversy Press [22] M. Hoffman F. R. Bach and D. M. Ble Onlne learnng for laen Drchle allocaon n Advances n Neural Informaon Processng Sysems 23 pp Curran Assocaes Inc [23] T. Broderck N. Boyd A. Wbsono A. C. Wlson and M.I. Jordan Sreamng varaonal Bayes n Advances n Neural Informaon Processng Sysems 26 pp Curran Assocaes Inc [24] M. Sao Onlne model selecon based on he varaonal Bayes Neural Compuaon vol. 13 no. 7 pp [25] K. Dedecus and V. Sečkárová Dynamc dffuson esmaon n exponenal famly models IEEE Sgnal Processng eers vol. 20 no. 11 pp Nov [26] D. A. Knowles and T. Mnka Non-conjugae varaonal message passng for mulnomal and bnary regresson n Advances n Neural Informaon Processng Sysems 24 pp Curran Assocaes Inc [27] M. P. Wand Fully smplfed mulvarae normal updaes n non-conjugae varaonal message passng Journal of Machne earnng Research vol. 15 pp

Exercise 8. Panel Data (Answers) (the values of any variable are correlated over time for the same individuals)

Exercise 8. Panel Data (Answers) (the values of any variable are correlated over time for the same individuals) ercse 8. Panel Daa Answers. Gven m a,.,. I follows ha Varm m s Varm Varm s - Covm, m s Wh anel daa s lkel ha Covm, m s > he vales of an varable are correlaed over me for he same ndvdals Wh ooled daa hen

More information

NoC Impact on Design Methodology. Characteristics of NoC-based design. Timing Closure in Traditional VLSI

NoC Impact on Design Methodology. Characteristics of NoC-based design. Timing Closure in Traditional VLSI NoC Impac on Desgn Mehodology Avnoam Kolodny NoC as Means o Handle Complexy (Oucome of Moore s Law) Prncples for dealng wh complexy: Absracon Herarchy egulary Desgn Mehodology 000 00 Termnals per module

More information

A study of volatility risk

A study of volatility risk Journal of Fnance and Accounng 2014; 2(1): 1-10 Publshed onlne January 30, 2014 (hp://www.scencepublshnggroup.com/j/jfa) do: 10.11648/j.jfa.20140201.11 A sudy of volaly rsk Kala Lama *, Jlan Faouz Graduae

More information

Kinematics. Overview. Forward Kinematics. Example: 2-Link Structure. Forward Kinematics. Forward Kinematics

Kinematics. Overview. Forward Kinematics. Example: 2-Link Structure. Forward Kinematics. Forward Kinematics Overvew Knemacs Tomas Funkouser Prnceon Unversy C0S 46, Sprng 004 Knemacs Consders only moon Deermned by posons, veloces, acceleraons Dynamcs Consders underlyng forces Compue moon from nal condons and

More information

Paul M. Sommers David U. Cha And Daniel P. Glatt. March 2010 MIDDLEBURY COLLEGE ECONOMICS DISCUSSION PAPER NO

Paul M. Sommers David U. Cha And Daniel P. Glatt. March 2010 MIDDLEBURY COLLEGE ECONOMICS DISCUSSION PAPER NO AN EMPIRICAL TEST OF BILL JAMES S PYTHAGOREAN FORMULA by Paul M. Sommers David U. Cha And Daniel P. Gla March 2010 MIDDLEBURY COLLEGE ECONOMICS DISCUSSION PAPER NO. 10-06 DEPARTMENT OF ECONOMICS MIDDLEBURY

More information

Authors: Christian Panzer, Gustav Resch, Reinhard Haas, Patrick Schumacher. Energy Economics Group, Vienna University of Technology

Authors: Christian Panzer, Gustav Resch, Reinhard Haas, Patrick Schumacher. Energy Economics Group, Vienna University of Technology Dervng fuure suppor shemes of RES, y onsderng he os evoluon of RES ehnologes a volale energy and raw maeral pres aompaned y ehnologal learnng mpas Auhors: Chrsan Panzer, Gusav Resh, Renhard Haas, Park

More information

THE PERFORMANCE OF ALTERNATIVE INTEREST RATE RISK MEASURES AND IMMUNIZATION STRATEGIES UNDER A HEATH-JARROW-MORTON FRAMEWORK

THE PERFORMANCE OF ALTERNATIVE INTEREST RATE RISK MEASURES AND IMMUNIZATION STRATEGIES UNDER A HEATH-JARROW-MORTON FRAMEWORK THE PERFORMANCE OF ALTERNATIVE INTEREST RATE RISK MEASURES AND IMMUNIZATION STRATEGIES UNDER A HEATH-JARROW-MORTON FRAMEWORK By Şenay Ağca Dsseraon submed o he faculy of Vrgna Polyechnc Insue and Sae Unversy

More information

2. Literature Review Theory of Investment behavior

2. Literature Review Theory of Investment behavior Effec of Equy Rsk Facors on he Reurn of Sock Porfolos of Companes Lsed a he Narob Secures Exchange n Kenya Beween 2009 and 2014 Mchael.N. Njogo 1* Edde Smyu 2 Sephen. T. Wahaka 3 Deparmen of Accounng and

More information

The influence of settlement on flood prevention capability of the floodcontrol wall along the Bund in Shanghai

The influence of settlement on flood prevention capability of the floodcontrol wall along the Bund in Shanghai IAG6 Paper number 8 The nfluence of selemen on flood prevenon capably of he floodconrol wall along he Bund n Shangha 3 BAO CHN, LIN-D YANG & HUI-FNG CHNG 3 Tong Unversy. (e-mal: chenbao@ong.edu.cn) Tong

More information

Interval Type-1 Non-Singleton Type-2 TSK Fuzzy Logic Systems Using the Kalman Filter - Back Propagation Hybrid Learning Mechanism

