Stochastic approximation vis-a-vis online learning for big data analytics

Size: px
Start display at page:

Download "Stochastic approximation vis-a-vis online learning for big data analytics"

Transcription

1 Sochasic approximaion vis-a-vis online learning for big daa analyics Konsaninos Slavakis, Seung-Jun Kim, Gonzalo Maeos, and Georgios B. Giannakis July 30, 2014 We live in an era of daa deluge, where daa ranslae o knowledge and can hus conribue in various direcions if harnessed and processed inelligenly. here is no doub ha signal processing (SP) is of uermos relevance o imely big daa applicaions such as real-ime medical imaging, smar ciies, nework sae visualizaion and anomaly deecion (e.g., in he power grid and he Inerne), healh informaics for personalized reamen, senimen analysis from online social media, web-based adverising, recommendaion sysems, sensor-empowered srucural healh monioring, and e-commerce fraud deecion, jus o name a few. Accordingly, abundan chances unfold o SP researchers and praciioners for fundamenal conribuions in big daa heory and pracice. Wih such big blessings however, come big challenges. he sheer volume and dimensionaliy of daa make i ofen impossible o run analyics and radiional bach inferenial mehods on sandalone processing unis. Wih regards o scalabiliy, online daa processing is well moivaed as he compuaional complexiy of joinly processing he enire daa-se as a bach is prohibiive. Furhermore, here are many applicaions in which daa hemselves are made available in a sreaming fashion, meaning ha smaller chunks of daa are acquired sequenially in ime, e.g., nodes of a large nework ransmiing small blocks of daa o a cenral uni coninuously and incoherenly in ime. As informaion sources unceasingly produce daa in real ime, analyics mus ofen be performed on-he-fly, ypically wihou a chance o revisi previous daa. In addiion, ofenimes big daa asks are subjec o sringen ime consrains, so ha a high-qualiy answer obained slowly via bach echniques can be less useful han a medium-qualiy answer ha is obained fas in an online fashion. RELEVANCE In his conex, his lecure noe presens recen advances in online learning for big daa analyics. I is demonsraed ha many of hese approaches, mosly developed wihin he machine learning discipline, have srong ies wih workhorse saisical SP ools such as sochasic approximaion (SA) and sochasic gradien (SG) algorihms. Imporan differences and novel aspecs are highlighed as well. A key message conveyed is ha e.g., Robbins-Monro s and Widrow s seminal works on SA, ha go back half a cenury, can play insrumenal roles in modern online learning asks for big daa analyics. Consequenly, ample opporuniies arise for he SP communiy o conribue in his growing and inherenly cross-disciplinary field, spanning muliple areas across science and engineering. Work in his paper was suppored by he NSF grans EARS , EAGER , and he MURI Gran AFOSR FA

2 he remainder of his lecure noe, which also serves as a supplemen o [1], is organized as follows. Basic principles of SA are reviewed firs, followed by a couple of examples. Sandard performance merics of SA algorihms are hen oulined, accompanied by a recen wis on performance analysis hrough convex analyic argumens. Sequenial schemes and daa skeching or sampling wih eminen poenial for big daa analyics are also delineaed. Finally, online learning approaches based on he powerful online convex opimizaion (OCO) framework are reviewed, where he links and differences vis-a-vis SA are highlighed. PREREQUISIES he required background includes basics of linear algebra, probabiliy heory, convex analysis, and sochasic opimizaion. SOCHASIC APPROXIMAION BASICS Consider he prooypical saisical learning problem in he realm of sochasic opimizaion (SO) [2, 3] where given a loss funcion f, one aims a minimizing he expeced loss E y { f (w; y)}, possibly augmened wih a complexiy-conrolling convex regularizer r(w), wih respec o (w.r..) a deerminisic parameer (weigh) vecor w W. An example of r(w) is he recenly popular sparsiy-promoing l 1 -norm of he p 1 vecor w where r(w) = w 1 := p i=1 w i. Expecaion E y { } is aken w.r.. he ypically unknown probabiliy disribuion of daa y describing, e.g., inpu-response pairs in a supervised learning seing, and W denoes a subse of some Euclidean space, inroduced here o cover general cases where consrains are imposed on w. In lieu of he aforemenioned disribuional informaion, given raining daa {y } =1 one can insead op for solving he empirical risk minimizaion problem min w W 1 f (w; y ) + r(w) (1) =1 which is an approximaion of is ensemble counerpar, namely min w W [E y { f (w; y)} + r(w)]. Beyond a purely learning paradigm, one should appreciae he generaliy offered by (1), since i can subsume, e.g., (consrained) maximum-likelihood problems wih f idenified as he log-likelihood funcion and daa assumed saisically independen. In big daa seings, can be huge, poenially infinie in a real-ime paradigm where idenifies ime insances of daa acquisiion. Moreover, he search space W can be excessively high-dimensional wih complex srucure. his observaion jusifies he inclusion of a regularizer in (1) o effecively reduce he dimensionaliy and/or size of W and yield parsimonious models ha are inerpreable and have saisfacory predicive performance. Unsurprisingly, here has been growing ineres over he las decade in devising scalable and fas online algorihms for big daa learning asks such as (1). he main premise of SO is cenered around solving he minimizaion ask [cf. (1)] min w R p [ϕ(w) := E y{ f (w; y)}] (2) wihou having E y { } available; see e.g., [3]. (Compared o (1) and is ensemble version, boh W and he regularizer r have been dropped here for breviy.) Key feaures presen in SO algorihms 2

3 are: (i) he daa comprise a sequence of eiher dependen vecors wih (asympoically) vanishing covariance, or, independen idenically disribued (i.i.d.) realizaions {y } =1 of y; and, (ii) given (w, y ), here is a means of obaining an unbiased sochasic gradien esimae f (w; y ), so ha E y { f (w; y )} = ϕ(w). For ϕ smooh, minimizing ϕ in (2) amouns o searching for a zero of Φ(w) := ϕ(w), i.e., a w 0 such ha (s..) Φ(w 0 ) = 0 [3]. he classical Newon-Raphson (N-R) algorihm provides he means o achieve his goal. For w scalar and wih denoing differeniaion, he sequence generaed by he recursion w k+1 := w k Φ(w k )/Φ (w k ) = w k ϕ (w k )/ϕ (w k ) converges under mild condiions o a roo of Φ(w), and hus o a minimizer of ϕ(w). An illusraion of he N-R ieraion can be seen in Figure 1: Newon-Raphson mehod for finding a Fig. 1. Saring from w 1 and using he derivaives {Φ (w k )} + k=1 w 0 s.. Φ(w 0 ) = 0. in he N-R ieraion, he resulan updaes {w k } + k=2 gradually approach w 0, where Φ(w 0 ) = 0. Such a simple recursion can be readily exended o he p 1 vecor case as w k+1 := w k H 1 ϕ (w k ) ϕ(w k ), where now H ϕ (w k ) sands for he p p Hessian marix of ϕ a w k wih (i, j)h enry 2 ϕ(w k )/( w i w j ). Clearly, he N-R algorihm canno be applied if E y { } is no available; e.g., if he probabiliy densiy funcion (pdf) of y is unknown, or, when compuing E y { } enails cumbersome inegraion over high-dimensional domains. o alleviae his burden, SA hrough he celebraed Robbins-Monro algorihm relies on he sequence of realizaions {y } and ingeniously uses he insananeous f (w ; y ) insead of he ensemble ϕ(w k ) (indexes have been changed from k o, for ime-adapive operaion). Wih µ denoing he sep-size, SA generaes he online (or sochasic) gradien descen (OGD) ieraion w +1 = w µ f (w ; y ) (3) which learns expecaions on-he-fly. his poin is beer illusraed by he following example. Online averaging as SA: he soluion of min w E y { w y 2 2 /2} is clearly w 0 = E y {y}. Following he SA raionale, consider f (w; y ) := w y 2 2 /2. he OGD ieraion is w +1 = w µ (w y ), and if w 1 := 0 as well as µ := 1/, simple mahemaical inducion yields w +1 = (1/) τ=1 y τ, which in accordance wih he law of large numbers converges o w 0 = E y {y} as + [3]. Several well-known adapive signal processing and online learning algorihms sem from OGD. LMS as SA: Consider for insance scalar d and vecor x processes which comprise he raining daa colleced in y := [d, x ], and le f (w; y ) := (d w x ) 2 /2, where sands for ransposiion. I can be readily verified ha f (w; y ) = (w x d )x, and applicaion of OGD yields w +1 = w µ (w x d )x, which is nohing bu he celebraed leas mean-squares (LMS) algorihm [3]. RLS as SA: he OGD class can be furher broadened by allowing marix sep-sizes {M } insead of scalar ones {µ } o obain w +1 = w M f (w ; y ). o highligh he poenial of his exension, consider (joinly) wide sense saionary {d, x } =1, wih C xx := E x {x x }, as well as r dx := E d,x {d x }. I urns ou ha he soluion of min w E d,x {(d w x ) 2 } is he linear minimum mean-square error esimaor w 0 = C 1 xx r dx. However, wihou knowing C xx one relies on he sample average esimae 3

