Playing against Hedge

Size: px
Start display at page:

Download "Playing against Hedge"

Transcription

1 In J Communcaons, ework and Sysem Scences, 4, 7, Publshed Onlne December 4 n ScRes hp://wwwscrporg/journal/jcns hp://dxdoorg/436/jcns475 Playng agans Hedge Mlades E Anagnosou, Mara A ambrou School of Elecrcal and Compuer Engneerng, aonal Techncal Unversy of Ahens, Ahens, Greece Deparmen of Shppng, Trade and Transpor, Unversy of he Aegean, Chos, Greece Emal: mlos@cenralnuagr, mlambrou@aegeangr Receved Ocober 4; revsed ovember 4; acceped December 4 Copyrgh 4 by auhors and Scenfc Research Publshng Inc Ths work s lcensed under he Creave Commons Arbuon Inernaonal cense (CC BY) hp://creavecommonsorg/lcenses/by/4/ Absrac Hedge has been proposed as an adapve scheme, whch gudes he player s hand n a mul-armed band full nformaon game Applcaons of hs game exs n nework pah selecon, load dsrbuon, and nework nerdcon We perform a wors case analyss of he Hedge algorhm by usng an adversary, who wll conssenly selec penales so as o mze he player s loss, assumng ha he adversary s penaly budge s lmed We furher explore he performance of bnary penales, and we prove ha he opmum bnary sraegy for he adversary s o make greedy decsons Keywords Hedge Algorhm, Adversary, Onlne Algorhm, Greedy Algorhm, Perodc Performance, Bnary Penales, Pah Selecon, ework Inerdcon Inroducon The problems of adapve nework pah selecon and load dsrbuon have ofen been consdered as games ha are played smulaneously and ndependenly by agens conrollng flows n a nework A possble absracon of hese and oher relaed problems s he band game In he mul-armed band game [] a player chooses one ou of sraeges (or machnes or opons or arms ) A loss or penaly (or a reward, whch can be s assgned o each sraegy (,,, ) modeled as a negave loss) = afer each round of he game An agen facng repeaed selecons wll possbly ry o explo he so far accumulaed experence A popular algorhm ha can gude he agen n each selecon round s he mulplcave updaes algorhm or Hedge In hs paper we calculae he wors possble performance of Hedge by usng he adversaral echnque, e we nvesgae he behavor of an nellgen adversary, who res o mze he player s cumulave loss In Secon we descrbe Hedge; n Secon we gve a rgorous formulaon of he adversary s problem; n Secon 3 we gve a recursve soluon; and n Secon 4 we presen sample numercal resuls Fnally, n Secon 5 we How o ce hs paper: Anagnosou, ME and ambrou, MA (4) Playng agans Hedge In J Communcaons, ework and Sysem Scences, 7, hp://dxdoorg/436/jcns475

2 M E Anagnosou, M A ambrou explore bnary adversaral sraeges Our man resul s ha he greedy adversaral sraegy s opmal among bnary sraeges The Band Game In a generalzed band game he player s allowed o play mxed sraeges, e o assgn a fracon p (such ha p = = ) of he oal be o opon, hereby geng a loss equal o = p = Alernavely, p can be nerpreed as a probably ha he player assgns he be on opon In he band verson only he oal loss s announced o he player, whle n he full nformaon verson he penaly vecor (,,, ) s announced A game consss of T rounds; a superscrp marks he h ( =,, T ) round Apparenly he player wll ry o mnmze he oal cumulave loss = p = = = () by conrollng he be dsrbuon, e by properly selecng he varables p We use he addonal assumpon ha he loss budge s lmed n each round by seng he consran = = Clearly a player s goal s o mnmze hs or her oal cumulave loss An exremely lucky player, or a player wh nsde nformaon, would selec he mnmum penaly opon n each round and would pu all hs or her be on hs opon, hereby T achevng a oal loss equal o mn The Hedge Algorhm = Que a few algorhmc soluons, whch wll gude he player s hand n he full nformaon game, have appeared n he leraure Freund and Schapre have proposed he Hedge algorhm [] for he full nformaon game Auer, Cesa-Banch, Freund and Schapre have proposed he Exp3 algorhm n [3] Allenberg-eeman and eeman proposed a Hedge varan, he G (Gan-oss) algorhm, for he full nformaon game wh gans and losses [4] Dan, Hayes, and Kakade have proposed he GeomercHedge algorhm n [5], and a modfcaon was proposed by Barle, Dan e al n [6] Recenly Cesa-Banch and ugos have proposed he ComBand algorhm for he band verson [7] A comparson can be found n [8] Hedge manans a vecor w = ( w, w,, w n ) of weghs, such ha w ( =,,, T, and =,,, ) In each round Hedge chooses he be allocaon accordng o he normalzed wegh p = w w When he opponen reveals he loss vecor of hs round, he nex round wegh w + s = deermned so as o reflec he loss resuls, e w + = wβ for some fxed β, such ha β In [9] Auer, Cesa-Banch, Freund and Schapre have proved ha he expeced Hedge performance and he expeced performance of he bes arm dffer a mos by O( Tln ) Freund and Schapre [] have gven a loss upper bound, whch relaes he oal cumulave loss wh he oal loss of he bes arm 3 Compeve Analyss The compeve analyss of an algorhm, whch n hs paper s Hedge, nvolves a comparson of s performance wh he performance of he opmal offlne algorhm In he band game he opmal offlne algorhm, e he opmal player s decsons gven he sequence of all penales n advance, s rval In a gven round he player can jus be everyhng on he opon wh he lowes penaly Accordng o S Iran and A Karln (n Secon 33 of []) a echnque n fndng bounds s o use an adversary who plays agans and concocs an npu, whch forces o ncur a hgh cos Usng an adversary s jus an llusrave way of sayng ha we ry o fnd he wors possble performance of an onlne algorhm In our analyss he adversary res o mze Hedge s oal loss by conrolng he penaly vecor (under a lmed budge) 498