Interval Type-1 Non-Singleton Type-2 TSK Fuzzy Logic Systems Using the Kalman Filter - Back Propagation Hybrid Learning Mechanism Inerval Type- Non-Sngleon Type- TSK Fuzzy Logc Sysems Usng he Kalman Fler - Bac Propagaon Hybrd Learnng Mechansm Gerardo M Mendez, Angeles Hernández, Marcela Casllo-Leal, Danel Loras, Insuo Tecnologco

More information

Geometrical Description of Signals GEOMETRICAL DESCRIPTION OF SIGNALS. Geometrical/Vectorial Representation. Coder. { } S i SOURCE CODER RECEIVER

Geometrical Description of Signals GEOMETRICAL DESCRIPTION OF SIGNALS. Geometrical/Vectorial Representation. Coder. { } S i SOURCE CODER RECEIVER UNIVERIÀ DEGLI UDI DI CAGLIARI Unversà degl ud d Cglr Corso d Lure Mgsrle n Ingegner Eleronc e delle elecomunczon UNIVERIÀ DEGLI UDI DI CAGLIARI Geomercl Descrpon of gnls n( { } OURCE CODER RECEIVER GEOMERICAL

More information

A Critical Analysis of the Technical Assumptions of the Standard Micro Portfolio Approach to Sovereign Debt Management

A Critical Analysis of the Technical Assumptions of the Standard Micro Portfolio Approach to Sovereign Debt Management Please ce hs paper as: Blommesen, H. J. and A. Hubg (2012), A Crcal Analyss of he echncal Assumpons of he Sandard Mcro Porfolo Approach o Soveregn Deb Managemen, OECD Workng Papers on Soveregn Borrowng

More information

REIT Markets and Rational Speculative Bubbles: An Empirical Investigation

REIT Markets and Rational Speculative Bubbles: An Empirical Investigation REIT Markes and Raonal Speculave Bubbles: An Emprcal Invesgaon George A. Waers Asssan Professor Deparmen of Economcs Illnos Sae Unversy Normal, IL 61790-4200 gawaer@lsu.edu 309-438-7301 and James E. Payne

More information

World Academy of Science, Engineering and Technology International Journal of Civil and Environmental Engineering Vol:4, No:10, 2010

World Academy of Science, Engineering and Technology International Journal of Civil and Environmental Engineering Vol:4, No:10, 2010 Evaluaon of he Dsplacemen-Based and he Force-Based Adapve Pushover Mehods n Sesmc Response Esmaon of Irregular Buldngs Consderng Torsonal Effecs R. Abbasna, F. Mohajer Nav, S. Zahedfar, and A. Tajk Absrac

More information

Research Division Federal Reserve Bank of St. Louis Working Paper Series

Research Division Federal Reserve Bank of St. Louis Working Paper Series Research Dvson Federal Reserve Bank of S. Lous Workng Paper Seres Inflaon: Do Expecaons Trump he Gap? Jeremy M. Pger and Rober H. Rasche Workng Paper 006-013A hp://research.slousfed.org/wp/006/006-013.pdf

More information

Productivity and Competitiveness: The Case of Football Teams Playing in the UEFA Champions League

Productivity and Competitiveness: The Case of Football Teams Playing in the UEFA Champions League Ahens Journal of Spors March 2016 Producvy and Compeveness: The Case of Fooball Teams Playng n he UEFA Champons League By Manuel Espa-Escuer Luca Isabel Garca-Cebran The purpose of hs sudy s o evaluae

More information

System GMM estimation with a small sample Marcelo Soto July 15, 2009

System GMM estimation with a small sample Marcelo Soto July 15, 2009 ysem GMM esmaon wh a small sample Marcelo oo July 15, 9 Barcelona Economcs Workng Paper eres Workng Paper nº 395 YTEM GMM ETIMATION WITH A MALL AMPLE Marcelo oo July 9 Properes of GMM esmaors for panel

More information

Time-Varying Correlations and Optimal Allocation in Emerging Market Equities for Australian Investors: A Study Using East European Depositary Receipts

Time-Varying Correlations and Optimal Allocation in Emerging Market Equities for Australian Investors: A Study Using East European Depositary Receipts Tme-Varyng Correlaons and Opmal Allocaon n Emergng Marke Eques for Ausralan Invesors: A Sudy Usng Eas European Deposary Receps Auhor Gupa, Rakesh, Jhendranahan, Thadavlll Publshed 2008 Journal Tle Inernaonal

More information

Identification of trend patterns related to the dynamics of competitive intelligence budgets (the case of Romanian software industry)

Identification of trend patterns related to the dynamics of competitive intelligence budgets (the case of Romanian software industry) Advances n Manageen & Appled Econocs, vol., no.3, 01, 133-16 ISSN: 179-7544 (prn verson), 179-755 (onlne) Scenpress Ld, 01 Idenfcaon of rend paerns relaed o he dynacs of copeve nellgence budges (he case

More information

The Impact of Demand Correlation on Bullwhip Effect in a Two-stage Supply Chain with Two Retailers

The Impact of Demand Correlation on Bullwhip Effect in a Two-stage Supply Chain with Two Retailers The Impac of Demand Correlaon on Bullwhp Effec n a Two-age Supply Chan wh Two Realer Janhua J Huafeng Je Zhang and Cucu eng Ana College of Economc and anagemen Shangha Jao Tong Unvery Shangha 0005 Chna

More information

CS 2750 Machine Learning. Lecture 4. Density estimation. CS 2750 Machine Learning. Announcements

CS 2750 Machine Learning. Lecture 4. Density estimation. CS 2750 Machine Learning. Announcements CS 75 Machne Learnng Lecture 4 ensty estmaton Mlos Hauskrecht mlos@cs.ptt.edu 539 Sennott Square CS 75 Machne Learnng Announcements Homework ue on Wednesday before the class Reports: hand n before the

More information

Betting Against Beta

Betting Against Beta Beng Agans Bea Andrea Frazzn and Lasse Heje Pedersen * Ths draf: February 4, 2013 Absrac. We presen a model wh leverage and margn consrans ha vary across nvesors and me. We fnd evdence conssen wh each

More information

PROGRAM ON HOUSING AND URBAN POLICY

PROGRAM ON HOUSING AND URBAN POLICY Insue of Busness and conomc esearch Fsher Cener for eal sae and Urban conomcs POGAM O OUSIG AD UBA POICY WOKIG PAP SIS WOKIG O. W-7 DGIG OUSIG ISK By Peer nglund Mn wang John M. Qugley December 2 These

More information

Using Rates of Change to Create a Graphical Model. LEARN ABOUT the Math. Create a speed versus time graph for Steve s walk to work.