4 ĈC := (1/) τ=1 x τ x 1 τ, and on OGD wih M := (1/)ĈC o obain w +1 = w 1 ĈC 1 +1 = ĈC [ĈC 1 x (w x d ) (4a) ] ĈC x +1 x+1ĉc /( + x+1ĉc x +1 ) (4b) where he marix inversion lemma is applied o carry ou efficienly he inversion in (4b). Recursions (4) comprise he well-known recursive leas-squares (RLS) algorihm [3]. PERFORMANCE OF SA ALGORIHMS Based on he samples {y }, SA algorihms produce esimaes {w } ha allow for esimaion, racking, and ou-of-sample inference asks, such as predicion. Performance analysis of SA schemes has leveraged advances in maringale and ordinary differenial equaion heories o esablish, e.g., in he saionary case, convergence of {w } o a ime-invarian w 0 in probabiliy, or wih probabiliy one, or in he mean-square sense [3]. In his saionary seing, convergence of OGD requires sep-sizes seleced o diminish wih a cerain rae. Specifically, {µ } mus saisfy (i) µ 0; (ii) lim µ = 0; and, (iii) =1 µ = +. Clearly, (i)-(iii) are saisfied for µ := 1/, which vanishes as +, bu no oo fas so ha (iii) enables {w } o reach asympoically he desired w 0. Deparing from he sandard roue of SA convergence analysis [3], recen resuls ake advanage of convexiy if i is presen in he objecive funcion. Specifically for convex coss, he OGD recursion (3) generalizes o: w +1 = P W [w µ f (w ; y )], where P W (w) := arg min w W w w 2 sands for he projecion mapping ono a closed and convex consrain se W. For ϕ differeniable and srongly convex wih index c > 0, i holds ha ϕ(w ) ϕ(w) + (w w) ϕ(w) + (c/2) w w 2 2, for all (w, w). Wih sep-sizes seleced as µ := µ/ wih µ > 1/(2c), and for bounded sochasic gradiens as in sup E w y{ f (w; y) 2 2 }, i can be verified ha he error E y{ w w }, where w 0 = arg min w W E y { f (w; y)}, saisfies he following finie-sample bound [2] E y { w w 0 2 2} Q(µ), wih Q(µ) := max { µ 2 2 /(2µc 1), w 1 w 0 2}. If in addiion ϕ is L-Lipschiz coninuous, i.e., ϕ(w) ϕ(w ) 2 L w w 2, w, w, hen a similar finie-sample bound holds also for he sequence of funcion values {ϕ(w )} [2] E y {ϕ(w ) ϕ(w 0 )} LQ(µ) 2 where expecaion is aken over {w } which involves sochasic gradiens. Performance analysis of SA algorihms deals wih convergence of {w }, whereas he online convex opimizaion framework oulined in a subsequen secion sars from (1), invokes less or no assumpions on he underlying pdfs, and assers convergence of he coss { f (w ; y )}, raher han primal variables. Recenly, SA was combined wih he alernaing direcion mehod of mulipliers (ADMM) which is aracive for off-line opimizaion of composie coss [4]. he resulan SA-ADMM solver [5] is suiable for online opimizaion of composie coss such as min w W [E y { f (w; y)} + r(w)], in a fully disribued fashion an operaional mode ha is highly desirable for big daa applicaions. 4

5 SEQUENIAL OPIMIZAION AND DAA SKECHING he imporance of sequenial opimizaion along wih he aracive operaion of random sampling (a.k.a. skeching) big daa will be illusraed in his subsecion in he conex of he familiar LS ask: [ ] 1 min w R p 2 X w d 2 2 = 1 1 =1 2 (x w d ) 2 (5) where X := [x 1,..., x ] denoes he p marix which gahers all available regressor or inpu vecors, and d := [d 1,...,d ] he 1 vecor of desired oupus (responses). Alhough irrelevan o he minimizaion in (5), he normalizaion wih is included o draw connecions wih (1). In his sense, he loss funcion becomes f (w; y ) = (x w d ) 2 /2, wih y := [d, x ], and is gradien f ( ; y ) is Lipschiz coninuous wih consan L = x 2 2. Differen from he previous discussion, here is fixed, and online means processing {d, x } =1 sequenially. Searching for a soluion w 0 of (5) requires eigen-decomposiion of X X, which incurs complexiy O( p 2 ). Alernaively, he sandard gradien descen recursion w k+1 = w k µ k (X X w k X d) enails O(p 2 ) compuaions per ieraion k. Boh cases are prohibiive in big daa seings where he number of samples,, is massive and/or he daa dimensionaliy, p, can be huge. o surmoun hese obsacles, solving for w 0 can rely on sub-sampling (a.k.a. skeching o obain a subse of) he rows of X, along wih he corresponding enries of d, o reduce complexiy w.r.., while visiing hem in a sequenial fashion ha scales linearly wih p. Kaczmarz s algorihm, a special case of he projecions ono convex ses (POCS) mehod [6], produces a sequence of esimaes {w k } o solve (5). For an arbirary iniial esimae w 1, he kh ieraion of Kaczmarz s algorihm selecs a row (k) of X, ogeher wih he corresponding enry d (k), and projecs he curren esimae w k ono he se of all minimizers H (k) := {w x(k) w = d (k) } of f (w; y (k) ), which is nohing bu a hyperplane (a closed and convex se). Hence, he (k + 1)s esimae is w k+1 := P H(k) (w k ) = w k x (k) w k d (k) x (k) 2 x (k) 2 (6) where P H(k) sands for he projecion mapping ono H (k). Noice here ha he complexiy of compuing P H(k) (w k ) scales linearly wih p. If every (d, x ) is visied infiniely ofen, hen under several condiions (6) converges o a soluion Figure 2: Kaczmarz s algorihm for hree hyperplanes {H } =1 3 wih non-empy inersecion {w 0 } = =1 3 H. Row (hyperplane) selecion affecs convergence rae; {w k } which alernaes beween H 1 and H 2 approaches w 0 faser han { w k } which is generaed via H 2,H 3. of (5) [6]. Visiing each (d, x ) a large number of imes is prohibiive wih big daa since can be excessively large. In conras, poor selecion of rows can slow down convergence; see Fig. 2. Neverheless, randomly drawing rows wih equal probabiliies has been shown empirically o accelerae convergence relaive o cyclic revisis of rows [7]. 5

6 Acceleraing SG via non-uniform sampling: In he noiseless case (X w = d), randomly drawing rows in proporion o heir Lipschiz consans L is known o provide finie-sample bounds of he form [7] E R { w k w 0 2 2} [ 1 κ(x) 2] k w1 w where κ(x) sands for he condiion number of X, and E R { } denoes expecaion w.r.. he disribuion over which {d, x } are seleced. he previous non-uniform sampling scheme yields beer convergence raes han hose resuling from uniform skeching [7]. More informaion on (non-)uniform skeching and is applicaion o SG descen mehods can be found in [8, 9]. LEARNING VIA ONLINE CONVEX OPIMIZAION Recenly, online learning approaches based on online convex opimizaion (OCO) framework have araced significan aenion, as hey do no require elaborae saisical models for daa and ye can provide robus performance guaranees. his is rue even under an adversarial seup, where he daa sequence {y } may be generaed sraegically in reacion o he learner s ieraes {w }, as in he humans-in-he-loop applicaions such as he web adverising opimizaion. he OCO framework can be viewed as a muli-round game beween a player (learner) and Figure 3: OCO as a muli-round game. an adversary [10]. In he conex of he learning formulaion in (1), he learner plays an acion w W in round, where W is assumed o be closed and convex. Based on he acion w ha he player ook, he adversary provides some feedback informaion F, manifesed in he daa (feaure) vecor y, based on which a convex loss funcion L : W R {+ } is consruced, such as L (w) := f (w; y ) + r(w). he learner hen suffers he loss a w, namely, L (w ). he overall process is depiced in Fig. 3. he learner s goal is o minimize he so-ermed regre R( ) over rounds, defined as R( ) := =1 L (w ) min w W =1 L (w) (7) which capures how much worse he learner performed cumulaively, compared o he case where a single bes acion is chosen wih he knowledge of he enire sequence of cos funcions {L } =1 in hindsigh. In paricular, OCO aims a producing a sequence {w }, which gives rise o sublinear regre, ha is he one wih R( )/ 0 as grows. Key quesion now for he learner is how o pick w in each round. OCO ALGORIHMS AND PERFORMANCE An imporan class of algorihms ha can achieve he desired sublinear regre bound is based on he online mirror descen (OMD) ieraion [11]. In a nushell, he mehod minimizes a firs-order 6