3 M E Anagnosou, M A ambrou 4 Inerpreaons and Applcaons In hs secon we offer some nerpreaons from he areas of ) communcaon neworks and ) ransporaon The general seng of course nvolves a number of opons or arms, whch mus be seleced by a player whou any knowledge of he fuure Band models have been used n que dverse decson makng suaons In [] He, Chen, Wand and u have used a band model for he mzaon of he revenue of a search engne provder, who charges for adversemens on a per-clck bass They have subsequenly defned he armed band problem wh shared nformaon ; arms are paroned n groups and loss nformaon s shared only among players usng arms of he same group In [] Park and ee have used a mul-armed band model for lane selecon n auomaed hghways and auonomous vehcles raffc conrol 4 Traffc oad Dsrbuon Ths frs applcaon example can ake mulple nerpreaons, whch always nvolve a selecon n a compeve envronmen, n whch compeon s lmed I can be seen as ) a pah selecon problem n neworkng, ) a ranspor means (mode) choce or pah selecon problem, 3) a compuaonal load dsrbuon problem, whch we menon n he end of hs secon Frsly, we descrbe he problem n he conex of neworkng Consder smlar ndependen pahs (n he smples case jus parallel lnks), whch jon a par of nodes, A raffc volume equal o Q s sen from o n consecuve me perods or rounds by Q, Q,, Q a populaon of agens Q s he same n each round, bu he allocaon of Q o pahs, e ( ) such ha Q = Q, s dfferen n each round An agen A produces a consan amoun of raffc equal = o A, such ha q Q, n T consecuve rounds, and allocaes a par equal o q ( q = q = ) o he h pah n round The average delay (or cos) experenced by A s raffc n he h round s proporonal o Qq =, f we assume a lnear delay (or cos) model near models are used for smplcy n nework analyss [3] and can be realsc f a nework resource sll operaes n he lnear regon of he delay vs load curve, eg when delay s calculaed n a lnk, whch operaes no very close o capacy Agen A ams a mnmzng he oal delay for s own raffc and may use Hedge o deermne he quanes q n round, assumng ha A knows he performance of s own raffc n each pah n he pas me perod oe ha he mum delay n a round occurs f A pus he whole q n a sngle pah ogeher wh he whole raffc of he compeon, e wh Q ; hen A s average delay n hs round equals Q On he conrary, f Q s evenly dsrbued n all pahs, A s allocaon decson does no really maer, as he average wll be equal o ( q q) ( Q ) = Q Of course he mnmum delay n a round wll occur f A pus he whole q n an empy pah, hereby achevng a zero delay The above problem can also be formulaed as a more general problem of dsrbung workload over a collecon of parallel resources (eg dsrbung jobs o parallel processors) A Blum and C Burch have used he followng movang scenaro n [4]: A process runs on some machne n an envronmen wh machnes n oal The process may move o a dfferen machne a he end of a me nerval The load, whch wll be found on a machne a me round s he penaly fel by he process 4 Inerdcon Alhough an adversary s usually a echncal (fconal) concep, whch serves he wors case analyss of onlne algorhms, n some envronmens a real adversary, who nenonally res o oppose a player, does exs An example s he nerdcon problem We presen a verson of he nerdcon problem n a nework secury conex An aacker aacks resources (eg launches a dsrbued denal of servce aack on nodes, servers, ec, see [5]) by sendng sreams of harmful packes o resource a a rae w (where =,, and w s consan) A defender assgns a defense mechansm of nensy (eg a fler ha s able o deec and avod harmful packes wh a probably proporonal o ) o resource A he end of a me nerval T, eg one day, boh he aacker and he defender revse he flows and he dsrbuon of defense mechansms o resources respecvely, 499

4 M E Anagnosou, M A ambrou based on pas performance Smlar nerpreaons exs n ransporaon nework envronmens, as n border and cusom conrol, ncludng llegal mmgraon conrol An nerdcon problem formulaon can be used n a marme ranspor secury conex: praes aack he vessels raversng a marme roue In [6] Vanek e al assgn he role of he player o he prae The prae operaes n rounds, sarng and fnshng n hs home por In each round he selecs a sea area (arm) o sal o and search for possble vcm vessels A parol force dsrbues he avalable escor resources o sea areas (arms), and prae gans are nversely proporonal o he srengh of he defender s forces on hs area aval forces reallocae her own resources o sea areas Problem Formulaon In hs paper we am a fndng he wors case performance of Hedge Effecvely, we ry o solve he followng problem: Problem Gven a number of opons, an nal normalzed wegh vecor w = ( w, w,, w ), and a T Hedge parameer β, fnd he sequence,,, ha mzes he player s oal cumulave loss where = (,, ) H ( β ) = p () = = s he penaly vecor n round ( =,,, T ) round penaly weghs p are updaed accordng o w w = w = w, p = τ τ β β = w =, such ha, and he h for =,, T and p = w Clearly he objecve funcon () s a funcon of a) he nal weghs = = (3) w, and b) he T varables T + + ndepen-, and c) β Due o he normalzaon of boh weghs and penales here are den varables n oal In he followng we use ( w,, w ;,,,,,, ) T T ( ;,, ) H β whenever s necessary o refer o hese varables w nsead of 3 Recurson or w and he adversary generaes penales Assumng ha a gven round sars wh weghs = ( w,, w ) = (,, ), he nex round wll wll sar wh weghs W ( w, ) = ( W ( w, ),, W ( w, )) = ( = ) j wjβ j= where w β W w,,,, (4) Then, he oal loss of a T round game, whch sars wh weghs w, can be wren as he sum of he losses of a sngle round game, whch sars wh weghs w, and a T round game, whch sars wh weghs W ( w, ) = W ( w, ),, W ( w, ), as follows: ( ) ( w ) = ( w ) + W ( w ) ( ) ;,,, ;, ;,, (5) T T T T T oe ha he erm, whch expresses he conrbuon of he las T rounds, depends only on he updaed weghs provded by he nal round Such a Markovan propery can be generalzed n he followng sense: A T + T round game can be seen as conssng of a T round game g followed by a T round game g, whose nal weghs are he fnal weghs of g, and no more deals abou g are passed o g ; w = w,, he followng recursve Assumng ha he soluon o Problem s T T formula for w can be derved from (5):,, 5