Using Rates of Change to Create a Graphical Model. LEARN ABOUT the Math. Create a speed versus time graph for Steve s walk to work. 2.4 Using Raes of Change o Creae a Graphical Model YOU WILL NEED graphing calculaor or graphing sofware GOAL Represen verbal descripions of raes of change using graphs. LEARN ABOUT he Mah Today Seve walked

More information

Interpreting Sinusoidal Functions

Interpreting Sinusoidal Functions 6.3 Inerpreing Sinusoidal Funcions GOAL Relae deails of sinusoidal phenomena o heir graphs. LEARN ABOUT he Mah Two sudens are riding heir bikes. A pebble is suck in he ire of each bike. The wo graphs show

More information

THE REMOVAL OF ONE-COMPONENT BUBBLES FROM GLASS MELTS IN A ROTATING CYLINDER

THE REMOVAL OF ONE-COMPONENT BUBBLES FROM GLASS MELTS IN A ROTATING CYLINDER Orgnal papers THE REMOVAL OF ONE-COMPONENT BUBBLES FROM GLASS MELTS IN A ROTATING CYLINDER LUBOMÍR NÌMEC, VLADISLAVA TONAROVÁ Laboraory of Inorganc Maerals, Jon Workplace of he Insue of Chemcal Technology

More information

A Probabilistic Approach to Worst Case Scenarios

A Probabilistic Approach to Worst Case Scenarios A Probabilisic Approach o Wors Case Scenarios A Probabilisic Approach o Wors Case Scenarios By Giovanni Barone-Adesi Universiy of Albera, Canada and Ciy Universiy Business School, London Frederick Bourgoin

More information

What the Puck? an exploration of Two-Dimensional collisions

What the Puck? an exploration of Two-Dimensional collisions Wha he Puck? an exploraion of Two-Dimensional collisions 1) Have you ever played 8-Ball pool and los he game because you scrached while aemping o sink he 8-Ball in a corner pocke? Skech he sho below: Each

More information

BISI Wear Dance Art Clothing. September 10, April 27, 2019

BISI Wear Dance Art Clothing. September 10, April 27, 2019 wwwbs-dancercom Elk Grove Vllage, IL 60007 Meacham Beserfeld) Phone: 8473636398 nfo@bs-dancercom brng shake 19 2018-20! Creave apparel, ye fashonable o DANCE n! We offer BISI dance are from head o oe for

More information

The Yen and The Competitiveness of Japanese Industries and Firms. March 4, 2008 (preliminary draft) Robert Dekle Department of Economics USC

The Yen and The Competitiveness of Japanese Industries and Firms. March 4, 2008 (preliminary draft) Robert Dekle Department of Economics USC 1 The Yen and The Compeveness of apanese Indusres and Frms March 4 2008 prelmnary draf Rober Dekle Deparmen of Economcs USC Kyoj Fukao Insue of Economc Research Hosubash Unversy Prepared for he ESRI Workshop

More information

Strategic Decision Making in Portfolio Management with Goal Programming Model

Strategic Decision Making in Portfolio Management with Goal Programming Model American Journal of Operaions Managemen and Informaion Sysems 06; (): 34-38 hp://www.sciencepublishinggroup.com//aomis doi: 0.648/.aomis.0600.4 Sraegic Decision Making in Porfolio Managemen wih Goal Programming

More information

Proportional Reasoning

Proportional Reasoning Proporional Reasoning Focus on Afer his lesson, you will be able o... solve problems using proporional reasoning use more han one mehod o solve proporional reasoning problems When you go snowboarding or

More information

English Premier League (EPL) Soccer Matches Prediction using An Adaptive Neuro-Fuzzy Inference System (ANFIS) for

English Premier League (EPL) Soccer Matches Prediction using An Adaptive Neuro-Fuzzy Inference System (ANFIS) for Englsh Premer League (EPL) Soccer Matches Predcton usng An Adaptve Neuro-Fuzzy Inference System (ANFIS) for Amadn, F. I 1 and Ob, J.C. 2 Department of Computer Scence, Unversty of Benn, Benn Cty. Ngera.

More information

Automatic air-main charging and pressure control system for compressed air supplies

Automatic air-main charging and pressure control system for compressed air supplies Auomaic air-main charging and pressure conrol sysem for compressed air supplies Type PCS A module from he sysem -vacorol Swiching on-off a compressed air uni in a compressed air supply generally akes place

More information

3.10 Convected Coordinates

3.10 Convected Coordinates Seco.0.0 Coveced Coordaes Some of he mpora resuls from secos.-.9 are ow re-epressed erms of coveced coordaes. As before, ay relaos epressed symbolc form hold also he coveced coordae sysem..0. The Sress