7 approximaion of L a he curren ierae w, while encouraging he search in he viciniy of w. Specifically, OMD compues he nex round ierae w +1 as w +1 = arg min w W (w w ) L (w ) + 1 µ D ψ(w, w ) (8) where L (w ) is a (sub)gradien of L a w, µ > 0 a learning rae parameer, and D ψ (w, v) is he Bregman divergence associaed wih a coninuously differeniable and srongly convex ψ, defined as D ψ (w, v) := ψ(w) ψ(v) (w v) ψ(v). (9) In he special case of using ψ(w) := w 2 2 /2, he corresponding D ψ(w, v) = w v 2 2 /2, and he OMD updae in (8) boils down o OGD [10], esablishing an immediae link beween OCO and SA. In general, a judicious choice of ψ can capure he srucure of he search space W, leading o an efficien updae formula for w. For example, when W is he probabiliy simplex, i.e., W := {w w i 0, i w i = 1}, seing ψ(w) := i w i logw i in (8) (9) yields he exponeniaed gradien algorihm, which obviaes he need o explicily impose he probabiliy simplex consrains [10]. COMID algorihm: While he OMD updae provides a compuaionally aracive soluion o (1), he linearizaion involved ofen defeas one of he purposes of he regularizer r, which is o promoe a priori known srucure in he soluion. For example, seing r(w) proporional o he l 1 -norm of w encourages sparsiy in w. o properly capure such a benefi, one has o respec he composie srucure of L, which decomposes ino he daa-dependen par f (w) := f (w; y ) and he invarian par r(w) [12, 13]. In paricular, he composie objecive mirror descen (COMID) algorihm relies on [12] w n+1 = arg min w W (w w ) f (w ) + r(w) + 1 µ D ψ(w, w ) (10) where i is seen ha he regularizer is no linearized. Boh COMID and OMD (which is a special case of COMID) can aain sublinear regre bounds. Specifically, R( ) = O( ) in general, and he bound becomes O(log ) when L is srongly convex [10, 12]. SA vis-a-vis OCO: Compared o he SA approaches, he OCO framework does no require sochasic models. his is a salien deparure from ypical SA seups, since he regre bounds are guaraneed even for {y } ha may have been generaed adversarially, i.e., wih y arbirary correlaed o pas acions {w τ } τ and pas daa {y τ } τ<. On he oher hand, he bounds perain o convergence of he sequence of coss raher han he ieraes {w } hemselves. Noneheless, building upon he flexibiliy offered by OCO, cerain limied feedback learning asks are feasible as elaboraed nex, where ineresingly, he SA ideas prove insrumenal once again. ONLINE LEARNING WIH BANDI FEEDBACK he bandi se-up of OCO refers o he case where he feedback F from he adversary does no explicily reveal he cos funcion L ( ), bu only he sample cos L (w ) due o acion w ; refer also o Fig. 3. For example, w may represen he adverising budge allocaed o differen media channels, and L (w ) he corresponding overall cos (e.g., he oal adverising expense minus he resuling 7

8 Figure 4: SA/SO vis-a-vis OCO: Feaures and implicaions. income). In his case, i may be difficul o know he explici form of L, bu L (w ) can be easily observed. he idea of bandi OCO is o esimae he necessary gradien using SA in he conex of OGD. Specifically, a key observaion is ha if one can evaluae a funcion f : R p R a w perurbed by a small δ v, where δ > 0 and v is uniformly disribued on he surface of a uni sphere, hen p δ f (w + δ v)v offers an unbiased esimae of he gradien a w of a locally smoohed version of f [14]. hus, plugging his noisy gradien direcly ino he OGD updae in he spiri of SA, one can sill esablish a sublinear regre bound. However, he bes bound found in [14] is O( 3/4 ), slower han he O( )-bound for he full informaion case, illusraing he price o pay for he lack of informaion. LESSONS LEARNED AND FUURE AVENUES his lecure noe offered a shor exposiion of recen advances in online learning for big daa analyics, highlighing heir differences and many similariies wih prominen saisical SP ools such as sochasic approximaion (SA) and sochasic opimizaion (SO) mehods. I was demonsraed ha he seminal Robbins-Monro algorihm, he workhorse behind several classical SP ools such as he LMS and RLS algorihms, carries rich poenial for solving large-scale learning asks under low compuaional budge. I was also explained ha sequenial or online learning schemes ogeher wih random sampling or daa skeching mehods are expeced o play a principal role in solving large-scale opimizaion asks. A shor descripion of he online convex opimizaion (OCO) framework revealed is flexibiliy on he variey of opimizaion asks ha can be accommodaed, including scenarios where daa are provided in an adversarial fashion, or wih limied feedback. Ye, such a flexibiliy comes a a price; OCO-based saisical analysis refers mosly o bounds of he regre cos. Based on he common ground beween OCO and SA, OCO can only benefi from he rich heoreical armory of SA, e.g., maringale heory, where resuls perain also o convergence of he primal (random) variables of he opimizaion ask a hand. Vice versa, SA can also profi from he powerful oolbox of convex analysis, he engine behind OCO, for esablishing srong analyical claims in he big daa conex. In closing, Fig. 4 depics he unique and complemenary srenghs SA, SO, and OCO offer o online learning, as well as adapive SP heory and big daa applicaions. Auhors: Konsaninos Slavakis (kslavaki@umn.edu) is a Research Assoc. Professor in he Dep. of Elecrical & Compuer Engineering and Digial echnology Cener, Univ. of Minnesoa, MN, USA; Seung-Jun Kim (sjkim@umbc.edu) is an Assis. Professor in he Dep. of Compuer Science 8

9 & Elecrical Engineering, Univ. of Maryland, Balimore Couny, MD, USA; Gonzalo Maeos is an Assis. Professor in he Dep. of Elecrical & Compuer Engineering, Univ. of Rocheser, Rocheser, NY, USA; Georgios B. Giannakis is a Professor in he Dep. of Elecrical & Compuer Engineering and Direcor of he Digial echnology Cener, Univ. of Minnesoa, Minneapolis, MN, USA. References [1] K. Slavakis, G. B. Giannakis, and G. Maeos, Modeling and opimizaion for big daa analyics, IEEE Signal Process. Magaz., vol. 31, Sep [2] A. Nemirovski, A. Judiski, G. Lan, and A. Shapiro, Robus sochasic approximaion approach o sochasic programming, SIAM J. Opim., vol. 19, no. 4, pp , [3] H. J. Kushner and G. G. Yin, Sochasic Approximaion Algorihms and Applicaions. New York: Springer, [4] D. P. Bersekas and J. N. sisiklis, Parallel and Disribued Compuaion: Numerical Mehods, 2nd ed. Ahena Scienific, [5] I. D. Schizas, G. Maeos, and G. B. Giannakis, Disribued LMS for consensus-based innework adapive processing, IEEE rans. Signal Process., vol. 57, no. 6, pp , [6] H. H. Bauschke and J. M. Borwein, On projecion algorihms for solving convex feasibiliy problems, SIAM Review, vol. 38, no. 3, pp , Sep [7]. Srohmer and R. Vershynin, A randomized Kaczmarz algorihm wih exponenial convergence, J. Fourier Anal. Appl., vol. 15, no. 2, pp , [8] D. Needell, N. Srebro, and R. Ward, Sochasic gradien descen and he randomized Kaczmarz algorihm, ArXiv e-prins, Feb [Online]. Available: arxiv: v2 [9] A. Nedić and D. P. Bersekas, Incremenal subgradien mehods for nondiffereniable opimizaion, SIAM J. Opim., vol. 12, pp , [10] S. Shalev-Shwarz, Online learning and online convex opimizaion, Foundaions and rends in Machine Learning, vol. 4, no. 2, pp , Mar [11] A. Beck and M. eboulle, Mirror descen and nonlinear projeced subgradien mehods for convex opimizaion, Operaional Research Leers, vol. 31, pp , [12] J. C. Duchi, S. Shalev-Shwarz, Y. Singer, and A. ewari, Composie objecive mirror descen, in Proc. Inl. Conf. Learning heory, Haifa: Israel, June [13] L. Xiao, Dual averaging mehods for regularized sochasic learning and online opimizaion, J. Machine Learning Research, vol. 11, pp , Oc [14] A. D. Flaxman, A.. Kalai, and H. B. McMahan, Online convex opimizaion in he bandi seing: Gradien descen wihou a gradien, in Proc. ACM-SIAM Symp. Discree Algorihms, Vancouver, Jan. 2005, pp