5 where = ( w) = ( w; ) + ( W ( w; )) M E Anagnosou, M A ambrou (6) s he penaly vecor chosen by he adversary n he nal round ; ; ; λ w = λ w,, λ w, where The opmal penales can be compued also recursvely e T ; λ wh weghs w ) The opmal penaly of he nal round w denoes he h opmal penaly of he h opon n he h round of a T round game (sarng opmzes (6) Therefore = + ( ) λ ; w w W w = s apparenly equal o he value of, whch arg ; ; (7) In all oher rounds =,,, T he opmal penales are such ha he oal loss of he res of he game s ( ) T mzed, e such ha T ;, T W w λ ( w ) s acheved Snce he oal loss ; T usng penales λ ( w ), he oal loss T ( ( ; W w, λ ( w ))) s realzed by usng T ; λ W w, λ ; ( w ) nsead Therefore ( ) ( ) ( T ) w s acheved by ; + ; λ w = λ W w, λ ; w =,,, (8) 4 Two Opon Games and umercal Resuls Ths secon we explo he recursve mehodology, whch has been presened n he prevous secon, n order o provde some numercal resuls for wo opon games We compare hese resuls wh avalable bounds n he leraure We consder =, e wo opon games We keep only he ndependen penales n he exended noaon and use he more compac verson ( w ;,,, ) As an example, he loss of a sngle round game s gven by = + ( )( ) w w w ; (9) T Also, snce he nal weghs are w = ( w, w), we smplfy he mum cumulave loss ( w) Assumng losses = =, he nex round wll wll sar wh weghs (, ) W( w ), where T, Then (6) s smplfed o ( w) w o W w and w W( w, β ) = wβ + β () = ( ; ) + ( (, )) w w W w () where = s he penaly chosen by he adversary for he frs opon n he nal round The eraon sars from ( w ), e he loss of a sngle round game In such game he adversary conrols a sngle penaly varable, as he loss s gven by (9) Apparenly he adversary wll choose bnary values, e = = = f w= w > ( w < ), and he mum oal loss s ( w) = { w, w}, e ( ) ( w) w, f w, = w, f w The graph of ( w ) appears as he lowes V-shaped curve n Fgure The fac ha he w s a pecewse lnear funcon of w wh a breakpon (e a sudden change n s slope), creaes even more break- w and so on Therefore, whle s possble o use he aforemenoned recurson n pons n ( w ), () 5

6 M E Anagnosou, M A ambrou T Fgure Plo of ( w) β =,3,,9 and T =,,, (mum loss n a T round game) vs w for order o fnd analycal expressons for he mum oal loss and he assocaed penales, he analyss becomes que complcaed even for small values of he number of rounds T (e n a T + round game) We om hs edous analyss and presen numercal resuls based on he recursve mehodology gven above T w s approxmaed by K + Insead we have mplemened a numercal compuaon based on () T samples n he nerval [, ], e by ( w), where =,,, K and w K funcons ( w; ) and W( w, ) are represened by ( K + ) samples ( m wn ; ) and W( m wn, ) where w = We have used K = Inally we creae ( w) (,,, K) use he resul as npu o () and creae ( ( w) ) Then we use he already calculaed and 3 T o calculae, hen and o calculae, and so on In Fgure we show ( w) he nal wegh w = w n games wh up o en rounds (,,) T ha he shape of ( w) s more neresng for unreasonably small values of β = In he same way he, = by usng (9) We n () as a funcon of T = for dfferen values of β Observe The opmal penales can be deermned by usng formulas (7) and (8) for = In Fgure we draw one of he curves of Fgure ogeher wh he respecve opmal penales The fnal round opmal penaly (e 3;3 3 λ ( w) n hs example) s ceran o be bnary, snce he adversary wll assgn = o he opon wh 3; 3; he greaes wegh facor However, he penales λ ( w) and λ ( w) of he frs wo games are clearly non-bnary 5 Bnary and Greedy Schemes The penaly values n he frs wo rounds n he example of Fgure prove ha he adversary s opmal penales are no necessarly bnary However, n hs example β s unnaurally close o, as n praccal Hedge mplemenaons β s chosen close o ; hs choce acheves a more gradual adapaon o losses Boh expermenal and analycal evdence show ha he opmal penales end rapdly o bnary values as β approaches Effecvely, seems ha resuls very close o opmum can be acheved by a bnary adversary, e an adversary ha wll resor o bnary values only On he oher hand he opmal adversaral polcy wh bnary penales can be found exhausvely as ( w) = (,, ) where S s a se of bnary vecors ( b b b ) ( w ) ;,, bn T S,,, such ha b = =, e only one componen equals T Apparenly, he complexy of hs calculaon grows wh However, n he followng we show ha he opmal bnary adversary s n fac he greedy adversary, The laer acheves bnary opmaly n lnear me A greedy adversary s eager o punsh he mum wegh opon as much as possble n each round Thus 5

7 M E Anagnosou, M A ambrou Fgure Plo of 3 w (mum oal loss of a 4 round game) vs w for β =, ogeher wh he opmal penales 3; λ (,,,3) = he adversary wll assgn exacly one un of penaly o he mum curren wegh opon, and zero penales o all oher opons Gven a suffcen number of rounds (say ), easy o see ha he weghs of an opon game are equalzed so ha any wo weghs p, p j are such ha p pj < β for When equalzaon s acheved, a perodc phenomenon sars and he greedy penales form a roaon scheme 5 Greedy Behavor We explore he greedy paern n a wo opon game ha can easly be generalzed o opons Assumng nal weghs w, w ( w = w) such ha w > > w, a greedy adversary wll choose = = = =, = ff wβ > w > wβ, where (havng assumed w > w ) A he wegh of he second opon becomes for he frs me greaer han he wegh of he frs opon, and a loss equal o s assgned o he second opon Therefore, n he nex sep + he weghs (before normalzaon) are w β and w β, or equvalenly w β and w for he second me In he nex round hey become w β and w agan, and n general hey oscllae beween hese wo pars perodcally Therefore he oal loss for n a par of subsequen rounds s equal o wβ w p = + wβ + wβ wβ + w The value of s deermned by he nally assumed nequaly, and snce ough o be neger = ( lnw lnw) lnβ The loss n he frs seps =,,, s equal o w + w β τ τ τ = wβ + w (3) 53