More information

DO BUBBLES AND TIME-VARYING RISK PREMIUMS AFFECT STOCK PRICES? A KALMAN FILTER APPROACH a

DO BUBBLES AND TIME-VARYING RISK PREMIUMS AFFECT STOCK PRICES? A KALMAN FILTER APPROACH a DO BUBBS AD TIM-VARYIG RISK RMIUMS AFFCT STOCK RICS? A KAMA FITR AROACH a Dr. -Tarn Cen, Acaema Snca, Tawan Dr. C. James Hueng, Unversy o Alabama, USA Dr. Cen-u Je n, aonal Tawan Unversy, Tawan ABSTRACT

More information

Time & Distance SAKSHI If an object travels the same distance (D) with two different speeds S 1 taking different times t 1

Time & Distance SAKSHI If an object travels the same distance (D) with two different speeds S 1 taking different times t 1 www.sakshieducaion.com Time & isance The raio beween disance () ravelled by an objec and he ime () aken by ha o ravel he disance is called he speed (S) of he objec. S = = S = Generally if he disance ()

More information

CHAPTER TEST REVIEW, LESSONS 4-1 TO 4-5

CHAPTER TEST REVIEW, LESSONS 4-1 TO 4-5 IB PHYSICS Name: DEVIL PHYSICS Perio: Dae: BADDEST CLASS ON CAMPUS CHAPTER TEST REVIEW, LESSONS 4- TO 4-5 S. Waer waves a he surface of a pon pass a floaing log of lengh L. The log is a res relaive o he

More information

CS 410/584, Algorithm Design & Analysis, Lecture Notes 5

CS 410/584, Algorithm Design & Analysis, Lecture Notes 5 CS 4/584,, Ford-Fulkeron Mehod Flow maximizaion in a nework (graph) wih capaciie Baic idea: Find a pah from ource o arge ha ill ha flow capaciy (augmening pah) Add he maximum flow allowed along hi pah

More information

APPLYING BI-OBJECTIVE SHORTEST PATH METHODS TO MODEL CYCLE ROUTE-CHOICE ABSTRACT

APPLYING BI-OBJECTIVE SHORTEST PATH METHODS TO MODEL CYCLE ROUTE-CHOICE ABSTRACT Andrea Rah, Chrs Van Houe, Judh Y.T. Wang, and Mahas Ehrgo APPLYING BI-OBJECTIVE SHORTEST PATH METHODS TO MODEL CYCLE ROUTE-CHOICE Andrea Rah 1, Chrs Van Houe 2, Judh Y. T. Wang 3, and Mahas Ehrgo 4 The

More information

Capacity Utilization Metrics Revisited: Delay Weighting vs Demand Weighting. Mark Hansen Chieh-Yu Hsiao University of California, Berkeley 01/29/04

Capacity Utilization Metrics Revisited: Delay Weighting vs Demand Weighting. Mark Hansen Chieh-Yu Hsiao University of California, Berkeley 01/29/04 Capaciy Uilizaion Merics Revisied: Delay Weighing vs Demand Weighing Mark Hansen Chieh-Yu Hsiao Universiy of California, Berkeley 01/29/04 1 Ouline Inroducion Exising merics examinaion Proposed merics

More information

The Comparison of Outlier Detection in Multiple Linear Regression

The Comparison of Outlier Detection in Multiple Linear Regression Poceedngs of he Wold Congess on Engneeng 2012 Vol I, July 4-6, 2012, London, U.K. The Compason of Oule Deecon n Mulple Lnea Regesson Pmpan Amphanhong Absac Fou Oule deecon appoaches n mulple lnea egessons

More information

AP Physics 1 Per. Unit 2 Homework. s av

AP Physics 1 Per. Unit 2 Homework. s av Name: Dae: AP Physics Per. Uni Homework. A car is driven km wes in hour and hen 7 km eas in hour. Eas is he posiive direcion. a) Wha is he average velociy and average speed of he car in km/hr? x km 3.3km/

More information

Morningstar Investor Return

Morningstar Investor Return Morningsar Invesor Reurn Morningsar Mehodology Paper March 3, 2009 2009 Morningsar, Inc. All righs reserved. The informaion in his documen is he propery of Morningsar, Inc. Reproducion or ranscripion by

More information

Analyzing the Bullwhip Effect in a Supply Chain with ARMA(1,1) Demand Using MMSE Forecasting

Analyzing the Bullwhip Effect in a Supply Chain with ARMA(1,1) Demand Using MMSE Forecasting Inernaona Journa on Advances n Informaon cences and ervce cences Voume Number March Anayzng he Buwhp Effec n a uppy han wh ARMA emand Usng MME Forecasng huanxu Wang choo of Economy and Managemen hangha

More information

Brand Selection and its Matrix Structure -Expansion to the Second Order Lag-

Brand Selection and its Matrix Structure -Expansion to the Second Order Lag- Journl of Compuons & Modellng vol.3 no.3 3 87-99 ISS: 79-765 (prn) 79-885 (onlne) Scenpress Ld 3 Brnd Selecon nd s Mrx Srucure -Expnson o he Second Order Lg- Kuhro Tkeysu nd Yuk Hguch src Focusng h consumers

More information

Stock Return Expectations in the Credit Market

Stock Return Expectations in the Credit Market Sock Reurn Expecaions in he Credi Marke Hans Bysröm * Sepember 016 In his paper we compue long-erm sock reurn expecaions (across he business cycle) for individual firms using informaion backed ou from

More information

Instruction Manual. Rugged PCB type. 1 Terminal Block. 2 Function. 3 Series Operation and Parallel Operation. 4 Assembling and Installation Method

Instruction Manual. Rugged PCB type. 1 Terminal Block. 2 Function. 3 Series Operation and Parallel Operation. 4 Assembling and Installation Method Rugged PCB ype Insrucion Manual 1 Terminal Block Funcion.1...4.5.6.7 Inpu volage range Inrush curren limiing Overcurren proecion Overvolage proecion Oupu volage adjusmen range Isolaion Remoe ON/OFF E9

More information

INSTRUCTIONS FOR USE. This file can only be used to produce a handout master:

INSTRUCTIONS FOR USE. This file can only be used to produce a handout master: INSTRUCTIONS OR USE This file can only be used o produce a handou maser: Use Prin from he ile menu o make a prinou of he es. You may no modify he conens of his file. IMPORTNT NOTICE: You may prin his es

More information

Zelio Control Measurement Relays RM4L Liquid Level Relays

Zelio Control Measurement Relays RM4L Liquid Level Relays Zelio Conrol Measuremen elays FNCTIONS These devices monior he levels of conducive liquids. They conrol he acuaion of pumps or valves o regulae levels; hey are also suiable for proecing submersible pumps

More information

ADDITIONAL INSTRUCTIONS FOR ISU SYNCHRONIZED SKATING TECHNICAL CONTROLLERS AND TECHNICAL SPECIALISTS

ADDITIONAL INSTRUCTIONS FOR ISU SYNCHRONIZED SKATING TECHNICAL CONTROLLERS AND TECHNICAL SPECIALISTS A ADDITIONAL INSTRUCTIONS FOR ISU SYNCHRONIZED SKATING TECHNICAL CONTROLLERS AND TECHNICAL SPECIALISTS DIFFICULTY GROUPS OF FEATURES 1 DEFINITIONS: Change of Rotaton: Refers to TURNS or LINKING STEPS rotatng

More information

Making Sense of Genetics Problems

Making Sense of Genetics Problems Bio 101 Ms. Bledsoe Making Sense of Geneics roblems Monohbrid crosses Le s sar wih somehing simle: crossing wo organisms and waching how one single rai comes ou in he offsring. Le s use eas, as Mendel

More information

Reduced drift, high accuracy stable carbon isotope ratio measurements using a reference gas with the Picarro 13 CO 2 G2101-i gas analyzer

Reduced drift, high accuracy stable carbon isotope ratio measurements using a reference gas with the Picarro 13 CO 2 G2101-i gas analyzer Reduced drft, hgh accuracy stable carbon sotope rato measurements usng a reference gas wth the Pcarro 13 CO 2 G2101- gas analyzer Chrs Rella, Ph.D. Drector of Research & Development Pcarro, Inc., Sunnyvale,

More information

Minimum Mean-Square Error (MMSE) and Linear MMSE (LMMSE) Estimation

Minimum Mean-Square Error (MMSE) and Linear MMSE (LMMSE) Estimation Minimum Mean-Square Error (MMSE) and Linear MMSE (LMMSE) Estimation Outline: MMSE estimation, Linear MMSE (LMMSE) estimation, Geometric formulation of LMMSE estimation and orthogonality principle. Reading:

More information

QUANTITATIVE FINANCE RESEARCH CENTRE. Optimal Time Series Momentum QUANTITATIVE FINANCE RESEARCH CENTRE QUANTITATIVE F INANCE RESEARCH CENTRE

QUANTITATIVE FINANCE RESEARCH CENTRE. Optimal Time Series Momentum QUANTITATIVE FINANCE RESEARCH CENTRE QUANTITATIVE F INANCE RESEARCH CENTRE QUANTITATIVE FINANCE RESEARCH CENTRE QUANTITATIVE F INANCE RESEARCH CENTRE QUANTITATIVE FINANCE RESEARCH CENTRE Research Paper 353 January 15 Opimal Time Series Momenum Xue-Zhong He, Kai Li and Youwei

More information

Evaluation of a Center Pivot Variable Rate Irrigation System

Evaluation of a Center Pivot Variable Rate Irrigation System Evaluaton of a Center Pvot Varable Rate Irrgaton System Ruxu Su Danel K. Fsher USDA-ARS Crop Producton Systems Research Unt, Stonevlle, Msssspp Abstrat: Unformty of water dstrbuton of a varable rate center

More information

PERSONAL VERSION. Readers are kindly asked to use the official publication in references.

PERSONAL VERSION. Readers are kindly asked to use the official publication in references. PSONAL VSION Ths s a so called personal verson auhors s anuscrp as acceped for publshng afer he revew process bu pror o fnal layou and copyedng of he arcle. Bu, H & Vrk, N S 204, ' Lqudy and asse prces:

More information

The t-test. What We Will Cover in This Section. A Research Situation

The t-test. What We Will Cover in This Section. A Research Situation The -es 1//008 P331 -ess 1 Wha We Will Cover in This Secion Inroducion One-sample -es. Power and effec size. Independen samples -es. Dependen samples -es. Key learning poins. 1//008 P331 -ess A Research

More information

On the Convergence of Bound Optimization Algorithms

On the Convergence of Bound Optimization Algorithms On the Convergence of ound Optmzaton lgorthms Ruslan Salakhutdnov Sam Rowes Unversty of Toronto 6 Kng s College Rd, MS 3G4, Canada rsalakhu,rowes@cs.toronto.edu Zoubn Ghahraman Gatsby Computatonal Neuroscence

More information

As time goes by - Using time series based decision tree induction to analyze the behaviour of opponent players

As time goes by - Using time series based decision tree induction to analyze the behaviour of opponent players As ime goes by - Using ime series based decision ree inducion o analyze he behaviour of opponen players Chrisian Drücker, Sebasian Hübner, Ubbo Visser, Hans-Georg Weland TZI - Cener for Compuing Technologies

More information

Homework 2. is unbiased if. Y is consistent if. c. in real life you typically get to sample many times.

Homework 2. is unbiased if. Y is consistent if. c. in real life you typically get to sample many times. Econ526 Mulile Choice. Homework 2 Choose he one ha bes comlees he saemen or answers he quesion. (1) An esimaor ˆ µ of he oulaion value µ is unbiased if a. ˆ µ = µ. b. has he smalles variance of all esimaors.