Real-time Particle Filters

Real-time Particle Filters Real-ime Paricle Filers Cody Kwok Dieer Fox Marina Meilă Dep. of Compuer Science & Engineering, Dep. of Saisics Universiy of Washingon Seale, WA 9895 ckwok,fox @cs.washingon.edu, mmp@sa.washingon.edu Absrac

More information

PROFIT TEST MODELLING IN LIFE ASSURANCE USING SPREADSHEETS PART ONE

PROFIT TEST MODELLING IN LIFE ASSURANCE USING SPREADSHEETS PART ONE Profi Tes Modelling in Life Assurance Using Spreadshees PROFIT TEST MODELLING IN LIFE ASSURANCE USING SPREADSHEETS PART ONE Erik Alm Peer Millingon 2004 Profi Tes Modelling in Life Assurance Using Spreadshees

More information

Accelerated Gradient Methods for Stochastic Optimization and Online Learning

Accelerated Gradient Methods for Stochastic Optimization and Online Learning Acceleraed Gradien Mehods for Sochasic Opimizaion and Online Learning Chonghai Hu, James T. Kwok, Weike Pan Deparmen of Compuer Science and Engineering Hong Kong Universiy of Science and Technology Clear

More information

TEMPORAL PATTERN IDENTIFICATION OF TIME SERIES DATA USING PATTERN WAVELETS AND GENETIC ALGORITHMS

TEMPORAL PATTERN IDENTIFICATION OF TIME SERIES DATA USING PATTERN WAVELETS AND GENETIC ALGORITHMS TEMPORAL PATTERN IDENTIFICATION OF TIME SERIES DATA USING PATTERN WAVELETS AND GENETIC ALGORITHMS RICHARD J. POVINELLI AND XIN FENG Deparmen of Elecrical and Compuer Engineering Marquee Universiy, P.O.

More information

Multiprocessor Systems-on-Chips

Multiprocessor Systems-on-Chips Par of: Muliprocessor Sysems-on-Chips Edied by: Ahmed Amine Jerraya and Wayne Wolf Morgan Kaufmann Publishers, 2005 2 Modeling Shared Resources Conex swiching implies overhead. On a processing elemen,

More information

DYNAMIC MODELS FOR VALUATION OF WRONGFUL DEATH PAYMENTS

DYNAMIC MODELS FOR VALUATION OF WRONGFUL DEATH PAYMENTS DYNAMIC MODELS FOR VALUATION OF WRONGFUL DEATH PAYMENTS Hong Mao, Shanghai Second Polyechnic Universiy Krzyszof M. Osaszewski, Illinois Sae Universiy Youyu Zhang, Fudan Universiy ABSTRACT Liigaion, exper

More information

A Note on Using the Svensson procedure to estimate the risk free rate in corporate valuation

A Note on Using the Svensson procedure to estimate the risk free rate in corporate valuation A Noe on Using he Svensson procedure o esimae he risk free rae in corporae valuaion By Sven Arnold, Alexander Lahmann and Bernhard Schwezler Ocober 2011 1. The risk free ineres rae in corporae valuaion

More information

The Transport Equation

The Transport Equation The Transpor Equaion Consider a fluid, flowing wih velociy, V, in a hin sraigh ube whose cross secion will be denoed by A. Suppose he fluid conains a conaminan whose concenraion a posiion a ime will be

More information

ANALYSIS AND COMPARISONS OF SOME SOLUTION CONCEPTS FOR STOCHASTIC PROGRAMMING PROBLEMS

ANALYSIS AND COMPARISONS OF SOME SOLUTION CONCEPTS FOR STOCHASTIC PROGRAMMING PROBLEMS ANALYSIS AND COMPARISONS OF SOME SOLUTION CONCEPTS FOR STOCHASTIC PROGRAMMING PROBLEMS R. Caballero, E. Cerdá, M. M. Muñoz and L. Rey () Deparmen of Applied Economics (Mahemaics), Universiy of Málaga,

More information

Morningstar Investor Return

Morningstar Investor Return Morningsar Invesor Reurn Morningsar Mehodology Paper Augus 31, 2010 2010 Morningsar, Inc. All righs reserved. The informaion in his documen is he propery of Morningsar, Inc. Reproducion or ranscripion

More information

Chapter 8: Regression with Lagged Explanatory Variables

Chapter 8: Regression with Lagged Explanatory Variables Chaper 8: Regression wih Lagged Explanaory Variables Time series daa: Y for =1,..,T End goal: Regression model relaing a dependen variable o explanaory variables. Wih ime series new issues arise: 1. One

More information

Task is a schedulable entity, i.e., a thread

Task is a schedulable entity, i.e., a thread Real-Time Scheduling Sysem Model Task is a schedulable eniy, i.e., a hread Time consrains of periodic ask T: - s: saring poin - e: processing ime of T - d: deadline of T - p: period of T Periodic ask T

More information

Term Structure of Prices of Asian Options

Term Structure of Prices of Asian Options Term Srucure of Prices of Asian Opions Jirô Akahori, Tsuomu Mikami, Kenji Yasuomi and Teruo Yokoa Dep. of Mahemaical Sciences, Risumeikan Universiy 1-1-1 Nojihigashi, Kusasu, Shiga 525-8577, Japan E-mail:

More information

Why Did the Demand for Cash Decrease Recently in Korea?

Why Did the Demand for Cash Decrease Recently in Korea? Why Did he Demand for Cash Decrease Recenly in Korea? Byoung Hark Yoo Bank of Korea 26. 5 Absrac We explores why cash demand have decreased recenly in Korea. The raio of cash o consumpion fell o 4.7% in

More information

Journal Of Business & Economics Research September 2005 Volume 3, Number 9

Journal Of Business & Economics Research September 2005 Volume 3, Number 9 Opion Pricing And Mone Carlo Simulaions George M. Jabbour, (Email: jabbour@gwu.edu), George Washingon Universiy Yi-Kang Liu, (yikang@gwu.edu), George Washingon Universiy ABSTRACT The advanage of Mone Carlo

More information

Random Walk in 1-D. 3 possible paths x vs n. -5 For our random walk, we assume the probabilities p,q do not depend on time (n) - stationary

Random Walk in 1-D. 3 possible paths x vs n. -5 For our random walk, we assume the probabilities p,q do not depend on time (n) - stationary Random Walk in -D Random walks appear in many cones: diffusion is a random walk process undersanding buffering, waiing imes, queuing more generally he heory of sochasic processes gambling choosing he bes

More information

The naive method discussed in Lecture 1 uses the most recent observations to forecast future values. That is, Y ˆ t + 1

The naive method discussed in Lecture 1 uses the most recent observations to forecast future values. That is, Y ˆ t + 1 Business Condiions & Forecasing Exponenial Smoohing LECTURE 2 MOVING AVERAGES AND EXPONENTIAL SMOOTHING OVERVIEW This lecure inroduces ime-series smoohing forecasing mehods. Various models are discussed,

More information

Distributed and Secure Computation of Convex Programs over a Network of Connected Processors

Distributed and Secure Computation of Convex Programs over a Network of Connected Processors DCDIS CONFERENCE GUELPH, ONTARIO, CANADA, JULY 2005 1 Disribued and Secure Compuaion of Convex Programs over a Newor of Conneced Processors Michael J. Neely Universiy of Souhern California hp://www-rcf.usc.edu/