8 M E Anagnosou, M A ambrou Therefore, for an even posve neger T he oal loss n T seps s wβ T wβ w τ = w + H β + τ + τ = wβ + w wβ + wβ wβ + w In a game wh more han wo opons s sraghforward o show ha n he seady (perodc) sae weghs end o become equal, e almos equal o, where s he number of opons Consequenly, he loss s gven by H T n a T round game β 5 Opmaly of he Greedy Behavor The followng proposon provdes a smple polynomal soluon o he problem of fndng he opmal bnary adversary Proposon The greedy sraegy s opmal for he adversary among all sraeges wh bnary penales Proof: Due o normalzaon of weghs and penales, n he proof we menon only opon weghs and n penales Assumng an nal wegh ω and penales,,, n he frs n rounds, he wegh, whch emerges before he (n + )h round s ωβ ( ωβ ω) +, where n = = Effecvely, wo opons are avalable o he adversary n each sep, eher ) o assgn a penaly equal o, whch wll produce an ncre- + ωβ ωβ ω ωβ ωβ + + ω or ) o menal loss equal o ( + ), and wll updae he wegh o assgn a zero penaly, whch wll produce a loss equal o ωβ ( ωβ ω) o ωβ ( ωβ x x + ω) Defne f ( x) ωβ ( ωβ + ω) + and an updaed wegh equal Ths looks lke a new game, n whch he adversary s he player The player s saus s deermned by a real f x f x, hs wll brng hm o number x, and possble rewards are f ( x ) and If he player ops for a new saus x + δ If he ops for f ( x), hs wll brng hm o x δ In our case f ( ) =, f ( + ) =, and f ω ξ s he roo of f ( x ) = (or f ( x) f ( x) = Moreover, f hen f ( x) for x ξ, and f ( x) for x ξ around ( ξ, ), e f ( ξ ) + x + f ξ x = f ξ =, and f ( x ) s concave n (,ξ ) convex n ( ξ, ) Assume ha ω, hen f = ω, and x < ξ, he greedy behavor s o move ( x ξ ) δ δ = oe also ha = ), I s easy o prove ha here s an odd symmery, whle s ξ If he curren saus of he player s x, and mes o he rgh, whch (unless T s oo shor) wll brng he player o a pon x such ha x ξ If x > ξ, hen f ( x) > > f ( x) and he greedy player mus choose f ( x ) and move back o x δ < ξ Effecvely, hs sars an oscllaon beween x δ and x, whch wll las unl he end of he game In he followng we prove ha hs behavor s opmal, n spe of he fac ha profs around ξ are low f x s never a The man dea behnd hs skech of proof s ha a rerea (wh consequen low profs good nvesmen for he fuure Assume x as he player s saus, and T seps (rounds) reman unl he end of he game, whle x+ Tδ < ξ The player execues M forward seps, e x = x + δ, =,,, M, wh rewards f ( x ) Then, M backward seps wh gans f ( x ) are execued; consequenly x s reached agan In he res of he game, e unl he T h sep, greedy selecons are made Ths course of evens s shown on curve (a) n Fgure 3, where he dos mark he rewards acheved (and some dos have been vercally dsplaced by a small amoun so as o be dsngushable from oher dos a he same poson) If greedy selecons had been made all he way, he course of evens would be as shown by curve (b) If y descrbes he saus of he adversary on he greedy curve (b) a he h sep and x he saus on curve (a), hen f ( x) = f ( y) for =,, M Furhermore, f ( x3m+ ) = f ( ym+ ) Therefore he dfference beween he cumulave reward on curve (b) and curve (a) s 54

9 M E Anagnosou, M A ambrou Fgure 3 Sample pahs of player behavor, whch are used n he proof of Proposon T T M M R = f y f x = f x + f x + + f x + = = M+ = = ( δ) ( δ) ( δ) T = M+ ( δ) ( δ) = f x+ M + f x f x+ M Effecvely we need o show ha R Frs, le us make some observaons and explore oher varaons of R oe ha R, as gven by (4), s posve f he cumulave reward from he back and forh movemen (n he frs M seps) s less han he reward n he las M seps However, as T ncreases, he poson of he las sep approaches ξ and can be shown ha he cumulave reward of he las M seps decreases Ths propery wll be proved laer, and s due o he convexy and monooncy properes of f When T furher ncreases, some of he very las M seps of he greedy behavor ener he phase of oscllaon around ξ, and for T suffcenly large, all M belong o he oscllaon phase oe, however, ha he oscllaon phase rewards are hose closer o /, whch s he lower lm of all greedy seps If he greedy algorhm s o be opmal, even he M oscllaory seps should brng a cumulave reward greaer han he orgnal back and forh movemen On he oher hand, f we prove hs las nequaly, hs wll also prove (4), whose las M seps brng more reward han he M oscllaory seps e ( ψ, ψ ) be he par of oscllaon pons around ξ, e ψ = x+ ( ξ x) δ δ and ψ = δ + ψ In he wors case, whch has jus been menoned, R = M ωβ + ωβ M + f x f x + M ψ ψ ωβ ωβ = M f ( x) f ( x Mδ ) ψ ψ + ωβ + ω ωβ + ω ψ ψ ψ ( ) ψ δ ωβ + ω ωβ + ω However, f ( x) f ( x+ Mδ ) can be seen as he sum of M erms f ( x δ) f ( x ( ) δ) + + +, for =, M We shall furher prove ha each of hese erms s smaller han he dfference nsde he bg parenheses, e ψ ψ ωβ ωβ f ( x+ δ) f ( x+ ( + ) δ) ψ ψ ωβ + ω ωβ + ω Ths s a consequence of he followng lemma: f x he followng nequaly s rue: emma For any concave funcon f ( x) f ( x x) f ( x x) f ( x x) Inequaly (5) holds because (4) (5) 55