More information

Evaluating Portfolio Policies: A Duality Approach

Evaluating Portfolio Policies: A Duality Approach OPERATIONS RESEARCH Vol. 54, No. 3, May June 26, pp. 45 418 issn 3-364X eissn 1526-5463 6 543 45 informs doi 1.1287/opre.16.279 26 INFORMS Evaluaing Porfolio Policies: A Dualiy Approach Marin B. Haugh

More information

Lifecycle Funds. T. Rowe Price Target Retirement Fund. Lifecycle Asset Allocation

Lifecycle Funds. T. Rowe Price Target Retirement Fund. Lifecycle Asset Allocation Lifecycle Funds Towards a Dynamic Asse Allocaion Framework for Targe Reiremen Funds: Geing Rid of he Dogma in Lifecycle Invesing Anup K. Basu Queensland Universiy of Technology The findings of he Mercer

More information

ANALYSIS OF RELIABILITY, MAINTENANCE AND RISK BASED INSPECTION OF PRESSURE SAFETY VALVES

ANALYSIS OF RELIABILITY, MAINTENANCE AND RISK BASED INSPECTION OF PRESSURE SAFETY VALVES ANALYSIS OF RELIABILITY, MAINTENANCE AND RISK BASED INSPECTION OF PRESSURE SAFETY VALVES Venilon Forunao Francisco Machado Mechanical Engineering Dep, Insiuo Superior Técnico, Av. Rovisco Pais, 049-00,

More information

Economics 487. Homework #4 Solution Key Portfolio Calculations and the Markowitz Algorithm

Economics 487. Homework #4 Solution Key Portfolio Calculations and the Markowitz Algorithm Economics 87 Homework # Soluion Key Porfolio Calculaions and he Markowiz Algorihm A. Excel Exercises: (10 poins) 1. Download he Excel file hw.xls from he class websie. This file conains monhly closing

More information

3. The amount to which $1,000 will grow in 5 years at a 6 percent annual interest rate compounded annually is

3. The amount to which $1,000 will grow in 5 years at a 6 percent annual interest rate compounded annually is 8 h Grade Eam - 00. Which one of he following saemens is rue? a) There is a larges negaive raional number. b) There is a larges negaive ineger. c) There is a smalles ineger. d) There is a smalles negaive

More information

Wladimir Andreff, Madeleine Andreff. To cite this version: HAL Id: halshs https://halshs.archives-ouvertes.

Wladimir Andreff, Madeleine Andreff. To cite this version: HAL Id: halshs https://halshs.archives-ouvertes. Economc predcon of spor performances from he Bejng Olympcs o he 2010 FIFA World Cup n Souh Afrca: he noon of surprsng sporng oucomes Wladmr Andreff Madelene Andreff To ce hs verson: Wladmr Andreff Madelene

More information

KINEMATICS IN ONE DIMENSION

KINEMATICS IN ONE DIMENSION chaper KINEMATICS IN ONE DIMENSION Secion 2.1 Displacemen Secion 2.2 Speed and Velociy 1. A paricle ravels along a curved pah beween wo poins P and Q as shown. The displacemen of he paricle does no depend

More information

ARMENIA: Second Education Quality and Relevance Project (APL2) Procurement Plan. As of March 15, Measu rement Unit.

ARMENIA: Second Education Quality and Relevance Project (APL2) Procurement Plan. As of March 15, Measu rement Unit. No According o he Cos Tables No of Packages Procuremen Mehod Review by WB PRIOR/ Pos Inviaion losure Auhorized Public Disclosure Auhorized Public Disclosure Auhorized Public Disclosure Auhorized Expeced

More information

A Liability Tracking Portfolio for Pension Fund Management

A Liability Tracking Portfolio for Pension Fund Management Proceedings of he 46h ISCIE Inernaional Symposium on Sochasic Sysems Theory and Is Applicaions Kyoo, Nov. 1-2, 214 A Liabiliy Tracking Porfolio for Pension Fund Managemen Masashi Ieda, Takashi Yamashia

More information

On the Convergence of Bound Optimization Algorithms

On the Convergence of Bound Optimization Algorithms On the Convergence of ound Optmzaton lgorthms Ruslan Salakhutdnov Sam Rowes Unversty of Toronto 6 Kng s College Rd, MS 3G4, Canada rsalakhu,rowes@cs.toronto.edu Zoubn Ghahraman Gatsby Computatonal Neuroscence

More information

2. JOMON WARE ROPE STYLES

2. JOMON WARE ROPE STYLES Proceedings of he IIEEJ Image Elecronics and Visual Compuing Workshop 2012 Kuching, Malaysia, November 21-24, 2012 A SIMULATION SYSTEM TO SYNTHESIZE ROPE ROLLING PATTERNS IN A VIRTUAL SPACE FOR RESEARCH

More information

HIGH RESOLUTION DEPTH IMAGE RECOVERY ALGORITHM USING GRAYSCALE IMAGE.

HIGH RESOLUTION DEPTH IMAGE RECOVERY ALGORITHM USING GRAYSCALE IMAGE. HIGH RESOLUTION DEPTH IMAGE RECOVERY ALGORITHM USING GRAYSCALE IMAGE Kazunori Uruma 1, Katsumi Konishi 2, Tomohiro Takahashi 1 and Toshihiro Furukawa 1 1 Graduate School of Engineering, Tokyo University

More information

The impact of foreign players on international football performance

The impact of foreign players on international football performance MPRA Munch Personal RePEc Archve The mpact of foregn players on nternatonal football performance Orhan Karaca Ekonomst Magazne, Research Department October 008 Onlne at http://mpra.ub.un-muenchen.de/11064/

More information

Stock (mis)pricing and Investment Dynamics in Africa

Stock (mis)pricing and Investment Dynamics in Africa Sock msprcng and nvesmen Dynamcs n Afrca Sad Aanda Musapha n 273 July 2017 Workng aper Seres Afrcan Developmen Bank Group Workng aper N o 273 Absrac The sudy ascerans he exen of msprcng n equy porfolos