More information

TSG-RAN Working Group 1 (Radio Layer 1) meeting #3 Nynashamn, Sweden 22 nd 26 th March 1999

TSG-RAN Working Group 1 (Radio Layer 1) meeting #3 Nynashamn, Sweden 22 nd 26 th March 1999 TSG-RAN Working Group 1 (Radio Layer 1) meeing #3 Nynashamn, Sweden 22 nd 26 h March 1999 RAN TSGW1#3(99)196 Agenda Iem: 9.1 Source: Tile: Documen for: Moorola Macro-diversiy for he PRACH Discussion/Decision

More information

Stochastic Optimal Control Problem for Life Insurance

Stochastic Optimal Control Problem for Life Insurance Sochasic Opimal Conrol Problem for Life Insurance s. Basukh 1, D. Nyamsuren 2 1 Deparmen of Economics and Economerics, Insiue of Finance and Economics, Ulaanbaaar, Mongolia 2 School of Mahemaics, Mongolian

More information

Chapter 7. Response of First-Order RL and RC Circuits

Chapter 7. Response of First-Order RL and RC Circuits Chaper 7. esponse of Firs-Order L and C Circuis 7.1. The Naural esponse of an L Circui 7.2. The Naural esponse of an C Circui 7.3. The ep esponse of L and C Circuis 7.4. A General oluion for ep and Naural

More information

USE OF EDUCATION TECHNOLOGY IN ENGLISH CLASSES

USE OF EDUCATION TECHNOLOGY IN ENGLISH CLASSES USE OF EDUCATION TECHNOLOGY IN ENGLISH CLASSES Mehme Nuri GÖMLEKSİZ Absrac Using educaion echnology in classes helps eachers realize a beer and more effecive learning. In his sudy 150 English eachers were

More information

The Grantor Retained Annuity Trust (GRAT)

The Grantor Retained Annuity Trust (GRAT) WEALTH ADVISORY Esae Planning Sraegies for closely-held, family businesses The Granor Reained Annuiy Trus (GRAT) An efficien wealh ransfer sraegy, paricularly in a low ineres rae environmen Family business

More information

Monte Carlo Observer for a Stochastic Model of Bioreactors

Monte Carlo Observer for a Stochastic Model of Bioreactors Mone Carlo Observer for a Sochasic Model of Bioreacors Marc Joannides, Irène Larramendy Valverde, and Vivien Rossi 2 Insiu de Mahémaiques e Modélisaion de Monpellier (I3M UMR 549 CNRS Place Eugène Baaillon

More information

Dynamic programming models and algorithms for the mutual fund cash balance problem

Dynamic programming models and algorithms for the mutual fund cash balance problem Submied o Managemen Science manuscrip Dynamic programming models and algorihms for he muual fund cash balance problem Juliana Nascimeno Deparmen of Operaions Research and Financial Engineering, Princeon

More information

Optimal Stock Selling/Buying Strategy with reference to the Ultimate Average

Optimal Stock Selling/Buying Strategy with reference to the Ultimate Average Opimal Sock Selling/Buying Sraegy wih reference o he Ulimae Average Min Dai Dep of Mah, Naional Universiy of Singapore, Singapore Yifei Zhong Dep of Mah, Naional Universiy of Singapore, Singapore July

More information

Measuring macroeconomic volatility Applications to export revenue data, 1970-2005

Measuring macroeconomic volatility Applications to export revenue data, 1970-2005 FONDATION POUR LES ETUDES ET RERS LE DEVELOPPEMENT INTERNATIONAL Measuring macroeconomic volailiy Applicaions o expor revenue daa, 1970-005 by Joël Cariolle Policy brief no. 47 March 01 The FERDI is a

More information

Large Scale Online Learning.

Large Scale Online Learning. Large Scale Online Learning. Léon Boou NEC Labs America Princeon NJ 08540 leon@boou.org Yann Le Cun NEC Labs America Princeon NJ 08540 yann@lecun.com Absrac We consider siuaions where raining daa is abundan

More information

PATHWISE PROPERTIES AND PERFORMANCE BOUNDS FOR A PERISHABLE INVENTORY SYSTEM

PATHWISE PROPERTIES AND PERFORMANCE BOUNDS FOR A PERISHABLE INVENTORY SYSTEM PATHWISE PROPERTIES AND PERFORMANCE BOUNDS FOR A PERISHABLE INVENTORY SYSTEM WILLIAM L. COOPER Deparmen of Mechanical Engineering, Universiy of Minnesoa, 111 Church Sree S.E., Minneapolis, MN 55455 billcoop@me.umn.edu

More information

Relationships between Stock Prices and Accounting Information: A Review of the Residual Income and Ohlson Models. Scott Pirie* and Malcolm Smith**

Relationships between Stock Prices and Accounting Information: A Review of the Residual Income and Ohlson Models. Scott Pirie* and Malcolm Smith** Relaionships beween Sock Prices and Accouning Informaion: A Review of he Residual Income and Ohlson Models Sco Pirie* and Malcolm Smih** * Inernaional Graduae School of Managemen, Universiy of Souh Ausralia

More information

A general decomposition formula for derivative prices in stochastic volatility models

A general decomposition formula for derivative prices in stochastic volatility models A general decomposiion formula for derivaive prices in sochasic volailiy models Elisa Alòs Universia Pompeu Fabra C/ Ramón rias Fargas, 5-7 85 Barcelona Absrac We see ha he price of an european call opion

More information

Analogue and Digital Signal Processing. First Term Third Year CS Engineering By Dr Mukhtiar Ali Unar

Analogue and Digital Signal Processing. First Term Third Year CS Engineering By Dr Mukhtiar Ali Unar Analogue and Digial Signal Processing Firs Term Third Year CS Engineering By Dr Mukhiar Ali Unar Recommended Books Haykin S. and Van Veen B.; Signals and Sysems, John Wiley& Sons Inc. ISBN: 0-7-380-7 Ifeachor

More information

How To Predict A Person'S Behavior

How To Predict A Person'S Behavior Informaion Theoreic Approaches for Predicive Models: Resuls and Analysis Monica Dinculescu Supervised by Doina Precup Absrac Learning he inernal represenaion of parially observable environmens has proven

More information

adaptive control; stochastic systems; certainty equivalence principle; long-term

adaptive control; stochastic systems; certainty equivalence principle; long-term COMMUICATIOS I IFORMATIO AD SYSTEMS c 2006 Inernaional Press Vol. 6, o. 4, pp. 299-320, 2006 003 ADAPTIVE COTROL OF LIEAR TIME IVARIAT SYSTEMS: THE BET O THE BEST PRICIPLE S. BITTATI AD M. C. CAMPI Absrac.

More information

SPEC model selection algorithm for ARCH models: an options pricing evaluation framework

SPEC model selection algorithm for ARCH models: an options pricing evaluation framework Applied Financial Economics Leers, 2008, 4, 419 423 SEC model selecion algorihm for ARCH models: an opions pricing evaluaion framework Savros Degiannakis a, * and Evdokia Xekalaki a,b a Deparmen of Saisics,

More information

Individual Health Insurance April 30, 2008 Pages 167-170

Individual Health Insurance April 30, 2008 Pages 167-170 Individual Healh Insurance April 30, 2008 Pages 167-170 We have received feedback ha his secion of he e is confusing because some of he defined noaion is inconsisen wih comparable life insurance reserve

More information

Chapter 1.6 Financial Management

Chapter 1.6 Financial Management Chaper 1.6 Financial Managemen Par I: Objecive ype quesions and answers 1. Simple pay back period is equal o: a) Raio of Firs cos/ne yearly savings b) Raio of Annual gross cash flow/capial cos n c) = (1

More information

Inventory Planning with Forecast Updates: Approximate Solutions and Cost Error Bounds

Inventory Planning with Forecast Updates: Approximate Solutions and Cost Error Bounds OPERATIONS RESEARCH Vol. 54, No. 6, November December 2006, pp. 1079 1097 issn 0030-364X eissn 1526-5463 06 5406 1079 informs doi 10.1287/opre.1060.0338 2006 INFORMS Invenory Planning wih Forecas Updaes:

More information

ARCH 2013.1 Proceedings

ARCH 2013.1 Proceedings Aricle from: ARCH 213.1 Proceedings Augus 1-4, 212 Ghislain Leveille, Emmanuel Hamel A renewal model for medical malpracice Ghislain Léveillé École d acuaria Universié Laval, Québec, Canada 47h ARC Conference