10 M E Anagnosou, M A ambrou ( ) f x f x+ x f x+ x f x+ x f ( x+ x) x x whch s a consequence of he mean value heorem sang ha here s a pon xx, + x such ha f ( φ ) = f ( x+ x) f ( x) x Also, here s a pon φ n ( x+ xx, + x) such ha ( φ ) φ n f = f x+ x f x+ x x However, f s a concave funcon, and s dervave s non-ncrea- + mples f ( φ ) f ( x x) f ( φ ) sng, herefore φ x x φ +, whch proves (6) In fac (5) can be easly generalzed o any same lengh nervals, even overlappng ones, e f x x, hen (6) f x f x + x f x f x + x (7) Due o (5) each successve equal lengh (e x ) nerval produces an ncremenal reward f x f x+ x, whch s smaller han he ncremenal reward of he nex nerval, and of all succeedng nervals, as long as f remans concave Effecvely, emma proves ha he ncremenal reward of he rghmos nerval, whch does no conan ξ, e he nerval ( ψ δψ, ), s he hghes among he rewards of all nervals of he same lengh, whch begn o he lef of ψ δ Unforunaely, our am was o prove (4), whch would be secured f f remaned concave n ψ, ψ, eg f ψ = ξ δ and ψ = ξ However hs s no rue, snce a ξ f urns from concave o convex Forunaely, he erm f ( ψ ) f ( ψ ), whch covers he nerval ( ψ, ψ ) can be seen as he sum of rewards relaed wh he concave f n (, ) cave f n ( ξ, ψ ) Due o he odd symmery around ξ, f ( ξ + ( ψ ξ) ) + f ( ξ ( ψ ξ) ) = f ( ξ), herefore f ( ψ) = f ( ξ) f ( ξ ψ), and f ( ψ) f ( ψ) = f ( ψ) f ( ξ) f ( ξ ψ) = f ( ψ) f ( ξ) + f ( ξ ψ) f ( ξ) However, due o he concavy of f, f ( ψ) f ( ξ) f ( ψ δ) f ( ξ δ), and f ( ξ ψ) f ( ξ) f ( ξ ψ ( ξ ψ) ) f ( ξ ( ξ ψ) ) f ( ξ δ) f ( ψ) f ( ψ) f ( ψ) f ( ψ δ) f ( ξ δ) + f ( ξ δ) f ( ψ) = f ( ψ δ) f ( ψ) Ths resul saes ha he nerval ( ψ, ψ+ δ), whch conans ξ, provdes hgher nerval ( ψ δψ, ), whch n urn s hgher han he f = Therefore ψ ξ and he con- f han he prevous of any prevous nerval of he same lengh Therefore we have seen so far ha a sequence of penales, whch begns a some x < ξ and nvolves one fold, can be reduced o a sequence whou any folds, and wh mproved oal reward, as shown n Fgure 4 In Fgure 4 a sequence of seps wh a sngle fold, whch sars a x and ends a x, s shown ogeher wh he respecve greedy sequence, whch sars a x and ends a x3 = Mδ + x If he sequence mus exend afer ξ, he addonal seps are oscllaon seps around ξ The res of hs proof s jus an applcaon of hs resul, so ha a sequence wh an arbrary number of folds can be reduced o an mproved reward foldless sequence Fgure 4 Reducon of a sequence of penales, whch conans a fold, o a sequence whou folds and wh mproved oal reward 56

11 M E Anagnosou, M A ambrou Suppose ha he nal poson of he game s a x, and ha x ξ (oherwse reverse he nal probables ω, ω) Suppose also ha he nal sequence does no exend beyond ψ, e does no reach ξ or nvolves a number of oscllaons around ξ Then ake he las fold and reduce as menoned, e by replacng wh an equal number of greedy seps a he end of he curren sequence If hese seps reach ξ, hey are oscllaon seps Repea he same sep, unl all folds have dsappeared (oscllaons do no coun as folds) If he orgnal sequence does exend beyond ξ, he approach s he same, bu he reader should noe ha he applcaon of hs algorhm wll fnally reduce he par, whch exends beyond ψ, o oscllaons beween ψ and ψ 6 Concluson We summarze he man resuls of hs paper: An wors performance (adversaral) analyss of he Hedge algorhm has been presened, under he assumpon of lmed penales per round A recursve expresson has been gven for he evaluaon of he mum oal cumulave loss; hs expresson can be exploed boh numercally and analycally However, bnary penaly schemes provde an excellen approxmaon o he opmal scheme, and, remarkably, he greedy bnary sraegy has been proved opmal among bnary schemes for he adversary References [] Robbns, H (95) Some Aspecs of he Sequenal Desgn of Expermens Bullen of he Amercan Mahemacal Socey, 58, hp://dxdoorg/9/s [] Freund, Y and Schapre, RE (997) A Decson-Theorec Generalzaon of On-ne earnng and an Applcaon o Boosng Journal of Compuer and Sysem Scences, 55, 9-39 hp://dxdoorg/6/jcss99754 [3] Auer, P, Cesa-Banch,, Freund, Y and Schapre, RE () The on-sochasc Mul-Armed Band Problem SIAM Journal on Compung, 3, hp://dxdoorg/37/s [4] Allenberg-eeman, C and eeman, B (4) Full Informaon Game wh Gans and osses Algorhmc earnng Theory: 5h Inernaonal Conference, 344, [5] Dan, V, Hayes, TP and Kakade, SM (8) The Prce of Band Informaon for Onlne Opmzaon In: Pla, JC, Koller, D, Snger, Y and Rowes, S, Eds, Advances n eural Informaon Processng Sysems, MIT Press, Cambrdge, [6] Barle, P, Dan, V, Hayes, T, Kakade, S, Rakhln, A and Tewar, A (8) Hgh-Probably Regre Bounds for Band Onlne near Opmzaon Proceedngs of nd Annual Conference on earnng Theory (COT), Helsnk [7] Cesa-Banch, and ugos, G () Combnaoral Bands Journal of Compuer and Sysem Scences, 78, 44-4 hp://dxdoorg/6/jjcss [8] Uchya, T, akamura, A and Kudo, M () Algorhms for Adversaral Band Problems wh Mulple Plays In: Huer, M, Sephan, F, Vovk, V and Zeugmann, T, Eds, Algorhmc earnng Theory, ecure oes n Arfcal Inellgence o 633, Sprnger, [9] Auer, P, Cesa-Banch,, Freund, Y and Schapre, RE (995) Gamblng n a Rgged Casno: The Adversaral Mul-Armed Band Problem Proceedngs of 36h Annual Symposum on Foundaons of Compuer Scence, Mlwaukee, 3-33 [] Hochbaum, DS (995) Approxmaon Algorhms for P-Hard Problems PWS Publshng Company, Boson [] He, D, Chen, W, Wang, and u, T-Y (3) Onlne earnng for Aucon Mechansm n Band Seng Decson Suppor Sysems, 56, hp://dxdoorg/6/jdss374 [] Park, C and ee, J () Inellgen Traffc Conrol Based on Mul-Armed Band and Wreless Schedulng Technques Inernaonal Conference on Advances n Vehcular Sysem, Technologes and Applcaons, Vence, 3-7 [3] Bersekas, DP (998) ework Opmzaon Ahena Scenfc, Belmon [4] Blum, A and Burch, C () On-ne earnng and he Mercal Task Sysem Problem Machne earnng, 39, hp://dxdoorg/3/a: [5] Cole, SJ and m, C (8) Algorhms for ework Inerdcon and Forfcaon Games Sprnger Opmzaon and Is Applcaons, 7, hp://dxdoorg/7/ _4 [6] Vanĕk, O, Jakob, M and Pĕchouček, M () Usng Agens o Improve Inernaonal Marme Transpor Secury IEEE Inellgen Sysems, 6, 9-95 hp://dxdoorg/9/mis3 57