More information

Bootstrapping Multilayer Neural Networks for Portfolio Construction

Bootstrapping Multilayer Neural Networks for Portfolio Construction Asia Pacific Managemen Review 17(2) (2012) 113-126 Boosrapping Mulilayer Neural Neworks for Porfolio Consrucion Chin-Sheng Huang a*, Zheng-Wei Lin b, Cheng-Wei Chen c www.apmr.managemen.ncku.edu.w a Deparmen

More information

Dynamics of market correlations: Taxonomy and portfolio analysis

Dynamics of market correlations: Taxonomy and portfolio analysis Dynamics of marke correlaions: Taxonomy and porfolio analysis J.-P. Onnela, A. Chakrabori, and K. Kaski Laboraory of Compuaional Engineering, Helsinki Universiy of Technology, P.O. Box 9203, FIN-02015

More information

CALCULATION OF EXPECTED SLIDING DISTANCE OF BREAKWATER CAISSON CONSIDERING VARIABILITY IN WAVE DIRECTION

CALCULATION OF EXPECTED SLIDING DISTANCE OF BREAKWATER CAISSON CONSIDERING VARIABILITY IN WAVE DIRECTION CALCULATION OF EXPECTED SLIDING DISTANCE OF BREAKWATER CAISSON CONSIDERING VARIABILITY IN WAVE DIRECTION SU YOUNG HONG School of Civil, Urban, and Geosysem Engineering, Seoul Naional Universiy, San 56-1,

More information

LSSVM Model for Penetration Depth Detection in Underwater Arc Welding Process

LSSVM Model for Penetration Depth Detection in Underwater Arc Welding Process ISS 1746-7659, England, UK Journal of Informaton and Computng Scence Vol. 5, o. 4, 2010, pp. 271-278 LSSVM Model for Penetraton Depth Detecton n Underwater Arc Weldng Process WeMn Zhang 1, 2, GuoRong Wang

More information

Simulation of Scattering Acoustic Field in Rod and Identify of. Ultrasonic Flaw Detecting Signal

Simulation of Scattering Acoustic Field in Rod and Identify of. Ultrasonic Flaw Detecting Signal 17h Word onference on Nondesrucive Tesing, 5-8 Oc 008, Shanghai, hina Simuaion of Scaering Acousic Fied in Rod and Idenify of Urasonic Faw Deecing Signa Ke-Yi YUAN 1, Wen-Ai SONG 1, Yi-Fang HEN 1 1 Deparmen

More information

The Current Account as A Dynamic Portfolio Choice Problem

The Current Account as A Dynamic Portfolio Choice Problem Public Disclosure Auhorized Policy Research Working Paper 486 WPS486 Public Disclosure Auhorized Public Disclosure Auhorized The Curren Accoun as A Dynamic Porfolio Choice Problem Taiana Didier Alexandre

More information

Transit Priority Strategies for Multiple Routes Under Headway-Based Operations

Transit Priority Strategies for Multiple Routes Under Headway-Based Operations Transi Prioriy Sraegies for Muliple Roues Under Headway-Based Operaions Yongjie Lin, Xianfeng Yang, Gang-Len Chang, and Nan Zou This paper presens a ransi signal prioriy (TSP) model designed o consider

More information

An Alternative Mathematical Model for Oxygen Transfer Evaluation in Clean Water

An Alternative Mathematical Model for Oxygen Transfer Evaluation in Clean Water An Alernaive Mahemaical Model for Oxygen Transfer Evaluaion in Clean Waer Yanjun (John) He 1, PE, BCEE 1 Kruger Inc., 41 Weson Parkway, Cary, NC 27513 Email: john.he@veolia.com ABSTRACT Energy consumpion

More information

The Great Recession in the U.K. Labour Market: A Transatlantic View

The Great Recession in the U.K. Labour Market: A Transatlantic View The Grea Recession in he U.K. Labour Marke: A Transalanic View Michael W. L. Elsby (Edinburgh, Michigan, NBER) Jennifer C. Smih (Warwick) Bank of England, 25 March 2011 U.K. and U.S. unemploymen U.K. unemploymen

More information

Gas Source Localisation by Constructing Concentration Gridmaps with a Mobile Robot

Gas Source Localisation by Constructing Concentration Gridmaps with a Mobile Robot Gas Source Localisaion by Consrucing Concenraion Gridmaps wih a Mobile Robo Achim Lilienhal 1, Tom Ducke 2 1 W.-Schickard-Ins. of Comp. Science, Universiy of Tübingen, D-72076 Tübingen, Germany lilien@informaik.uni-uebingen.de

More information

Reproducing laboratory-scale rip currents on a barred beach by a Boussinesq wave model

Reproducing laboratory-scale rip currents on a barred beach by a Boussinesq wave model See discussions, sas, and auhor profiles for his publicaion a: hps://www.researchgae.ne/publicaion/9977 Reproducing laboraory-scale rip currens on a barred beach by a Boussinesq wave model Aricle in Journal

More information

Basic Systematic Experiments and New Type Child Unit of Anchor Climber: Swarm Type Wall Climbing Robot System

Basic Systematic Experiments and New Type Child Unit of Anchor Climber: Swarm Type Wall Climbing Robot System 2008 IEEE Inernaional Conference on Roboics and Auomaion Pasadena, CA, USA, May 19-23, 2008 Basic Sysemaic Eperimens and New Type Child Uni of Anchor Climber: Swarm Type Wall Climbing Robo Sysem Masaaka

More information

Simulation based approach for measuring concentration risk

Simulation based approach for measuring concentration risk MPRA Munich Personal RePEc Archive Simulaion based approach for measuring concenraion risk Kim, Joocheol and Lee, Duyeol UNSPECIFIED February 27 Online a hp://mpra.ub.uni-muenchen.de/2968/ MPRA Paper No.