More information

DETERMINISTIC INVENTORY MODEL FOR ITEMS WITH TIME VARYING DEMAND, WEIBULL DISTRIBUTION DETERIORATION AND SHORTAGES KUN-SHAN WU

DETERMINISTIC INVENTORY MODEL FOR ITEMS WITH TIME VARYING DEMAND, WEIBULL DISTRIBUTION DETERIORATION AND SHORTAGES KUN-SHAN WU Yugoslav Journal of Operaions Research 2 (22), Number, 6-7 DEERMINISIC INVENORY MODEL FOR IEMS WIH IME VARYING DEMAND, WEIBULL DISRIBUION DEERIORAION AND SHORAGES KUN-SHAN WU Deparmen of Bussines Adminisraion

More information

The Application of Multi Shifts and Break Windows in Employees Scheduling

The Application of Multi Shifts and Break Windows in Employees Scheduling The Applicaion of Muli Shifs and Brea Windows in Employees Scheduling Evy Herowai Indusrial Engineering Deparmen, Universiy of Surabaya, Indonesia Absrac. One mehod for increasing company s performance

More information

Appendix D Flexibility Factor/Margin of Choice Desktop Research

Appendix D Flexibility Factor/Margin of Choice Desktop Research Appendix D Flexibiliy Facor/Margin of Choice Deskop Research Cheshire Eas Council Cheshire Eas Employmen Land Review Conens D1 Flexibiliy Facor/Margin of Choice Deskop Research 2 Final Ocober 2012 \\GLOBAL.ARUP.COM\EUROPE\MANCHESTER\JOBS\200000\223489-00\4

More information

THE SMART grid vision aims at capitalizing on information

THE SMART grid vision aims at capitalizing on information IEEE TRANSACTIONS ON SMART GRID, VOL. 6, NO. 3, MAY 2015 1303 Real-Time Load Elasiciy Tracking and Pricing for Elecric Vehicle Charging Nasim Yahya Solani, Suden Member, IEEE, Seung-Jun Kim, Senior Member,

More information

Online Learning with Sample Path Constraints

Online Learning with Sample Path Constraints Journal of Machine Learning Research 0 (2009) 569-590 Submied 7/08; Revised /09; Published 3/09 Online Learning wih Sample Pah Consrains Shie Mannor Deparmen of Elecrical and Compuer Engineering McGill

More information

Mathematics in Pharmacokinetics What and Why (A second attempt to make it clearer)

Mathematics in Pharmacokinetics What and Why (A second attempt to make it clearer) Mahemaics in Pharmacokineics Wha and Why (A second aemp o make i clearer) We have used equaions for concenraion () as a funcion of ime (). We will coninue o use hese equaions since he plasma concenraions

More information

How To Calculate Price Elasiciy Per Capia Per Capi

How To Calculate Price Elasiciy Per Capia Per Capi Price elasiciy of demand for crude oil: esimaes for 23 counries John C.B. Cooper Absrac This paper uses a muliple regression model derived from an adapaion of Nerlove s parial adjusmen model o esimae boh

More information

Principal components of stock market dynamics. Methodology and applications in brief (to be updated ) Andrei Bouzaev, bouzaev@ya.

Principal components of stock market dynamics. Methodology and applications in brief (to be updated ) Andrei Bouzaev, bouzaev@ya. Principal componens of sock marke dynamics Mehodology and applicaions in brief o be updaed Andrei Bouzaev, bouzaev@ya.ru Why principal componens are needed Objecives undersand he evidence of more han one

More information

The Greek financial crisis: growing imbalances and sovereign spreads. Heather D. Gibson, Stephan G. Hall and George S. Tavlas

The Greek financial crisis: growing imbalances and sovereign spreads. Heather D. Gibson, Stephan G. Hall and George S. Tavlas The Greek financial crisis: growing imbalances and sovereign spreads Heaher D. Gibson, Sephan G. Hall and George S. Tavlas The enry The enry of Greece ino he Eurozone in 2001 produced a dividend in he

More information

Bayesian Filtering with Online Gaussian Process Latent Variable Models

Bayesian Filtering with Online Gaussian Process Latent Variable Models Bayesian Filering wih Online Gaussian Process Laen Variable Models Yali Wang Laval Universiy yali.wang.1@ulaval.ca Marcus A. Brubaker TTI Chicago mbrubake@cs.orono.edu Brahim Chaib-draa Laval Universiy

More information

Duration and Convexity ( ) 20 = Bond B has a maturity of 5 years and also has a required rate of return of 10%. Its price is $613.

Duration and Convexity ( ) 20 = Bond B has a maturity of 5 years and also has a required rate of return of 10%. Its price is $613. Graduae School of Business Adminisraion Universiy of Virginia UVA-F-38 Duraion and Convexiy he price of a bond is a funcion of he promised paymens and he marke required rae of reurn. Since he promised

More information

The option pricing framework

The option pricing framework Chaper 2 The opion pricing framework The opion markes based on swap raes or he LIBOR have become he larges fixed income markes, and caps (floors) and swapions are he mos imporan derivaives wihin hese markes.

More information

UNDERSTANDING THE DEATH BENEFIT SWITCH OPTION IN UNIVERSAL LIFE POLICIES. Nadine Gatzert

UNDERSTANDING THE DEATH BENEFIT SWITCH OPTION IN UNIVERSAL LIFE POLICIES. Nadine Gatzert UNDERSTANDING THE DEATH BENEFIT SWITCH OPTION IN UNIVERSAL LIFE POLICIES Nadine Gazer Conac (has changed since iniial submission): Chair for Insurance Managemen Universiy of Erlangen-Nuremberg Lange Gasse

More information

Working Paper No. 482. Net Intergenerational Transfers from an Increase in Social Security Benefits

Working Paper No. 482. Net Intergenerational Transfers from an Increase in Social Security Benefits Working Paper No. 482 Ne Inergeneraional Transfers from an Increase in Social Securiy Benefis By Li Gan Texas A&M and NBER Guan Gong Shanghai Universiy of Finance and Economics Michael Hurd RAND Corporaion

More information

Single-machine Scheduling with Periodic Maintenance and both Preemptive and. Non-preemptive jobs in Remanufacturing System 1

Single-machine Scheduling with Periodic Maintenance and both Preemptive and. Non-preemptive jobs in Remanufacturing System 1 Absrac number: 05-0407 Single-machine Scheduling wih Periodic Mainenance and boh Preempive and Non-preempive jobs in Remanufacuring Sysem Liu Biyu hen Weida (School of Economics and Managemen Souheas Universiy

More information

MTH6121 Introduction to Mathematical Finance Lesson 5

MTH6121 Introduction to Mathematical Finance Lesson 5 26 MTH6121 Inroducion o Mahemaical Finance Lesson 5 Conens 2.3 Brownian moion wih drif........................... 27 2.4 Geomeric Brownian moion........................... 28 2.5 Convergence of random

More information

Inductance and Transient Circuits

Inductance and Transient Circuits Chaper H Inducance and Transien Circuis Blinn College - Physics 2426 - Terry Honan As a consequence of Faraday's law a changing curren hrough one coil induces an EMF in anoher coil; his is known as muual

More information

Automatic measurement and detection of GSM interferences

Automatic measurement and detection of GSM interferences Auomaic measuremen and deecion of GSM inerferences Poor speech qualiy and dropped calls in GSM neworks may be caused by inerferences as a resul of high raffic load. The radio nework analyzers from Rohde

More information

E0 370 Statistical Learning Theory Lecture 20 (Nov 17, 2011)

E0 370 Statistical Learning Theory Lecture 20 (Nov 17, 2011) E0 370 Saisical Learning Theory Lecure 0 (ov 7, 0 Online Learning from Expers: Weighed Majoriy and Hedge Lecurer: Shivani Agarwal Scribe: Saradha R Inroducion In his lecure, we will look a he problem of

More information

Performance Center Overview. Performance Center Overview 1

Performance Center Overview. Performance Center Overview 1 Performance Cener Overview Performance Cener Overview 1 ODJFS Performance Cener ce Cener New Performance Cener Model Performance Cener Projec Meeings Performance Cener Execuive Meeings Performance Cener