12

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

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

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

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

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

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

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

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

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

Tight results for Next Fit and Worst Fit with resource augmentation

Tight results for Next Fit and Worst Fit with resource augmentation Tgh resuls for Nex F and Wors F wh resource augmenaon Joan Boyar Leah Epsen Asaf Levn Asrac I s well known ha he wo smple algorhms for he classc n packng prolem, NF and WF oh have an approxmaon rao of

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

Modeling and Solving of Multi-Product Inventory Lot-Sizing with Supplier Selection under Quantity Discounts

Modeling and Solving of Multi-Product Inventory Lot-Sizing with Supplier Selection under Quantity Discounts nernaonal ournal of Appled Engneerng Research SSN 0973-4562 Volume 13, Number 10 (2018) pp. 8708-8713 Modelng and Solvng of Mul-Produc nvenory Lo-Szng wh Suppler Selecon under Quany Dscouns Naapa anchanaruangrong

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

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

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

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

CS 268: Packet Scheduling

CS 268: Packet Scheduling Pace Schedulng Decde when and wha pace o send on oupu ln - Usually mplemened a oupu nerface CS 68: Pace Schedulng flow Ion Soca March 9, 004 Classfer flow flow n Buffer managemen Scheduler soca@cs.bereley.edu

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

Performance Analysis for a Network having Standby Redundant Unit with Waiting in Repair

Performance Analysis for a Network having Standby Redundant Unit with Waiting in Repair TECHNI Inernaonal Journal of Compung Scence Communcaon Technologes VOL.5 NO. July 22 (ISSN 974-3375 erformance nalyss for a Nework havng Sby edundan Un wh ang n epar Jendra Sngh 2 abns orwal 2 Deparmen

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

UNIVERSITAT AUTÒNOMA DE BARCELONA MARCH 2017 EXAMINATION

UNIVERSITAT AUTÒNOMA DE BARCELONA MARCH 2017 EXAMINATION INTERNATIONAL TRADE T. J. KEHOE UNIVERSITAT AUTÒNOMA DE BARCELONA MARCH 27 EXAMINATION Please answer wo of he hree quesons. You can consul class noes, workng papers, and arcles whle you are workng on he

More information

Comparison of Differences between Power Means 1

Comparison of Differences between Power Means 1 In. Journal of Mah. Analyss, Vol. 7, 203, no., 5-55 Comparson of Dfferences beween Power Means Chang-An Tan, Guanghua Sh and Fe Zuo College of Mahemacs and Informaon Scence Henan Normal Unversy, 453007,

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

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

On computing differential transform of nonlinear non-autonomous functions and its applications

On computing differential transform of nonlinear non-autonomous functions and its applications On compung dfferenal ransform of nonlnear non-auonomous funcons and s applcaons Essam. R. El-Zahar, and Abdelhalm Ebad Deparmen of Mahemacs, Faculy of Scences and Humanes, Prnce Saam Bn Abdulazz Unversy,

More information

Department of Economics University of Toronto

Department of Economics University of Toronto Deparmen of Economcs Unversy of Torono ECO408F M.A. Economercs Lecure Noes on Heeroskedascy Heeroskedascy o Ths lecure nvolves lookng a modfcaons we need o make o deal wh he regresson model when some of

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

Existence and Uniqueness Results for Random Impulsive Integro-Differential Equation

Existence and Uniqueness Results for Random Impulsive Integro-Differential Equation Global Journal of Pure and Appled Mahemacs. ISSN 973-768 Volume 4, Number 6 (8), pp. 89-87 Research Inda Publcaons hp://www.rpublcaon.com Exsence and Unqueness Resuls for Random Impulsve Inegro-Dfferenal

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

HEAT CONDUCTION PROBLEM IN A TWO-LAYERED HOLLOW CYLINDER BY USING THE GREEN S FUNCTION METHOD

HEAT CONDUCTION PROBLEM IN A TWO-LAYERED HOLLOW CYLINDER BY USING THE GREEN S FUNCTION METHOD Journal of Appled Mahemacs and Compuaonal Mechancs 3, (), 45-5 HEAT CONDUCTION PROBLEM IN A TWO-LAYERED HOLLOW CYLINDER BY USING THE GREEN S FUNCTION METHOD Sansław Kukla, Urszula Sedlecka Insue of Mahemacs,

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

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

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

Testing a new idea to solve the P = NP problem with mathematical induction

Testing a new idea to solve the P = NP problem with mathematical induction Tesng a new dea o solve he P = NP problem wh mahemacal nducon Bacground P and NP are wo classes (ses) of languages n Compuer Scence An open problem s wheher P = NP Ths paper ess a new dea o compare he

More information

Approximate Analytic Solution of (2+1) - Dimensional Zakharov-Kuznetsov(Zk) Equations Using Homotopy

Approximate Analytic Solution of (2+1) - Dimensional Zakharov-Kuznetsov(Zk) Equations Using Homotopy Arcle Inernaonal Journal of Modern Mahemacal Scences, 4, (): - Inernaonal Journal of Modern Mahemacal Scences Journal homepage: www.modernscenfcpress.com/journals/jmms.aspx ISSN: 66-86X Florda, USA Approxmae

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

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

Multi-priority Online Scheduling with Cancellations

Multi-priority Online Scheduling with Cancellations Submed o Operaons Research manuscrp (Please, provde he manuscrp number!) Auhors are encouraged o subm new papers o INFORMS journals by means of a syle fle emplae, whch ncludes he journal le. However, use

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

Online Supplement for Dynamic Multi-Technology. Production-Inventory Problem with Emissions Trading

Online Supplement for Dynamic Multi-Technology. Production-Inventory Problem with Emissions Trading Onlne Supplemen for Dynamc Mul-Technology Producon-Invenory Problem wh Emssons Tradng by We Zhang Zhongsheng Hua Yu Xa and Baofeng Huo Proof of Lemma For any ( qr ) Θ s easy o verfy ha he lnear programmng

More information

Should Exact Index Numbers have Standard Errors? Theory and Application to Asian Growth

Should Exact Index Numbers have Standard Errors? Theory and Application to Asian Growth Should Exac Index umbers have Sandard Errors? Theory and Applcaon o Asan Growh Rober C. Feensra Marshall B. Rensdorf ovember 003 Proof of Proposon APPEDIX () Frs, we wll derve he convenonal Sao-Vara prce

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

Li An-Ping. Beijing , P.R.China

Li An-Ping. Beijing , P.R.China A New Type of Cpher: DICING_csb L An-Png Bejng 100085, P.R.Chna apl0001@sna.com Absrac: In hs paper, we wll propose a new ype of cpher named DICING_csb, whch s derved from our prevous sream cpher DICING.