More information

RECOMMENDATION FOR INTERCHANGEABLE STUD BOLTS AND TAP END STUDS FOR API SPEC 6A FLANGES

RECOMMENDATION FOR INTERCHANGEABLE STUD BOLTS AND TAP END STUDS FOR API SPEC 6A FLANGES Issue Dae: June 6 15 Revision B June 2010 RECOMMENDAION FOR INERCHANGEABE UD BO AND A END UD FOR AI EC 6A FANGE ECHNICA REOR R501 Revision B AWHEM publicaions may be use by anyone esiring o o so. Every

More information

Explore Graphs of Linear Relations. 1. a) Use a method of your choice to determine how much water will be needed each day of a seven-day cruise.

Explore Graphs of Linear Relations. 1. a) Use a method of your choice to determine how much water will be needed each day of a seven-day cruise. . Graphing Linear Relaions Focus on Aer his lesson, ou will be able o graph linear relaions mach equaions o linear relaions wih graphs solve problems b graphing a linear relaion and analsing he graph Tina

More information

MODEL SELECTION FOR VALUE-AT-RISK: UNIVARIATE AND MULTIVARIATE APPROACHES SANG JIN LEE

MODEL SELECTION FOR VALUE-AT-RISK: UNIVARIATE AND MULTIVARIATE APPROACHES SANG JIN LEE MODEL SELECTION FOR VALUE-AT-RISK: UNIVARIATE AND MULTIVARIATE APPROACHES By SANG JIN LEE Bachelor of Science in Mahemaics Yonsei Universiy Seoul, Republic of Korea 999 Maser of Business Adminisraion Yonsei

More information

2018 GIRLS DISTRICT-SPECIFIC PLAYER DEVELOPMENT GUIDE

2018 GIRLS DISTRICT-SPECIFIC PLAYER DEVELOPMENT GUIDE 2018 GIRLS DISTRICT-SPECIFIC PLAYER DEVELOPMENT GUIDE GENERAL OVERVIEW USA Hockey Grls Player Development Dstrct-Specfc Gude The USA Hockey Grls Player Development Dstrct-Specfc Gude outlnes the 2018 grls

More information

Type Control action Setpoint range Air Weight Volume flow % capacity I n /h kg. Pressure diff. 1) Pa

Type Control action Setpoint range Air Weight Volume flow % capacity I n /h kg. Pressure diff. 1) Pa 7.0/ RL 0 & 0: neumaic volume-flow conroller Used in conjuncion wih an orifice plae or a dnamic pressure sensor and a pneumaic damper drive for conrolling he air volume in air-condiioning ssems. For fixed,

More information

Overview. Do white-tailed tailed and mule deer compete? Ecological Definitions (Birch 1957): Mule and white-tailed tailed deer potentially compete.

Overview. Do white-tailed tailed and mule deer compete? Ecological Definitions (Birch 1957): Mule and white-tailed tailed deer potentially compete. COMPETITION BETWEEN MULE AND WHITE- TAILED DEER METAPOPULATIONS IN NORTH-CENTRAL WASHINGTON E. O. Garon, Kris Hennings : Fish and Wildlife Dep., Univ. of Idaho, Moscow, ID 83844 Maureen Murphy, and Seve

More information

Chapter : Linear Motion 1

Chapter : Linear Motion 1 Te: Chaper 2.1-2.4 Think and Eplain: 1-3 Think and Sole: --- Chaper 2.1-2.4: Linear Moion 1 NAME: Vocabulary: disance, displacemen, ime, consan speed, consan elociy, aerage, insananeous, magniude, ecor,

More information

The safe ships trajectory in a restricted area

The safe ships trajectory in a restricted area Scienific Journals Mariime Universiy of Szczecin Zeszyy Naukowe Akademia Morska w Szczecinie 214, 39(111) pp. 122 127 214, 39(111) s. 122 127 ISSN 1733-867 The safe ships rajecory in a resriced area Zbigniew

More information

Mixture Models & EM. Nicholas Ruozzi University of Texas at Dallas. based on the slides of Vibhav Gogate

Mixture Models & EM. Nicholas Ruozzi University of Texas at Dallas. based on the slides of Vibhav Gogate Mixture Models & EM Nicholas Ruozzi University of Texas at Dallas based on the slides of Vibhav Gogate Previously We looked at -means and hierarchical clustering as mechanisms for unsupervised learning

More information

OPTIMIZATION OF PRESSURE HULLS OF COMPOSITE MATERIALS

OPTIMIZATION OF PRESSURE HULLS OF COMPOSITE MATERIALS OPTIMIZATION OF PRESSURE HULLS OF COMPOSITE MATERIALS J.Franco a, A.Corz a*.a.peña b a Materal Composte Group. Unversdad de Cadz. Avda/Ramon Puyol s/n. 11205 Algecras (Span) *alcorz@caltech.es b Calpe

More information

Development of Urban Public Transit Network Structure Integrating Multi-Class Public Transit Lines and Transfer Hubs

Development of Urban Public Transit Network Structure Integrating Multi-Class Public Transit Lines and Transfer Hubs Developmen of Urban Public Transi Nework Srucure Inegraing Muli-Class Public Transi Lines and Transfer Hubs Zhenbao Wang 1, Anyan Chen 2 1College of Civil Engineering, Hebei Universiy of Engineering Handan,

More information

Name Class Date. Step 2: Rearrange the acceleration equation to solve for final speed. a v final v initial v. final v initial v.

Name Class Date. Step 2: Rearrange the acceleration equation to solve for final speed. a v final v initial v. final v initial v. Skills Workshee Mah Skills Acceleraion Afer you sudy each sample problem and soluion, work ou he pracice problems on a separae shee of paper. Wrie your answers in he spaces provided. In 1970, Don Big Daddy

More information