More information

MODEL AND ALGORITHMS FOR THE REAL TIME MANAGEMENT OF RESIDENTIAL ELECTRICITY DEMAND. A. Barbato, G. Carpentieri

MODEL AND ALGORITHMS FOR THE REAL TIME MANAGEMENT OF RESIDENTIAL ELECTRICITY DEMAND. A. Barbato, G. Carpentieri MODEL AND ALGORITHMS FOR THE REAL TIME MANAGEMENT OF RESIDENTIAL ELECTRICITY DEMAND A. Barbao, G. Carpenieri Poliecnico di Milano, Diparimeno di Eleronica e Informazione, Email: barbao@ele.polimi.i, giuseppe.carpenieri@mail.polimi.i

More information

Distributing Human Resources among Software Development Projects 1

Distributing Human Resources among Software Development Projects 1 Disribuing Human Resources among Sofware Developmen Proecs Macario Polo, María Dolores Maeos, Mario Piaini and rancisco Ruiz Summary This paper presens a mehod for esimaing he disribuion of human resources

More information

CPU Provisioning Algorithms for Service Differentiation in Cloud-based Environments

CPU Provisioning Algorithms for Service Differentiation in Cloud-based Environments CPU Provisioning Algorihms for Service Differeniaion in Cloud-based Environmens Kosas Kasalis, Georgios S. Paschos, Yannis Viniois, Leandros Tassiulas Absrac This work focuses on he design, analysis and

More information

Vector Autoregressions (VARs): Operational Perspectives

Vector Autoregressions (VARs): Operational Perspectives Vecor Auoregressions (VARs): Operaional Perspecives Primary Source: Sock, James H., and Mark W. Wason, Vecor Auoregressions, Journal of Economic Perspecives, Vol. 15 No. 4 (Fall 2001), 101-115. Macroeconomericians

More information

An Online Learning-based Framework for Tracking

An Online Learning-based Framework for Tracking An Online Learning-based Framework for Tracking Kamalika Chaudhuri Compuer Science and Engineering Universiy of California, San Diego La Jolla, CA 9293 Yoav Freund Compuer Science and Engineering Universiy

More information

AP Calculus AB 2013 Scoring Guidelines

AP Calculus AB 2013 Scoring Guidelines AP Calculus AB 1 Scoring Guidelines The College Board The College Board is a mission-driven no-for-profi organizaion ha connecs sudens o college success and opporuniy. Founded in 19, he College Board was

More information

Optimal Investment and Consumption Decision of Family with Life Insurance

Optimal Investment and Consumption Decision of Family with Life Insurance Opimal Invesmen and Consumpion Decision of Family wih Life Insurance Minsuk Kwak 1 2 Yong Hyun Shin 3 U Jin Choi 4 6h World Congress of he Bachelier Finance Sociey Torono, Canada June 25, 2010 1 Speaker

More information

THE FIRM'S INVESTMENT DECISION UNDER CERTAINTY: CAPITAL BUDGETING AND RANKING OF NEW INVESTMENT PROJECTS

THE FIRM'S INVESTMENT DECISION UNDER CERTAINTY: CAPITAL BUDGETING AND RANKING OF NEW INVESTMENT PROJECTS VII. THE FIRM'S INVESTMENT DECISION UNDER CERTAINTY: CAPITAL BUDGETING AND RANKING OF NEW INVESTMENT PROJECTS The mos imporan decisions for a firm's managemen are is invesmen decisions. While i is surely

More information

Sprotego Analysis and the Big Daa

Sprotego Analysis and the Big Daa ONLINE SKETCHING FOR BIG DATA SUBSPACE LEARNING Moreza Mardani and Georgios B. Giannakis Dep. of ECE and Digiial Technology Cener, Universiy of Minnesoa ABSTRACT Skeching a.k.a. subsampling high-dimensional

More information

Analysis of Pricing and Efficiency Control Strategy between Internet Retailer and Conventional Retailer

Analysis of Pricing and Efficiency Control Strategy between Internet Retailer and Conventional Retailer Recen Advances in Business Managemen and Markeing Analysis of Pricing and Efficiency Conrol Sraegy beween Inerne Reailer and Convenional Reailer HYUG RAE CHO 1, SUG MOO BAE and JOG HU PARK 3 Deparmen of

More information

4 Convolution. Recommended Problems. x2[n] 1 2[n]

4 Convolution. Recommended Problems. x2[n] 1 2[n] 4 Convoluion Recommended Problems P4.1 This problem is a simple example of he use of superposiion. Suppose ha a discree-ime linear sysem has oupus y[n] for he given inpus x[n] as shown in Figure P4.1-1.

More information

DOES TRADING VOLUME INFLUENCE GARCH EFFECTS? SOME EVIDENCE FROM THE GREEK MARKET WITH SPECIAL REFERENCE TO BANKING SECTOR

DOES TRADING VOLUME INFLUENCE GARCH EFFECTS? SOME EVIDENCE FROM THE GREEK MARKET WITH SPECIAL REFERENCE TO BANKING SECTOR Invesmen Managemen and Financial Innovaions, Volume 4, Issue 3, 7 33 DOES TRADING VOLUME INFLUENCE GARCH EFFECTS? SOME EVIDENCE FROM THE GREEK MARKET WITH SPECIAL REFERENCE TO BANKING SECTOR Ahanasios

More information

Making a Faster Cryptanalytic Time-Memory Trade-Off

Making a Faster Cryptanalytic Time-Memory Trade-Off Making a Faser Crypanalyic Time-Memory Trade-Off Philippe Oechslin Laboraoire de Securié e de Crypographie (LASEC) Ecole Polyechnique Fédérale de Lausanne Faculé I&C, 1015 Lausanne, Swizerland philippe.oechslin@epfl.ch

More information

II.1. Debt reduction and fiscal multipliers. dbt da dpbal da dg. bal

II.1. Debt reduction and fiscal multipliers. dbt da dpbal da dg. bal Quarerly Repor on he Euro Area 3/202 II.. Deb reducion and fiscal mulipliers The deerioraion of public finances in he firs years of he crisis has led mos Member Saes o adop sizeable consolidaion packages.

More information

Answer, Key Homework 2 David McIntyre 45123 Mar 25, 2004 1

Answer, Key Homework 2 David McIntyre 45123 Mar 25, 2004 1 Answer, Key Homework 2 Daid McInyre 4123 Mar 2, 2004 1 This prin-ou should hae 1 quesions. Muliple-choice quesions may coninue on he ne column or page find all choices before making your selecion. The

More information

Hedging with Forwards and Futures

Hedging with Forwards and Futures Hedging wih orwards and uures Hedging in mos cases is sraighforward. You plan o buy 10,000 barrels of oil in six monhs and you wish o eliminae he price risk. If you ake he buy-side of a forward/fuures

More information

Advise on the development of a Learning Technologies Strategy at the Leopold-Franzens-Universität Innsbruck

Advise on the development of a Learning Technologies Strategy at the Leopold-Franzens-Universität Innsbruck Advise on he developmen of a Learning Technologies Sraegy a he Leopold-Franzens-Universiä Innsbruck Prof. Dr. Rob Koper Open Universiy of he Neherlands Educaional Technology Experise Cener Conex - Period

More information

cooking trajectory boiling water B (t) microwave 0 2 4 6 8 101214161820 time t (mins)

cooking trajectory boiling water B (t) microwave 0 2 4 6 8 101214161820 time t (mins) Alligaor egg wih calculus We have a large alligaor egg jus ou of he fridge (1 ) which we need o hea o 9. Now here are wo accepable mehods for heaing alligaor eggs, one is o immerse hem in boiling waer

More information

Stock Trading with Recurrent Reinforcement Learning (RRL) CS229 Application Project Gabriel Molina, SUID 5055783

Stock Trading with Recurrent Reinforcement Learning (RRL) CS229 Application Project Gabriel Molina, SUID 5055783 Sock raing wih Recurren Reinforcemen Learning (RRL) CS9 Applicaion Projec Gabriel Molina, SUID 555783 I. INRODUCION One relaively new approach o financial raing is o use machine learning algorihms o preic

More information

AP Calculus BC 2010 Scoring Guidelines

AP Calculus BC 2010 Scoring Guidelines AP Calculus BC Scoring Guidelines The College Board The College Board is a no-for-profi membership associaion whose mission is o connec sudens o college success and opporuniy. Founded in, he College Board