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

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

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

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

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

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

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

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

Mechanics Physics 151

Mechanics Physics 151 Mechancs Physcs 5 Lecure 0 Canoncal Transformaons (Chaper 9) Wha We Dd Las Tme Hamlon s Prncple n he Hamlonan formalsm Dervaon was smple δi δ Addonal end-pon consrans pq H( q, p, ) d 0 δ q ( ) δq ( ) δ

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

First-order piecewise-linear dynamic circuits

First-order piecewise-linear dynamic circuits Frs-order pecewse-lnear dynamc crcus. Fndng he soluon We wll sudy rs-order dynamc crcus composed o a nonlnear resse one-por, ermnaed eher by a lnear capacor or a lnear nducor (see Fg.. Nonlnear resse one-por

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

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

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

Algorithmic models of human decision making in Gaussian multi-armed bandit problems

Algorithmic models of human decision making in Gaussian multi-armed bandit problems Algorhmc models of human decson makng n Gaussan mul-armed band problems Paul Reverdy, Vabhav Srvasava and Naom E. Leonard Absrac We consder a heursc Bayesan algorhm as a model of human decson makng n mul-armed

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

Epistemic Game Theory: Online Appendix

Epistemic Game Theory: Online Appendix Epsemc Game Theory: Onlne Appendx Edde Dekel Lucano Pomao Marcano Snscalch July 18, 2014 Prelmnares Fx a fne ype srucure T I, S, T, β I and a probably µ S T. Le T µ I, S, T µ, βµ I be a ype srucure ha

More information

, t 1. Transitions - this one was easy, but in general the hardest part is choosing the which variables are state and control variables

, t 1. Transitions - this one was easy, but in general the hardest part is choosing the which variables are state and control variables Opmal Conrol Why Use I - verss calcls of varaons, opmal conrol More generaly More convenen wh consrans (e.g., can p consrans on he dervaves More nsghs no problem (a leas more apparen han hrogh calcls of

More information

The Dynamic Programming Models for Inventory Control System with Time-varying Demand

The Dynamic Programming Models for Inventory Control System with Time-varying Demand The Dynamc Programmng Models for Invenory Conrol Sysem wh Tme-varyng Demand Truong Hong Trnh (Correspondng auhor) The Unversy of Danang, Unversy of Economcs, Venam Tel: 84-236-352-5459 E-mal: rnh.h@due.edu.vn

More information

Boosted LMS-based Piecewise Linear Adaptive Filters

Boosted LMS-based Piecewise Linear Adaptive Filters 016 4h European Sgnal Processng Conference EUSIPCO) Boosed LMS-based Pecewse Lnear Adapve Flers Darush Kar and Iman Marvan Deparmen of Elecrcal and Elecroncs Engneerng Blken Unversy, Ankara, Turkey {kar,

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

SOME NOISELESS CODING THEOREMS OF INACCURACY MEASURE OF ORDER α AND TYPE β

SOME NOISELESS CODING THEOREMS OF INACCURACY MEASURE OF ORDER α AND TYPE β SARAJEVO JOURNAL OF MATHEMATICS Vol.3 (15) (2007), 137 143 SOME NOISELESS CODING THEOREMS OF INACCURACY MEASURE OF ORDER α AND TYPE β M. A. K. BAIG AND RAYEES AHMAD DAR Absrac. In hs paper, we propose

More information

Lecture 2 M/G/1 queues. M/G/1-queue

Lecture 2 M/G/1 queues. M/G/1-queue Lecure M/G/ queues M/G/-queue Posson arrval process Arbrary servce me dsrbuon Sngle server To deermne he sae of he sysem a me, we mus now The number of cusomers n he sysems N() Tme ha he cusomer currenly

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

ON THE WEAK LIMITS OF SMOOTH MAPS FOR THE DIRICHLET ENERGY BETWEEN MANIFOLDS

ON THE WEAK LIMITS OF SMOOTH MAPS FOR THE DIRICHLET ENERGY BETWEEN MANIFOLDS ON THE WEA LIMITS OF SMOOTH MAPS FOR THE DIRICHLET ENERGY BETWEEN MANIFOLDS FENGBO HANG Absrac. We denfy all he weak sequenal lms of smooh maps n W (M N). In parcular, hs mples a necessary su cen opologcal

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

Including the ordinary differential of distance with time as velocity makes a system of ordinary differential equations.

Including the ordinary differential of distance with time as velocity makes a system of ordinary differential equations. Soluons o Ordnary Derenal Equaons An ordnary derenal equaon has only one ndependen varable. A sysem o ordnary derenal equaons consss o several derenal equaons each wh he same ndependen varable. An eample

More information

On the Boyd- Kuramoto Model : Emergence in a Mathematical Model for Adversarial C2 Systems

On the Boyd- Kuramoto Model : Emergence in a Mathematical Model for Adversarial C2 Systems On he oyd- Kuramoo Model : Emergence n a Mahemacal Model for Adversaral C2 Sysems Alexander Kallonas DSTO, Jon Operaons Dvson C2 Processes: many are cycles! oyd s Observe-Oren-Decde-Ac Loop: Snowden s

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

Appendix to Online Clustering with Experts

Appendix to Online Clustering with Experts A Appendx o Onlne Cluserng wh Expers Furher dscusson of expermens. Here we furher dscuss expermenal resuls repored n he paper. Ineresngly, we observe ha OCE (and n parcular Learn- ) racks he bes exper

More information

Dynamic Team Decision Theory

Dynamic Team Decision Theory Dynamc Team Decson Theory EECS 558 Proec Repor Shruvandana Sharma and Davd Shuman December, 005 I. Inroducon Whle he sochasc conrol problem feaures one decson maker acng over me, many complex conrolled

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

Dual Approximate Dynamic Programming for Large Scale Hydro Valleys

Dual Approximate Dynamic Programming for Large Scale Hydro Valleys Dual Approxmae Dynamc Programmng for Large Scale Hydro Valleys Perre Carpener and Jean-Phlppe Chanceler 1 ENSTA ParsTech and ENPC ParsTech CMM Workshop, January 2016 1 Jon work wh J.-C. Alas, suppored

More information

Attribute Reduction Algorithm Based on Discernibility Matrix with Algebraic Method GAO Jing1,a, Ma Hui1, Han Zhidong2,b

Attribute Reduction Algorithm Based on Discernibility Matrix with Algebraic Method GAO Jing1,a, Ma Hui1, Han Zhidong2,b Inernaonal Indusral Informacs and Compuer Engneerng Conference (IIICEC 05) Arbue educon Algorhm Based on Dscernbly Marx wh Algebrac Mehod GAO Jng,a, Ma Hu, Han Zhdong,b Informaon School, Capal Unversy

More information

Political Economy of Institutions and Development: Problem Set 2 Due Date: Thursday, March 15, 2019.