More information

Real Time Bid Optimization with Smooth Budget Delivery in Online Advertising

Real Time Bid Optimization with Smooth Budget Delivery in Online Advertising Real Time Bid Opimizaion wih Smooh Budge Delivery in Online Adverising Kuang-Chih Lee Ali Jalali Ali Dasdan Turn Inc. 835 Main Sree, Redwood Ciy, CA 94063 {klee,ajalali,adasdan}@urn.com ABSTRACT Today,

More information

Statistical Analysis with Little s Law. Supplementary Material: More on the Call Center Data. by Song-Hee Kim and Ward Whitt

Statistical Analysis with Little s Law. Supplementary Material: More on the Call Center Data. by Song-Hee Kim and Ward Whitt Saisical Analysis wih Lile s Law Supplemenary Maerial: More on he Call Cener Daa by Song-Hee Kim and Ward Whi Deparmen of Indusrial Engineering and Operaions Research Columbia Universiy, New York, NY 17-99

More information

AP Calculus AB 2010 Scoring Guidelines

AP Calculus AB 2010 Scoring Guidelines AP Calculus AB 1 Scoring Guidelines The College Board The College Board is a no-for-profi membership associaion whose mission is o connec sudens o college success and opporuniy. Founded in 1, he College

More information

Dependent Interest and Transition Rates in Life Insurance

Dependent Interest and Transition Rates in Life Insurance Dependen Ineres and ransiion Raes in Life Insurance Krisian Buchard Universiy of Copenhagen and PFA Pension January 28, 2013 Absrac In order o find marke consisen bes esimaes of life insurance liabiliies

More information

Communication Networks II Contents

Communication Networks II Contents 3 / 1 -- Communicaion Neworks II (Görg) -- www.comnes.uni-bremen.de Communicaion Neworks II Conens 1 Fundamenals of probabiliy heory 2 Traffic in communicaion neworks 3 Sochasic & Markovian Processes (SP

More information

Pricing Fixed-Income Derivaives wih he Forward-Risk Adjused Measure Jesper Lund Deparmen of Finance he Aarhus School of Business DK-8 Aarhus V, Denmark E-mail: jel@hha.dk Homepage: www.hha.dk/~jel/ Firs

More information

BALANCE OF PAYMENTS. First quarter 2008. Balance of payments

BALANCE OF PAYMENTS. First quarter 2008. Balance of payments BALANCE OF PAYMENTS DATE: 2008-05-30 PUBLISHER: Balance of Paymens and Financial Markes (BFM) Lena Finn + 46 8 506 944 09, lena.finn@scb.se Camilla Bergeling +46 8 506 942 06, camilla.bergeling@scb.se

More information

4. International Parity Conditions

4. International Parity Conditions 4. Inernaional ariy ondiions 4.1 urchasing ower ariy he urchasing ower ariy ( heory is one of he early heories of exchange rae deerminaion. his heory is based on he concep ha he demand for a counry's currency

More information

Bayesian model comparison with un-normalised likelihoods

Bayesian model comparison with un-normalised likelihoods Saisics and Compuing manuscrip No. (will be insered by he edior) Bayesian model comparison wih un-normalised likelihoods Richard G. Everi Adam M. Johansen Ellen Rowing Melina Evdemon-Hogan Each of hese

More information

Evolutionary building of stock trading experts in real-time systems

Evolutionary building of stock trading experts in real-time systems Evoluionary building of sock rading expers in real-ime sysems Jerzy J. Korczak Universié Louis Paseur Srasbourg, France Email: jjk@dp-info.u-srasbg.fr Absrac: This paper addresses he problem of consrucing

More information

Forecasting and Information Sharing in Supply Chains Under Quasi-ARMA Demand

Forecasting and Information Sharing in Supply Chains Under Quasi-ARMA Demand Forecasing and Informaion Sharing in Supply Chains Under Quasi-ARMA Demand Avi Giloni, Clifford Hurvich, Sridhar Seshadri July 9, 2009 Absrac In his paper, we revisi he problem of demand propagaion in

More information

Analysis of Planck and the Equilibrium ofantis in Tropical Physics

Analysis of Planck and the Equilibrium ofantis in Tropical Physics Emergence of Fokker-Planck Dynamics wihin a Closed Finie Spin Sysem H. Niemeyer(*), D. Schmidke(*), J. Gemmer(*), K. Michielsen(**), H. de Raed(**) (*)Universiy of Osnabrück, (**) Supercompuing Cener Juelich

More information

Markov Chain Modeling of Policy Holder Behavior in Life Insurance and Pension

Markov Chain Modeling of Policy Holder Behavior in Life Insurance and Pension Markov Chain Modeling of Policy Holder Behavior in Life Insurance and Pension Lars Frederik Brand Henriksen 1, Jeppe Woemann Nielsen 2, Mogens Seffensen 1, and Chrisian Svensson 2 1 Deparmen of Mahemaical

More information

A UNIFIED APPROACH TO MATHEMATICAL OPTIMIZATION AND LAGRANGE MULTIPLIER THEORY FOR SCIENTISTS AND ENGINEERS

A UNIFIED APPROACH TO MATHEMATICAL OPTIMIZATION AND LAGRANGE MULTIPLIER THEORY FOR SCIENTISTS AND ENGINEERS A UNIFIED APPROACH TO MATHEMATICAL OPTIMIZATION AND LAGRANGE MULTIPLIER THEORY FOR SCIENTISTS AND ENGINEERS RICHARD A. TAPIA Appendix E: Differeniaion in Absrac Spaces I should be no surprise ha he differeniaion

More information

A Production-Inventory System with Markovian Capacity and Outsourcing Option

A Production-Inventory System with Markovian Capacity and Outsourcing Option OPERATIONS RESEARCH Vol. 53, No. 2, March April 2005, pp. 328 349 issn 0030-364X eissn 1526-5463 05 5302 0328 informs doi 10.1287/opre.1040.0165 2005 INFORMS A Producion-Invenory Sysem wih Markovian Capaciy

More information

Tracking of Multiple Moving Sources Using Recursive EM Algorithm

Tracking of Multiple Moving Sources Using Recursive EM Algorithm EURASIP Journal on Applied Signal Processing 24:18, 1 11 c 24 Hindawi Publishing Corporaion Tracking of Muliple Moving Sources Using Recursive EM Algorihm Pei-Jung Chung Deparmen of Elecrical Engineering

More information

Economics Honors Exam 2008 Solutions Question 5

Economics Honors Exam 2008 Solutions Question 5 Economics Honors Exam 2008 Soluions Quesion 5 (a) (2 poins) Oupu can be decomposed as Y = C + I + G. And we can solve for i by subsiuing in equaions given in he quesion, Y = C + I + G = c 0 + c Y D + I

More information

Maintaining Multi-Modality through Mixture Tracking

Maintaining Multi-Modality through Mixture Tracking Mainaining Muli-Modaliy hrough Mixure Tracking Jaco Vermaak, Arnaud Douce Cambridge Universiy Engineering Deparmen Cambridge, CB2 1PZ, UK Parick Pérez Microsof Research Cambridge, CB3 0FB, UK Absrac In

More information

A Scalable and Lightweight QoS Monitoring Technique Combining Passive and Active Approaches

A Scalable and Lightweight QoS Monitoring Technique Combining Passive and Active Approaches A Scalable and Lighweigh QoS Monioring Technique Combining Passive and Acive Approaches On he Mahemaical Formulaion of CoMPACT Monior Masai Aida, Naoo Miyoshi and Keisue Ishibashi NTT Informaion Sharing

More information

MACROECONOMIC FORECASTS AT THE MOF A LOOK INTO THE REAR VIEW MIRROR

MACROECONOMIC FORECASTS AT THE MOF A LOOK INTO THE REAR VIEW MIRROR MACROECONOMIC FORECASTS AT THE MOF A LOOK INTO THE REAR VIEW MIRROR The firs experimenal publicaion, which summarised pas and expeced fuure developmen of basic economic indicaors, was published by he Minisry

More information

Supplementary Appendix for Depression Babies: Do Macroeconomic Experiences Affect Risk-Taking?

Supplementary Appendix for Depression Babies: Do Macroeconomic Experiences Affect Risk-Taking? Supplemenary Appendix for Depression Babies: Do Macroeconomic Experiences Affec Risk-Taking? Ulrike Malmendier UC Berkeley and NBER Sefan Nagel Sanford Universiy and NBER Sepember 2009 A. Deails on SCF

More information