Political Economy of Institutions and Development: Problem Set 2 Due Date: Thursday, March 15, 2019. Polcal Economy of Insuons and Developmen: 14.773 Problem Se 2 Due Dae: Thursday, March 15, 2019. Please answer Quesons 1, 2 and 3. Queson 1 Consder an nfne-horzon dynamc game beween wo groups, an ele and

More information

P R = P 0. The system is shown on the next figure:

P R = P 0. The system is shown on the next figure: TPG460 Reservor Smulaon 08 page of INTRODUCTION TO RESERVOIR SIMULATION Analycal and numercal soluons of smple one-dmensonal, one-phase flow equaons As an nroducon o reservor smulaon, we wll revew he smples

More information

EP2200 Queuing theory and teletraffic systems. 3rd lecture Markov chains Birth-death process - Poisson process. Viktoria Fodor KTH EES

EP2200 Queuing theory and teletraffic systems. 3rd lecture Markov chains Birth-death process - Poisson process. Viktoria Fodor KTH EES EP Queung heory and eleraffc sysems 3rd lecure Marov chans Brh-deah rocess - Posson rocess Vora Fodor KTH EES Oulne for oday Marov rocesses Connuous-me Marov-chans Grah and marx reresenaon Transen and

More information

APOC #232 Capacity Planning for Fault-Tolerant All-Optical Network

APOC #232 Capacity Planning for Fault-Tolerant All-Optical Network APOC #232 Capacy Plannng for Faul-Toleran All-Opcal Nework Mchael Kwok-Shng Ho and Kwok-wa Cheung Deparmen of Informaon ngneerng The Chnese Unversy of Hong Kong Shan, N.T., Hong Kong SAR, Chna -mal: kwcheung@e.cuhk.edu.hk

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

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

Appendix H: Rarefaction and extrapolation of Hill numbers for incidence data

Appendix H: Rarefaction and extrapolation of Hill numbers for incidence data Anne Chao Ncholas J Goell C seh lzabeh L ander K Ma Rober K Colwell and Aaron M llson 03 Rarefacon and erapolaon wh ll numbers: a framewor for samplng and esmaon n speces dversy sudes cology Monographs

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

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

Bernoulli process with 282 ky periodicity is detected in the R-N reversals of the earth s magnetic field

Bernoulli process with 282 ky periodicity is detected in the R-N reversals of the earth s magnetic field Submed o: Suden Essay Awards n Magnecs Bernoull process wh 8 ky perodcy s deeced n he R-N reversals of he earh s magnec feld Jozsef Gara Deparmen of Earh Scences Florda Inernaonal Unversy Unversy Park,

More information

Implementation of Quantized State Systems in MATLAB/Simulink

Implementation of Quantized State Systems in MATLAB/Simulink SNE T ECHNICAL N OTE Implemenaon of Quanzed Sae Sysems n MATLAB/Smulnk Parck Grabher, Mahas Rößler 2*, Bernhard Henzl 3 Ins. of Analyss and Scenfc Compung, Venna Unversy of Technology, Wedner Haupsraße

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

[ ] 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

Optimal environmental charges under imperfect compliance

Optimal environmental charges under imperfect compliance ISSN 1 746-7233, England, UK World Journal of Modellng and Smulaon Vol. 4 (28) No. 2, pp. 131-139 Opmal envronmenal charges under mperfec complance Dajn Lu 1, Ya Wang 2 Tazhou Insue of Scence and Technology,

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

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

ABSTRACT KEYWORDS. Bonus-malus systems, frequency component, severity component. 1. INTRODUCTION

ABSTRACT KEYWORDS. Bonus-malus systems, frequency component, severity component. 1. INTRODUCTION EERAIED BU-MAU YTEM ITH A FREQUECY AD A EVERITY CMET A IDIVIDUA BAI I AUTMBIE IURACE* BY RAHIM MAHMUDVAD AD HEI HAAI ABTRACT Frangos and Vronos (2001) proposed an opmal bonus-malus sysems wh a frequency

More information

MANY real-world applications (e.g. production

MANY real-world applications (e.g. production Barebones Parcle Swarm for Ineger Programmng Problems Mahamed G. H. Omran, Andres Engelbrech and Ayed Salman Absrac The performance of wo recen varans of Parcle Swarm Opmzaon (PSO) when appled o Ineger

More information

Pendulum Dynamics. = Ft tangential direction (2) radial direction (1)

Pendulum Dynamics. = Ft tangential direction (2) radial direction (1) Pendulum Dynams Consder a smple pendulum wh a massless arm of lengh L and a pon mass, m, a he end of he arm. Assumng ha he fron n he sysem s proporonal o he negave of he angenal veloy, Newon s seond law

More information

A Novel Efficient Stopping Criterion for BICM-ID System

A Novel Efficient Stopping Criterion for BICM-ID System A Novel Effcen Soppng Creron for BICM-ID Sysem Xao Yng, L Janpng Communcaon Unversy of Chna Absrac Ths paper devses a novel effcen soppng creron for b-nerleaved coded modulaon wh erave decodng (BICM-ID)

More information

Lecture 18: The Laplace Transform (See Sections and 14.7 in Boas)

Lecture 18: The Laplace Transform (See Sections and 14.7 in Boas) Lecure 8: The Lalace Transform (See Secons 88- and 47 n Boas) Recall ha our bg-cure goal s he analyss of he dfferenal equaon, ax bx cx F, where we emloy varous exansons for he drvng funcon F deendng on

More information

Increasing the Probablility of Timely and Correct Message Delivery in Road Side Unit Based Vehicular Communcation

Increasing the Probablility of Timely and Correct Message Delivery in Road Side Unit Based Vehicular Communcation Halmsad Unversy For he Developmen of Organsaons Producs and Qualy of Lfe. Increasng he Probablly of Tmely and Correc Message Delvery n Road Sde Un Based Vehcular Communcaon Magnus Jonsson Krsna Kuner and

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